Implement minidump support for RISC-V.

Submitted by:	Mitchell Horne <mhorne063@gmail.com>
Differential Revision:	https://reviews.freebsd.org/D18320

sys/riscv/riscv/minidump_machdep.c

@@ -1,5 +1,7 @@
 /*-
  * Copyright (c) 2006 Peter Wemm
+ * Copyright (c) 2015 The FreeBSD Foundation
+ * Copyright (c) 2019 Mitchell Horne
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -53,9 +55,381 @@ __FBSDID("$FreeBSD$");
 CTASSERT(sizeof(struct kerneldumpheader) == 512);
 CTASSERT(sizeof(*vm_page_dump) == 8);
 
+uint64_t *vm_page_dump;
+int vm_page_dump_size;
+
+static struct kerneldumpheader kdh;
+
+/* Handle chunked writes. */
+static size_t fragsz;
+static void *dump_va;
+static size_t counter, progress, dumpsize;
+
+static uint64_t tmpbuffer[PAGE_SIZE / sizeof(uint64_t)];
+
+static struct {
+	int min_per;
+	int max_per;
+	int visited;
+} progress_track[10] = {
+	{  0,  10, 0},
+	{ 10,  20, 0},
+	{ 20,  30, 0},
+	{ 30,  40, 0},
+	{ 40,  50, 0},
+	{ 50,  60, 0},
+	{ 60,  70, 0},
+	{ 70,  80, 0},
+	{ 80,  90, 0},
+	{ 90, 100, 0}
+};
+
+static void
+report_progress(size_t progress, size_t dumpsize)
+{
+	int sofar, i;
+
+	sofar = 100 - ((progress * 100) / dumpsize);
+	for (i = 0; i < nitems(progress_track); i++) {
+		if (sofar < progress_track[i].min_per ||
+		    sofar > progress_track[i].max_per)
+			continue;
+		if (progress_track[i].visited)
+			return;
+		progress_track[i].visited = 1;
+		printf("..%d%%", sofar);
+		return;
+	}
+}
+
+static bool
+is_dumpable(vm_paddr_t pa)
+{
+	vm_page_t m;
+	int i;
+
+	if ((m = vm_phys_paddr_to_vm_page(pa)) != NULL)
+		return ((m->flags & PG_NODUMP) == 0);
+
+	for (i = 0; dump_avail[i] != 0 || dump_avail[i + 1] != 0; i += 2) {
+		if (pa >= dump_avail[i] && pa < dump_avail[i + 1])
+			return (true);
+	}
+	return (false);
+}
+
+static int
+blk_flush(struct dumperinfo *di)
+{
+	int error;
+
+	if (fragsz == 0)
+		return (0);
+
+	error = dump_append(di, dump_va, 0, fragsz);
+	fragsz = 0;
+	return (error);
+}
+
+/*
+ * Write a block of data to the dump file.
+ *
+ * Caller can provide data through a pointer or by specifying its
+ * physical address.
+ *
+ * XXX writes using pa should be no larger than PAGE_SIZE.
+ */
+static int
+blk_write(struct dumperinfo *di, char *ptr, vm_paddr_t pa, size_t sz)
+{
+	size_t len;
+	int error, c;
+	u_int maxdumpsz;
+
+	maxdumpsz = min(di->maxiosize, MAXDUMPPGS * PAGE_SIZE);
+	if (maxdumpsz == 0)	/* seatbelt */
+		maxdumpsz = PAGE_SIZE;
+	error = 0;
+	if ((sz % PAGE_SIZE) != 0) {
+		printf("size not page aligned\n");
+		return (EINVAL);
+	}
+	if (ptr != NULL && pa != 0) {
+		printf("cant have both va and pa!\n");
+		return (EINVAL);
+	}
+	if ((((uintptr_t)pa) % PAGE_SIZE) != 0) {
+		printf("address not page aligned %#lx\n", (uintptr_t)pa);
+		return (EINVAL);
+	}
+	if (ptr != NULL) {
+		/*
+		 * If we're doing a virtual dump, flush any
+		 * pre-existing pa pages.
+		 */
+		error = blk_flush(di);
+		if (error != 0)
+			return (error);
+	}
+	while (sz) {
+		len = maxdumpsz - fragsz;
+		if (len > sz)
+			len = sz;
+		counter += len;
+		progress -= len;
+		if (counter >> 22) {
+			report_progress(progress, dumpsize);
+			counter &= (1 << