jimzah/freebsd-src
1
0
Fork 0
mirror of https://github.com/freebsd/freebsd-src.git synced 2024-11-30 04:22:44 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity
f39aebddfc
freebsd-src/sys/i386/scsi/aic7xxx.c

3941 lines
102 KiB
C
Raw Normal View History

My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
* Generic driver for the aic7xxx based adaptec SCSI controllers
* Product specific probe and attach routines can be found in:
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
* i386/eisa/aic7770.c 27/284X and aic7770 motherboard controllers
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
* pci/aic7870.c 3940, 2940, aic7880, aic7870, aic7860,
* and aic7850 controllers
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
*
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
* Copyright (c) 1994, 1995, 1996, 1997 Justin T. Gibbs.
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
* All rights reserved.
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice immediately at the beginning of the file, without modification,
* this list of conditions, and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission.
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*
Drop the number of allowed tags back down to 8. Pluto uses a higher value which mistakenly got committed. Fix two bugs in the ahc_reset_device code: Limit search for SCBs to process to those that are active and are not queued for done processing. It's okay for an SCB to not have a waiting next SCB.
1997-04-10 20:14:58 +01:00
* $Id: aic7xxx.c,v 1.114 1997/04/07 18:32:47 gibbs Exp $
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*/
/*
* TODO:
* Implement Target Mode
*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
* A few notes on features of the driver.
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
*
* SCB paging takes advantage of the fact that devices stay disconnected
* from the bus a relatively long time and that while they're disconnected,
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
* having the SCBs for these transactions down on the host adapter is of
* little use. Instead of leaving this idle SCB down on the card we copy
* it back up into kernel memory and reuse the SCB slot on the card to
* schedule another transaction. This can be a real payoff when doing random
* I/O to tagged queueing devices since there are more transactions active at
* once for the device to sort for optimal seek reduction. The algorithm goes
* like this...
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
* The sequencer maintains two lists of its hardware SCBs. The first is the
* singly linked free list which tracks all SCBs that are not currently in
* use. The second is the doubly linked disconnected list which holds the
* SCBs of transactions that are in the disconnected state sorted most
* recently disconnected first. When the kernel queues a transaction to
* the card, a hardware SCB to "house" this transaction is retrieved from
* either of these two lists. If the SCB came from the disconnected list,
* a check is made to see if any data transfer or SCB linking (more on linking
* in a bit) information has been changed since it was copied from the host
* and if so, DMAs the SCB back up before it can be used. Once a hardware
* SCB has been obtained, the SCB is DMAed from the host. Before any work
* can begin on this SCB, the sequencer must ensure that either the SCB is
* for a tagged transaction or the target is not already working on another
* non-tagged transaction. If a conflict arises in the non-tagged case, the
* sequencer finds the SCB for the active transactions and sets the SCB_LINKED
* field in that SCB to this next SCB to execute. To facilitate finding
* active non-tagged SCBs, the last four bytes of up to the first four hardware
* SCBs serve as a storage area for the currently active SCB ID for each
* target.
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
* When a device reconnects, a search is made of the hardware SCBs to find
* the SCB for this transaction. If the search fails, a hardware SCB is
* pulled from either the free or disconnected SCB list and the proper
* SCB is DMAed from the host. If the SCB_ABORTED control bit is set
* in the control byte of the SCB while it was disconnected, the sequencer
* will send an abort or abort tag message to the target during the
* reconnection and signal the kernel that the abort was successfull.
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
* When a command completes, a check for non-zero status and residuals is
* made. If either of these conditions exists, the SCB is DMAed back up to
* the host so that it can interpret this information. Additionally, in the
* case of bad status, the sequencer generates a special interrupt and pauses
* itself. This allows the host to setup a request sense command if it
* chooses for this target synchronously with the error so that sense
* information isn't lost.
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
*
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*/
#include <sys/param.h>
#include <sys/systm.h>
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__NetBSD__)
#include <sys/device.h>
#include <machine/bus.h>
#include <machine/intr.h>
#endif /* defined(__NetBSD__) */
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
#include <sys/malloc.h>
#include <sys/buf.h>
#include <sys/proc.h>
Replaced #includes of <sys/user.h> by less gross headers, usually <sys/vm.h>. Many device drivers need only the definition of vtophys() from vm. Added nearby #includes of <sys/conf.h> where appropriate.
1995-12-06 23:52:35 +00:00
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
#include <scsi/scsi_all.h>
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
#include <scsi/scsi_message.h>
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__NetBSD__)
#include <scsi/scsi_debug.h>
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
#include <scsi/scsiconf.h>
Removed nested #includes of <scsi/scsi_debug.h> and <scsi/scsi_driver.h> from <scsi/scsiconf.h> and fixed everything that depended on them.
1997-03-23 06:33:55 +00:00
#include <scsi/scsi_debug.h>
Replaced #includes of <sys/user.h> by less gross headers, usually <sys/vm.h>. Many device drivers need only the definition of vtophys() from vm. Added nearby #includes of <sys/conf.h> where appropriate.
1995-12-06 23:52:35 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
Compile cleanly -Wall. Remove some code that was an attempt to make the 294x probe more stable, but since the guess was wrong, go back to the clearer aproach I used the first time around.
1995-03-07 08:59:28 +00:00
#include <machine/clock.h>
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#endif
Replaced #includes of <sys/user.h> by less gross headers, usually <sys/vm.h>. Many device drivers need only the definition of vtophys() from vm. Added nearby #includes of <sys/conf.h> where appropriate.
1995-12-06 23:52:35 +00:00
#include <vm/vm.h>
Untangled the vm.h include file spaghetti.
1995-12-07 12:48:31 +00:00
#include <vm/vm_param.h>
#include <vm/pmap.h>
Replaced #includes of <sys/user.h> by less gross headers, usually <sys/vm.h>. Many device drivers need only the definition of vtophys() from vm. Added nearby #includes of <sys/conf.h> where appropriate.
1995-12-06 23:52:35 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
Point include to i386/scsi
1995-01-13 02:27:08 +00:00
#include <i386/scsi/aic7xxx.h>
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
#include <dev/aic7xxx/sequencer.h>
#include <aic7xxx_reg.h>
#include <aic7xxx_seq.h>
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#endif /* defined(__FreeBSD__) */
#if defined(__NetBSD__)
#include <dev/ic/aic7xxxreg.h>
#include <dev/ic/aic7xxxvar.h>
#define bootverbose 1
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
#if DEBUGTARGET < 0 /* Negative numbers for disabling cause warnings */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
#define DEBUGTARGET 17
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#endif
#endif /* defined(__NetBSD__) */
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
#include <sys/kernel.h>
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
#define MAX(a,b) (((a) > (b)) ? (a) : (b))
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
#define MIN(a,b) (((a) < (b)) ? (a) : (b))
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
#define ALL_TARGETS -1
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
#define ALL_LUNS -1
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
#define ALL_CHANNELS '\0'
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 08:06:02 +01:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
-Wall fixes.
1995-11-06 05:21:13 +00:00
u_long ahc_unit = 0;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
#ifdef AHC_DEBUG
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
static int ahc_debug = AHC_DEBUG;
#endif
#ifdef AHC_BROKEN_CACHE
int ahc_broken_cache = 1;
/*
* "wbinvd" cause writing back whole cache (both CPU internal & external)
* to memory, so that the instruction takes a lot of time.
* This makes machine slow.
*/
#define INVALIDATE_CACHE() __asm __volatile("wbinvd")
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/**** bit definitions for SCSIDEF ****/
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
#define HSCSIID 0x07 /* our SCSI ID */
#define HWSCSIID 0x0f /* our SCSI ID if Wide Bus */
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
static void ahcminphys __P((struct buf *bp));
static int32_t ahc_scsi_cmd __P((struct scsi_xfer *xs));
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
static void ahc_run_waiting_queue __P((struct ahc_softc *ahc));
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
static struct scb *
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
ahc_get_scb __P((struct ahc_softc *ahc, u_int32_t flags));
static void ahc_free_scb __P((struct ahc_softc *ahc, struct scb *scb));
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
static struct scb *
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
ahc_alloc_scb __P((struct ahc_softc *ahc));
static inline void pause_sequencer __P((struct ahc_softc *ahc));
static inline void unpause_sequencer __P((struct ahc_softc *ahc,
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
int unpause_always));
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
static inline void restart_sequencer __P((struct ahc_softc *ahc));
Completed function declarations and added prototypes. Sorted prototypes.
1995-12-15 23:49:42 +00:00
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
static int timeouts_work; /*
* Boolean that lets the driver know
* that it can safely use timeouts
* for event scheduling. Timeouts
* don't work early in the boot
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
* process.
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
*/
#define AHC_BUSRESET_DELAY 1000 /* Reset delay in us */
#define AHC_BUSSETTLE_DELAY (100 * 1000)/* Delay before taking commands
* after a reset in us
*/
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static struct scsi_adapter ahc_switch =
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
ahc_scsi_cmd,
ahcminphys,
NULL,
NULL,
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
NULL,
"ahc",
{ 0, 0 }
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#endif
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
};
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static struct scsi_device ahc_dev =
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
NULL, /* Use default error handler */
NULL, /* have a queue, served by this */
NULL, /* have no async handler */
NULL, /* Use default 'done' routine */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
"ahc",
0,
{ 0, 0 }
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
};
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
static inline void
pause_sequencer(ahc)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
{
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, HCNTRL, ahc->pause);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
/*
* Since the sequencer can disable pausing in a critical section, we
* must loop until it actually stops.
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
while ((ahc_inb(ahc, HCNTRL) & PAUSE) == 0)
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
static inline void
unpause_sequencer(ahc, unpause_always)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
int unpause_always;
{
if (unpause_always
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
|| (ahc_inb(ahc, INTSTAT) & (SCSIINT | SEQINT | BRKADRINT)) == 0)
ahc_outb(ahc, HCNTRL, ahc->unpause);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
* Restart the sequencer program from address zero
*/
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
static inline void
restart_sequencer(ahc)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
{
do {
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SEQCTL, SEQRESET|FASTMODE);
} while ((ahc_inb(ahc, SEQADDR0) != 0)
|| (ahc_inb(ahc, SEQADDR1) != 0));
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
unpause_sequencer(ahc, /*unpause_always*/TRUE);
}
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
#if defined(__FreeBSD__)
#define IS_SCSIBUS_B(ahc, sc_link) \
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
(((u_int32_t)((sc_link)->fordriver) & SELBUSB) != 0)
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
#else /* NetBSD/OpenBSD */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#define IS_SCSIBUS_B(ahc, sc_link) \
((sc_link)->scsibus == (ahc)->sc_link_b.scsibus)
#endif
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
static u_int8_t ahc_abort_wscb __P((struct ahc_softc *ahc, struct scb *scbp,
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t scbpos, u_int8_t prev,
u_int32_t xs_error));
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
static void ahc_done __P((struct ahc_softc *ahc, struct scb *scbp));
static void ahc_handle_seqint __P((struct ahc_softc *ahc, u_int8_t intstat));
static void ahc_handle_scsiint __P((struct ahc_softc *ahc,
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t intstat));
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
static void ahc_handle_devreset __P((struct ahc_softc *ahc,
struct scb *scb));
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
static void ahc_loadseq __P((struct ahc_softc *ahc));
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
static struct patch *
ahc_next_patch __P((struct patch *cur_patch, int options,
int instrptr));
static void ahc_download_instr __P((struct ahc_softc *ahc, int options,
int instrptr));
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
static int ahc_match_scb __P((struct scb *scb, int target, char channel,
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
int lun, u_int8_t tag));
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
static int ahc_poll __P((struct ahc_softc *ahc, int wait));
Completed function declarations and added prototypes. Sorted prototypes.
1995-12-15 23:49:42 +00:00
#ifdef AHC_DEBUG
static void ahc_print_scb __P((struct scb *scb));
#endif
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
static u_int8_t ahc_find_scb __P((struct ahc_softc *ahc, struct scb *scb));
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
static int ahc_search_qinfifo __P((struct ahc_softc *ahc, int target,
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
char channel, int lun, u_int8_t tag,
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
u_int32_t flags, u_int32_t xs_error,
int requeue));
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
static int ahc_reset_channel __P((struct ahc_softc *ahc, char channel,
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
u_int32_t xs_error, int initiate_reset));
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
static int ahc_reset_device __P((struct ahc_softc *ahc, int target,
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
char channel, int lun, u_int8_t tag,
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
u_int32_t xs_error));
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
static u_int8_t ahc_rem_scb_from_disc_list __P((struct ahc_softc *ahc,
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
u_int8_t scbptr));
static void ahc_add_curscb_to_free_list __P((struct ahc_softc *ahc));
static void ahc_clear_intstat __P((struct ahc_softc *ahc));
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
static void ahc_reset_current_bus __P((struct ahc_softc *ahc));
static void ahc_run_done_queue __P((struct ahc_softc *ahc));
Fix an infinite loop caused by calling ahc_run_done queue while the driver is waiting a bus settle delay. There should really be a facility for the controller driver to "freeze" its queue during recovery operations which would make all of this gymnastics unnecessary.
1997-04-07 19:32:47 +01:00
static void ahc_untimeout_done_queue __P((struct ahc_softc *ahc));
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
static void ahc_scsirate __P((struct ahc_softc* ahc, u_int8_t *scsirate,
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
u_int8_t *period, u_int8_t *offset,
char channel, int target));
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
Completed function declarations and added prototypes. Sorted prototypes.
1995-12-15 23:49:42 +00:00
static timeout_t
ahc_timeout;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#elif defined(__NetBSD__)
static void ahc_timeout __P((void *));
#endif
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
static timeout_t
ahc_first_timeout;
static timeout_t
ahc_busreset_complete;
static timeout_t
ahc_busreset_settle_complete;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
static u_int8_t ahc_index_busy_target __P((struct ahc_softc *ahc, int target,
char channel, int unbusy));
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
When we perform an "automatic request sense", we issue an untagged command. The sequencer expects untagged transactions to have the SCBID of the transaction in the "busy target" array. So, ensure that the busy entry is up to date for the target in this case. The new identify code in the sequencer that performs additional sanity checking got caught up when a tagged transaction created an untagged request sense. In ahc_handle_seqint, ensure that the target ID is taken from the right place. In the case of a selection, the ID is in SCSIID. In the case of a reconnection it is found in SELID.
1997-03-01 06:50:41 +00:00
static void ahc_busy_target __P((struct ahc_softc *ahc, int target,
char channel, u_int8_t scbid));
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
static void ahc_construct_sdtr __P((struct ahc_softc *ahc, int start_byte,
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t period, u_int8_t offset));
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
static void ahc_construct_wdtr __P((struct ahc_softc *ahc, int start_byte,
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
u_int8_t bus_width));
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
static void ahc_calc_residual __P((struct scb *scb));
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
char *ahc_name(ahc)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
{
static char name[10];
sprintf(name, "ahc%d", ahc->unit);
return (name);
}
#elif defined(__NetBSD__)
struct cfdriver ahc_cd = {
NULL, "ahc", DV_DULL
};
#endif
Completed function declarations and added prototypes. Sorted prototypes.
1995-12-15 23:49:42 +00:00
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
#ifdef AHC_DEBUG
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static void
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
ahc_print_scb(scb)
struct scb *scb;
{
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
struct hardware_scb *hscb = scb->hscb;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
printf("scb:%p control:0x%x tcl:0x%x cmdlen:%d cmdpointer:0x%lx\n",
scb,
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
hscb->control,
hscb->tcl,
hscb->cmdlen,
hscb->cmdpointer );
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
printf(" datlen:%d data:0x%lx segs:0x%x segp:0x%lx\n",
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
hscb->datalen,
hscb->data,
hscb->SG_segment_count,
hscb->SG_list_pointer);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
printf(" sg_addr:%lx sg_len:%ld\n",
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
hscb->ahc_dma[0].addr,
hscb->ahc_dma[0].len);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
static struct {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t errno;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
char *errmesg;
} hard_error[] = {
Compile cleanly -Wall. Remove some code that was an attempt to make the 294x probe more stable, but since the guess was wrong, go back to the clearer aproach I used the first time around.
1995-03-07 08:59:28 +00:00
{ ILLHADDR, "Illegal Host Access" },
{ ILLSADDR, "Illegal Sequencer Address referrenced" },
{ ILLOPCODE, "Illegal Opcode in sequencer program" },
{ PARERR, "Sequencer Ram Parity Error" }
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
};
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
* Valid SCSIRATE values. (p. 3-17)
* Provides a mapping of tranfer periods in ns to the proper value to
* stick in the scsiscfr reg to use that transfer rate.
*/
static struct {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
int sxfr;
Properly deal with the Ultra series of adapters. We should now understand the new seeprom format and negotiate up to 20MHz sync if set in SCSI-Select. Reduce the complexity of the timeout code by running it at splhigh(). Fix a bug that caused rescheduled timeouts at 0 clock ticks in the future causing an infinite loop. Obtained from: Timeout bug noticed by David Greenman and wcarchive.
1995-10-26 23:57:18 +00:00
/* Rates in Ultra mode have bit 8 of sxfr set */
#define ULTRA_SXFR 0x100
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t period; /* Period to send to SCSI target */
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
char *rate;
} ahc_syncrates[] = {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
{ 0x100, 12, "20.0" },
{ 0x110, 15, "16.0" },
{ 0x120, 18, "13.4" },
{ 0x000, 25, "10.0" },
{ 0x010, 31, "8.0" },
{ 0x020, 37, "6.67" },
{ 0x030, 43, "5.7" },
{ 0x040, 50, "5.0" },
{ 0x050, 56, "4.4" },
{ 0x060, 62, "4.0" },
{ 0x070, 68, "3.6" }
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
};
static int ahc_num_syncrates =
sizeof(ahc_syncrates) / sizeof(ahc_syncrates[0]);
/*
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
* Allocate a controller structure for a new device and initialize it.
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *
ahc_alloc(unit, iobase, maddr, type, flags, scb_data)
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
int unit;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int32_t iobase;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#elif defined(__NetBSD__)
void
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
ahc_construct(ahc, bc, ioh, maddr, type, flags)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
bus_chipset_tag_t bc;
bus_io_handle_t ioh;
#endif
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
vm_offset_t maddr;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
ahc_type type;
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-06 00:52:03 +01:00
ahc_flag flags;
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct scb_data *scb_data;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
/*
* find unit and check we have that many defined
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
size_t alloc_size;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
/*
* Allocate a storage area for us
*/
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
if (scb_data == NULL)
/*
* We are not sharing SCB space with another controller
* so allocate our own SCB data space.
*/
alloc_size = sizeof(struct full_ahc_softc);
else
alloc_size = sizeof(struct ahc_softc);
ahc = malloc(alloc_size, M_DEVBUF, M_NOWAIT);
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
if (!ahc) {
printf("ahc%d: cannot malloc!\n", unit);
return NULL;
}
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
bzero(ahc, alloc_size);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#endif
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
if (scb_data == NULL) {
struct full_ahc_softc* full_softc = (struct full_ahc_softc*)ahc;
ahc->scb_data = &full_softc->scb_data_storage;
STAILQ_INIT(&ahc->scb_data->free_scbs);
} else
ahc->scb_data = scb_data;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
STAILQ_INIT(&ahc->waiting_scbs);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
ahc->unit = unit;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#endif
#if defined(__FreeBSD__)
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
ahc->baseport = iobase;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#elif defined(__NetBSD__)
ahc->sc_bc = bc;
ahc->sc_ioh = ioh;
#endif
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
ahc->maddr = (volatile u_int8_t *)maddr;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
ahc->type = type;
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-06 00:52:03 +01:00
ahc->flags = flags;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc->unpause = (ahc_inb(ahc, HCNTRL) & IRQMS) | INTEN;
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
ahc->pause = ahc->unpause | PAUSE;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
ahc->busreset_args.ahc = ahc;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
ahc->busreset_args.bus = 'A';
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
ahc->busreset_args_b.ahc = ahc;
ahc->busreset_args_b.bus = 'B';
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
return (ahc);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#endif
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
void
ahc_free(ahc)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
{
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
free(ahc, M_DEVBUF);
return;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
void
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_reset(ahc)
struct ahc_softc *ahc;
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t hcntrl;
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
int wait;
/* Retain the IRQ type accross the chip reset */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
hcntrl = (ahc_inb(ahc, HCNTRL) & IRQMS) | INTEN;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, HCNTRL, CHIPRST | PAUSE);
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
/*
* Ensure that the reset has finished
*/
wait = 1000;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
while (--wait && !(ahc_inb(ahc, HCNTRL) & CHIPRSTACK))
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
DELAY(1000);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (wait == 0) {
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
printf("%s: WARNING - Failed chip reset! "
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
"Trying to initialize anyway.\n", ahc_name(ahc));
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
}
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, HCNTRL, hcntrl | PAUSE);
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
* Look up the valid period to SCSIRATE conversion in our table.
*/
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static void
Enable/Disable Ultra mode on a per target basis. This allows the driver to use the full range of settings from 3.6-20MHz on any target. Remove all Ultra settings except for the top three that are documented to work. This fixes some problem reports with the last revision of the driver since at least the 5.7MHz entry doesn't work in Ultra mode.
1996-05-21 19:37:25 +01:00
ahc_scsirate(ahc, scsirate, period, offset, channel, target )
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
u_int8_t *scsirate;
u_int8_t *period;
u_int8_t *offset;
char channel;
int target;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
int i;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
u_int32_t ultra_enb_addr;
u_int8_t sxfrctl0;
u_int8_t ultra_enb;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
i = ahc_num_syncrates; /* Default to async */
if (*period >= ahc_syncrates[0].period && *offset != 0) {
for (i = 0; i < ahc_num_syncrates; i++) {
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
if (*period <= ahc_syncrates[i].period) {
Enable/Disable Ultra mode on a per target basis. This allows the driver to use the full range of settings from 3.6-20MHz on any target. Remove all Ultra settings except for the top three that are documented to work. This fixes some problem reports with the last revision of the driver since at least the 5.7MHz entry doesn't work in Ultra mode.
1996-05-21 19:37:25 +01:00
/*
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
* Watch out for Ultra speeds when ultra is not
* enabled and vice-versa.
Enable/Disable Ultra mode on a per target basis. This allows the driver to use the full range of settings from 3.6-20MHz on any target. Remove all Ultra settings except for the top three that are documented to work. This fixes some problem reports with the last revision of the driver since at least the 5.7MHz entry doesn't work in Ultra mode.
1996-05-21 19:37:25 +01:00
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (!(ahc->type & AHC_ULTRA)
&& (ahc_syncrates[i].sxfr & ULTRA_SXFR)) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
/*
* This should only happen if the
* drive is the first to negotiate
* and chooses a high rate. We'll
* just move down the table util
* we hit a non ultra speed.
*/
continue;
}
*scsirate = (ahc_syncrates[i].sxfr & 0xF0)
| (*offset & 0x0f);
*period = ahc_syncrates[i].period;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (bootverbose) {
printf("%s: target %d synchronous at "
"%sMHz, offset = 0x%x\n",
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
ahc_name(ahc), target,
ahc_syncrates[i].rate, *offset );
}
break;
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-06 00:52:03 +01:00
}
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
}
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
if (i >= ahc_num_syncrates) {
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
/* Use asynchronous transfers. */
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
*scsirate = 0;
*period = 0;
*offset = 0;
Bring in bug fix from 'SCSI' branch.
1996-10-06 23:50:56 +01:00
if (bootverbose)
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
printf("%s: target %d using asynchronous transfers\n",
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
ahc_name(ahc), target );
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-06 00:52:03 +01:00
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
/*
* Ensure Ultra mode is set properly for
* this target.
*/
ultra_enb_addr = ULTRA_ENB;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (channel == 'B' || target > 7)
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
ultra_enb_addr++;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ultra_enb = ahc_inb(ahc, ultra_enb_addr);
sxfrctl0 = ahc_inb(ahc, SXFRCTL0);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
if (*scsirate != 0 && ahc_syncrates[i].sxfr & ULTRA_SXFR) {
ultra_enb |= 0x01 << (target & 0x07);
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
sxfrctl0 |= FAST20;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
ultra_enb &= ~(0x01 << (target & 0x07));
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
sxfrctl0 &= ~FAST20;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
}
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, ultra_enb_addr, ultra_enb);
ahc_outb(ahc, SXFRCTL0, sxfrctl0);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__NetBSD__)
int
ahcprint(aux, name)
void *aux;
char *name;
{
if (name != NULL)
printf("%s: scsibus ", name);
return UNCONF;
}
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
* Attach all the sub-devices we can find
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
*/
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
int
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
ahc_attach(ahc)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-24 00:03:17 +01:00
struct scsibus_data *scbus;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#ifdef AHC_BROKEN_CACHE
if (cpu_class == CPUCLASS_386) /* doesn't have "wbinvd" instruction */
ahc_broken_cache = 0;
#endif
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-24 00:03:17 +01:00
/*
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
* fill in the prototype scsi_links.
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-24 00:03:17 +01:00
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
ahc->sc_link.adapter_unit = ahc->unit;
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-24 00:03:17 +01:00
ahc->sc_link.adapter_targ = ahc->our_id;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
ahc->sc_link.fordriver = 0;
#elif defined(__NetBSD__)
ahc->sc_link.adapter_target = ahc->our_id;
#endif
Use the new adapter_softc field in the scsi_link structure so that these drivers don't need to maintain an array of configured units. The bt driver still needs to because ISA interrupt handlers take a unit number. :(
1996-01-07 19:24:36 +00:00
ahc->sc_link.adapter_softc = ahc;
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-24 00:03:17 +01:00
ahc->sc_link.adapter = &ahc_switch;
Increase the DELAY used in ahc_poll. This is only used during intial probe of the device and shouldn't affect normal operation. It seems that some version of the aic7870 are extreamly slow in clearing their interrupt register so the added delay is required to ensure we don't poll a stale interrupt value. Clean up debugging printfs and stuff superflous output into #ifdef AHC_DEBUG. Ensure that sdtrpending and wdtrpending are cleared in all cases after we recieve an SDTR or WDTR message. With this commit, the 294x should work or at least be much closer to working properly.
1995-03-17 23:58:27 +00:00
ahc->sc_link.opennings = 2;
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-24 00:03:17 +01:00
ahc->sc_link.device = &ahc_dev;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
ahc->sc_link.flags = DEBUGLEVEL;
Register with devconfg so we show up in lsdev.
1994-11-18 20:34:30 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc->type & AHC_TWIN) {
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
/* Configure the second scsi bus */
ahc->sc_link_b = ahc->sc_link;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
ahc->sc_link_b.adapter_targ = ahc->our_id_b;
ahc->sc_link_b.adapter_bus = 1;
ahc->sc_link_b.fordriver = (void *)SELBUSB;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#elif defined(__NetBSD__)
ahc->sc_link_b.adapter_target = ahc->our_id_b;
#endif
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-24 00:03:17 +01:00
/*
* Prepare the scsibus_data area for the upperlevel
* scsi code.
*/
scbus = scsi_alloc_bus();
if(!scbus)
return 0;
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
scbus->adapter_link = (ahc->flags & AHC_CHANNEL_B_PRIMARY) ?
&ahc->sc_link_b : &ahc->sc_link;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc->type & AHC_WIDE)
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-24 00:03:17 +01:00
scbus->maxtarg = 15;
/*
* ask the adapter what subunits are present
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (bootverbose)
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
printf("ahc%d: Probing channel %c\n", ahc->unit,
(ahc->flags & AHC_CHANNEL_B_PRIMARY) ? 'B' : 'A');
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-24 00:03:17 +01:00
scsi_attachdevs(scbus);
scbus = NULL; /* Upper-level SCSI code owns this now */
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc->type & AHC_TWIN) {
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-24 00:03:17 +01:00
scbus = scsi_alloc_bus();
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (!scbus)
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-24 00:03:17 +01:00
return 0;
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
scbus->adapter_link = (ahc->flags & AHC_CHANNEL_B_PRIMARY) ?
&ahc->sc_link : &ahc->sc_link_b;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc->type & AHC_WIDE)
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-24 00:03:17 +01:00
scbus->maxtarg = 15;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (bootverbose)
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
printf("ahc%d: Probing Channel %c\n", ahc->unit,
(ahc->flags & AHC_CHANNEL_B_PRIMARY) ? 'A': 'B');
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-24 00:03:17 +01:00
scsi_attachdevs(scbus);
scbus = NULL; /* Upper-level SCSI code owns this now */
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#elif defined(__NetBSD__)
/*
* XXX - Update MI SCSI code
*
* if(ahc->type & AHC_WIDE)
* max target of both channel A and B = 15;
*/
/*
* ask the adapter what subunits are present
*/
if ((ahc->flags & AHC_CHANNEL_B_PRIMARY) == 0) {
/* make IS_SCSIBUS_B() == false, while probing channel A */
ahc->sc_link_b.scsibus = 0xff;
if (ahc->type & AHC_TWIN)
printf("%s: Probing channel A\n", ahc_name(ahc));
config_found((void *)ahc, &ahc->sc_link, ahcprint);
if (ahc->type & AHC_TWIN) {
printf("%s: Probing channel B\n", ahc_name(ahc));
config_found((void *)ahc, &ahc->sc_link_b, ahcprint);
}
} else {
/*
* if implementation of IS_SCSIBUS_B() is changed to use
* ahc->sc_link.scsibus, then "ahc->sc_link.scsibus = 0xff;"
* is needed, here.
*/
/* assert(ahc->type & AHC_TWIN); */
printf("%s: Probing channel B\n", ahc_name(ahc));
config_found((void *)ahc, &ahc->sc_link_b, ahcprint);
printf("%s: Probing channel A\n", ahc_name(ahc));
config_found((void *)ahc, &ahc->sc_link, ahcprint);
}
#endif
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-24 00:03:17 +01:00
return 1;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
/*
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
* Catch an interrupt from the adapter
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
Make PCI interrupt handlers return void like everybody else does. Reviewed by: davidg
1996-01-23 21:48:28 +00:00
void
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#elif defined (__NetBSD__)
int
#endif
Make PCI interrupt handlers return void like everybody else does. Reviewed by: davidg
1996-01-23 21:48:28 +00:00
ahc_intr(arg)
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
void *arg;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t intstat;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
ahc = (struct ahc_softc *)arg;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
intstat = ahc_inb(ahc, INTSTAT);
Major overhaul of the aic7xxx driver: - catch the interrupt type (EDGE/LEVEL) before chip reset instead of guessing the right type. - Add pause variable to the ahc struct to better handle the different interrupt types and pausing the sequencer. - CLRINTSTAT -> CLRSCSIINT: This is a documented bit in the CLRINT register in newer Adaptec documentation, so use their name for it. - Report valid residual byte counts. - Don't mess with the target scratch areas > id 8 on single, narrow, channel devices. The BIOS does a checksum of this area and can flip out if we zero it out. - Initialize the sequencer FLAGS scratch ram variable in the single channel devices to 0. This was the cause of the annoying warning where we would get a cmdcmplt the first time we did any type of transfer negotiation with no valid scb. It also fixes the problem that looked like the INTSTAT register wasn't clearing fast enough. This only showed up on 294x cards, not motherboard aic7870s. - Add the AHC_AIC7870 type and use it as the superset of aic7870 based controllers. - clear the sync offset section of the targ scratch area so that we default to asyncronous transfers. This was only a problem for wide controllers because there was a scenario where the offset wouldn't get updated before a data(out/in) phase would occur. This required some change in the sequencer code since we were depending on this field to hold the rate to negotiate. - allow sync and wide negotiated commands to be tagged (the sequencer now handles this properly).
1995-03-31 14:54:41 +01:00
/*
* Is this interrupt for me? or for
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
* someone who is sharing my interrupt?
Major overhaul of the aic7xxx driver: - catch the interrupt type (EDGE/LEVEL) before chip reset instead of guessing the right type. - Add pause variable to the ahc struct to better handle the different interrupt types and pausing the sequencer. - CLRINTSTAT -> CLRSCSIINT: This is a documented bit in the CLRINT register in newer Adaptec documentation, so use their name for it. - Report valid residual byte counts. - Don't mess with the target scratch areas > id 8 on single, narrow, channel devices. The BIOS does a checksum of this area and can flip out if we zero it out. - Initialize the sequencer FLAGS scratch ram variable in the single channel devices to 0. This was the cause of the annoying warning where we would get a cmdcmplt the first time we did any type of transfer negotiation with no valid scb. It also fixes the problem that looked like the INTSTAT register wasn't clearing fast enough. This only showed up on 294x cards, not motherboard aic7870s. - Add the AHC_AIC7870 type and use it as the superset of aic7870 based controllers. - clear the sync offset section of the targ scratch area so that we default to asyncronous transfers. This was only a problem for wide controllers because there was a scenario where the offset wouldn't get updated before a data(out/in) phase would occur. This required some change in the sequencer code since we were depending on this field to hold the rate to negotiate. - allow sync and wide negotiated commands to be tagged (the sequencer now handles this properly).
1995-03-31 14:54:41 +01:00
*/
if (!(intstat & INT_PEND))
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
Make PCI interrupt handlers return void like everybody else does. Reviewed by: davidg
1996-01-23 21:48:28 +00:00
return;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#elif defined(__NetBSD__)
return 0;
#endif
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (intstat & CMDCMPLT) {
struct scb *scb;
u_int8_t scb_index;
u_int8_t qoutcnt;
int int_cleared;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
int_cleared = 0;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
while (qoutcnt = (ahc_inb(ahc, QOUTCNT) & ahc->qcntmask)) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
for (; qoutcnt > 0; qoutcnt--) {
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
scb_index = ahc_inb(ahc, QOUTFIFO);
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
scb = ahc->scb_data->scbarray[scb_index];
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (!scb || !(scb->flags & SCB_ACTIVE)) {
printf("%s: WARNING "
"no command for scb %d "
"(cmdcmplt)\nQOUTCNT == %d\n",
ahc_name(ahc), scb_index,
qoutcnt);
continue;
}
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
/*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
* Save off the residual if there is one.
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (scb->hscb->residual_SG_segment_count != 0)
ahc_calc_residual(scb);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
if ((scb->flags & SCB_QUEUED_ABORT) != 0) {
/* Have to clean up any possible
* entries in the waiting queue and
* QINFIFO.
*/
int target;
char channel;
int lun;
u_int8_t tag;
tag = SCB_LIST_NULL;
target = scb->xs->sc_link->target;
lun = scb->xs->sc_link->lun;
channel = (scb->hscb->tcl & SELBUSB)
? 'B': 'A';
if (scb->hscb->control & TAG_ENB)
tag = scb->hscb->tag;
ahc_reset_device(ahc,
target,
channel,
lun,
tag,
scb->xs->error);
ahc_run_done_queue(ahc);
}
Fix a nasty bug that meant a QUEUE_FULL status would result in a lost SCB. This is probably a main reason for the recent reports of timeouts.
1997-03-24 05:05:18 +00:00
ahc_done(ahc, scb);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
}
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, CLRINT, CLRCMDINT);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
int_cleared++;
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (int_cleared == 0)
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, CLRINT, CLRCMDINT);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
}
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
if (intstat & BRKADRINT) {
/*
* We upset the sequencer :-(
* Lookup the error message
*/
int i, error, num_errors;
error = ahc_inb(ahc, ERROR);
num_errors = sizeof(hard_error)/sizeof(hard_error[0]);
for (i = 0; error != 1 && i < num_errors; i++)
error >>= 1;
printf("%s: brkadrint, %s at seqaddr = 0x%x\n",
ahc_name(ahc), hard_error[i].errmesg,
(ahc_inb(ahc, SEQADDR1) << 8) |
ahc_inb(ahc, SEQADDR0));
ahc_reset_device(ahc, ALL_TARGETS, ALL_CHANNELS, ALL_LUNS,
SCB_LIST_NULL, XS_DRIVER_STUFFUP);
ahc_run_done_queue(ahc);
}
if (intstat & SEQINT)
ahc_handle_seqint(ahc, intstat);
if (intstat & SCSIINT)
ahc_handle_scsiint(ahc, intstat);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc->waiting_scbs.stqh_first != NULL)
ahc_run_waiting_queue(ahc);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
#if defined(__NetBSD__)
return 1;
#endif
}
static void
ahc_handle_seqint(ahc, intstat)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
u_int8_t intstat;
{
struct scb *scb;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int16_t targ_mask;
When we perform an "automatic request sense", we issue an untagged command. The sequencer expects untagged transactions to have the SCBID of the transaction in the "busy target" array. So, ensure that the busy entry is up to date for the target in this case. The new identify code in the sequencer that performs additional sanity checking got caught up when a tagged transaction created an untagged request sense. In ahc_handle_seqint, ensure that the target ID is taken from the right place. In the case of a selection, the ID is in SCSIID. In the case of a reconnection it is found in SELID.
1997-03-01 06:50:41 +00:00
u_int8_t target;
int scratch_offset;
char channel;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
if ((ahc_inb(ahc, SEQ_FLAGS) & RESELECTED) != 0)
When we perform an "automatic request sense", we issue an untagged command. The sequencer expects untagged transactions to have the SCBID of the transaction in the "busy target" array. So, ensure that the busy entry is up to date for the target in this case. The new identify code in the sequencer that performs additional sanity checking got caught up when a tagged transaction created an untagged request sense. In ahc_handle_seqint, ensure that the target ID is taken from the right place. In the case of a selection, the ID is in SCSIID. In the case of a reconnection it is found in SELID.
1997-03-01 06:50:41 +00:00
target = ahc_inb(ahc, SELID);
else
target = ahc_inb(ahc, SCSIID);
target = (target >> 4) & 0x0f;
scratch_offset = target;
channel = ahc_inb(ahc, SBLKCTL) & SELBUSB ? 'B': 'A';
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
if (channel == 'B')
scratch_offset += 8;
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
targ_mask = (0x01 << scratch_offset);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
switch (intstat & SEQINT_MASK) {
case NO_MATCH:
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
{
/*
* This could be for a normal abort request.
* Figure out the SCB that we were trying to find
* and only give an error if we didn't ask for this
* to happen.
*/
u_int8_t scb_index;
u_int8_t busy_scbid;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
busy_scbid = ahc_index_busy_target(ahc, target, channel,
/*unbusy*/FALSE);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
scb_index = ahc_inb(ahc, ARG_1);
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
if (scb_index == SCB_LIST_NULL)
/* Untagged Request */
scb_index = busy_scbid;
if (scb_index < ahc->scb_data->numscbs) {
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
scb = ahc->scb_data->scbarray[scb_index];
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
if (scb->hscb->control & ABORT_SCB) {
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
/*
* We expected this. Let the busfree
* handler take care of this when we
* the abort is finially sent.
* Set IDENTIFY_SEEN so that the busfree
* handler knows that there is an SCB to
* cleanup.
*/
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, SEQ_FLAGS, ahc_inb(ahc, SEQ_FLAGS)
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
| IDENTIFY_SEEN);
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
sc_print_addr(scb->xs->sc_link);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
printf("reconnect SCB abort successfull\n");
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
break;
}
}
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
/*
* We expect a busfree to occur, so don't bother to interrupt
* when it happens - we've already handled the error.
*/
ahc_outb(ahc, SIMODE1, ahc_inb(ahc, SIMODE1) & ~ENBUSFREE);
busy_scbid = ahc_index_busy_target(ahc, target, channel,
/*unbusy*/TRUE);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
printf("%s:%c:%d: no active SCB for reconnecting "
"target - issuing ABORT\n",
ahc_name(ahc), channel, target);
printf("SAVED_TCL == 0x%x\n",
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_inb(ahc, SAVED_TCL));
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
break;
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
}
case NO_MATCH_BUSY:
{
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
/*
* XXX Leave this as a panic for the time being since
* it indicates a bug in the timeout code for this
* to happen.
*/
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
u_int8_t scb_index;
u_int8_t busy_scbindex;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
scb_index = ahc_inb(ahc, CUR_SCBID);
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
scb = ahc->scb_data->scbarray[scb_index];
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
panic("%s:%c:%d: Target busy link failure.\n",
ahc_name(ahc), (scb->hscb->tcl & SELBUSB) ? 'B' : 'A',
scb->xs->sc_link->target);
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
case SEND_REJECT:
{
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
u_int8_t rejbyte = ahc_inb(ahc, REJBYTE);
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
printf("%s:%c:%d: Warning - unknown message received from "
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
"target (0x%x). Rejecting\n",
ahc_name(ahc), channel, target, rejbyte);
break;
}
case NO_IDENT:
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
{
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
/*
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
* The reconnecting target either did not send an identify
* message, or did, but we didn't find and SCB to match and
* before it could respond to our ATN/abort, it hit a dataphase.
* The only safe thing to do is to blow it away with a bus
* reset.
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
*/
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
int found;
printf("%s:%c:%d: Target did not send an IDENTIFY message. "
"LASTPHASE = 0x%x, SAVED_TCL == 0x%x\n",
ahc_name(ahc), channel, target, ahc_inb(ahc, LASTPHASE),
ahc_inb(ahc, SAVED_TCL));
found = ahc_reset_channel(ahc, channel, XS_TIMEOUT,
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
/*initiate reset*/TRUE);
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
printf("%s: Issued Channel %c Bus Reset. "
"%d SCBs aborted\n", ahc_name(ahc), channel, found);
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
break;
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
case BAD_PHASE:
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
if (ahc_inb(ahc, LASTPHASE) == P_BUSFREE) {
printf("%s:%c:%d: Missed busfree.\n", ahc_name(ahc),
channel, target);
restart_sequencer(ahc);
} else {
printf("%s:%c:%d: unknown scsi bus phase. Attempting "
"to continue\n", ahc_name(ahc), channel, target);
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
break;
case EXTENDED_MSG:
{
u_int8_t message_length;
u_int8_t message_code;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
u_int8_t scb_index;
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
message_length = ahc_inb(ahc, MSGIN_EXT_LEN);
message_code = ahc_inb(ahc, MSGIN_EXT_OPCODE);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
scb_index = ahc_inb(ahc, SCB_TAG);
scb = ahc->scb_data->scbarray[scb_index];
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
switch (message_code) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
case MSG_EXT_SDTR:
{
u_int8_t period;
u_int8_t offset;
If, during an SDTR negotiation, the target comes back with a response that is too low for the aic7xxx chip to handle with sync transfers, negotiate async transfers.
1996-10-06 20:43:36 +01:00
u_int8_t saved_offset;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
u_int8_t targ_scratch;
u_int8_t maxoffset;
u_int8_t rate;
if (message_length != MSG_EXT_SDTR_LEN) {
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, RETURN_1, SEND_REJ);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
break;
}
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
period = ahc_inb(ahc, MSGIN_EXT_BYTES);
saved_offset = ahc_inb(ahc, MSGIN_EXT_BYTES + 1);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
targ_scratch = ahc_inb(ahc, TARG_SCRATCH
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
+ scratch_offset);
if (targ_scratch & WIDEXFER)
maxoffset = MAX_OFFSET_16BIT;
else
maxoffset = MAX_OFFSET_8BIT;
If, during an SDTR negotiation, the target comes back with a response that is too low for the aic7xxx chip to handle with sync transfers, negotiate async transfers.
1996-10-06 20:43:36 +01:00
offset = MIN(saved_offset, maxoffset);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
ahc_scsirate(ahc, &rate, &period, &offset,
channel, target);
/* Preserve the WideXfer flag */
targ_scratch = rate | (targ_scratch & WIDEXFER);
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
/*
* Update both the target scratch area and the
* current SCSIRATE.
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, TARG_SCRATCH + scratch_offset,
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
targ_scratch);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCSIRATE, targ_scratch);
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
If, during an SDTR negotiation, the target comes back with a response that is too low for the aic7xxx chip to handle with sync transfers, negotiate async transfers.
1996-10-06 20:43:36 +01:00
/*
* See if we initiated Sync Negotiation
* and didn't have to fall down to async
* transfers.
*/
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
if ((scb->flags & SCB_MSGOUT_SDTR) != 0) {
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
/* We started it */
Move the include opt_aic7xxx in aic7xxx.h so that all of the driver files get it automatically. The AHC_FORCE_PIO option wasn't having any effect because the PCI probe code didn't include this file. Fix some problems with the new sync and wide negotiation code. First off, go back to async transfers by using a message reject again. The SCSI II and III spec indicate that if a target's response to an initiater does not suit (i.e. its too low), then performing a message reject is the appropriate response. If, on the other hand, the initiator begins the negotiation and we want to go async, we will send back an SDTR message with a 0 period and offset. Also fix a really bad negotiation problem caused by a missing "break". This would usually hit people that had "smart" wide devices that immediately attempt sync negotiation after a successful wide negotiation. 2.2 Candidate.
1996-11-05 07:57:29 +00:00
if (saved_offset == offset) {
/*
* Don't send an SDTR back to
* the target
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, RETURN_1, 0);
Move the include opt_aic7xxx in aic7xxx.h so that all of the driver files get it automatically. The AHC_FORCE_PIO option wasn't having any effect because the PCI probe code didn't include this file. Fix some problems with the new sync and wide negotiation code. First off, go back to async transfers by using a message reject again. The SCSI II and III spec indicate that if a target's response to an initiater does not suit (i.e. its too low), then performing a message reject is the appropriate response. If, on the other hand, the initiator begins the negotiation and we want to go async, we will send back an SDTR message with a 0 period and offset. Also fix a really bad negotiation problem caused by a missing "break". This would usually hit people that had "smart" wide devices that immediately attempt sync negotiation after a successful wide negotiation. 2.2 Candidate.
1996-11-05 07:57:29 +00:00
} else
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
/* Went too low - force async */
ahc_outb(ahc, RETURN_1, SEND_REJ);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
} else {
/*
* Send our own SDTR in reply
Add $Ids. Use DMA for downloading SCBs to the sequencer. Implement SDTR, WDTR, and message reject handlers so they don't need to exist in the sequencer. All three of these cases are not on the critical path, so it makes little sense to use up precious sequencer ram for them.
1995-01-16 16:33:47 +00:00
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
printf("Sending SDTR!!\n");
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
ahc_construct_sdtr(ahc, /*start_byte*/0,
period, offset);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, RETURN_1, SEND_MSG);
Add $Ids. Use DMA for downloading SCBs to the sequencer. Implement SDTR, WDTR, and message reject handlers so they don't need to exist in the sequencer. All three of these cases are not on the critical path, so it makes little sense to use up precious sequencer ram for them.
1995-01-16 16:33:47 +00:00
}
Move the include opt_aic7xxx in aic7xxx.h so that all of the driver files get it automatically. The AHC_FORCE_PIO option wasn't having any effect because the PCI probe code didn't include this file. Fix some problems with the new sync and wide negotiation code. First off, go back to async transfers by using a message reject again. The SCSI II and III spec indicate that if a target's response to an initiater does not suit (i.e. its too low), then performing a message reject is the appropriate response. If, on the other hand, the initiator begins the negotiation and we want to go async, we will send back an SDTR message with a 0 period and offset. Also fix a really bad negotiation problem caused by a missing "break". This would usually hit people that had "smart" wide devices that immediately attempt sync negotiation after a successful wide negotiation. 2.2 Candidate.
1996-11-05 07:57:29 +00:00
ahc->needsdtr &= ~targ_mask;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
break;
}
case MSG_EXT_WDTR:
{
u_int8_t scratch, bus_width;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
if (message_length != MSG_EXT_WDTR_LEN) {
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, RETURN_1, SEND_REJ);
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
break;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
}
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
bus_width = ahc_inb(ahc, MSGIN_EXT_BYTES);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
scratch = ahc_inb(ahc, TARG_SCRATCH + scratch_offset);
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
if ((scb->flags & SCB_MSGOUT_WDTR) != 0) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
/*
* Don't send a WDTR back to the
* target, since we asked first.
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, RETURN_1, 0);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
switch (bus_width){
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
case BUS_8_BIT:
scratch &= 0x7f;
break;
case BUS_16_BIT:
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (bootverbose)
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
printf("%s: target %d using "
"16Bit transfers\n",
ahc_name(ahc), target);
scratch |= WIDEXFER;
break;
case BUS_32_BIT:
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
/*
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
* How can we do 32bit transfers
* on a 16bit bus?
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, RETURN_1, SEND_REJ);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
printf("%s: target %d requested 32Bit "
"transfers. Rejecting...\n",
ahc_name(ahc), target);
break;
default:
Add $Ids. Use DMA for downloading SCBs to the sequencer. Implement SDTR, WDTR, and message reject handlers so they don't need to exist in the sequencer. All three of these cases are not on the critical path, so it makes little sense to use up precious sequencer ram for them.
1995-01-16 16:33:47 +00:00
break;
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
} else {
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
/*
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
* Send our own WDTR in reply
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
switch (bus_width) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
case BUS_8_BIT:
scratch &= 0x7f;
break;
case BUS_32_BIT:
case BUS_16_BIT:
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc->type & AHC_WIDE) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
/* Negotiate 16_BITS */
bus_width = BUS_16_BIT;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (bootverbose)
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
printf("%s: target %d "
"using 16Bit "
"transfers\n",
ahc_name(ahc),
target);
scratch |= WIDEXFER;
} else
bus_width = BUS_8_BIT;
break;
default:
break;
}
ahc_construct_wdtr(ahc, /*start_byte*/0,
bus_width);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, RETURN_1, SEND_MSG);
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
ahc->needwdtr &= ~targ_mask;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, TARG_SCRATCH + scratch_offset, scratch);
ahc_outb(ahc, SCSIRATE, scratch);
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
break;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
}
default:
/* Unknown extended message. Reject it. */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, RETURN_1, SEND_REJ);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
}
Move the include opt_aic7xxx in aic7xxx.h so that all of the driver files get it automatically. The AHC_FORCE_PIO option wasn't having any effect because the PCI probe code didn't include this file. Fix some problems with the new sync and wide negotiation code. First off, go back to async transfers by using a message reject again. The SCSI II and III spec indicate that if a target's response to an initiater does not suit (i.e. its too low), then performing a message reject is the appropriate response. If, on the other hand, the initiator begins the negotiation and we want to go async, we will send back an SDTR message with a 0 period and offset. Also fix a really bad negotiation problem caused by a missing "break". This would usually hit people that had "smart" wide devices that immediately attempt sync negotiation after a successful wide negotiation. 2.2 Candidate.
1996-11-05 07:57:29 +00:00
break;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
}
case REJECT_MSG:
{
/*
* What we care about here is if we had an
* outstanding SDTR or WDTR message for this
* target. If we did, this is a signal that
* the target is refusing negotiation.
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t targ_scratch;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
u_int8_t scb_index;
scb_index = ahc_inb(ahc, SCB_TAG);
scb = ahc->scb_data->scbarray[scb_index];
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
targ_scratch = ahc_inb(ahc, TARG_SCRATCH
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
+ scratch_offset);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
if ((scb->flags & SCB_MSGOUT_WDTR) != 0) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
/* note 8bit xfers and clear flag */
targ_scratch &= 0x7f;
ahc->needwdtr &= ~targ_mask;
printf("%s:%c:%d: refuses WIDE negotiation. Using "
"8bit transfers\n", ahc_name(ahc),
channel, target);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
} else if ((scb->flags & SCB_MSGOUT_SDTR) != 0) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
/* note asynch xfers and clear flag */
targ_scratch &= 0xf0;
ahc->needsdtr &= ~targ_mask;
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
printf("%s:%c:%d: refuses synchronous negotiation. "
"Using asynchronous transfers\n",
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
ahc_name(ahc),
channel, target);
} else {
Major overhaul of the aic7xxx driver: - catch the interrupt type (EDGE/LEVEL) before chip reset instead of guessing the right type. - Add pause variable to the ahc struct to better handle the different interrupt types and pausing the sequencer. - CLRINTSTAT -> CLRSCSIINT: This is a documented bit in the CLRINT register in newer Adaptec documentation, so use their name for it. - Report valid residual byte counts. - Don't mess with the target scratch areas > id 8 on single, narrow, channel devices. The BIOS does a checksum of this area and can flip out if we zero it out. - Initialize the sequencer FLAGS scratch ram variable in the single channel devices to 0. This was the cause of the annoying warning where we would get a cmdcmplt the first time we did any type of transfer negotiation with no valid scb. It also fixes the problem that looked like the INTSTAT register wasn't clearing fast enough. This only showed up on 294x cards, not motherboard aic7870s. - Add the AHC_AIC7870 type and use it as the superset of aic7870 based controllers. - clear the sync offset section of the targ scratch area so that we default to asyncronous transfers. This was only a problem for wide controllers because there was a scenario where the offset wouldn't get updated before a data(out/in) phase would occur. This required some change in the sequencer code since we were depending on this field to hold the rate to negotiate. - allow sync and wide negotiated commands to be tagged (the sequencer now handles this properly).
1995-03-31 14:54:41 +01:00
/*
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
* Otherwise, we ignore it.
Major overhaul of the aic7xxx driver: - catch the interrupt type (EDGE/LEVEL) before chip reset instead of guessing the right type. - Add pause variable to the ahc struct to better handle the different interrupt types and pausing the sequencer. - CLRINTSTAT -> CLRSCSIINT: This is a documented bit in the CLRINT register in newer Adaptec documentation, so use their name for it. - Report valid residual byte counts. - Don't mess with the target scratch areas > id 8 on single, narrow, channel devices. The BIOS does a checksum of this area and can flip out if we zero it out. - Initialize the sequencer FLAGS scratch ram variable in the single channel devices to 0. This was the cause of the annoying warning where we would get a cmdcmplt the first time we did any type of transfer negotiation with no valid scb. It also fixes the problem that looked like the INTSTAT register wasn't clearing fast enough. This only showed up on 294x cards, not motherboard aic7870s. - Add the AHC_AIC7870 type and use it as the superset of aic7870 based controllers. - clear the sync offset section of the targ scratch area so that we default to asyncronous transfers. This was only a problem for wide controllers because there was a scenario where the offset wouldn't get updated before a data(out/in) phase would occur. This required some change in the sequencer code since we were depending on this field to hold the rate to negotiate. - allow sync and wide negotiated commands to be tagged (the sequencer now handles this properly).
1995-03-31 14:54:41 +01:00
*/
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
#ifdef AHC_DEBUG
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc_debug & AHC_SHOWMISC)
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
printf("%s:%c:%d: Message reject -- ignored\n",
ahc_name(ahc), channel, target);
#endif
break;
}
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, TARG_SCRATCH + scratch_offset, targ_scratch);
ahc_outb(ahc, SCSIRATE, targ_scratch);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
break;
}
case BAD_STATUS:
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t scb_index;
struct scsi_xfer *xs;
struct hardware_scb *hscb;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/*
* The sequencer will notify us when a command
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
* has an error that would be of interest to
* the kernel. This allows us to leave the sequencer
* running in the common case of command completes
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
* without error. The sequencer will already have
* dma'd the SCB back up to us, so we can reference
* the in kernel copy directly.
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
*/
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
scb_index = ahc_inb(ahc, SCB_TAG);
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
scb = ahc->scb_data->scbarray[scb_index];
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
hscb = scb->hscb;
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
/*
* Set the default return value to 0 (don't
* send sense). The sense code will change
* this if needed and this reduces code
* duplication.
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, RETURN_1, 0);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
if (!(scb && (scb->flags & SCB_ACTIVE))) {
printf("%s:%c:%d: ahc_intr - referenced scb "
"not valid during seqint 0x%x scb(%d)\n",
ahc_name(ahc),
channel, target, intstat,
scb_index);
goto clear;
}
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
xs = scb->xs;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
xs->status = hscb->status;
switch (hscb->status){
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
case SCSI_OK:
printf("%s: Interrupted for staus of"
" 0???\n", ahc_name(ahc));
break;
case SCSI_CHECK:
#ifdef AHC_DEBUG
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc_debug & AHC_SHOWSENSE) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
sc_print_addr(xs->sc_link);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
printf("SCB %d: requests Check Status\n",
scb->hscb->tag);
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 08:06:02 +01:00
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
#endif
if ((xs->error == XS_NOERROR)
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
&& !(scb->flags & SCB_SENSE)) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
struct ahc_dma_seg *sg = scb->ahc_dma;
struct scsi_sense *sc = &(scb->sense_cmd);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/*
* Save off the residual if there is one.
*/
if (hscb->residual_SG_segment_count != 0)
ahc_calc_residual(scb);
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
#ifdef AHC_DEBUG
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc_debug & AHC_SHOWSENSE) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
sc_print_addr(xs->sc_link);
printf("Sending Sense\n");
}
#endif
#if defined(__FreeBSD__)
sc->op_code = REQUEST_SENSE;
#elif defined(__NetBSD__)
sc->opcode = REQUEST_SENSE;
#endif
sc->byte2 = xs->sc_link->lun << 5;
sc->length = sizeof(struct scsi_sense_data);
sc->control = 0;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
sg->addr = vtophys(&xs->sense);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
sg->len = sizeof(struct scsi_sense_data);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
hscb->control &= DISCENB;
hscb->status = 0;
hscb->SG_segment_count = 1;
hscb->SG_list_pointer = vtophys(sg);
hscb->data = sg->addr;
Move the include opt_aic7xxx in aic7xxx.h so that all of the driver files get it automatically. The AHC_FORCE_PIO option wasn't having any effect because the PCI probe code didn't include this file. Fix some problems with the new sync and wide negotiation code. First off, go back to async transfers by using a message reject again. The SCSI II and III spec indicate that if a target's response to an initiater does not suit (i.e. its too low), then performing a message reject is the appropriate response. If, on the other hand, the initiator begins the negotiation and we want to go async, we will send back an SDTR message with a 0 period and offset. Also fix a really bad negotiation problem caused by a missing "break". This would usually hit people that had "smart" wide devices that immediately attempt sync negotiation after a successful wide negotiation. 2.2 Candidate.
1996-11-05 07:57:29 +00:00
/* Maintain SCB_LINKED_NEXT */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
hscb->datalen &= 0xFF000000;
hscb->datalen |= sg->len;
hscb->cmdpointer = vtophys(sc);
hscb->cmdlen = sizeof(*sc);
Fix a bug in the reporting of residuals. The code was relying on the SG_COUNT filed in the hardware SCB not changing during the course of a transaction. Since the sequencer now DMAs the hardware SCB back up to the host when it detects a residual, this is no longer the case. I added a field to the "software" scb to mirror this information and it is now used for doing the residual calculation.
1997-02-11 17:10:37 +00:00
scb->sg_count = hscb->SG_segment_count;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
scb->flags |= SCB_SENSE;
When we perform an "automatic request sense", we issue an untagged command. The sequencer expects untagged transactions to have the SCBID of the transaction in the "busy target" array. So, ensure that the busy entry is up to date for the target in this case. The new identify code in the sequencer that performs additional sanity checking got caught up when a tagged transaction created an untagged request sense. In ahc_handle_seqint, ensure that the target ID is taken from the right place. In the case of a selection, the ID is in SCSIID. In the case of a reconnection it is found in SELID.
1997-03-01 06:50:41 +00:00
/*
* Ensure the target is busy since this
* will be an untagged request.
*/
ahc_busy_target(ahc, target, channel,
hscb->tag);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, RETURN_1, SEND_SENSE);
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
/*
* Ensure we have enough time to actually
* retrieve the sense.
*/
untimeout(ahc_timeout, (caddr_t)scb);
Changed timeout for requesting sense from 100ms to 1 second. Submitted by: gibbs
1997-02-14 03:13:37 +00:00
timeout(ahc_timeout, (caddr_t)scb, hz);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
break;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
}
Detect and report dataphase overruns. Put the adapter into 'Bit Bucket' mode when this occurs and allow the target to complete the transaction. Force a retry on overruns since they are usually caused by termination or cable problems.
1996-06-09 18:33:18 +01:00
/*
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
* Clear the SCB_SENSE Flag and have
* the sequencer do a normal command
* complete with either a "DRIVER_STUFFUP"
* error or whatever other error condition
* we already had.
Detect and report dataphase overruns. Put the adapter into 'Bit Bucket' mode when this occurs and allow the target to complete the transaction. Force a retry on overruns since they are usually caused by termination or cable problems.
1996-06-09 18:33:18 +01:00
*/
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
scb->flags &= ~SCB_SENSE;
if (xs->error == XS_NOERROR)
xs->error = XS_DRIVER_STUFFUP;
Detect and report dataphase overruns. Put the adapter into 'Bit Bucket' mode when this occurs and allow the target to complete the transaction. Force a retry on overruns since they are usually caused by termination or cable problems.
1996-06-09 18:33:18 +01:00
break;
Be even more careful in how we manipulate the QOUTQCNT variable. Now we do reset it from the QOUTCNT register inside a pause/unpause. This now happens once per command complete interrupt in the paging case (one interrupt can be for multiple completed commands). I may introduce a counter and do a lazy update in the future, similar to what is done with the QINCNT. Enhance the QUEUE FULL condition handling so that the number of openings will be reduced. This has become more important now that the driver is faster. This code really belongs in the gerneric SCSI layer, as will be the case once 3.0 gets the code from the 'SCSI' branch. Add some #if 0'd out trace code I've been using to help debug sequencer problems. Fix the SCB paging problem that I was seeing. This was only on my 7850 controller and stems from the fact that its QINFIFO can only handle 3bit SCB identifiers. This means that you can only have 8 transactions open at a time with the current paging scheme to these controllers. The code added to enforce this is generic in that it tests for the number of relevent bits that the QINFIFO can store and adjusts the max accordingly. It may be possible to come up with a scheme that allows for more than 8 commands at a time, but I don't know that it is worth the effort simply to fix a low end card. The aic7880 still can do 255. This problem may be related to what Andrey was seeing since I don't have n aic7770 rev E chip here to test on, but as soon as someone probes one of these cards with this new code, the dmesg output will tell the whole story.
1996-11-16 01:19:14 +00:00
case SCSI_QUEUE_FULL:
if (scb->hscb->control & TAG_ENB) {
/*
* The upper level SCSI code in 3.0
* handles this properly...
*/
struct scsi_link *sc_link;
sc_link = xs->sc_link;
if (sc_link->active > 2
&& sc_link->opennings != 0) {
/* truncate the opennings */
sc_link->opennings = 0;
sc_print_addr(sc_link);
printf("Tagged openings reduced to "
"%d\n", sc_link->active);
}
/*
* XXX requeue this unconditionally.
*/
Fix a nasty bug that meant a QUEUE_FULL status would result in a lost SCB. This is probably a main reason for the recent reports of timeouts.
1997-03-24 05:05:18 +00:00
scb->xs->retries++;
scb->xs->error = XS_BUSY;
Be even more careful in how we manipulate the QOUTQCNT variable. Now we do reset it from the QOUTCNT register inside a pause/unpause. This now happens once per command complete interrupt in the paging case (one interrupt can be for multiple completed commands). I may introduce a counter and do a lazy update in the future, similar to what is done with the QINCNT. Enhance the QUEUE FULL condition handling so that the number of openings will be reduced. This has become more important now that the driver is faster. This code really belongs in the gerneric SCSI layer, as will be the case once 3.0 gets the code from the 'SCSI' branch. Add some #if 0'd out trace code I've been using to help debug sequencer problems. Fix the SCB paging problem that I was seeing. This was only on my 7850 controller and stems from the fact that its QINFIFO can only handle 3bit SCB identifiers. This means that you can only have 8 transactions open at a time with the current paging scheme to these controllers. The code added to enforce this is generic in that it tests for the number of relevent bits that the QINFIFO can store and adjusts the max accordingly. It may be possible to come up with a scheme that allows for more than 8 commands at a time, but I don't know that it is worth the effort simply to fix a low end card. The aic7880 still can do 255. This problem may be related to what Andrey was seeing since I don't have n aic7770 rev E chip here to test on, but as soon as someone probes one of these cards with this new code, the dmesg output will tell the whole story.
1996-11-16 01:19:14 +00:00
break;
}
/* Else treat as if it is a BUSY condition */
scb->hscb->status = SCSI_BUSY;
/* Fall Through... */
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
case SCSI_BUSY:
xs->error = XS_BUSY;
sc_print_addr(xs->sc_link);
printf("Target Busy\n");
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
break;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
default:
sc_print_addr(xs->sc_link);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
printf("unexpected targ_status: %x\n", hscb->status);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
xs->error = XS_DRIVER_STUFFUP;
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
break;
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
break;
}
case AWAITING_MSG:
{
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
int scb_index;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
u_int8_t message_offset;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
scb_index = ahc_inb(ahc, SCB_TAG);
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
scb = ahc->scb_data->scbarray[scb_index];
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
/*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
* This SCB had MK_MESSAGE set in its control byte,
* informing the sequencer that we wanted to send a
* special message to this target.
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
*/
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
message_offset = ahc_inb(ahc, MSG_LEN);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
if (scb->flags & SCB_DEVICE_RESET) {
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, MSG_OUT, MSG_BUS_DEV_RESET);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, MSG_LEN, 1);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
sc_print_addr(scb->xs->sc_link);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
printf("Bus Device Reset Message Sent\n");
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
} else if (scb->flags & SCB_ABORT) {
if ((scb->hscb->control & TAG_ENB) != 0)
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, MSG_OUT + message_offset,
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
MSG_ABORT_TAG);
else
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, MSG_OUT + message_offset,
MSG_ABORT);
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
ahc_outb(ahc, MSG_LEN, message_offset + 1);
sc_print_addr(scb->xs->sc_link);
printf("Abort Message Sent\n");
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
} else if (scb->flags & SCB_MSGOUT_WDTR) {
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
ahc_construct_wdtr(ahc, message_offset, BUS_16_BIT);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
} else if (scb->flags & SCB_MSGOUT_SDTR) {
int sxfr;
int i;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
u_int16_t ultraenable;
u_int8_t target_scratch;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
/* Pull the user defined setting */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
target_scratch = ahc_inb(ahc, TARG_SCRATCH
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
+ scratch_offset);
sxfr = target_scratch & SXFR;
Remove some unnecessary overhead in the command complete processing. It should be nearly impossible to overflow the QOUTFIFO (worst case 9 command have to complete with at least 6 of them requiring paging on an aic7850), so don't take the additional PIO hit to guard against this condition. If we don't see our interrupt in time, the system has bigger problems elsewhere. If this ever does happen, the timeout handler will notice and retry the command. Remove the ABORT_TAG sequencer interrupt handler. This condition can't happen with the new SCB paging scheme. Fix a few bugs noticed by Dan Eischen <deischen@iworks.InterWorks.org> that could prevent ULTRA from being negotiated to drives above ID 7 and also could allow an SCB to be passed to ahc_done twice during error recovery. Fix a bug notice by Rory Bolt <rory@atBackup.com>. It turns out that a sequencer reset will actually start the sequencer running regardless of the state of the pause bit. This could lead to strange problems with loading the sequencer.
1997-01-22 18:05:31 +00:00
ultraenable = ahc_inb(ahc, ULTRA_ENB)
| (ahc_inb(ahc, ULTRA_ENB + 1) << 8);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
if (ultraenable & targ_mask)
/* Want an ultra speed in the table */
sxfr |= 0x100;
for (i = 0; i < ahc_num_syncrates; i++)
Bring in bug fix from 'SCSI' branch.
1996-10-06 23:50:56 +01:00
if (sxfr == ahc_syncrates[i].sxfr)
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
break;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
ahc_construct_sdtr(ahc, message_offset,
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
ahc_syncrates[i].period,
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
(target_scratch & WIDEXFER) ?
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
MAX_OFFSET_16BIT : MAX_OFFSET_8BIT);
} else
panic("ahc_intr: AWAITING_MSG for an SCB that "
"does not have a waiting message");
break;
}
case DATA_OVERRUN:
{
/*
* When the sequencer detects an overrun, it
* sets STCNT to 0x00ffffff and allows the
* target to complete its transfer in
* BITBUCKET mode.
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
u_int8_t scbindex = ahc_inb(ahc, SCB_TAG);
Fix a bug in the reporting of residuals. The code was relying on the SG_COUNT filed in the hardware SCB not changing during the course of a transaction. Since the sequencer now DMAs the hardware SCB back up to the host when it detects a residual, this is no longer the case. I added a field to the "software" scb to mirror this information and it is now used for doing the residual calculation.
1997-02-11 17:10:37 +00:00
u_int8_t lastphase = ahc_inb(ahc, LASTPHASE);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
u_int32_t overrun;
Fix a bug in the reporting of residuals. The code was relying on the SG_COUNT filed in the hardware SCB not changing during the course of a transaction. Since the sequencer now DMAs the hardware SCB back up to the host when it detects a residual, this is no longer the case. I added a field to the "software" scb to mirror this information and it is now used for doing the residual calculation.
1997-02-11 17:10:37 +00:00
int i;
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
scb = ahc->scb_data->scbarray[scbindex];
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
overrun = ahc_inb(ahc, STCNT)
| (ahc_inb(ahc, STCNT + 1) << 8)
| (ahc_inb(ahc, STCNT + 2) << 16);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
overrun = 0x00ffffff - overrun;
sc_print_addr(scb->xs->sc_link);
Fix a bug in the reporting of residuals. The code was relying on the SG_COUNT filed in the hardware SCB not changing during the course of a transaction. Since the sequencer now DMAs the hardware SCB back up to the host when it detects a residual, this is no longer the case. I added a field to the "software" scb to mirror this information and it is now used for doing the residual calculation.
1997-02-11 17:10:37 +00:00
printf("data overrun of %d bytes detected in %s phase."
" Tag == 0x%x. Forcing a retry.\n", overrun,
lastphase == P_DATAIN ? "Data-In" : "Data-Out",
scb->hscb->tag);
sc_print_addr(scb->xs->sc_link);
printf("%s seen Data Phase. Length = %d. NumSGs = %d.\n",
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_inb(ahc, SEQ_FLAGS) & DPHASE ? "Have" : "Haven't",
Fix a bug in the reporting of residuals. The code was relying on the SG_COUNT filed in the hardware SCB not changing during the course of a transaction. Since the sequencer now DMAs the hardware SCB back up to the host when it detects a residual, this is no longer the case. I added a field to the "software" scb to mirror this information and it is now used for doing the residual calculation.
1997-02-11 17:10:37 +00:00
scb->xs->datalen, scb->sg_count);
for (i = 0; i < scb->sg_count; i++) {
printf("sg[%d] - Addr 0x%x : Length %d\n",
i,
scb->ahc_dma[i].addr,
scb->ahc_dma[i].len);
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
/*
* Set this and it will take affect when the
* target does a command complete.
*/
scb->xs->error = XS_DRIVER_STUFFUP;
break;
}
#if NOT_YET
/* XXX Fill these in later */
case MESG_BUFFER_BUSY:
break;
case MSGIN_PHASEMIS:
break;
Be even more careful in how we manipulate the QOUTQCNT variable. Now we do reset it from the QOUTCNT register inside a pause/unpause. This now happens once per command complete interrupt in the paging case (one interrupt can be for multiple completed commands). I may introduce a counter and do a lazy update in the future, similar to what is done with the QINCNT. Enhance the QUEUE FULL condition handling so that the number of openings will be reduced. This has become more important now that the driver is faster. This code really belongs in the gerneric SCSI layer, as will be the case once 3.0 gets the code from the 'SCSI' branch. Add some #if 0'd out trace code I've been using to help debug sequencer problems. Fix the SCB paging problem that I was seeing. This was only on my 7850 controller and stems from the fact that its QINFIFO can only handle 3bit SCB identifiers. This means that you can only have 8 transactions open at a time with the current paging scheme to these controllers. The code added to enforce this is generic in that it tests for the number of relevent bits that the QINFIFO can store and adjusts the max accordingly. It may be possible to come up with a scheme that allows for more than 8 commands at a time, but I don't know that it is worth the effort simply to fix a low end card. The aic7880 still can do 255. This problem may be related to what Andrey was seeing since I don't have n aic7770 rev E chip here to test on, but as soon as someone probes one of these cards with this new code, the dmesg output will tell the whole story.
1996-11-16 01:19:14 +00:00
#endif
#if 0
case SCB_TRACE_POINT:
{
/*
* Print out the bus phase
*/
char *phase;
u_int8_t scbindex = ahc_inb(ahc, SCB_TAG);
u_int8_t lastphase = ahc_inb(ahc, LASTPHASE);
scb = ahc->scb_data->scbarray[scbindex];
sc_print_addr(scb->xs->sc_link);
switch (lastphase) {
case P_DATAOUT:
phase = "Data-Out";
break;
case P_DATAIN:
phase = "Data-In";
break;
case P_COMMAND:
phase = "Command";
break;
case P_MESGOUT:
phase = "Message-Out";
break;
case P_STATUS:
phase = "Status";
break;
case P_MESGIN:
phase = "Message-In";
break;
default:
phase = "busfree";
break;
}
printf("- %s\n", phase);
break;
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#endif
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
case ABORT_CMDCMPLT:
/* This interrupt serves to pause the sequencer
* until we can clean up the QOUTFIFO allowing us
* to hanle any abort SCBs that may have completed
* yet still have an SCB in the QINFIFO or
* waiting for selection queue. By the time we get
* here, we should have already cleaned up the
* queues, so all we need to do is unpause the sequencer.
*/
break;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
default:
printf("ahc_intr: seqint, "
"intstat == 0x%x, scsisigi = 0x%x\n",
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
intstat, ahc_inb(ahc, SCSISIGI));
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
break;
}
clear:
/*
* Clear the upper byte that holds SEQINT status
* codes and clear the SEQINT bit.
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, CLRINT, CLRSEQINT);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
/*
* The sequencer is paused immediately on
* a SEQINT, so we should restart it when
* we're done.
*/
unpause_sequencer(ahc, /*unpause_always*/TRUE);
Fixed the type of ahcintr(). The type of an ISA interrupt handler is incompatible with the type of a PCI interrupt handler. A new entry point `ahc_pci_intr()' is used for PCI. ISA and PCI interrupts are penalized equally (:-) by calling a common handler `ahc_intr()'. This should be reorganized. Some strings now name the wrong function...
1995-11-04 14:43:30 +00:00
}
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 22:14:45 +01:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
static void
ahc_handle_scsiint(ahc, intstat)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t intstat;
{
u_int8_t scb_index;
u_int8_t status;
struct scb *scb;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
scb_index = ahc_inb(ahc, SCB_TAG);
status = ahc_inb(ahc, SSTAT1);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
if (scb_index < ahc->scb_data->numscbs) {
scb = ahc->scb_data->scbarray[scb_index];
if ((scb->flags & SCB_ACTIVE) == 0)
scb = NULL;
} else
scb = NULL;
if ((status & SCSIRSTI) != 0) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
char channel;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
channel = (ahc_inb(ahc, SBLKCTL) & SELBUSB) ? 'B' : 'A';
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
printf("%s: Someone reset channel %c\n",
ahc_name(ahc), channel);
ahc_reset_channel(ahc,
channel,
XS_BUSY,
/* Initiate Reset */FALSE);
scb = NULL;
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
} else if ((status & BUSFREE) != 0 && (status & SELTO) == 0) {
/*
* First look at what phase we were last in.
* If its message out, chances are pretty good
* that the busfree was in response to one of
* our abort requests.
*/
u_int8_t lastphase = ahc_inb(ahc, LASTPHASE);
u_int8_t target = (ahc_inb(ahc, SCSIID) >> 4) & 0x0f;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
u_int8_t scbptr = ahc_inb(ahc, SCBPTR);
u_int8_t scb_control = ahc_inb(ahc, SCB_CONTROL);
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
char channel = ahc_inb(ahc, SBLKCTL) & SELBUSB ? 'B': 'A';
int printerror = 1;
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, SCSISEQ, 0);
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
if (lastphase != P_BUSFREE) {
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
u_int8_t flags = ahc_inb(ahc, SEQ_FLAGS)
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
& (IDENTIFY_SEEN|RESELECTED);
if (flags == 0 || flags == (IDENTIFY_SEEN|RESELECTED)) {
/*
* We have an SCB we have to clean up.
*/
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
if ((scb_control & DISCONNECTED) != 0)
ahc_rem_scb_from_disc_list(ahc, scbptr);
else {
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
ahc_add_curscb_to_free_list(ahc);
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
}
/* Did we ask for this?? */
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
if ((lastphase == P_MESGOUT
|| lastphase == P_MESGIN) && scb != NULL) {
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
if (scb->flags & SCB_DEVICE_RESET) {
ahc_handle_devreset(ahc, scb);
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
scb = NULL;
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
printerror = 0;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
} else if (scb->flags & SCB_ABORT) {
Don't rely on AAP(Auto Access Pause) when queueing SCBs to the card. This will increase the overhead of queueing a command, but some recent bug reports make me believe that AAP isn't really working and that we are losing some SCBs from the input queue. Hopefully this will cure that problem. Fix some bugs in the error recovery code. Mainly these could cause us to inadvertantly forget to untimeout an SCB that was recovered causing later confusion.
1997-02-18 04:25:31 +00:00
struct hardware_scb *hscb;
u_int8_t tag;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
int lun;
Don't rely on AAP(Auto Access Pause) when queueing SCBs to the card. This will increase the overhead of queueing a command, but some recent bug reports make me believe that AAP isn't really working and that we are losing some SCBs from the input queue. Hopefully this will cure that problem. Fix some bugs in the error recovery code. Mainly these could cause us to inadvertantly forget to untimeout an SCB that was recovered causing later confusion.
1997-02-18 04:25:31 +00:00
hscb = scb->hscb;
tag = SCB_LIST_NULL;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
lun = hscb->tcl & 0x7;
sc_print_addr(scb->xs->sc_link);
printf("SCB %d - Abort "
"Completed.\n",
scbptr);
Don't rely on AAP(Auto Access Pause) when queueing SCBs to the card. This will increase the overhead of queueing a command, but some recent bug reports make me believe that AAP isn't really working and that we are losing some SCBs from the input queue. Hopefully this will cure that problem. Fix some bugs in the error recovery code. Mainly these could cause us to inadvertantly forget to untimeout an SCB that was recovered causing later confusion.
1997-02-18 04:25:31 +00:00
if (hscb->control & TAG_ENB)
tag = scb->hscb->tag;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
ahc_reset_device(ahc,
target,
channel,
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
lun,
Don't rely on AAP(Auto Access Pause) when queueing SCBs to the card. This will increase the overhead of queueing a command, but some recent bug reports make me believe that AAP isn't really working and that we are losing some SCBs from the input queue. Hopefully this will cure that problem. Fix some bugs in the error recovery code. Mainly these could cause us to inadvertantly forget to untimeout an SCB that was recovered causing later confusion.
1997-02-18 04:25:31 +00:00
tag,
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
XS_TIMEOUT);
ahc_run_done_queue(ahc);
Remove some unnecessary overhead in the command complete processing. It should be nearly impossible to overflow the QOUTFIFO (worst case 9 command have to complete with at least 6 of them requiring paging on an aic7850), so don't take the additional PIO hit to guard against this condition. If we don't see our interrupt in time, the system has bigger problems elsewhere. If this ever does happen, the timeout handler will notice and retry the command. Remove the ABORT_TAG sequencer interrupt handler. This condition can't happen with the new SCB paging scheme. Fix a few bugs noticed by Dan Eischen <deischen@iworks.InterWorks.org> that could prevent ULTRA from being negotiated to drives above ID 7 and also could allow an SCB to be passed to ahc_done twice during error recovery. Fix a bug notice by Rory Bolt <rory@atBackup.com>. It turns out that a sequencer reset will actually start the sequencer running regardless of the state of the pause bit. This could lead to strange problems with loading the sequencer.
1997-01-22 18:05:31 +00:00
scb = NULL;
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
printerror = 0;
}
}
}
}
if (printerror != 0) {
if (scb != NULL) {
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
u_int8_t next;
ahc_index_busy_target(ahc, target,
channel, /*unbusy*/TRUE);
ahc_outb(ahc, SCBPTR, scbptr);
next = ahc_inb(ahc, SCB_LINKED_NEXT);
if (next != SCB_LIST_NULL) {
/*
* Re-queue the waiting SCB via the
* waiting list.
*/
struct scb *next_scb;
next_scb =
ahc->scb_data->scbarray[next];
STAILQ_INSERT_HEAD(&ahc->waiting_scbs,
next_scb, links);
next_scb->flags |= SCB_WAITINGQ;
}
When not using SCB paging, we can always directly index the SCB of interest either by looking it up in the array of pending, per target, untagged transactions, or by using the tag value passed in during the identify. The old code only direct indexed for tagged transactions. This makes the "findSCB" routine only necessary when SCB paging is enabled, so appropriately conditionalize it. This greatly simplifies the non SCB paging code flow.
1997-04-04 05:09:29 +01:00
scb->xs->error = XS_TIMEOUT;
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
sc_print_addr(scb->xs->sc_link);
ahc_done(ahc, scb);
Remove some unnecessary overhead in the command complete processing. It should be nearly impossible to overflow the QOUTFIFO (worst case 9 command have to complete with at least 6 of them requiring paging on an aic7850), so don't take the additional PIO hit to guard against this condition. If we don't see our interrupt in time, the system has bigger problems elsewhere. If this ever does happen, the timeout handler will notice and retry the command. Remove the ABORT_TAG sequencer interrupt handler. This condition can't happen with the new SCB paging scheme. Fix a few bugs noticed by Dan Eischen <deischen@iworks.InterWorks.org> that could prevent ULTRA from being negotiated to drives above ID 7 and also could allow an SCB to be passed to ahc_done twice during error recovery. Fix a bug notice by Rory Bolt <rory@atBackup.com>. It turns out that a sequencer reset will actually start the sequencer running regardless of the state of the pause bit. This could lead to strange problems with loading the sequencer.
1997-01-22 18:05:31 +00:00
scb = NULL;
When not using SCB paging, we can always directly index the SCB of interest either by looking it up in the array of pending, per target, untagged transactions, or by using the tag value passed in during the identify. The old code only direct indexed for tagged transactions. This makes the "findSCB" routine only necessary when SCB paging is enabled, so appropriately conditionalize it. This greatly simplifies the non SCB paging code flow.
1997-04-04 05:09:29 +01:00
} else {
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
printf("%s: ", ahc_name(ahc));
When not using SCB paging, we can always directly index the SCB of interest either by looking it up in the array of pending, per target, untagged transactions, or by using the tag value passed in during the identify. The old code only direct indexed for tagged transactions. This makes the "findSCB" routine only necessary when SCB paging is enabled, so appropriately conditionalize it. This greatly simplifies the non SCB paging code flow.
1997-04-04 05:09:29 +01:00
}
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
printf("Unexpected busfree. LASTPHASE == 0x%x\n"
"SEQADDR == 0x%x\n",
lastphase, (ahc_inb(ahc, SEQADDR1) << 8)
| ahc_inb(ahc, SEQADDR0));
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
}
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
ahc_outb(ahc, SIMODE1, ahc_inb(ahc, SIMODE1) & ~ENBUSFREE);
Clear the busfree interrupt when one occurs, after a SELTO, or a bus reset.
1996-12-03 17:06:00 +00:00
ahc_outb(ahc, CLRSINT1, CLRBUSFREE);
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
ahc_outb(ahc, CLRINT, CLRSCSIINT);
Drop the number of allowed tags back down to 8. Pluto uses a higher value which mistakenly got committed. Fix two bugs in the ahc_reset_device code: Limit search for SCBs to process to those that are active and are not queued for done processing. It's okay for an SCB to not have a waiting next SCB.
1997-04-10 20:14:58 +01:00
restart_sequencer(ahc);
Fix a bug in the selection timeout handler that was introduced when the selection loop was merged with the poll_for_work loop. We cannot assume that the SCB for the selection timeout is the current SCB. Instead we must look at the SCB at the head of the waiting for selection list. This fixes part of a problem reported by David Malone, but does not explain why he was getting selection timeouts in the first place.
1997-04-05 22:41:13 +01:00
} else if ((status & SELTO) != 0) {
struct scsi_xfer *xs;
u_int8_t scbptr;
u_int8_t nextscb;
scbptr = ahc_inb(ahc, WAITING_SCBH);
ahc_outb(ahc, SCBPTR, scbptr);
scb_index = ahc_inb(ahc, SCB_TAG);
if (scb_index < ahc->scb_data->numscbs) {
scb = ahc->scb_data->scbarray[scb_index];
if ((scb->flags & SCB_ACTIVE) == 0)
scb = NULL;
} else
scb = NULL;
if (scb == NULL) {
printf("%s: ahc_intr - referenced scb not "
"valid during SELTO scb(%d)\n",
ahc_name(ahc), scb_index);
} else {
/*
* XXX If we queued an abort tag, go clean up the
* disconnected list.
*/
xs = scb->xs;
xs->error = XS_SELTIMEOUT;
/*
* Clear any pending messages for the timed out
* target, and mark the target as free
*/
ahc_outb(ahc, MSG_LEN, 0);
ahc_index_busy_target(ahc, xs->sc_link->target,
IS_SCSIBUS_B(ahc, xs->sc_link) ? 'B' : 'A',
/*unbusy*/TRUE);
ahc_outb(ahc, SCB_CONTROL, 0);
/* Shift the waiting Q forward. */
nextscb = ahc_inb(ahc, SCB_NEXT);
ahc_outb(ahc, WAITING_SCBH, nextscb);
ahc_add_curscb_to_free_list(ahc);
}
/* Stop the selection */
ahc_outb(ahc, SCSISEQ, 0);
ahc_outb(ahc, CLRSINT1, CLRSELTIMEO|CLRBUSFREE);
ahc_outb(ahc, CLRINT, CLRSCSIINT);
restart_sequencer(ahc);
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
} else if (scb == NULL) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
printf("%s: ahc_intr - referenced scb not "
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
"valid during scsiint 0x%x scb(%d)\n"
"SIMODE0 = 0x%x, SIMODE1 = 0x%x, SSTAT0 = 0x%x\n"
"SEQADDR = 0x%x\n", ahc_name(ahc),
status, scb_index, ahc_inb(ahc, SIMODE0),
ahc_inb(ahc, SIMODE1), ahc_inb(ahc, SSTAT0),
(ahc_inb(ahc, SEQADDR1) << 8)
| ahc_inb(ahc, SEQADDR0));
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, CLRSINT1, status);
ahc_outb(ahc, CLRINT, CLRSCSIINT);
Drop the number of allowed tags back down to 8. Pluto uses a higher value which mistakenly got committed. Fix two bugs in the ahc_reset_device code: Limit search for SCBs to process to those that are active and are not queued for done processing. It's okay for an SCB to not have a waiting next SCB.
1997-04-10 20:14:58 +01:00
unpause_sequencer(ahc, /*unpause_always*/TRUE);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
scb = NULL;
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
} else if ((status & SCSIPERR) != 0) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/*
* Determine the bus phase and
* queue an appropriate message
*/
char *phase;
u_int8_t mesg_out = MSG_NOOP;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
u_int8_t lastphase = ahc_inb(ahc, LASTPHASE);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
struct scsi_xfer *xs;
xs = scb->xs;
sc_print_addr(xs->sc_link);
switch (lastphase) {
case P_DATAOUT:
phase = "Data-Out";
break;
case P_DATAIN:
phase = "Data-In";
mesg_out = MSG_INITIATOR_DET_ERR;
break;
case P_COMMAND:
phase = "Command";
break;
case P_MESGOUT:
phase = "Message-Out";
break;
case P_STATUS:
phase = "Status";
mesg_out = MSG_INITIATOR_DET_ERR;
break;
case P_MESGIN:
phase = "Message-In";
mesg_out = MSG_PARITY_ERROR;
break;
default:
phase = "unknown";
break;
}
printf("parity error during %s phase.\n", phase);
/*
* We've set the hardware to assert ATN if we
* get a parity error on "in" phases, so all we
* need to do is stuff the message buffer with
* the appropriate message. "In" phases have set
* mesg_out to something other than MSG_NOP.
*/
if (mesg_out != MSG_NOOP) {
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, MSG_OUT, mesg_out);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, MSG_LEN, 1);
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
scb = NULL;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else
/*
* Should we allow the target to make
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
* this decision for us? If we get a
* sense request from the drive, we will
* not fetch it since xs->error != XS_NOERROR.
* perhaps we need two error fields in the
* xs structure?
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
*/
xs->error = XS_DRIVER_STUFFUP;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
ahc_outb(ahc, CLRSINT1, CLRSCSIPERR);
ahc_outb(ahc, CLRINT, CLRSCSIINT);
Drop the number of allowed tags back down to 8. Pluto uses a higher value which mistakenly got committed. Fix two bugs in the ahc_reset_device code: Limit search for SCBs to process to those that are active and are not queued for done processing. It's okay for an SCB to not have a waiting next SCB.
1997-04-10 20:14:58 +01:00
unpause_sequencer(ahc, /*unpause_always*/TRUE);
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
} else {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
sc_print_addr(scb->xs->sc_link);
printf("Unknown SCSIINT. Status = 0x%x\n", status);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, CLRSINT1, status);
ahc_outb(ahc, CLRINT, CLRSCSIINT);
Drop the number of allowed tags back down to 8. Pluto uses a higher value which mistakenly got committed. Fix two bugs in the ahc_reset_device code: Limit search for SCBs to process to those that are active and are not queued for done processing. It's okay for an SCB to not have a waiting next SCB.
1997-04-10 20:14:58 +01:00
unpause_sequencer(ahc, /*unpause_always*/TRUE);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
scb = NULL;
}
if (scb != NULL) {
/* We want to process the command */
ahc_done(ahc, scb);
}
}
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
static void
ahc_handle_devreset(ahc, scb)
struct ahc_softc *ahc;
struct scb *scb;
{
u_int16_t targ_mask;
u_int8_t targ_scratch;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
int target = scb->xs->sc_link->target;
int lun = scb->xs->sc_link->lun;
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
int scratch_offset = target;
char channel = (scb->hscb->tcl & SELBUSB) ? 'B': 'A';
int found;
if (channel == 'B')
scratch_offset += 8;
targ_mask = (0x01 << scratch_offset);
/*
* Go back to async/narrow transfers and
* renegotiate.
*/
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
ahc_index_busy_target(ahc, target, channel, /*unbusy*/TRUE);
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
ahc->needsdtr |= ahc->needsdtr_orig & targ_mask;
ahc->needwdtr |= ahc->needwdtr_orig & targ_mask;
ahc->sdtrpending &= ~targ_mask;
ahc->wdtrpending &= ~targ_mask;
targ_scratch = ahc_inb(ahc, TARG_SCRATCH + scratch_offset);
targ_scratch &= SXFR;
ahc_outb(ahc, TARG_SCRATCH + scratch_offset, targ_scratch);
sc_print_addr(scb->xs->sc_link);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
found = ahc_reset_device(ahc, target, channel, lun,
SCB_LIST_NULL, XS_NOERROR);
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
printf("Bus Device Reset delivered. %d SCBs aborted\n", found);
ahc_run_done_queue(ahc);
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
* We have a scb which has been processed by the
* adaptor, now we look to see how the operation
* went.
*/
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static void
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
ahc_done(ahc, scb)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
struct scb *scb;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
struct scsi_xfer *xs = scb->xs;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
SC_DEBUG(xs->sc_link, SDEV_DB2, ("ahc_done\n"));
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
/*
* If the recovery SCB completes, we have to be
* out of our timeout.
*/
if (scb->flags & SCB_RECOVERY_SCB) {
ahc->in_timeout = FALSE;
sc_print_addr(scb->xs->sc_link);
printf("no longer in timeout\n");
}
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
/*
* Put the results of the operation
* into the xfer and call whoever started it
*/
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
/* Don't override the error value. */
if (xs->error == XS_NOERROR
&& (scb->flags & SCB_SENSE) != 0)
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
xs->error = XS_SENSE;
Fix some more problems in the recovery code. Cleanup of the disconnected list was broken in the SCB paging case (confusion of NULLand SCB_LIST_NULL) Implement a clean mechanism for determining that we have exited the timeout state and test for this in ahc_done instead of all over the place. Bring back the use of AAP (Auto Access Pause) I don't think it was the true cause of the bus hangs people were reporting. We want to reset the bus if we've been through an Abort action, not if we are a recovery SCB (one implies the other, but not vice-versa).
1997-02-19 01:00:51 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__FreeBSD__)
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
if ((xs->flags & SCSI_ERR_OK) && !(xs->error == XS_SENSE)) {
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/* All went correctly OR errors expected */
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
xs->error = XS_NOERROR;
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#elif defined(__NetBSD__)
/*
* Since NetBSD doesn't have error ignoring operation mode
* (SCSI_ERR_OK in FreeBSD), we don't have to care this case.
*/
#endif
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
xs->flags |= ITSDONE;
Fix the abort code in the BUS DEVICE RESET case. Needed to set cmdlen to 0. Ensure that queued commands are not touched by the abort code by setting the SCB status to indicate what queue it is in. Fix deadlocks when using SCB paging by using SCBs from the assigned_scbs queue or an SCB that completed during the same interrupt if needed. Don't ever use insl to pull SCBs from any of the controllers. You can only do 8bit PIO reads. This only affected SCB paging. With this checkin, SCB paging works quite a bit better, but I still have some problems with it that may be caused by a firmware problem in my PD1800s. It seems that using a tag number higher than the maximum number of tags allowed by the device, confuses it. For example, if I queue two commands, tagged 3 and 36, it never reconnects for tag 36.
1996-04-28 20:21:20 +01:00
#ifdef AHC_TAGENABLE
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/*
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
* This functionality will be provided by the generic SCSI layer
* in FreeBSD 3.0.
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
*/
if (xs->cmd->opcode == INQUIRY && xs->error == XS_NOERROR) {
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
struct scsi_inquiry_data *inq_data;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int16_t mask = 0x01 << (xs->sc_link->target |
(scb->hscb->tcl & 0x08));
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
/*
* Sneak a look at the results of the SCSI Inquiry
* command and see if we can do Tagged queing. This
* should really be done by the higher level drivers.
*/
inq_data = (struct scsi_inquiry_data *)xs->data;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if ((inq_data->flags & SID_CmdQue)
&& !(ahc->tagenable & mask)) {
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
printf("%s: target %d Tagged Queuing Device\n",
ahc_name(ahc), xs->sc_link->target);
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
ahc->tagenable |= mask;
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
if (ahc->scb_data->maxhscbs >= 16
|| (ahc->flags & AHC_PAGESCBS)) {
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
/* Default to 8 tags */
Drop the number of allowed tags back down to 8. Pluto uses a higher value which mistakenly got committed. Fix two bugs in the ahc_reset_device code: Limit search for SCBs to process to those that are active and are not queued for done processing. It's okay for an SCB to not have a waiting next SCB.
1997-04-10 20:14:58 +01:00
xs->sc_link->opennings += 6;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else {
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
/*
* Default to 4 tags on whimpy
* cards that don't have much SCB
* space and can't page. This prevents
* a single device from hogging all
* slots. We should really have a better
* way of providing fairness.
*/
xs->sc_link->opennings += 2;
}
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
}
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
#endif /* AHC_TAGENABLE */
Remove some unnecessary overhead in the command complete processing. It should be nearly impossible to overflow the QOUTFIFO (worst case 9 command have to complete with at least 6 of them requiring paging on an aic7850), so don't take the additional PIO hit to guard against this condition. If we don't see our interrupt in time, the system has bigger problems elsewhere. If this ever does happen, the timeout handler will notice and retry the command. Remove the ABORT_TAG sequencer interrupt handler. This condition can't happen with the new SCB paging scheme. Fix a few bugs noticed by Dan Eischen <deischen@iworks.InterWorks.org> that could prevent ULTRA from being negotiated to drives above ID 7 and also could allow an SCB to be passed to ahc_done twice during error recovery. Fix a bug notice by Rory Bolt <rory@atBackup.com>. It turns out that a sequencer reset will actually start the sequencer running regardless of the state of the pause bit. This could lead to strange problems with loading the sequencer.
1997-01-22 18:05:31 +00:00
if ((scb->flags & SCB_MSGOUT_WDTR|SCB_MSGOUT_SDTR) != 0) {
/*
* Turn off the pending flags for any DTR messages
* regardless of whether they completed successfully
* or not. This ensures that we don't have lingering
* state after we abort an SCB.
*/
u_int16_t mask;
mask = (0x01 << (xs->sc_link->target
| (IS_SCSIBUS_B(ahc, xs->sc_link) ? SELBUSB : 0)));
if (scb->flags & SCB_MSGOUT_WDTR)
ahc->wdtrpending &= ~mask;
if (scb->flags & SCB_MSGOUT_SDTR)
ahc->sdtrpending &= ~mask;
}
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
untimeout(ahc_timeout, (caddr_t)scb);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
ahc_free_scb(ahc, scb);
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
/*
* If we're polling, we rely on the ITS_DONE flag in the xs structure
* to know that the command has completed. Unfortunately, scsi_done
* can cause the same xs to get requeued putting us in an infinite
* loop. So, we defer the scsi_done call until the poll routine exits
* its loop. I hate the way this works.
*/
if ((xs->flags & SCSI_NOMASK) == 0)
scsi_done(xs);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
/*
* Start the board, ready for normal operation
*/
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
int
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
ahc_init(ahc)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
93cx6.c: Style nit. Backslashes in macro weren't aligned. aic7xxx.c: Preserve the value of STPWEN in SXFRCTL1 during initialization. STPWEN controls low byte termination and is setup by the PCI probe front end.
1997-01-24 21:59:32 +00:00
u_int8_t scsi_conf, sblkctl, sxfrctl1, i;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
u_int16_t ultraenable = 0;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
int max_targ = 15;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
* Assume we have a board at this stage and it has been reset.
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*/
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
/* Handle the SCBPAGING option */
#ifndef AHC_SCBPAGING_ENABLE
ahc->flags &= ~AHC_PAGESCBS;
#endif
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
/* Determine channel configuration and who we are on the scsi bus. */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
switch ((sblkctl = ahc_inb(ahc, SBLKCTL) & 0x0a)) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
case 0:
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc->our_id = (ahc_inb(ahc, SCSICONF) & HSCSIID);
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
ahc->flags &= ~AHC_CHANNEL_B_PRIMARY;
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
if ((ahc->type & AHC_39X) != 0) {
char channel = 'A';
if ((ahc->flags & (AHC_CHNLB|AHC_CHNLC)) != 0)
channel = ahc->flags & AHC_CHNLB ? 'B' : 'C';
printf("Channel %c, SCSI Id=%d, ", channel,
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-06 00:52:03 +01:00
ahc->our_id);
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
} else
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-06 00:52:03 +01:00
printf("Single Channel, SCSI Id=%d, ", ahc->our_id);
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, SEQ_FLAGS, ahc->flags & AHC_PAGESCBS);
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
break;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
case 2:
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc->our_id = (ahc_inb(ahc, SCSICONF + 1) & HWSCSIID);
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
ahc->flags &= ~AHC_CHANNEL_B_PRIMARY;
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
if ((ahc->type & AHC_39X) != 0) {
char channel = 'A';
if ((ahc->flags & (AHC_CHNLB|AHC_CHNLC)) != 0)
channel = ahc->flags & AHC_CHNLB ? 'B' : 'C';
printf("Wide Channel %c, SCSI Id=%d, ", channel,
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-06 00:52:03 +01:00
ahc->our_id);
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
} else
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-06 00:52:03 +01:00
printf("Wide Channel, SCSI Id=%d, ", ahc->our_id);
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
ahc->type |= AHC_WIDE;
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, SEQ_FLAGS, WIDE_BUS | (ahc->flags & AHC_PAGESCBS));
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
break;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
case 8:
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc->our_id = (ahc_inb(ahc, SCSICONF) & HSCSIID);
ahc->our_id_b = (ahc_inb(ahc, SCSICONF + 1) & HSCSIID);
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
printf("Twin Channel, A SCSI Id=%d, B SCSI Id=%d, ",
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
ahc->our_id, ahc->our_id_b);
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
ahc->type |= AHC_TWIN;
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, SEQ_FLAGS, TWIN_BUS | (ahc->flags & AHC_PAGESCBS));
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
break;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
default:
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
printf(" Unsupported adapter type. Ignoring\n");
return(-1);
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/* Determine the number of SCBs and initialize them */
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
if (ahc->scb_data->maxhscbs == 0) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/* SCB 0 heads the free list */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, FREE_SCBH, 0);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
for (i = 0; i < AHC_SCB_MAX; i++) {
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCBPTR, i);
ahc_outb(ahc, SCB_CONTROL, i);
if(ahc_inb(ahc, SCB_CONTROL) != i)
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
break;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCBPTR, 0);
if(ahc_inb(ahc, SCB_CONTROL) != 0)
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
break;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCBPTR, i);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/* Clear the control byte. */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCB_CONTROL, 0);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/* Set the next pointer */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCB_NEXT, i+1);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
/* Make the tag number invalid */
ahc_outb(ahc, SCB_TAG, SCB_LIST_NULL);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/* No Busy non-tagged targets yet */
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, SCB_BUSYTARGETS, SCB_LIST_NULL);
ahc_outb(ahc, SCB_BUSYTARGETS + 1, SCB_LIST_NULL);
ahc_outb(ahc, SCB_BUSYTARGETS + 2, SCB_LIST_NULL);
ahc_outb(ahc, SCB_BUSYTARGETS + 3, SCB_LIST_NULL);
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/* Make that the last SCB terminates the free list */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCBPTR, i-1);
ahc_outb(ahc, SCB_NEXT, SCB_LIST_NULL);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
/* Ensure we clear the 0 SCB's control byte. */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCBPTR, 0);
ahc_outb(ahc, SCB_CONTROL, 0);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
ahc->scb_data->maxhscbs = i;
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
}
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
if ((ahc->scb_data->maxhscbs < AHC_SCB_MAX)
&& (ahc->flags & AHC_PAGESCBS)) {
Be even more careful in how we manipulate the QOUTQCNT variable. Now we do reset it from the QOUTCNT register inside a pause/unpause. This now happens once per command complete interrupt in the paging case (one interrupt can be for multiple completed commands). I may introduce a counter and do a lazy update in the future, similar to what is done with the QINCNT. Enhance the QUEUE FULL condition handling so that the number of openings will be reduced. This has become more important now that the driver is faster. This code really belongs in the gerneric SCSI layer, as will be the case once 3.0 gets the code from the 'SCSI' branch. Add some #if 0'd out trace code I've been using to help debug sequencer problems. Fix the SCB paging problem that I was seeing. This was only on my 7850 controller and stems from the fact that its QINFIFO can only handle 3bit SCB identifiers. This means that you can only have 8 transactions open at a time with the current paging scheme to these controllers. The code added to enforce this is generic in that it tests for the number of relevent bits that the QINFIFO can store and adjusts the max accordingly. It may be possible to come up with a scheme that allows for more than 8 commands at a time, but I don't know that it is worth the effort simply to fix a low end card. The aic7880 still can do 255. This problem may be related to what Andrey was seeing since I don't have n aic7770 rev E chip here to test on, but as soon as someone probes one of these cards with this new code, the dmesg output will tell the whole story.
1996-11-16 01:19:14 +00:00
u_int8_t max_scbid = 255;
/* Determine the number of valid bits in the FIFOs */
ahc_outb(ahc, QINFIFO, max_scbid);
max_scbid = ahc_inb(ahc, QINFIFO);
ahc->scb_data->maxscbs = MIN(AHC_SCB_MAX, max_scbid + 1);
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
printf("%d/%d SCBs\n", ahc->scb_data->maxhscbs,
ahc->scb_data->maxscbs);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else {
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
ahc->scb_data->maxscbs = ahc->scb_data->maxhscbs;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
ahc->flags &= ~AHC_PAGESCBS;
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
printf("%d SCBs\n", ahc->scb_data->maxhscbs);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
}
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
#ifdef AHC_DEBUG
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc_debug & AHC_SHOWMISC) {
printf("%s: hardware scb %d bytes; kernel scb %d bytes; "
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
"ahc_dma %d bytes\n",
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
ahc_name(ahc),
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
sizeof(struct hardware_scb),
sizeof(struct scb),
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
sizeof(struct ahc_dma_seg));
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
}
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
#endif /* AHC_DEBUG */
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Properly deal with the Ultra series of adapters. We should now understand the new seeprom format and negotiate up to 20MHz sync if set in SCSI-Select. Reduce the complexity of the timeout code by running it at splhigh(). Fix a bug that caused rescheduled timeouts at 0 clock ticks in the future causing an infinite loop. Obtained from: Timeout bug noticed by David Greenman and wcarchive.
1995-10-26 23:57:18 +00:00
/* Set the SCSI Id, SXFRCTL0, SXFRCTL1, and SIMODE1, for both channels*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc->type & AHC_TWIN) {
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
/*
Initialize SIMODE1, SCSIID, and SXFRCTL1 in ahc_init instead of at each reconnection/selection with a target.
1995-01-22 00:48:39 +00:00
* The device is gated to channel B after a chip reset,
* so set those values first
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCSIID, ahc->our_id_b);
scsi_conf = ahc_inb(ahc, SCSICONF + 1);
93cx6.c: Style nit. Backslashes in macro weren't aligned. aic7xxx.c: Preserve the value of STPWEN in SXFRCTL1 during initialization. STPWEN controls low byte termination and is setup by the PCI probe front end.
1997-01-24 21:59:32 +00:00
sxfrctl1 = ahc_inb(ahc, SXFRCTL1);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SXFRCTL1, (scsi_conf & (ENSPCHK|STIMESEL))
93cx6.c: Style nit. Backslashes in macro weren't aligned. aic7xxx.c: Preserve the value of STPWEN in SXFRCTL1 during initialization. STPWEN controls low byte termination and is setup by the PCI probe front end.
1997-01-24 21:59:32 +00:00
|(sxfrctl1 & STPWEN)
|ENSTIMER|ACTNEGEN);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SIMODE1, ENSELTIMO|ENSCSIRST|ENSCSIPERR);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc->type & AHC_ULTRA)
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
ahc_outb(ahc, SXFRCTL0, DFON|SPIOEN|FAST20);
Initialize SXFRCTL0 with DFON|SPIOEN|ULTRAEN.
1995-10-29 05:57:48 +00:00
else
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SXFRCTL0, DFON|SPIOEN);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (scsi_conf & RESET_SCSI) {
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
/* Reset the bus */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (bootverbose)
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
printf("%s: Resetting Channel B\n",
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
ahc_name(ahc));
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
ahc_reset_current_bus(ahc);
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
}
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
Initialize SIMODE1, SCSIID, and SXFRCTL1 in ahc_init instead of at each reconnection/selection with a target.
1995-01-22 00:48:39 +00:00
/* Select Channel A */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SBLKCTL, 0);
Initialize SIMODE1, SCSIID, and SXFRCTL1 in ahc_init instead of at each reconnection/selection with a target.
1995-01-22 00:48:39 +00:00
}
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCSIID, ahc->our_id);
scsi_conf = ahc_inb(ahc, SCSICONF);
93cx6.c: Style nit. Backslashes in macro weren't aligned. aic7xxx.c: Preserve the value of STPWEN in SXFRCTL1 during initialization. STPWEN controls low byte termination and is setup by the PCI probe front end.
1997-01-24 21:59:32 +00:00
sxfrctl1 = ahc_inb(ahc, SXFRCTL1);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SXFRCTL1, (scsi_conf & (ENSPCHK|STIMESEL))
93cx6.c: Style nit. Backslashes in macro weren't aligned. aic7xxx.c: Preserve the value of STPWEN in SXFRCTL1 during initialization. STPWEN controls low byte termination and is setup by the PCI probe front end.
1997-01-24 21:59:32 +00:00
|(sxfrctl1 & STPWEN)
|ENSTIMER|ACTNEGEN);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SIMODE1, ENSELTIMO|ENSCSIRST|ENSCSIPERR);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc->type & AHC_ULTRA)
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
ahc_outb(ahc, SXFRCTL0, DFON|SPIOEN|FAST20);
Initialize SXFRCTL0 with DFON|SPIOEN|ULTRAEN.
1995-10-29 05:57:48 +00:00
else
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SXFRCTL0, DFON|SPIOEN);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (scsi_conf & RESET_SCSI) {
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
/* Reset the bus */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (bootverbose)
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
printf("%s: Resetting Channel A\n", ahc_name(ahc));
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
ahc_reset_current_bus(ahc);
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 17:21:05 +01:00
}
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
* Look at the information that board initialization or
* the board bios has left us. In the lower four bits of each
* target's scratch space any value other than 0 indicates
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
* that we should initiate synchronous transfers. If it's zero,
* the user or the BIOS has decided to disable synchronous
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
* negotiation to that target so we don't activate the needsdtr
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
* flag.
*/
Add $Ids. Use DMA for downloading SCBs to the sequencer. Implement SDTR, WDTR, and message reject handlers so they don't need to exist in the sequencer. All three of these cases are not on the critical path, so it makes little sense to use up precious sequencer ram for them.
1995-01-16 16:33:47 +00:00
ahc->needsdtr_orig = 0;
ahc->needwdtr_orig = 0;
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 22:14:45 +01:00
/* Grab the disconnection disable table and invert it for our needs */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc->flags & AHC_USEDEFAULTS) {
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
printf("%s: Host Adapter Bios disabled. Using default SCSI "
"device parameters\n", ahc_name(ahc));
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 22:14:45 +01:00
ahc->discenable = 0xff;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc->discenable = ~((ahc_inb(ahc, DISC_DSB + 1) << 8)
| ahc_inb(ahc, DISC_DSB));
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 22:14:45 +01:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (!(ahc->type & (AHC_WIDE|AHC_TWIN)))
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-06 00:52:03 +01:00
max_targ = 7;
Major overhaul of the aic7xxx driver: - catch the interrupt type (EDGE/LEVEL) before chip reset instead of guessing the right type. - Add pause variable to the ahc struct to better handle the different interrupt types and pausing the sequencer. - CLRINTSTAT -> CLRSCSIINT: This is a documented bit in the CLRINT register in newer Adaptec documentation, so use their name for it. - Report valid residual byte counts. - Don't mess with the target scratch areas > id 8 on single, narrow, channel devices. The BIOS does a checksum of this area and can flip out if we zero it out. - Initialize the sequencer FLAGS scratch ram variable in the single channel devices to 0. This was the cause of the annoying warning where we would get a cmdcmplt the first time we did any type of transfer negotiation with no valid scb. It also fixes the problem that looked like the INTSTAT register wasn't clearing fast enough. This only showed up on 294x cards, not motherboard aic7870s. - Add the AHC_AIC7870 type and use it as the superset of aic7870 based controllers. - clear the sync offset section of the targ scratch area so that we default to asyncronous transfers. This was only a problem for wide controllers because there was a scenario where the offset wouldn't get updated before a data(out/in) phase would occur. This required some change in the sequencer code since we were depending on this field to hold the rate to negotiate. - allow sync and wide negotiated commands to be tagged (the sequencer now handles this properly).
1995-03-31 14:54:41 +01:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
for (i = 0; i <= max_targ; i++) {
u_int8_t target_settings;
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
if (ahc->flags & AHC_USEDEFAULTS) {
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
target_settings = 0; /* 10MHz/20MHz */
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 22:14:45 +01:00
ahc->needsdtr_orig |= (0x01 << i);
Add $Ids. Use DMA for downloading SCBs to the sequencer. Implement SDTR, WDTR, and message reject handlers so they don't need to exist in the sequencer. All three of these cases are not on the critical path, so it makes little sense to use up precious sequencer ram for them.
1995-01-16 16:33:47 +00:00
ahc->needwdtr_orig |= (0x01 << i);
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
if (ahc->type & AHC_ULTRA)
ultraenable |= (0x01 << i);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else {
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 22:14:45 +01:00
/* Take the settings leftover in scratch RAM. */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
target_settings = ahc_inb(ahc, TARG_SCRATCH + i);
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 22:14:45 +01:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (target_settings & 0x0f) {
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 22:14:45 +01:00
ahc->needsdtr_orig |= (0x01 << i);
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
/*Default to a asynchronous transfers(0 offset)*/
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 22:14:45 +01:00
target_settings &= 0xf0;
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (target_settings & 0x80) {
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 22:14:45 +01:00
ahc->needwdtr_orig |= (0x01 << i);
/*
* We'll set the Wide flag when we
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
* are successful with Wide negotiation.
* Turn it off for now so we aren't
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 22:14:45 +01:00
* confused.
*/
target_settings &= 0x7f;
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc->type & AHC_ULTRA) {
Enable/Disable Ultra mode on a per target basis. This allows the driver to use the full range of settings from 3.6-20MHz on any target. Remove all Ultra settings except for the top three that are documented to work. This fixes some problem reports with the last revision of the driver since at least the 5.7MHz entry doesn't work in Ultra mode.
1996-05-21 19:37:25 +01:00
/*
* Enable Ultra for any target that
* has a valid ultra syncrate setting.
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t rate = target_settings & 0x70;
if (rate == 0x00 || rate == 0x10 ||
rate == 0x20 || rate == 0x40) {
if (rate == 0x40) {
Correct a botched commit from yesturday. It helps to bring over the right patch file.<sigh>
1996-05-23 16:02:18 +01:00
/* Treat 10MHz specially */
target_settings &= ~0x70;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else
Correct a botched commit from yesturday. It helps to bring over the right patch file.<sigh>
1996-05-23 16:02:18 +01:00
ultraenable |= (0x01 << i);
}
Enable/Disable Ultra mode on a per target basis. This allows the driver to use the full range of settings from 3.6-20MHz on any target. Remove all Ultra settings except for the top three that are documented to work. This fixes some problem reports with the last revision of the driver since at least the 5.7MHz entry doesn't work in Ultra mode.
1996-05-21 19:37:25 +01:00
}
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
}
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, TARG_SCRATCH+i,target_settings);
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
}
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
/*
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
* If we are not a WIDE device, forget WDTR. This
* makes the driver work on some cards that don't
* leave these fields cleared when the BIOS is not
* installed.
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if ((ahc->type & AHC_WIDE) == 0)
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
ahc->needwdtr_orig = 0;
Compile cleanly -Wall. Remove some code that was an attempt to make the 294x probe more stable, but since the guess was wrong, go back to the clearer aproach I used the first time around.
1995-03-07 08:59:28 +00:00
ahc->needsdtr = ahc->needsdtr_orig;
Add $Ids. Use DMA for downloading SCBs to the sequencer. Implement SDTR, WDTR, and message reject handlers so they don't need to exist in the sequencer. All three of these cases are not on the critical path, so it makes little sense to use up precious sequencer ram for them.
1995-01-16 16:33:47 +00:00
ahc->needwdtr = ahc->needwdtr_orig;
Keep track of when we actually are awaiting a reply to an SDTR or WDTR message instead of relying on the fact that we are scheduled to send them. The old method worked 99.9% of the time, but someone reported some periferals that did MSG_REJECT at odd times (sometimes before we could send an SDTR or WDTR) that we would construe as the response to an SDTR or WDTR message. This also removes a possible race condition where after a bus reset (the result of a command time out not during intial probe time), we might queue two commands both requesting SDTR, WDTR or both.
1995-02-03 17:15:12 +00:00
ahc->sdtrpending = 0;
ahc->wdtrpending = 0;
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
ahc->tagenable = 0;
Fix the abort code in the BUS DEVICE RESET case. Needed to set cmdlen to 0. Ensure that queued commands are not touched by the abort code by setting the SCB status to indicate what queue it is in. Fix deadlocks when using SCB paging by using SCBs from the assigned_scbs queue or an SCB that completed during the same interrupt if needed. Don't ever use insl to pull SCBs from any of the controllers. You can only do 8bit PIO reads. This only affected SCB paging. With this checkin, SCB paging works quite a bit better, but I still have some problems with it that may be caused by a firmware problem in my PD1800s. It seems that using a tag number higher than the maximum number of tags allowed by the device, confuses it. For example, if I queue two commands, tagged 3 and 36, it never reconnects for tag 36.
1996-04-28 20:21:20 +01:00
ahc->orderedtag = 0;
Increase the DELAY used in ahc_poll. This is only used during intial probe of the device and shouldn't affect normal operation. It seems that some version of the aic7870 are extreamly slow in clearing their interrupt register so the added delay is required to ensure we don't poll a stale interrupt value. Clean up debugging printfs and stuff superflous output into #ifdef AHC_DEBUG. Ensure that sdtrpending and wdtrpending are cleared in all cases after we recieve an SDTR or WDTR message. With this commit, the 294x should work or at least be much closer to working properly.
1995-03-17 23:58:27 +00:00
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, ULTRA_ENB, ultraenable & 0xff);
ahc_outb(ahc, ULTRA_ENB + 1, (ultraenable >> 8) & 0xff);
Enable/Disable Ultra mode on a per target basis. This allows the driver to use the full range of settings from 3.6-20MHz on any target. Remove all Ultra settings except for the top three that are documented to work. This fixes some problem reports with the last revision of the driver since at least the 5.7MHz entry doesn't work in Ultra mode.
1996-05-21 19:37:25 +01:00
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
#ifdef AHC_DEBUG
/* How did we do? */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc_debug & AHC_SHOWMISC)
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
printf("NEEDSDTR == 0x%x\nNEEDWDTR == 0x%x\n"
"DISCENABLE == 0x%x\n", ahc->needsdtr,
ahc->needwdtr, ahc->discenable);
#endif
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
/*
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
* Set the number of available hardware SCBs
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCBCOUNT, ahc->scb_data->maxhscbs);
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
/*
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
* 2's compliment of maximum tag value
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
*/
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
i = ahc->scb_data->maxscbs;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, COMP_SCBCOUNT, -i & 0xff);
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/*
* Allocate enough "hardware scbs" to handle
* the maximum number of concurrent transactions
* we can have active. We have to use contigmalloc
* if this array crosses a page boundary since the
* sequencer depends on this array being physically
* contiguous.
*/
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
if (ahc->scb_data->hscbs == NULL) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
size_t array_size;
u_int32_t hscb_physaddr;
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
array_size = ahc->scb_data->maxscbs*sizeof(struct hardware_scb);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (array_size > PAGE_SIZE) {
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
ahc->scb_data->hscbs = (struct hardware_scb *)
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
contigmalloc(array_size, M_DEVBUF,
M_NOWAIT, 0ul, 0xffffffff,
PAGE_SIZE, 0x10000);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else {
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
ahc->scb_data->hscbs = (struct hardware_scb *)
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
malloc(array_size, M_DEVBUF, M_NOWAIT);
}
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
if (ahc->scb_data->hscbs == NULL) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
printf("%s: unable to allocate hardware SCB array. "
"Failing attach\n");
return (-1);
}
Bzero the kernel scb array after it is allocated otherwise the control byte used on the first transaction on an SCB is indeterminate. Spaces -> tabs.
1996-11-07 06:39:44 +00:00
/* At least the control byte of each hscb needs to be zeroed */
bzero(ahc->scb_data->hscbs, array_size);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/* Tell the sequencer where it can find the hscb array. */
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
hscb_physaddr = vtophys(ahc->scb_data->hscbs);
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, HSCB_ADDR, hscb_physaddr & 0xFF);
ahc_outb(ahc, HSCB_ADDR + 1, (hscb_physaddr >> 8)& 0xFF);
ahc_outb(ahc, HSCB_ADDR + 2, (hscb_physaddr >> 16)& 0xFF);
ahc_outb(ahc, HSCB_ADDR + 3, (hscb_physaddr >> 24)& 0xFF);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
}
/*
* Q-Full-Count. Some cards have more Q space
* then SCBs.
*/
if (ahc->type & AHC_AIC7770) {
ahc->qfullcount = 4;
ahc->qcntmask = 0x07;
} else if (ahc->type & AHC_AIC7850) {
ahc->qfullcount = 8;
ahc->qcntmask = 0x0f;
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
} else if (ahc->scb_data->maxhscbs == 255) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/* 7870/7880 with external SRAM */
ahc->qfullcount = 255;
ahc->qcntmask = 0xff;
} else {
/* 7870/7880 */
ahc->qfullcount = 16;
ahc->qcntmask = 0x1f;
}
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
/*
* QCount mask to deal with broken aic7850s that
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
* sporadically get garbage in the upper bits of
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
* their QCount registers.
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, QCNTMASK, ahc->qcntmask);
Additions to support the WAITING_SCB list that the sequencer maintains. It is the kernel driver's responsibility to do the list manipulation whenever a selection timeout or a request sense occurs. Print out the interrupt type that the device has been set to. It seems that one of the Asus motherboards botches this and David thought a diagnostic would be nice. Fix a bug in my diagnostic code that David found. Reviewed by: Wcarchive and David Greenman
1995-04-27 18:47:17 +01:00
/* We don't have any waiting selections */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, WAITING_SCBH, SCB_LIST_NULL);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
/* Our disconnection list is empty too */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, DISCONNECTED_SCBH, SCB_LIST_NULL);
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
/* Message out buffer starts empty */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, MSG_LEN, 0x00);
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
Compile cleanly -Wall. Remove some code that was an attempt to make the 294x probe more stable, but since the guess was wrong, go back to the clearer aproach I used the first time around.
1995-03-07 08:59:28 +00:00
/*
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
* Load the Sequencer program and Enable the adapter
* in "fast" mode.
Compile cleanly -Wall. Remove some code that was an attempt to make the 294x probe more stable, but since the guess was wrong, go back to the clearer aproach I used the first time around.
1995-03-07 08:59:28 +00:00
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (bootverbose)
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
printf("%s: Downloading Sequencer Program...",
ahc_name(ahc));
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
ahc_loadseq(ahc);
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (bootverbose)
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-06 00:52:03 +01:00
printf("Done\n");
Compile cleanly -Wall. Remove some code that was an attempt to make the 294x probe more stable, but since the guess was wrong, go back to the clearer aproach I used the first time around.
1995-03-07 08:59:28 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
unpause_sequencer(ahc, /*unpause_always*/TRUE);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
/* Notify us when the system can handle timeouts */
timeout(ahc_first_timeout, NULL, 0);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
* Note that we are going and return (to probe)
*/
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
ahc->flags |= AHC_INIT;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
return (0);
}
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static void
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
ahcminphys(bp)
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
struct buf *bp;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
/*
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
* Even though the card can transfer up to 16megs per command
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
* we are limited by the number of segments in the dma segment
* list that we can hold. The worst case is that all pages are
* discontinuous physically, hense the "page per segment" limit
* enforced here.
*/
Enable/Disable Ultra mode on a per target basis. This allows the driver to use the full range of settings from 3.6-20MHz on any target. Remove all Ultra settings except for the top three that are documented to work. This fixes some problem reports with the last revision of the driver since at least the 5.7MHz entry doesn't work in Ultra mode.
1996-05-21 19:37:25 +01:00
if (bp->b_bcount > ((AHC_NSEG - 1) * PAGE_SIZE)) {
bp->b_bcount = ((AHC_NSEG - 1) * PAGE_SIZE);
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#if defined(__NetBSD__)
minphys(bp);
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
/*
* start a scsi operation given the command and
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
* the data address, target, and lun all of which
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
* are stored in the scsi_xfer struct
*/
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
static int32_t
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
ahc_scsi_cmd(xs)
struct scsi_xfer *xs;
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
struct scb *scb;
struct hardware_scb *hscb;
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int16_t mask;
int flags;
int s;
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
ahc = (struct ahc_softc *)xs->sc_link->adapter_softc;
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 21:02:37 +01:00
mask = (0x01 << (xs->sc_link->target
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
| (IS_SCSIBUS_B(ahc, xs->sc_link) ? SELBUSB : 0)));
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 21:02:37 +01:00
SC_DEBUG(xs->sc_link, SDEV_DB2, ("ahc_scsi_cmd\n"));
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
flags = xs->flags;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
if (ahc->in_reset != 0) {
/*
* If we are still in a reset, send the transaction
* back indicating that it was aborted due to a
* reset.
*/
if ((IS_SCSIBUS_B(ahc, xs->sc_link)
&& (ahc->in_reset & CHANNEL_B_RESET) != 0)
|| (!IS_SCSIBUS_B(ahc, xs->sc_link)
&& (ahc->in_reset & CHANNEL_A_RESET) != 0)) {
/* Ick, but I don't want it to abort this */
xs->retries++;
xs->error = XS_BUSY;
return(COMPLETE);
}
}
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 21:02:37 +01:00
/*
* get an scb to use. If the transfer
* is from a buf (possibly from interrupt time)
* then we can't allow it to sleep
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if ((scb = ahc_get_scb(ahc, flags)) == NULL) {
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 21:02:37 +01:00
xs->error = XS_DRIVER_STUFFUP;
return (TRY_AGAIN_LATER);
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
hscb = scb->hscb;
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 21:02:37 +01:00
SC_DEBUG(xs->sc_link, SDEV_DB3, ("start scb(%p)\n", scb));
scb->xs = xs;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 21:02:37 +01:00
/*
* Put all the arguments for the xfer in the scb
*/
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
if (ahc->discenable & mask) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
hscb->control |= DISCENB;
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
if (ahc->tagenable & mask)
Don't rely on AAP(Auto Access Pause) when queueing SCBs to the card. This will increase the overhead of queueing a command, but some recent bug reports make me believe that AAP isn't really working and that we are losing some SCBs from the input queue. Hopefully this will cure that problem. Fix some bugs in the error recovery code. Mainly these could cause us to inadvertantly forget to untimeout an SCB that was recovered causing later confusion.
1997-02-18 04:25:31 +00:00
hscb->control |= MSG_SIMPLE_Q_TAG;
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
if (ahc->orderedtag & mask) {
/* XXX this should be handled by the upper SCSI layer */
printf("Ordered Tag sent\n");
hscb->control |= MSG_ORDERED_Q_TAG;
ahc->orderedtag &= ~mask;
}
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
}
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (flags & SCSI_RESET) {
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
scb->flags |= SCB_DEVICE_RESET;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
hscb->control |= MK_MESSAGE;
} else if ((ahc->needwdtr & mask) && !(ahc->wdtrpending & mask)) {
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
ahc->wdtrpending |= mask;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
hscb->control |= MK_MESSAGE;
scb->flags |= SCB_MSGOUT_WDTR;
} else if((ahc->needsdtr & mask) && !(ahc->sdtrpending & mask)) {
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
ahc->sdtrpending |= mask;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
hscb->control |= MK_MESSAGE;
scb->flags |= SCB_MSGOUT_SDTR;
}
Be even more careful in how we manipulate the QOUTQCNT variable. Now we do reset it from the QOUTCNT register inside a pause/unpause. This now happens once per command complete interrupt in the paging case (one interrupt can be for multiple completed commands). I may introduce a counter and do a lazy update in the future, similar to what is done with the QINCNT. Enhance the QUEUE FULL condition handling so that the number of openings will be reduced. This has become more important now that the driver is faster. This code really belongs in the gerneric SCSI layer, as will be the case once 3.0 gets the code from the 'SCSI' branch. Add some #if 0'd out trace code I've been using to help debug sequencer problems. Fix the SCB paging problem that I was seeing. This was only on my 7850 controller and stems from the fact that its QINFIFO can only handle 3bit SCB identifiers. This means that you can only have 8 transactions open at a time with the current paging scheme to these controllers. The code added to enforce this is generic in that it tests for the number of relevent bits that the QINFIFO can store and adjusts the max accordingly. It may be possible to come up with a scheme that allows for more than 8 commands at a time, but I don't know that it is worth the effort simply to fix a low end card. The aic7880 still can do 255. This problem may be related to what Andrey was seeing since I don't have n aic7770 rev E chip here to test on, but as soon as someone probes one of these cards with this new code, the dmesg output will tell the whole story.
1996-11-16 01:19:14 +00:00
#if 0
/* Set the trace flag if this is the target we want to trace */
if (ahc->unit == 2 && xs->sc_link->target == 3)
hscb->control |= TRACE_SCB;
#endif
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
hscb->tcl = ((xs->sc_link->target << 4) & 0xF0)
| (IS_SCSIBUS_B(ahc,xs->sc_link)? SELBUSB : 0)
| (xs->sc_link->lun & 0x07);
hscb->cmdlen = xs->cmdlen;
hscb->cmdpointer = vtophys(xs->cmd);
Major overhaul of the aic7xxx driver: - catch the interrupt type (EDGE/LEVEL) before chip reset instead of guessing the right type. - Add pause variable to the ahc struct to better handle the different interrupt types and pausing the sequencer. - CLRINTSTAT -> CLRSCSIINT: This is a documented bit in the CLRINT register in newer Adaptec documentation, so use their name for it. - Report valid residual byte counts. - Don't mess with the target scratch areas > id 8 on single, narrow, channel devices. The BIOS does a checksum of this area and can flip out if we zero it out. - Initialize the sequencer FLAGS scratch ram variable in the single channel devices to 0. This was the cause of the annoying warning where we would get a cmdcmplt the first time we did any type of transfer negotiation with no valid scb. It also fixes the problem that looked like the INTSTAT register wasn't clearing fast enough. This only showed up on 294x cards, not motherboard aic7870s. - Add the AHC_AIC7870 type and use it as the superset of aic7870 based controllers. - clear the sync offset section of the targ scratch area so that we default to asyncronous transfers. This was only a problem for wide controllers because there was a scenario where the offset wouldn't get updated before a data(out/in) phase would occur. This required some change in the sequencer code since we were depending on this field to hold the rate to negotiate. - allow sync and wide negotiated commands to be tagged (the sequencer now handles this properly).
1995-03-31 14:54:41 +01:00
xs->resid = 0;
Merge RELENG_2_0_5 into HEAD
1995-06-11 20:33:05 +01:00
xs->status = 0;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/* Only use S/G if non-zero length */
if (xs->datalen) {
int seg;
u_int32_t datalen;
vm_offset_t vaddr;
u_int32_t paddr;
u_int32_t nextpaddr;
struct ahc_dma_seg *sg;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
seg = 0;
datalen = xs->datalen;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
vaddr = (vm_offset_t)xs->data;
paddr = vtophys(vaddr);
sg = scb->ahc_dma;
hscb->SG_list_pointer = vtophys(sg);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
while ((datalen > 0) && (seg < AHC_NSEG)) {
/* put in the base address and length */
sg->addr = paddr;
sg->len = 0;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
/* do it at least once */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
nextpaddr = paddr;
while ((datalen > 0) && (paddr == nextpaddr)) {
u_int32_t size;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
/*
* This page is contiguous (physically)
* with the the last, just extend the
* length
*/
/* how far to the end of the page */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
nextpaddr = (paddr & (~PAGE_MASK)) + PAGE_SIZE;
/*
* Compute the maximum size
*/
size = nextpaddr - paddr;
if (size > datalen)
size = datalen;
sg->len += size;
vaddr += size;
datalen -= size;
if (datalen > 0)
paddr = vtophys(vaddr);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
}
/*
* next page isn't contiguous, finish the seg
*/
seg++;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
sg++;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
hscb->SG_segment_count = seg;
Fix a bug in the reporting of residuals. The code was relying on the SG_COUNT filed in the hardware SCB not changing during the course of a transaction. Since the sequencer now DMAs the hardware SCB back up to the host when it detects a residual, this is no longer the case. I added a field to the "software" scb to mirror this information and it is now used for doing the residual calculation.
1997-02-11 17:10:37 +00:00
scb->sg_count = hscb->SG_segment_count;
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 22:14:45 +01:00
/* Copy the first SG into the data pointer area */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
hscb->data = scb->ahc_dma->addr;
hscb->datalen = scb->ahc_dma->len | (SCB_LIST_NULL << 24);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
if (datalen) {
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 22:14:45 +01:00
/* there's still data, must have run out of segs! */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
printf("%s: ahc_scsi_cmd: more than %d DMA segs\n",
ahc_name(ahc), AHC_NSEG);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
xs->error = XS_DRIVER_STUFFUP;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
ahc_free_scb(ahc, scb);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
return (COMPLETE);
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
#ifdef AHC_BROKEN_CACHE
if (ahc_broken_cache)
INVALIDATE_CACHE();
#endif
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else {
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 08:06:02 +01:00
/*
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
* No data xfer, use non S/G values
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 08:06:02 +01:00
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
hscb->SG_segment_count = 0;
Fix a bug in the reporting of residuals. The code was relying on the SG_COUNT filed in the hardware SCB not changing during the course of a transaction. Since the sequencer now DMAs the hardware SCB back up to the host when it detects a residual, this is no longer the case. I added a field to the "software" scb to mirror this information and it is now used for doing the residual calculation.
1997-02-11 17:10:37 +00:00
scb->sg_count = hscb->SG_segment_count;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
hscb->SG_list_pointer = 0;
hscb->data = 0;
hscb->datalen = (SCB_LIST_NULL << 24);
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 08:06:02 +01:00
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
#ifdef AHC_DEBUG
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if((ahc_debug & AHC_SHOWSCBS) && (xs->sc_link->target == DEBUGTARGET))
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
ahc_print_scb(scb);
#endif
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
s = splbio();
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
STAILQ_INSERT_TAIL(&ahc->waiting_scbs, scb, links);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
scb->flags |= SCB_ACTIVE|SCB_WAITINGQ;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
ahc_run_waiting_queue(ahc);
if ((flags & SCSI_NOMASK) == 0) {
timeout(ahc_timeout, (caddr_t)scb, (xs->timeout * hz) / 1000);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
splx(s);
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
return (SUCCESSFULLY_QUEUED);
}
/*
* If we can't use interrupts, poll for completion
*/
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
SC_DEBUG(xs->sc_link, SDEV_DB3, ("cmd_poll\n"));
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
do {
if (ahc_poll(ahc, xs->timeout)) {
if (!(xs->flags & SCSI_SILENT))
printf("cmd fail\n");
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
ahc_timeout(scb);
break;
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} while ((xs->flags & ITSDONE) == 0); /* a non command complete intr */
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
scsi_done(xs);
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
splx(s);
return (COMPLETE);
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/*
* Look for space in the QINFIFO and queue as many SCBs in the waiting
* queue as possible. Assumes that it is called at splbio().
*/
static void
ahc_run_waiting_queue(ahc)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
{
struct scb *scb;
Fix a nasty bug that meant a QUEUE_FULL status would result in a lost SCB. This is probably a main reason for the recent reports of timeouts.
1997-03-24 05:05:18 +00:00
/*
* On aic78X0 chips, we rely on Auto Access Pause (AAP)
* instead of doing an explicit pause/unpause.
*/
if ((ahc->type & AHC_AIC78X0) == 0)
pause_sequencer(ahc);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
while ((scb = ahc->waiting_scbs.stqh_first) != NULL) {
if (ahc->curqincnt >= ahc->qfullcount) {
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc->curqincnt = ahc_inb(ahc, QINCNT) & ahc->qcntmask;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (ahc->curqincnt >= ahc->qfullcount)
/* Still no space */
break;
}
STAILQ_REMOVE_HEAD(&ahc->waiting_scbs, links);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
scb->flags &= ~SCB_WAITINGQ;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, QINFIFO, scb->hscb->tag);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if ((ahc->flags & AHC_PAGESCBS) != 0)
/*
* We only care about this statistic when paging
* since it is impossible to overflow the qinfifo
* in the non-paging case.
*/
ahc->curqincnt++;
}
Fix a nasty bug that meant a QUEUE_FULL status would result in a lost SCB. This is probably a main reason for the recent reports of timeouts.
1997-03-24 05:05:18 +00:00
if ((ahc->type & AHC_AIC78X0) == 0)
unpause_sequencer(ahc, /*Unpause always*/FALSE);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
* An scb (and hence an scb entry on the board) is put onto the
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
* free list.
*/
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static void
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
ahc_free_scb(ahc, scb)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
struct scb *scb;
{
struct hardware_scb *hscb;
int opri;
hscb = scb->hscb;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Don't arbitrarily set SCSI_NOSLEEP. It is now handled correctly by the higher level scsi code. Spls should never be conditionalized, so don't do so here. Restructure the get_scb routine so that we can't get into an infinite loop if the ccbs are exhausted and we are are called with SCSI_NOSLEEP set. Other driver maintainer's that based their scb allocation routines on Julian's code should look at these changes and implement them for their driver. The aic7xxx driver inspired these changes because early revs of the aic7770 chips have so few SCBs that you can actually run out. If you have a rev C or aic7770 (as is reported by the driver probe) and had more than 2 drives, you could get into an infinite loop when using up all of the SCBs. Since the driver will only allow two SCBs per device and I only had two devices, I never saw this problem on my Rev C card. Bzero only 19 bytes of the scb instead of 2k (ack!). This was a hold over from when a struct SCB only contained the information downloaded to the board, but we now store kernel driver data in there as well. This greatly lowers the overhead for small transactions (I get ~1MB/sec for dds with a 512 byte block size). Submitted by: John Dyson with the aic7xxx specific optimization by me
1995-04-23 23:04:58 +01:00
opri = splbio();
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 21:02:37 +01:00
/* Clean up for the next user */
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
scb->flags = SCB_FREE;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
hscb->control = 0;
hscb->status = 0;
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 21:02:37 +01:00
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
STAILQ_INSERT_HEAD(&ahc->scb_data->free_scbs, scb, links);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (scb->links.stqe_next == NULL) {
/*
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
* If there were no SCBs available, wake anybody waiting
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
* for one to come free.
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
*/
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
wakeup((caddr_t)&ahc->scb_data->free_scbs);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
}
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 21:02:37 +01:00
#ifdef AHC_DEBUG
ahc->activescbs--;
#endif
Don't arbitrarily set SCSI_NOSLEEP. It is now handled correctly by the higher level scsi code. Spls should never be conditionalized, so don't do so here. Restructure the get_scb routine so that we can't get into an infinite loop if the ccbs are exhausted and we are are called with SCSI_NOSLEEP set. Other driver maintainer's that based their scb allocation routines on Julian's code should look at these changes and implement them for their driver. The aic7xxx driver inspired these changes because early revs of the aic7770 chips have so few SCBs that you can actually run out. If you have a rev C or aic7770 (as is reported by the driver probe) and had more than 2 drives, you could get into an infinite loop when using up all of the SCBs. Since the driver will only allow two SCBs per device and I only had two devices, I never saw this problem on my Rev C card. Bzero only 19 bytes of the scb instead of 2k (ack!). This was a hold over from when a struct SCB only contained the information downloaded to the board, but we now store kernel driver data in there as well. This greatly lowers the overhead for small transactions (I get ~1MB/sec for dds with a 512 byte block size). Submitted by: John Dyson with the aic7xxx specific optimization by me
1995-04-23 23:04:58 +01:00
splx(opri);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
* Get a free scb, either one already assigned to a hardware slot
* on the adapter or one that will require an SCB to be paged out before
* use. If there are none, see if we can allocate a new SCB. Otherwise
* either return an error or sleep.
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*/
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static struct scb *
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
ahc_get_scb(ahc, flags)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int32_t flags;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
struct scb *scbp;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
int opri;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Don't arbitrarily set SCSI_NOSLEEP. It is now handled correctly by the higher level scsi code. Spls should never be conditionalized, so don't do so here. Restructure the get_scb routine so that we can't get into an infinite loop if the ccbs are exhausted and we are are called with SCSI_NOSLEEP set. Other driver maintainer's that based their scb allocation routines on Julian's code should look at these changes and implement them for their driver. The aic7xxx driver inspired these changes because early revs of the aic7770 chips have so few SCBs that you can actually run out. If you have a rev C or aic7770 (as is reported by the driver probe) and had more than 2 drives, you could get into an infinite loop when using up all of the SCBs. Since the driver will only allow two SCBs per device and I only had two devices, I never saw this problem on my Rev C card. Bzero only 19 bytes of the scb instead of 2k (ack!). This was a hold over from when a struct SCB only contained the information downloaded to the board, but we now store kernel driver data in there as well. This greatly lowers the overhead for small transactions (I get ~1MB/sec for dds with a 512 byte block size). Submitted by: John Dyson with the aic7xxx specific optimization by me
1995-04-23 23:04:58 +01:00
opri = splbio();
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
/*
* If we can and have to, sleep waiting for one to come free
* but only if we can't allocate a new one.
*/
while (1) {
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
if ((scbp = ahc->scb_data->free_scbs.stqh_first)) {
STAILQ_REMOVE_HEAD(&ahc->scb_data->free_scbs, links);
} else if(ahc->scb_data->numscbs < ahc->scb_data->maxscbs) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
scbp = ahc_alloc_scb(ahc);
if (scbp == NULL)
printf("%s: Can't malloc SCB\n", ahc_name(ahc));
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
} else if ((flags & SCSI_NOSLEEP) == 0) {
tsleep((caddr_t)&ahc->scb_data->free_scbs, PRIBIO,
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
"ahcscb", 0);
continue;
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
}
break;
}
Don't arbitrarily set SCSI_NOSLEEP. It is now handled correctly by the higher level scsi code. Spls should never be conditionalized, so don't do so here. Restructure the get_scb routine so that we can't get into an infinite loop if the ccbs are exhausted and we are are called with SCSI_NOSLEEP set. Other driver maintainer's that based their scb allocation routines on Julian's code should look at these changes and implement them for their driver. The aic7xxx driver inspired these changes because early revs of the aic7770 chips have so few SCBs that you can actually run out. If you have a rev C or aic7770 (as is reported by the driver probe) and had more than 2 drives, you could get into an infinite loop when using up all of the SCBs. Since the driver will only allow two SCBs per device and I only had two devices, I never saw this problem on my Rev C card. Bzero only 19 bytes of the scb instead of 2k (ack!). This was a hold over from when a struct SCB only contained the information downloaded to the board, but we now store kernel driver data in there as well. This greatly lowers the overhead for small transactions (I get ~1MB/sec for dds with a 512 byte block size). Submitted by: John Dyson with the aic7xxx specific optimization by me
1995-04-23 23:04:58 +01:00
#ifdef AHC_DEBUG
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 21:02:37 +01:00
if (scbp) {
Don't arbitrarily set SCSI_NOSLEEP. It is now handled correctly by the higher level scsi code. Spls should never be conditionalized, so don't do so here. Restructure the get_scb routine so that we can't get into an infinite loop if the ccbs are exhausted and we are are called with SCSI_NOSLEEP set. Other driver maintainer's that based their scb allocation routines on Julian's code should look at these changes and implement them for their driver. The aic7xxx driver inspired these changes because early revs of the aic7770 chips have so few SCBs that you can actually run out. If you have a rev C or aic7770 (as is reported by the driver probe) and had more than 2 drives, you could get into an infinite loop when using up all of the SCBs. Since the driver will only allow two SCBs per device and I only had two devices, I never saw this problem on my Rev C card. Bzero only 19 bytes of the scb instead of 2k (ack!). This was a hold over from when a struct SCB only contained the information downloaded to the board, but we now store kernel driver data in there as well. This greatly lowers the overhead for small transactions (I get ~1MB/sec for dds with a 512 byte block size). Submitted by: John Dyson with the aic7xxx specific optimization by me
1995-04-23 23:04:58 +01:00
ahc->activescbs++;
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
if((ahc_debug & AHC_SHOWSCBCNT)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
&& (ahc->activescbs == ahc->scb_data->maxhscbs))
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
printf("%s: Max SCBs active\n", ahc_name(ahc));
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
}
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 21:02:37 +01:00
#endif
Don't arbitrarily set SCSI_NOSLEEP. It is now handled correctly by the higher level scsi code. Spls should never be conditionalized, so don't do so here. Restructure the get_scb routine so that we can't get into an infinite loop if the ccbs are exhausted and we are are called with SCSI_NOSLEEP set. Other driver maintainer's that based their scb allocation routines on Julian's code should look at these changes and implement them for their driver. The aic7xxx driver inspired these changes because early revs of the aic7770 chips have so few SCBs that you can actually run out. If you have a rev C or aic7770 (as is reported by the driver probe) and had more than 2 drives, you could get into an infinite loop when using up all of the SCBs. Since the driver will only allow two SCBs per device and I only had two devices, I never saw this problem on my Rev C card. Bzero only 19 bytes of the scb instead of 2k (ack!). This was a hold over from when a struct SCB only contained the information downloaded to the board, but we now store kernel driver data in there as well. This greatly lowers the overhead for small transactions (I get ~1MB/sec for dds with a 512 byte block size). Submitted by: John Dyson with the aic7xxx specific optimization by me
1995-04-23 23:04:58 +01:00
splx(opri);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 04:15:31 +01:00
return (scbp);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
static struct scb *
ahc_alloc_scb(ahc)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
{
static struct ahc_dma_seg *next_sg_array = NULL;
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
static int sg_arrays_free = 0;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
struct scb *newscb;
newscb = (struct scb *) malloc(sizeof(struct scb), M_DEVBUF, M_NOWAIT);
if (newscb != NULL) {
bzero(newscb, sizeof(struct scb));
if (next_sg_array == NULL) {
size_t alloc_size = sizeof(struct ahc_dma_seg)
* AHC_NSEG;
sg_arrays_free = PAGE_SIZE / alloc_size;
alloc_size *= sg_arrays_free;
if (alloc_size == 0)
panic("%s: SG list doesn't fit in a page",
ahc_name(ahc));
next_sg_array = (struct ahc_dma_seg *)
malloc(alloc_size, M_DEVBUF, M_NOWAIT);
}
if (next_sg_array != NULL) {
struct hardware_scb *hscb;
newscb->ahc_dma = next_sg_array;
sg_arrays_free--;
if (sg_arrays_free == 0)
next_sg_array = NULL;
else
next_sg_array = &next_sg_array[AHC_NSEG];
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
hscb = &ahc->scb_data->hscbs[ahc->scb_data->numscbs];
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
newscb->hscb = hscb;
hscb->control = 0;
hscb->status = 0;
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
hscb->tag = ahc->scb_data->numscbs;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
hscb->residual_data_count[2] = 0;
hscb->residual_data_count[1] = 0;
hscb->residual_data_count[0] = 0;
hscb->residual_SG_segment_count = 0;
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
ahc->scb_data->numscbs++;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/*
* Place in the scbarray
* Never is removed.
*/
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
ahc->scb_data->scbarray[hscb->tag] = newscb;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else {
free(newscb, M_DEVBUF);
newscb = NULL;
}
}
return newscb;
}
Remove some unnecessary overhead in the command complete processing. It should be nearly impossible to overflow the QOUTFIFO (worst case 9 command have to complete with at least 6 of them requiring paging on an aic7850), so don't take the additional PIO hit to guard against this condition. If we don't see our interrupt in time, the system has bigger problems elsewhere. If this ever does happen, the timeout handler will notice and retry the command. Remove the ABORT_TAG sequencer interrupt handler. This condition can't happen with the new SCB paging scheme. Fix a few bugs noticed by Dan Eischen <deischen@iworks.InterWorks.org> that could prevent ULTRA from being negotiated to drives above ID 7 and also could allow an SCB to be passed to ahc_done twice during error recovery. Fix a bug notice by Rory Bolt <rory@atBackup.com>. It turns out that a sequencer reset will actually start the sequencer running regardless of the state of the pause bit. This could lead to strange problems with loading the sequencer.
1997-01-22 18:05:31 +00:00
static void
ahc_loadseq(ahc)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
int options;
struct patch *cur_patch;
int i;
int downloaded;
options = 1; /* Code for all options */
downloaded = 0;
if ((ahc->type & AHC_ULTRA) != 0)
options |= ULTRA;
if ((ahc->type & AHC_TWIN) != 0)
options |= TWIN_CHANNEL;
if (ahc->scb_data->maxscbs > ahc->scb_data->maxhscbs)
options |= SCB_PAGING;
cur_patch = patches;
Remove some unnecessary overhead in the command complete processing. It should be nearly impossible to overflow the QOUTFIFO (worst case 9 command have to complete with at least 6 of them requiring paging on an aic7850), so don't take the additional PIO hit to guard against this condition. If we don't see our interrupt in time, the system has bigger problems elsewhere. If this ever does happen, the timeout handler will notice and retry the command. Remove the ABORT_TAG sequencer interrupt handler. This condition can't happen with the new SCB paging scheme. Fix a few bugs noticed by Dan Eischen <deischen@iworks.InterWorks.org> that could prevent ULTRA from being negotiated to drives above ID 7 and also could allow an SCB to be passed to ahc_done twice during error recovery. Fix a bug notice by Rory Bolt <rory@atBackup.com>. It turns out that a sequencer reset will actually start the sequencer running regardless of the state of the pause bit. This could lead to strange problems with loading the sequencer.
1997-01-22 18:05:31 +00:00
ahc_outb(ahc, SEQCTL, PERRORDIS|LOADRAM);
ahc_outb(ahc, SEQADDR0, 0);
ahc_outb(ahc, SEQADDR1, 0);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
for(i = 0; i < sizeof(seqprog)/4; i++) {
cur_patch = ahc_next_patch(cur_patch, options, i);
if (cur_patch && cur_patch->begin <= i && cur_patch->end > i)
/* Skip this instruction for this configuration */
continue;
ahc_download_instr(ahc, options, i);
downloaded++;
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Remove some unnecessary overhead in the command complete processing. It should be nearly impossible to overflow the QOUTFIFO (worst case 9 command have to complete with at least 6 of them requiring paging on an aic7850), so don't take the additional PIO hit to guard against this condition. If we don't see our interrupt in time, the system has bigger problems elsewhere. If this ever does happen, the timeout handler will notice and retry the command. Remove the ABORT_TAG sequencer interrupt handler. This condition can't happen with the new SCB paging scheme. Fix a few bugs noticed by Dan Eischen <deischen@iworks.InterWorks.org> that could prevent ULTRA from being negotiated to drives above ID 7 and also could allow an SCB to be passed to ahc_done twice during error recovery. Fix a bug notice by Rory Bolt <rory@atBackup.com>. It turns out that a sequencer reset will actually start the sequencer running regardless of the state of the pause bit. This could lead to strange problems with loading the sequencer.
1997-01-22 18:05:31 +00:00
ahc_outb(ahc, SEQCTL, FASTMODE);
ahc_outb(ahc, SEQADDR0, 0);
ahc_outb(ahc, SEQADDR1, 0);
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
if (bootverbose)
printf("%s: %d instructions downloaded\n", ahc_name(ahc),
downloaded);
}
static struct patch *
ahc_next_patch(cur_patch, options, instrptr)
struct patch *cur_patch;
int options;
int instrptr;
{
while(cur_patch != NULL) {
if (((cur_patch->options & options) != 0
&& cur_patch->negative == FALSE)
|| ((cur_patch->options & options) == 0
&& cur_patch->negative == TRUE)
|| (instrptr >= cur_patch->end)) {
/*
* Either we want to keep this section of code,
* or we have consumed this patch. Skip to the
* next patch.
*/
cur_patch++;
if (cur_patch->options == 0)
/* Out of patches */
cur_patch = NULL;
} else
/* Found an okay patch */
break;
}
return (cur_patch);
}
static void
ahc_download_instr(ahc, options, instrptr)
struct ahc_softc *ahc;
int options;
int instrptr;
{
u_int8_t opcode;
struct ins_format3 *instr;
instr = (struct ins_format3 *)&seqprog[instrptr * 4];
/* Pull the opcode */
opcode = instr->opcode_addr >> 1;
switch (opcode) {
case AIC_OP_JMP:
case AIC_OP_JC:
case AIC_OP_JNC:
case AIC_OP_CALL:
case AIC_OP_JNE:
case AIC_OP_JNZ:
case AIC_OP_JE:
case AIC_OP_JZ:
{
int address_offset;
struct ins_format3 new_instr;
u_int address;
struct patch *patch;
int i;
address_offset = 0;
new_instr = *instr; /* structure copy */
address = new_instr.address;
address |= (new_instr.opcode_addr & ADDR_HIGH_BIT) << 8;
for (i = 0; i < sizeof(patches)/sizeof(patches[0]); i++) {
patch = &patches[i];
if (((patch->options & options) == 0
&& patch->negative == FALSE)
|| ((patch->options & options) != 0
&& patch->negative == TRUE)) {
if (address >= patch->end)
address_offset +=
patch->end - patch->begin;
}
}
address -= address_offset;
new_instr.address = address & 0xFF;
new_instr.opcode_addr &= ~ADDR_HIGH_BIT;
new_instr.opcode_addr |= (address >> 8) & ADDR_HIGH_BIT;
ahc_outsb(ahc, SEQRAM, &new_instr.immediate, 4);
break;
}
case AIC_OP_OR:
case AIC_OP_AND:
case AIC_OP_XOR:
case AIC_OP_ADD:
case AIC_OP_ADC:
case AIC_OP_ROL:
ahc_outsb(ahc, SEQRAM, &instr->immediate, 4);
break;
default:
panic("Unknown opcode encountered in seq program");
break;
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
/*
* Function to poll for command completion when
* interrupts are disabled (crash dumps)
*/
static int
ahc_poll(ahc, wait)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
int wait; /* in msec */
{
while (--wait) {
DELAY(1000);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
if (ahc_inb(ahc, INTSTAT) & INT_PEND)
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
break;
} if (wait == 0) {
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
printf("%s: board is not responding\n", ahc_name(ahc));
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
return (EIO);
}
ahc_intr((void *)ahc);
return (0);
}
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
/*
* Handler to register that the system is capable of
* delivering timeouts. We register this timeout in
* ahc_init and it should go off as soon as we're through
* the boot process.
*/
static void
ahc_first_timeout(arg)
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
void *arg;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
{
timeouts_work = 1;
}
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static void
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
ahc_timeout(arg)
void *arg;
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
{
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
struct scb *scb = (struct scb *)arg;
struct ahc_softc *ahc;
int s, found;
u_int8_t bus_state;
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
char channel;
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
ahc = (struct ahc_softc *)scb->xs->sc_link->adapter_softc;
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
s = splbio();
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
/*
* Ensure that the card doesn't do anything
* behind our back. Also make sure that we
* didn't "just" miss an interrupt that would
* affect this timeout.
*/
do {
ahc_intr(ahc);
pause_sequencer(ahc);
} while (ahc_inb(ahc, INTSTAT) & INT_PEND);
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
if (!(scb->flags & SCB_ACTIVE)) {
/* Previous timeout took care of me already */
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
printf("Timedout SCB handled by another timeout\n");
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
unpause_sequencer(ahc, /*unpause_always*/TRUE);
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
splx(s);
return;
}
if (ahc->in_timeout) {
/*
* Some other SCB has started a recovery operation
* and is still working on cleaning things up.
*/
When not using SCB paging, we can always directly index the SCB of interest either by looking it up in the array of pending, per target, untagged transactions, or by using the tag value passed in during the identify. The old code only direct indexed for tagged transactions. This makes the "findSCB" routine only necessary when SCB paging is enabled, so appropriately conditionalize it. This greatly simplifies the non SCB paging code flow.
1997-04-04 05:09:29 +01:00
if ((scb->flags & SCB_RECOVERY_SCB) == 0) {
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
/*
* This is not the SCB that started this timeout
* processing. Give this scb another lifetime so
* that it can continue once we deal with the
* timeout.
*/
scb->flags |= SCB_TIMEDOUT;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
sc_print_addr(scb->xs->sc_link);
printf("SCB 0x%x timedout while recovery in progress\n",
scb->hscb->tag);
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
timeout(ahc_timeout, (caddr_t)scb,
(scb->xs->timeout * hz) / 1000);
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
unpause_sequencer(ahc, /*unpause_always*/TRUE);
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
splx(s);
return;
}
}
ahc->in_timeout = TRUE;
sc_print_addr(scb->xs->sc_link);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
printf("SCB 0x%x - timed out ", scb->hscb->tag);
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
/*
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
* Take a snapshot of the bus state and print out
* some information so we can track down driver bugs.
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
bus_state = ahc_inb(ahc, LASTPHASE);
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
Be even more careful in how we manipulate the QOUTQCNT variable. Now we do reset it from the QOUTCNT register inside a pause/unpause. This now happens once per command complete interrupt in the paging case (one interrupt can be for multiple completed commands). I may introduce a counter and do a lazy update in the future, similar to what is done with the QINCNT. Enhance the QUEUE FULL condition handling so that the number of openings will be reduced. This has become more important now that the driver is faster. This code really belongs in the gerneric SCSI layer, as will be the case once 3.0 gets the code from the 'SCSI' branch. Add some #if 0'd out trace code I've been using to help debug sequencer problems. Fix the SCB paging problem that I was seeing. This was only on my 7850 controller and stems from the fact that its QINFIFO can only handle 3bit SCB identifiers. This means that you can only have 8 transactions open at a time with the current paging scheme to these controllers. The code added to enforce this is generic in that it tests for the number of relevent bits that the QINFIFO can store and adjusts the max accordingly. It may be possible to come up with a scheme that allows for more than 8 commands at a time, but I don't know that it is worth the effort simply to fix a low end card. The aic7880 still can do 255. This problem may be related to what Andrey was seeing since I don't have n aic7770 rev E chip here to test on, but as soon as someone probes one of these cards with this new code, the dmesg output will tell the whole story.
1996-11-16 01:19:14 +00:00
switch(bus_state)
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
case P_DATAOUT:
printf("in dataout phase");
break;
case P_DATAIN:
printf("in datain phase");
break;
case P_COMMAND:
printf("in command phase");
break;
case P_MESGOUT:
printf("in message out phase");
break;
case P_STATUS:
printf("in status phase");
break;
case P_MESGIN:
printf("in message in phase");
break;
Clear the busfree interrupt when one occurs, after a SELTO, or a bus reset.
1996-12-03 17:06:00 +00:00
case P_BUSFREE:
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
printf("while idle, LASTPHASE == 0x%x",
bus_state);
Clear the busfree interrupt when one occurs, after a SELTO, or a bus reset.
1996-12-03 17:06:00 +00:00
break;
default:
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/*
* We aren't in a valid phase, so assume we're
* idle.
*/
Clear the busfree interrupt when one occurs, after a SELTO, or a bus reset.
1996-12-03 17:06:00 +00:00
printf("invalid phase, LASTPHASE == 0x%x",
bus_state);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
break;
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
}
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
printf(", SCSISIGI == 0x%x\n", ahc_inb(ahc, SCSISIGI));
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
printf("SEQADDR == 0x%x\n", ahc_inb(ahc, SEQADDR0)
| (ahc_inb(ahc, SEQADDR1) << 8));
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
/* Decide our course of action */
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
channel = (scb->hscb->tcl & SELBUSB) ? 'B': 'A';
Fix some more problems in the recovery code. Cleanup of the disconnected list was broken in the SCB paging case (confusion of NULLand SCB_LIST_NULL) Implement a clean mechanism for determining that we have exited the timeout state and test for this in ahc_done instead of all over the place. Bring back the use of AAP (Auto Access Pause) I don't think it was the true cause of the bus hangs people were reporting. We want to reset the bus if we've been through an Abort action, not if we are a recovery SCB (one implies the other, but not vice-versa).
1997-02-19 01:00:51 +00:00
if (scb->flags & SCB_ABORT) {
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
/*
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
* Been down this road before.
* Do a full bus reset.
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
*/
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
bus_reset:
scb->flags |= SCB_RECOVERY_SCB;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
found = ahc_reset_channel(ahc, channel, XS_TIMEOUT,
/*Initiate Reset*/TRUE);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
printf("%s: Issued Channel %c Bus Reset. "
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
"%d SCBs aborted\n", ahc_name(ahc), channel, found);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else if ((scb->hscb->control & TAG_ENB) != 0
&& (scb->flags & SCB_SENTORDEREDTAG) == 0) {
Fix the abort code in the BUS DEVICE RESET case. Needed to set cmdlen to 0. Ensure that queued commands are not touched by the abort code by setting the SCB status to indicate what queue it is in. Fix deadlocks when using SCB paging by using SCBs from the assigned_scbs queue or an SCB that completed during the same interrupt if needed. Don't ever use insl to pull SCBs from any of the controllers. You can only do 8bit PIO reads. This only affected SCB paging. With this checkin, SCB paging works quite a bit better, but I still have some problems with it that may be caused by a firmware problem in my PD1800s. It seems that using a tag number higher than the maximum number of tags allowed by the device, confuses it. For example, if I queue two commands, tagged 3 and 36, it never reconnects for tag 36.
1996-04-28 20:21:20 +01:00
/*
* We could be starving this command
* try sending an ordered tag command
* to the target we come from.
*/
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
u_int16_t mask;
mask = (0x01 << (scb->xs->sc_link->target
| (IS_SCSIBUS_B(ahc, scb->xs->sc_link) ?
SELBUSB : 0)));
Don't rely on AAP(Auto Access Pause) when queueing SCBs to the card. This will increase the overhead of queueing a command, but some recent bug reports make me believe that AAP isn't really working and that we are losing some SCBs from the input queue. Hopefully this will cure that problem. Fix some bugs in the error recovery code. Mainly these could cause us to inadvertantly forget to untimeout an SCB that was recovered causing later confusion.
1997-02-18 04:25:31 +00:00
scb->flags |= SCB_SENTORDEREDTAG|SCB_RECOVERY_SCB;
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
ahc->orderedtag |= mask;
When not using SCB paging, we can always directly index the SCB of interest either by looking it up in the array of pending, per target, untagged transactions, or by using the tag value passed in during the identify. The old code only direct indexed for tagged transactions. This makes the "findSCB" routine only necessary when SCB paging is enabled, so appropriately conditionalize it. This greatly simplifies the non SCB paging code flow.
1997-04-04 05:09:29 +01:00
timeout(ahc_timeout, (caddr_t)scb, (5 * hz));
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
unpause_sequencer(ahc, /*unpause_always*/TRUE);
Fix the abort code in the BUS DEVICE RESET case. Needed to set cmdlen to 0. Ensure that queued commands are not touched by the abort code by setting the SCB status to indicate what queue it is in. Fix deadlocks when using SCB paging by using SCBs from the assigned_scbs queue or an SCB that completed during the same interrupt if needed. Don't ever use insl to pull SCBs from any of the controllers. You can only do 8bit PIO reads. This only affected SCB paging. With this checkin, SCB paging works quite a bit better, but I still have some problems with it that may be caused by a firmware problem in my PD1800s. It seems that using a tag number higher than the maximum number of tags allowed by the device, confuses it. For example, if I queue two commands, tagged 3 and 36, it never reconnects for tag 36.
1996-04-28 20:21:20 +01:00
printf("Ordered Tag queued\n");
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else {
Remove trailing whitespace.
1995-05-30 09:16:23 +01:00
/*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
* Send an Abort Message:
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
* The target that is holding up the bus may not
* be the same as the one that triggered this timeout
* (different commands have different timeout lengths).
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
* Our strategy here is to queue an abort message
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
* to the timed out target if it is disconnected.
* Otherwise, if we have an active target we stuff the
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
* message buffer with an abort message and assert ATN
* in the hopes that the target will let go of the bus
* and go to the mesgout phase. If this fails, we'll
* get another timeout 2 seconds later which will attempt
* a bus reset.
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t saved_scbptr;
u_int8_t active_scb_index;
struct scb *active_scb;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
saved_scbptr = ahc_inb(ahc, SCBPTR);
active_scb_index = ahc_inb(ahc, SCB_TAG);
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
active_scb = ahc->scb_data->scbarray[active_scb_index];
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
if (bus_state != P_BUSFREE) {
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
if (active_scb_index >= ahc->scb_data->numscbs) {
/* Go "immediatly" to the bus reset */
sc_print_addr(scb->xs->sc_link);
printf("SCB %d: Yucky Immediate reset. "
"Flags = 0x%x\n", scb->hscb->tag,
scb->flags);
goto bus_reset;
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/* Send the abort to the active SCB */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, MSG_LEN, 1);
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, MSG_OUT,
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
(active_scb->hscb->control & TAG_ENB) == 0 ?
MSG_ABORT : MSG_ABORT_TAG);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCSISIGO, bus_state|ATNO);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
sc_print_addr(active_scb->xs->sc_link);
printf("abort message in message buffer\n");
Don't rely on AAP(Auto Access Pause) when queueing SCBs to the card. This will increase the overhead of queueing a command, but some recent bug reports make me believe that AAP isn't really working and that we are losing some SCBs from the input queue. Hopefully this will cure that problem. Fix some bugs in the error recovery code. Mainly these could cause us to inadvertantly forget to untimeout an SCB that was recovered causing later confusion.
1997-02-18 04:25:31 +00:00
active_scb->flags |= SCB_ABORT|SCB_RECOVERY_SCB;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (active_scb != scb) {
untimeout(ahc_timeout,
(caddr_t)active_scb);
/* Give scb a new lease on life */
timeout(ahc_timeout, (caddr_t)scb,
(scb->xs->timeout * hz) / 1000);
}
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
timeout(ahc_timeout, (caddr_t)active_scb,
Don't rely on AAP(Auto Access Pause) when queueing SCBs to the card. This will increase the overhead of queueing a command, but some recent bug reports make me believe that AAP isn't really working and that we are losing some SCBs from the input queue. Hopefully this will cure that problem. Fix some bugs in the error recovery code. Mainly these could cause us to inadvertantly forget to untimeout an SCB that was recovered causing later confusion.
1997-02-18 04:25:31 +00:00
(200 * hz) / 1000);
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
unpause_sequencer(ahc, /*unpause_always*/TRUE);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else {
int disconnected;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
u_int8_t hscb_index;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
disconnected = FALSE;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
hscb_index = ahc_find_scb(ahc, scb);
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
if (hscb_index == SCB_LIST_NULL) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
disconnected = TRUE;
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
} else {
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCBPTR, hscb_index);
if (ahc_inb(ahc, SCB_CONTROL) & DISCONNECTED)
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
disconnected = TRUE;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 22:29:27 +01:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (disconnected) {
/* Simply set the ABORT_SCB control bit */
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
scb->hscb->control |= ABORT_SCB|MK_MESSAGE;
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
if ((ahc->flags & AHC_PAGESCBS) == 0)
scb->hscb->control &= ~DISCONNECTED;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
scb->flags |= SCB_QUEUED_ABORT
| SCB_ABORT|SCB_RECOVERY_SCB;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
if (hscb_index != SCB_LIST_NULL) {
u_int8_t scb_control;
scb_control = ahc_inb(ahc, SCB_CONTROL);
ahc_outb(ahc, SCB_CONTROL,
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
scb_control | MK_MESSAGE
| ABORT_SCB);
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
}
/*
* Actually re-queue this SCB in case we can
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
* select the device before it reconnects. If
* the transaction we want to abort is not
* tagged, unbusy it first so that we don't
* get held back from sending the command.
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
*/
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
if ((scb->hscb->control & TAG_ENB) == 0) {
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
int target;
int lun;
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
target = scb->xs->sc_link->target;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
lun = scb->xs->sc_link->lun;
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
ahc_search_qinfifo(ahc, target,
channel,
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
lun,
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
SCB_LIST_NULL,
0, 0,
/*requeue*/TRUE);
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
}
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
sc_print_addr(scb->xs->sc_link);
printf("Queueing an Abort SCB\n");
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
STAILQ_INSERT_HEAD(&ahc->waiting_scbs, scb,
links);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
scb->flags |= SCB_WAITINGQ;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
timeout(ahc_timeout, (caddr_t)scb,
Fix some more problems in the recovery code. Cleanup of the disconnected list was broken in the SCB paging case (confusion of NULLand SCB_LIST_NULL) Implement a clean mechanism for determining that we have exited the timeout state and test for this in ahc_done instead of all over the place. Bring back the use of AAP (Auto Access Pause) I don't think it was the true cause of the bus hangs people were reporting. We want to reset the bus if we've been through an Abort action, not if we are a recovery SCB (one implies the other, but not vice-versa).
1997-02-19 01:00:51 +00:00
(200 * hz) / 1000);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
ahc_outb(ahc, SCBPTR, saved_scbptr);
Now that we use AAP, we have to explicitly unpause the sequencer when queueing an abort SCB.
1997-04-04 20:36:04 +01:00
/*
* ahc_run_waiting_queue may unpause us
* so do this last.
*/
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
ahc_run_waiting_queue(ahc);
Now that we use AAP, we have to explicitly unpause the sequencer when queueing an abort SCB.
1997-04-04 20:36:04 +01:00
/*
* If we are using AAP, ahc_run_waiting_queue
* will not unpause us, so ensure we are
* unpaused.
*/
unpause_sequencer(ahc, /*unpause_always*/FALSE);
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
} else {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/* Go "immediatly" to the bus reset */
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
sc_print_addr(scb->xs->sc_link);
printf("SCB %d: Immediate reset. "
"Flags = 0x%x\n", scb->hscb->tag,
scb->flags);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
goto bus_reset;
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
}
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
}
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
}
Properly deal with the Ultra series of adapters. We should now understand the new seeprom format and negotiate up to 20MHz sync if set in SCSI-Select. Reduce the complexity of the timeout code by running it at splhigh(). Fix a bug that caused rescheduled timeouts at 0 clock ticks in the future causing an infinite loop. Obtained from: Timeout bug noticed by David Greenman and wcarchive.
1995-10-26 23:57:18 +00:00
splx(s);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/*
* Look through the SCB array of the card and attempt to find the
* hardware SCB that corresponds to the passed in SCB. Return
* SCB_LIST_NULL if unsuccessful. This routine assumes that the
* card is already paused.
*/
static u_int8_t
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
ahc_find_scb(ahc, scb)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
struct scb *scb;
{
u_int8_t saved_scbptr;
u_int8_t curindex;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
saved_scbptr = ahc_inb(ahc, SCBPTR);
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
for (curindex = 0; curindex < ahc->scb_data->maxhscbs; curindex++) {
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCBPTR, curindex);
if (ahc_inb(ahc, SCB_TAG) == scb->hscb->tag)
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
break;
}
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCBPTR, saved_scbptr);
Fix a fencepost error in ahc_find_scb that could cause us to wrongfully find an SCB still down on the card that was paged out. This only affects error recovery. Submitted by: Daniel M. Eischen <deischen@iworks.InterWorks.org>
1997-03-24 17:42:25 +00:00
if (curindex >= ahc->scb_data->maxhscbs)
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
curindex = SCB_LIST_NULL;
return curindex;
}
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
static int
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
ahc_search_qinfifo(ahc, target, channel, lun, tag, flags, xs_error, requeue)
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
struct ahc_softc *ahc;
int target;
char channel;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
int lun;
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
u_int8_t tag;
u_int32_t flags;
u_int32_t xs_error;
int requeue;
{
u_int8_t saved_queue[AHC_SCB_MAX];
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
int queued = ahc_inb(ahc, QINCNT) & ahc->qcntmask;
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
int i;
int found;
struct scb *scbp;
STAILQ_HEAD(, scb) removed_scbs;
Initialization of a variable got lost in the last commit when I moved a piece of code into a subroutine.
1997-02-03 17:24:25 +00:00
found = 0;
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
STAILQ_INIT(&removed_scbs);
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
for (i = 0; i < (queued - found); i++) {
saved_queue[i] = ahc_inb(ahc, QINFIFO);
scbp = ahc->scb_data->scbarray[saved_queue[i]];
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
if (ahc_match_scb(scbp, target, channel, lun, tag)) {
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
/*
* We found an scb that needs to be removed.
*/
if (requeue) {
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
STAILQ_INSERT_HEAD(&removed_scbs, scbp, links);
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
} else {
Fix some more problems in the recovery code. Cleanup of the disconnected list was broken in the SCB paging case (confusion of NULLand SCB_LIST_NULL) Implement a clean mechanism for determining that we have exited the timeout state and test for this in ahc_done instead of all over the place. Bring back the use of AAP (Auto Access Pause) I don't think it was the true cause of the bus hangs people were reporting. We want to reset the bus if we've been through an Abort action, not if we are a recovery SCB (one implies the other, but not vice-versa).
1997-02-19 01:00:51 +00:00
scbp->flags |= flags;
scbp->flags &= ~SCB_ACTIVE;
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
scbp->xs->error = xs_error;
}
i--;
found++;
}
}
/* Now put the saved scbs back. */
for (queued = 0; queued < i; queued++)
ahc_outb(ahc, QINFIFO, saved_queue[queued]);
if (requeue) {
while ((scbp = removed_scbs.stqh_first) != NULL) {
STAILQ_REMOVE_HEAD(&removed_scbs, links);
STAILQ_INSERT_HEAD(&ahc->waiting_scbs, scbp, links);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
scbp->flags |= SCB_WAITINGQ;
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
}
}
return found;
}
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
/*
* The device at the given target/channel has been reset. Abort
* all active and queued scbs for that target/channel.
*/
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static int
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
ahc_reset_device(ahc, target, channel, lun, tag, xs_error)
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
struct ahc_softc *ahc;
int target;
char channel;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
int lun;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
u_int8_t tag;
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
u_int32_t xs_error;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
{
struct scb *scbp;
u_char active_scb;
int i = 0;
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
int found;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
/* restore this when we're done */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
active_scb = ahc_inb(ahc, SCBPTR);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
/*
Fix an oversight in the handling of non-tagged abort requests. We need to search the QINFIFO to remove any possible command that is waiting otherwise our abort request may not be held up still waiting for the first command to complete.
1997-02-03 16:29:07 +00:00
* Remove any entries from the Queue-In FIFO.
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
*/
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
found = ahc_search_qinfifo(ahc, target, channel, lun, tag,
ahc_search_qinfo->ahc_search_qinfifo ULTRAENB->FAST20 Add a missing ahc_run_done_queue if a BRKADDRINT occurs. This should never happen (haven't heard of one happening), but it was still a bug. Brought the ordered tag sending code up into the tag code to be clearer. If we decide we should send an ordered tag, only do so for the target that timed out instead of all targets. Initialize the STAILQ in ahc_serach__qinfifo. This was causing a panic during some recovery operations. Remove the unused varable maxtarget.
1997-02-09 03:26:56 +00:00
SCB_ABORTED|SCB_QUEUED_FOR_DONE, xs_error,
/*requeue*/FALSE);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
/*
* Search waiting for selection list.
*/
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t next, prev;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
next = ahc_inb(ahc, WAITING_SCBH); /* Start at head of list. */
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
prev = SCB_LIST_NULL;
while (next != SCB_LIST_NULL) {
Fix some more problems in the recovery code. Cleanup of the disconnected list was broken in the SCB paging case (confusion of NULLand SCB_LIST_NULL) Implement a clean mechanism for determining that we have exited the timeout state and test for this in ahc_done instead of all over the place. Bring back the use of AAP (Auto Access Pause) I don't think it was the true cause of the bus hangs people were reporting. We want to reset the bus if we've been through an Abort action, not if we are a recovery SCB (one implies the other, but not vice-versa).
1997-02-19 01:00:51 +00:00
u_int8_t scb_index;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCBPTR, next);
Fix some more problems in the recovery code. Cleanup of the disconnected list was broken in the SCB paging case (confusion of NULLand SCB_LIST_NULL) Implement a clean mechanism for determining that we have exited the timeout state and test for this in ahc_done instead of all over the place. Bring back the use of AAP (Auto Access Pause) I don't think it was the true cause of the bus hangs people were reporting. We want to reset the bus if we've been through an Abort action, not if we are a recovery SCB (one implies the other, but not vice-versa).
1997-02-19 01:00:51 +00:00
scb_index = ahc_inb(ahc, SCB_TAG);
if (scb_index >= ahc->scb_data->numscbs) {
panic("Waiting List inconsistency. "
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
"SCB index == %d, yet numscbs == %d.",
Fix some more problems in the recovery code. Cleanup of the disconnected list was broken in the SCB paging case (confusion of NULLand SCB_LIST_NULL) Implement a clean mechanism for determining that we have exited the timeout state and test for this in ahc_done instead of all over the place. Bring back the use of AAP (Auto Access Pause) I don't think it was the true cause of the bus hangs people were reporting. We want to reset the bus if we've been through an Abort action, not if we are a recovery SCB (one implies the other, but not vice-versa).
1997-02-19 01:00:51 +00:00
scb_index, ahc->scb_data->numscbs);
}
scbp = ahc->scb_data->scbarray[scb_index];
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
if (ahc_match_scb(scbp, target, channel, lun, tag)) {
u_int8_t linked_next;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
next = ahc_abort_wscb(ahc, scbp, next, prev,
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
xs_error);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
linked_next = ahc_inb(ahc, SCB_LINKED_NEXT);
if (linked_next != SCB_LIST_NULL) {
struct scb *next_scb;
/*
* Re-queue the waiting SCB via the
* waiting list.
*/
next_scb =
ahc->scb_data->scbarray[linked_next];
if (!ahc_match_scb(next_scb, target,
channel, lun, tag)) {
STAILQ_INSERT_HEAD(&ahc->waiting_scbs,
next_scb,
links);
next_scb->flags |= SCB_WAITINGQ;
}
}
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
found++;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else {
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
prev = next;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
next = ahc_inb(ahc, SCB_NEXT);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
}
}
}
/*
* Go through the entire SCB array now and look for
* commands for this target that are active. These
* are other (most likely tagged) commands that
* were disconnected when the reset occured.
*/
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
for (i = 0; i < ahc->scb_data->numscbs; i++) {
scbp = ahc->scb_data->scbarray[i];
Drop the number of allowed tags back down to 8. Pluto uses a higher value which mistakenly got committed. Fix two bugs in the ahc_reset_device code: Limit search for SCBs to process to those that are active and are not queued for done processing. It's okay for an SCB to not have a waiting next SCB.
1997-04-10 20:14:58 +01:00
if ((scbp->flags & SCB_ACTIVE) != 0
&& (scbp->flags & SCB_QUEUED_FOR_DONE) == 0
&& ahc_match_scb(scbp, target, channel, lun, tag)) {
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
u_int8_t busy_scbid;
Fix some more problems in the recovery code. Cleanup of the disconnected list was broken in the SCB paging case (confusion of NULLand SCB_LIST_NULL) Implement a clean mechanism for determining that we have exited the timeout state and test for this in ahc_done instead of all over the place. Bring back the use of AAP (Auto Access Pause) I don't think it was the true cause of the bus hangs people were reporting. We want to reset the bus if we've been through an Abort action, not if we are a recovery SCB (one implies the other, but not vice-versa).
1997-02-19 01:00:51 +00:00
scbp->flags |= SCB_ABORTED|SCB_QUEUED_FOR_DONE;
scbp->flags &= ~SCB_ACTIVE;
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
scbp->xs->error = xs_error;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
found++;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
if ((scbp->flags & SCB_WAITINGQ) != 0) {
STAILQ_REMOVE(&ahc->waiting_scbs, scbp, scb,
links);
scbp->flags &= ~SCB_WAITINGQ;
}
/* Ensure the target is "free" */
busy_scbid = ahc_index_busy_target(ahc, target,
(scbp->hscb->tcl & SELBUSB) ?
'B' : 'A', /*unbusy*/FALSE);
if (busy_scbid != SCB_LIST_NULL) {
struct scb *busy_scb;
struct scb *next_scb;
u_int8_t next_scbid;
busy_scb = ahc->scb_data->scbarray[busy_scbid];
if (!ahc_match_scb(busy_scb, target,
channel, lun, tag))
continue;
ahc_index_busy_target(ahc, target,
(scbp->hscb->tcl & SELBUSB) ?
'B' : 'A', /*unbusy*/TRUE);
next_scbid = busy_scb->hscb->datalen >> 24;
if (next_scbid == SCB_LIST_NULL) {
busy_scbid = ahc_find_scb(ahc, busy_scb);
if (busy_scbid == SCB_LIST_NULL)
panic("Couldn't find busy SCB");
ahc_outb(ahc, SCBPTR, busy_scbid);
next_scbid = ahc_inb(ahc,
SCB_LINKED_NEXT);
}
Drop the number of allowed tags back down to 8. Pluto uses a higher value which mistakenly got committed. Fix two bugs in the ahc_reset_device code: Limit search for SCBs to process to those that are active and are not queued for done processing. It's okay for an SCB to not have a waiting next SCB.
1997-04-10 20:14:58 +01:00
if (next_scbid != SCB_LIST_NULL) {
next_scb = ahc->scb_data->scbarray[next_scbid];
if (!ahc_match_scb(next_scb, target,
channel, lun, tag)) {
STAILQ_INSERT_HEAD(&ahc->waiting_scbs,
next_scb, links);
next_scb->flags |= SCB_WAITINGQ;
}
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
}
}
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
}
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
}
/*
* Go through the disconnected list and remove any entries we
* have queued for completion, 0'ing their control byte too.
*/
{
u_int8_t next, prev;
next = ahc_inb(ahc, DISCONNECTED_SCBH);
prev = SCB_LIST_NULL;
while (next != SCB_LIST_NULL) {
Fix some more problems in the recovery code. Cleanup of the disconnected list was broken in the SCB paging case (confusion of NULLand SCB_LIST_NULL) Implement a clean mechanism for determining that we have exited the timeout state and test for this in ahc_done instead of all over the place. Bring back the use of AAP (Auto Access Pause) I don't think it was the true cause of the bus hangs people were reporting. We want to reset the bus if we've been through an Abort action, not if we are a recovery SCB (one implies the other, but not vice-versa).
1997-02-19 01:00:51 +00:00
u_int8_t scb_index;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
ahc_outb(ahc, SCBPTR, next);
Fix some more problems in the recovery code. Cleanup of the disconnected list was broken in the SCB paging case (confusion of NULLand SCB_LIST_NULL) Implement a clean mechanism for determining that we have exited the timeout state and test for this in ahc_done instead of all over the place. Bring back the use of AAP (Auto Access Pause) I don't think it was the true cause of the bus hangs people were reporting. We want to reset the bus if we've been through an Abort action, not if we are a recovery SCB (one implies the other, but not vice-versa).
1997-02-19 01:00:51 +00:00
scb_index = ahc_inb(ahc, SCB_TAG);
if (scb_index >= ahc->scb_data->numscbs) {
panic("Disconnected List inconsistency. "
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
"SCB index == %d, yet numscbs == %d.",
Fix some more problems in the recovery code. Cleanup of the disconnected list was broken in the SCB paging case (confusion of NULLand SCB_LIST_NULL) Implement a clean mechanism for determining that we have exited the timeout state and test for this in ahc_done instead of all over the place. Bring back the use of AAP (Auto Access Pause) I don't think it was the true cause of the bus hangs people were reporting. We want to reset the bus if we've been through an Abort action, not if we are a recovery SCB (one implies the other, but not vice-versa).
1997-02-19 01:00:51 +00:00
scb_index, ahc->scb_data->numscbs);
}
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
scbp = ahc->scb_data->scbarray[scb_index];
if (ahc_match_scb(scbp, target, channel, lun, tag)) {
next = ahc_rem_scb_from_disc_list(ahc, next);
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
} else {
prev = next;
next = ahc_inb(ahc, SCB_NEXT);
}
}
}
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCBPTR, active_scb);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
return found;
}
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
static u_int8_t
ahc_rem_scb_from_disc_list(ahc, scbptr)
struct ahc_softc *ahc;
u_int8_t scbptr;
{
u_int8_t next;
u_int8_t prev;
ahc_outb(ahc, SCBPTR, scbptr);
next = ahc_inb(ahc, SCB_NEXT);
prev = ahc_inb(ahc, SCB_PREV);
ahc_outb(ahc, SCB_CONTROL, 0);
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
ahc_add_curscb_to_free_list(ahc);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
if (prev != SCB_LIST_NULL) {
ahc_outb(ahc, SCBPTR, prev);
ahc_outb(ahc, SCB_NEXT, next);
} else
ahc_outb(ahc, DISCONNECTED_SCBH, next);
if (next != SCB_LIST_NULL) {
ahc_outb(ahc, SCBPTR, next);
ahc_outb(ahc, SCB_PREV, prev);
}
return next;
}
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
static void
ahc_add_curscb_to_free_list(ahc)
struct ahc_softc *ahc;
{
/* Invalidate the tag so that ahc_find_scb doesn't think it's active */
ahc_outb(ahc, SCB_TAG, SCB_LIST_NULL);
ahc_outb(ahc, SCB_NEXT, ahc_inb(ahc, FREE_SCBH));
ahc_outb(ahc, FREE_SCBH, ahc_inb(ahc, SCBPTR));
}
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
/*
* Manipulate the waiting for selection list and return the
* scb that follows the one that we remove.
*/
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static u_char
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
ahc_abort_wscb (ahc, scbp, scbpos, prev, xs_error)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
struct scb *scbp;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t scbpos;
u_int8_t prev;
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
u_int32_t xs_error;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t curscb, next;
int target = ((scbp->hscb->tcl >> 4) & 0x0f);
char channel = (scbp->hscb->tcl & SELBUSB) ? 'B' : 'A';
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
/*
* Select the SCB we want to abort and
* pull the next pointer out of it.
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
curscb = ahc_inb(ahc, SCBPTR);
ahc_outb(ahc, SCBPTR, scbpos);
next = ahc_inb(ahc, SCB_NEXT);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
/* Clear the necessary fields */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCB_CONTROL, 0);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
ahc_index_busy_target(ahc, target, channel, /*unbusy*/TRUE);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
ahc_add_curscb_to_free_list(ahc);
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
/* update the waiting list */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (prev == SCB_LIST_NULL)
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
/* First in the list */
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, WAITING_SCBH, next);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
else {
/*
* Select the scb that pointed to us
* and update its next pointer.
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCBPTR, prev);
ahc_outb(ahc, SCB_NEXT, next);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
}
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
/*
* Point us back at the original scb position
* and inform the SCSI system that the command
* has been aborted.
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCBPTR, curscb);
Fix some more problems in the recovery code. Cleanup of the disconnected list was broken in the SCB paging case (confusion of NULLand SCB_LIST_NULL) Implement a clean mechanism for determining that we have exited the timeout state and test for this in ahc_done instead of all over the place. Bring back the use of AAP (Auto Access Pause) I don't think it was the true cause of the bus hangs people were reporting. We want to reset the bus if we've been through an Abort action, not if we are a recovery SCB (one implies the other, but not vice-versa).
1997-02-19 01:00:51 +00:00
scbp->flags |= SCB_ABORTED|SCB_QUEUED_FOR_DONE;
scbp->flags &= ~SCB_ACTIVE;
Don't rely on AAP(Auto Access Pause) when queueing SCBs to the card. This will increase the overhead of queueing a command, but some recent bug reports make me believe that AAP isn't really working and that we are losing some SCBs from the input queue. Hopefully this will cure that problem. Fix some bugs in the error recovery code. Mainly these could cause us to inadvertantly forget to untimeout an SCB that was recovered causing later confusion.
1997-02-18 04:25:31 +00:00
scbp->xs->error = xs_error;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
return next;
}
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
static u_int8_t
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
ahc_index_busy_target(ahc, target, channel, unbusy)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
int target;
char channel;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
int unbusy;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t active_scb;
u_int8_t info_scb;
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
u_int8_t busy_scbid;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int32_t scb_offset;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
info_scb = target / 4;
if (channel == 'B')
info_scb += 2;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
active_scb = ahc_inb(ahc, SCBPTR);
ahc_outb(ahc, SCBPTR, info_scb);
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
scb_offset = SCB_BUSYTARGETS + (target & 0x03);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
busy_scbid = ahc_inb(ahc, scb_offset);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
if (unbusy)
ahc_outb(ahc, scb_offset, SCB_LIST_NULL);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SCBPTR, active_scb);
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
return busy_scbid;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
}
When we perform an "automatic request sense", we issue an untagged command. The sequencer expects untagged transactions to have the SCBID of the transaction in the "busy target" array. So, ensure that the busy entry is up to date for the target in this case. The new identify code in the sequencer that performs additional sanity checking got caught up when a tagged transaction created an untagged request sense. In ahc_handle_seqint, ensure that the target ID is taken from the right place. In the case of a selection, the ID is in SCSIID. In the case of a reconnection it is found in SELID.
1997-03-01 06:50:41 +00:00
static void
ahc_busy_target(ahc, target, channel, scbid)
struct ahc_softc *ahc;
int target;
char channel;
u_int8_t scbid;
{
u_int8_t active_scb;
u_int8_t info_scb;
u_int8_t busy_scbid;
u_int32_t scb_offset;
info_scb = target / 4;
if (channel == 'B')
info_scb += 2;
active_scb = ahc_inb(ahc, SCBPTR);
ahc_outb(ahc, SCBPTR, info_scb);
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
scb_offset = SCB_BUSYTARGETS + (target & 0x03);
When we perform an "automatic request sense", we issue an untagged command. The sequencer expects untagged transactions to have the SCBID of the transaction in the "busy target" array. So, ensure that the busy entry is up to date for the target in this case. The new identify code in the sequencer that performs additional sanity checking got caught up when a tagged transaction created an untagged request sense. In ahc_handle_seqint, ensure that the target ID is taken from the right place. In the case of a selection, the ID is in SCSIID. In the case of a reconnection it is found in SELID.
1997-03-01 06:50:41 +00:00
ahc_outb(ahc, scb_offset, scbid);
ahc_outb(ahc, SCBPTR, active_scb);
return;
}
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
static void
ahc_clear_intstat(ahc)
struct ahc_softc *ahc;
{
/* Clear any interrupt conditions this may have caused */
ahc_outb(ahc, CLRSINT0, CLRSELDO|CLRSELDI|CLRSELINGO);
ahc_outb(ahc, CLRSINT1, CLRSELTIMEO|CLRATNO|CLRSCSIRSTI
|CLRBUSFREE|CLRSCSIPERR|CLRPHASECHG|
CLRREQINIT);
ahc_outb(ahc, CLRINT, CLRSCSIINT);
}
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static void
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
ahc_reset_current_bus(ahc)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
{
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
u_int8_t scsiseq;
u_int8_t sblkctl;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
ahc_outb(ahc, SIMODE1, ahc_inb(ahc, SIMODE1) & ~ENSCSIRST);
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
scsiseq = ahc_inb(ahc, SCSISEQ);
ahc_outb(ahc, SCSISEQ, scsiseq | SCSIRSTO);
if (timeouts_work != 0) {
struct ahc_busreset_args *args;
sblkctl = ahc_inb(ahc, SBLKCTL);
args = (sblkctl & SELBUSB) ? &ahc->busreset_args_b
: &ahc->busreset_args;
ahc->in_reset |= (args->bus == 'A' ? CHANNEL_A_RESET
: CHANNEL_B_RESET);
timeout(ahc_busreset_complete, args,
MAX((AHC_BUSRESET_DELAY * hz) / 1000000, 1));
} else {
DELAY(AHC_BUSRESET_DELAY);
/* Turn off the bus reset */
ahc_outb(ahc, SCSISEQ, scsiseq & ~SCSIRSTO);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
ahc_clear_intstat(ahc);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
/* Re-enable reset interrupts */
ahc_outb(ahc, SIMODE1, ahc_inb(ahc, SIMODE1) | ENSCSIRST);
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
/* Wait a minimal bus settle delay */
DELAY(AHC_BUSSETTLE_DELAY);
Fix an infinite loop caused by calling ahc_run_done queue while the driver is waiting a bus settle delay. There should really be a facility for the controller driver to "freeze" its queue during recovery operations which would make all of this gymnastics unnecessary.
1997-04-07 19:32:47 +01:00
ahc_run_done_queue(ahc);
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
}
}
static void
ahc_busreset_complete(arg)
void *arg;
{
struct ahc_busreset_args *args;
struct ahc_softc *ahc;
int s;
u_int8_t scsiseq;
u_int8_t sblkctl;
args = (struct ahc_busreset_args *)arg;
ahc = (struct ahc_softc *)args->ahc;
s = splbio();
pause_sequencer(ahc);
/* Save the current block control state */
sblkctl = ahc_inb(ahc, SBLKCTL);
/* Switch to the right bus */
if (args->bus == 'A')
ahc_outb(ahc, SBLKCTL, sblkctl & ~SELBUSB);
else
ahc_outb(ahc, SBLKCTL, sblkctl | SELBUSB);
/* Turn off the bus reset */
printf("Clearing bus reset\n");
scsiseq = ahc_inb(ahc, SCSISEQ);
ahc_outb(ahc, SCSISEQ, scsiseq & ~SCSIRSTO);
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
ahc_clear_intstat(ahc);
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
/* Re-enable reset interrupts */
ahc_outb(ahc, SIMODE1, ahc_inb(ahc, SIMODE1) | ENSCSIRST);
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
/*
* Set a 100ms timeout to clear our in_reset flag.
* It will be *at-least* that long before any targets
* respond to real commands on the bus.
*/
timeout(ahc_busreset_settle_complete, args,
(AHC_BUSSETTLE_DELAY * hz) / 1000000);
/* Restore the original channel */
ahc_outb(ahc, SBLKCTL, sblkctl);
unpause_sequencer(ahc, /*unpause_always*/FALSE);
splx(s);
}
static void
ahc_busreset_settle_complete(arg)
void *arg;
{
struct ahc_busreset_args *args;
struct ahc_softc *ahc;
int s;
args = (struct ahc_busreset_args *)arg;
ahc = (struct ahc_softc *)args->ahc;
/* Clear the reset flag */
s = splbio();
printf("Clearing 'in-reset' flag\n");
ahc->in_reset &= (args->bus == 'A' ? ~CHANNEL_A_RESET
: ~CHANNEL_B_RESET);
Fix an infinite loop caused by calling ahc_run_done queue while the driver is waiting a bus settle delay. There should really be a facility for the controller driver to "freeze" its queue during recovery operations which would make all of this gymnastics unnecessary.
1997-04-07 19:32:47 +01:00
ahc_run_done_queue(ahc);
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
splx(s);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
}
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static int
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
ahc_reset_channel(ahc, channel, xs_error, initiate_reset)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
char channel;
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
u_int32_t xs_error;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
int initiate_reset;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int32_t offset, offset_max;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
int found;
int target;
u_int8_t sblkctl;
char cur_channel;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
pause_sequencer(ahc);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
/*
* Clean up all the state information for the
* pending transactions on this bus.
*/
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
found = ahc_reset_device(ahc, ALL_TARGETS, channel, ALL_LUNS,
SCB_LIST_NULL, xs_error);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (channel == 'B') {
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
ahc->needsdtr |= (ahc->needsdtr_orig & 0xff00);
ahc->sdtrpending &= 0x00ff;
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
ahc->orderedtag &= 0x00ff;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
offset = TARG_SCRATCH + 8;
offset_max = TARG_SCRATCH + 16;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else if (ahc->type & AHC_WIDE){
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
ahc->needsdtr = ahc->needsdtr_orig;
ahc->needwdtr = ahc->needwdtr_orig;
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
ahc->orderedtag = 0;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
ahc->sdtrpending = 0;
ahc->wdtrpending = 0;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
offset = TARG_SCRATCH;
offset_max = TARG_SCRATCH + 16;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else {
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
ahc->needsdtr |= (ahc->needsdtr_orig & 0x00ff);
ahc->sdtrpending &= 0xff00;
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
ahc->orderedtag &= 0xff00;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
offset = TARG_SCRATCH;
offset_max = TARG_SCRATCH + 8;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
for (; offset < offset_max; offset++) {
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
/*
* Revert to async/narrow transfers
* until we renegotiate.
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
u_int8_t targ_scratch;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
targ_scratch = ahc_inb(ahc, offset);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
targ_scratch &= SXFR;
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, offset, targ_scratch);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
}
/*
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
* Reset the bus if we are initiating this reset and
* restart/unpause the sequencer
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
sblkctl = ahc_inb(ahc, SBLKCTL);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
cur_channel = (sblkctl & SELBUSB) ? 'B' : 'A';
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (cur_channel != channel) {
/* Case 1: Command for another bus is active
* Stealthily reset the other bus without
* upsetting the current bus.
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
*/
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SBLKCTL, sblkctl ^ SELBUSB);
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
ahc_outb(ahc, SIMODE1, ahc_inb(ahc, SIMODE1) & ~ENBUSFREE);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (initiate_reset)
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
ahc_reset_current_bus(ahc);
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, SCSISEQ, 0);
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
ahc_clear_intstat(ahc);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, SBLKCTL, sblkctl);
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
unpause_sequencer(ahc, /*unpause_always*/FALSE);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
} else {
/* Case 2: A command from this bus is active or we're idle */
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
ahc_outb(ahc, SIMODE1, ahc_inb(ahc, SIMODE1) & ~ENBUSFREE);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (initiate_reset)
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 08:19:59 +01:00
ahc_reset_current_bus(ahc);
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, SCSISEQ, 0);
Functionalize some code that was repeated throughout the driver. Fix a bug in the initialization of the busreset_args that left the B channel args unitialized and the A channel ones initialized to B's vales. Oops. If we get a NO_IDENT sequencer interrupt (the reconnecting target didn't issue an identify or botched it), reset the bus instead of panicing. We should be able to recover from this error. In the AWAITING_MSG handler, order messages by severity. Since the message we send is based on a flag on the SCB, it is possible, during error recovery, to get more than one flag set. This is fine since any time a new flag is set, it is meant to take us to a more draconian level of recovery. This also ensures that we don't lose any "history" of what the command has gone through. When we reset the bus, reset the "send ordered tag" bitmask. Clear some additional interrupt status when we perform a bus reset.
1997-02-28 03:58:21 +00:00
ahc_clear_intstat(ahc);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
restart_sequencer(ahc);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
}
Fix an infinite loop caused by calling ahc_run_done queue while the driver is waiting a bus settle delay. There should really be a facility for the controller driver to "freeze" its queue during recovery operations which would make all of this gymnastics unnecessary.
1997-04-07 19:32:47 +01:00
/*
* Untimeout our scbs now in case we have to delay our done
* processing.
*/
ahc_untimeout_done_queue(ahc);
if (initiate_reset == 0) {
/*
* If we initiated the reset, we'll run the queue
* once our bus-settle delay has expired.
*/
ahc_run_done_queue(ahc);
}
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
return found;
}
Fix an infinite loop caused by calling ahc_run_done queue while the driver is waiting a bus settle delay. There should really be a facility for the controller driver to "freeze" its queue during recovery operations which would make all of this gymnastics unnecessary.
1997-04-07 19:32:47 +01:00
static void
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
ahc_run_done_queue(ahc)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
{
int i;
struct scb *scbp;
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
for (i = 0; i < ahc->scb_data->numscbs; i++) {
scbp = ahc->scb_data->scbarray[i];
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
if (scbp->flags & SCB_QUEUED_FOR_DONE)
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
ahc_done(ahc, scbp);
}
}
Fix an infinite loop caused by calling ahc_run_done queue while the driver is waiting a bus settle delay. There should really be a facility for the controller driver to "freeze" its queue during recovery operations which would make all of this gymnastics unnecessary.
1997-04-07 19:32:47 +01:00
static void
ahc_untimeout_done_queue(ahc)
struct ahc_softc *ahc;
{
int i;
struct scb *scbp;
for (i = 0; i < ahc->scb_data->numscbs; i++) {
scbp = ahc->scb_data->scbarray[i];
if (scbp->flags & SCB_QUEUED_FOR_DONE)
untimeout(ahc_timeout, (caddr_t)scbp);
}
}
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static int
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
ahc_match_scb (scb, target, channel, lun, tag)
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
struct scb *scb;
int target;
char channel;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
int lun;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
u_int8_t tag;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
{
White space cleanup and other cosmetic style changes. Fix a few panics during error recovery: 1) Stupid mistake in the "no SCB match handler" where I was using the wrong variable (busy_scbid instead of scb_index). 2) Unbusy the target of an abort request if the command we are trying to abort is an untagged transaction. If we don't, we get a fatal NO_MATCH_BUSY condition which "should never happen". 3) When an abort completes, turn off ahc->in_timeout or else the next timeout will hit the protective "scb timesout again" panic. 4) Fix a typo that caused the requeued "abort" SCB to have its TAG_ENB and disconnect bits to be cleared (missing ~) so that devices would complain about overlapped commands. Be sure to turn off the unexpected busfree interrupt after we do a bus reset since we are expecting the bus to go free in that case. Return XS_TIMEOUT instead of XS_DRIVERSTUFFUP in certain scenarios. XS_TIMEOUT allows for retries, XS_DRIVERSTUFFUP does not. Allow commands with SDTR and WDTR negotiation to be tagged. The SCSI II spec says that you probably should not do this for fear of hitting bogus devices. The driver did this in the past for almost two years without any problem, and not doing it causes problems during error recovery to a tag capable device as the number of openings is higher than two and we'll start sending it tagged commands causing "overlapped commands attempted" type errors. The real fix needs to happen in the generic SCSI layer which can limit the number and type of transactions to a device during error recovery efficiently. Give ourselves at least 100ms to perform a request sense instead of relying on the original timeout to be long enough to complete this new command as well as the one that generated the condition. Removed some redundant code.
1997-02-03 02:16:16 +00:00
int targ = (scb->hscb->tcl >> 4) & 0x0f;
char chan = (scb->hscb->tcl & SELBUSB) ? 'B' : 'A';
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
int slun = scb->hscb->tcl & 0x07;
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
int match;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
match = ((chan == channel) || (channel == ALL_CHANNELS));
if (match != 0)
match = ((targ == target) || (target == ALL_TARGETS));
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
if (match != 0)
match = ((lun == slun) || (lun == ALL_LUNS));
Add 1997 to my copyright. If we can, use timeouts instead of DELAYs when dealing with a bus reset. This prevents us from holding up the whole machine for a noticible amount of time (especially for a real time app). Make a pass over the timeout/error handling code. Aborts are more reliable. We actually handle parity errors correctly now instead of locking up the bus. Added code to properly clean up disconnected SCBs down on the card during error handling. Improved robustness in several areas. If we are using defaults, but are an Ultra card, negotiate at 20MHz instead of 10. Don't attempt to handle any commands for 100ms after a reset has occured. This is the minimum time before a target will respond to selection. Also disable the busfree interrupt before doing a bus reset. This prevents the driver from getting confused by an "unexpected busfree".
1997-01-29 05:27:03 +00:00
if (match != 0)
match = ((tag == scb->hscb->tag) || (tag == SCB_LIST_NULL));
return match;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 09:25:36 +01:00
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
static void
ahc_construct_sdtr(ahc, start_byte, period, offset)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
int start_byte;
u_int8_t period;
u_int8_t offset;
{
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, MSG_OUT + start_byte, MSG_EXTENDED);
ahc_outb(ahc, MSG_OUT + 1 + start_byte, MSG_EXT_SDTR_LEN);
ahc_outb(ahc, MSG_OUT + 2 + start_byte, MSG_EXT_SDTR);
ahc_outb(ahc, MSG_OUT + 3 + start_byte, period);
ahc_outb(ahc, MSG_OUT + 4 + start_byte, offset);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, MSG_LEN, start_byte + 5);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
}
static void
ahc_construct_wdtr(ahc, start_byte, bus_width)
Add basic support for the 398X cards as multi-channel SCSI host adapters. This involves expanding the support of the SEEPROM routines to deal with the larger SEEPROMs on these cards and providing a mechanism to share SCB arrays between multiple controllers. Most of the 398X support came from Dan Eischer. ahc_data -> ahc_softc Clean up some more type bogons I missed from the last pass. Be more clear when handing the NO_MATCH condition. NO_MATCH can also happen when the sequencer encounters an SCB we've asked to be aborted.
1996-10-28 06:10:02 +00:00
struct ahc_softc *ahc;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
int start_byte;
u_int8_t bus_width;
{
Adapt to some changes in the register definitions. Clear the selection enable in SCSISEQ during error recovery to deal with the way the sequencer leaves selections enabled now. Add code to perform "patching" during sequencer program download. Spelling fixes obtained from NetBSD.
1997-03-16 07:12:07 +00:00
ahc_outb(ahc, MSG_OUT + start_byte, MSG_EXTENDED);
ahc_outb(ahc, MSG_OUT + 1 + start_byte, MSG_EXT_WDTR_LEN);
ahc_outb(ahc, MSG_OUT + 2 + start_byte, MSG_EXT_WDTR);
ahc_outb(ahc, MSG_OUT + 3 + start_byte, bus_width);
Clean up the memory mapped/Programmed I/O stuff so that the driver completely uses one or the other. This required some changes to the ahc_reset() function, and how early the probes had to allocate their softc. Turn the AHC_IN/OUT* macros into inline functions and lowercase their names to indicate this change. Geting AHC_OUTSB to work as a macro doing conditional memory mapped I/O would have been too gross. Stop setting STPWEN in the main driver and let the PCI front end do it instead. It knows better. Add the clearing of the QOUTQCNT variable during command complete processing in the SCB paging case. Go back to doing unconditional retries for the QUEUE FULL status condition. This is really a kludge, but the code to handle it properly is on the SCSI branch and will not make it into 2.2.
1996-11-11 05:24:46 +00:00
ahc_outb(ahc, MSG_LEN, start_byte + 4);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 17:38:45 +01:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
static void
ahc_calc_residual(scb)
struct scb *scb;
{
struct scsi_xfer *xs;
struct hardware_scb *hscb;
int resid_sgs;
xs = scb->xs;
hscb = scb->hscb;
Fix numerous problems with the abort/recovery code. Highlights include fixing a race condition in how SDTR and WDTR negotiation are handled, fixes for multi-lun non-tagged device recovery, and ensuring that the timedout scbs in the waiting queue are cleaned up. Fix a problem with SCB paging that caused bogus residuals to be reported.
1997-02-25 03:05:35 +00:00
/*
* If the disconnected flag is still set, this is bogus
* residual information left over from a sequencer
* pagin/pageout, so ignore this case.
*/
if ((scb->hscb->control & DISCONNECTED) == 0
&& (scb->flags & SCB_SENSE) == 0) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
/*
* Remainder of the SG where the transfer
* stopped.
*/
xs->resid = (hscb->residual_data_count[2] <<16) |
(hscb->residual_data_count[1] <<8) |
(hscb->residual_data_count[0]);
/*
* Add up the contents of all residual
* SG segments that are after the SG where
* the transfer stopped.
*/
Fix a bug in the reporting of residuals. The code was relying on the SG_COUNT filed in the hardware SCB not changing during the course of a transaction. Since the sequencer now DMAs the hardware SCB back up to the host when it detects a residual, this is no longer the case. I added a field to the "software" scb to mirror this information and it is now used for doing the residual calculation.
1997-02-11 17:10:37 +00:00
resid_sgs = scb->hscb->residual_SG_segment_count - 1;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
while (resid_sgs > 0) {
int sg;
Fix a bug in the reporting of residuals. The code was relying on the SG_COUNT filed in the hardware SCB not changing during the course of a transaction. Since the sequencer now DMAs the hardware SCB back up to the host when it detects a residual, this is no longer the case. I added a field to the "software" scb to mirror this information and it is now used for doing the residual calculation.
1997-02-11 17:10:37 +00:00
sg = scb->sg_count - resid_sgs;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
xs->resid += scb->ahc_dma[sg].len;
resid_sgs--;
}
#if defined(__FreeBSD__)
xs->flags |= SCSI_RESID_VALID;
#elif defined(__NetBSD__)
/* XXX - Update to do this right */
#endif
}
/*
Add code to the SCSIINT handler for unexpected busfrees and handling immediate SCBs in response to a busfree. When re-queueing an SCB that returned with QUEUE FULL status, reset its timeout. Ensure that aborted SCBs have an error code set in there xs before it gets passed back up with scsi_done. Fix a few KNF nits.
1996-11-22 08:28:45 +00:00
* Clean out the residual information in this SCB for its
* next consumer.
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 07:42:53 +01:00
*/
hscb->residual_data_count[2] = 0;
hscb->residual_data_count[1] = 0;
hscb->residual_data_count[0] = 0;
hscb->residual_SG_segment_count = 0;
#ifdef AHC_DEBUG
if (ahc_debug & AHC_SHOWMISC) {
sc_print_addr(xs->sc_link);
printf("Handled Residual of %ld bytes\n" ,xs->resid);
}
#endif
}
Reference in New Issue Copy Permalink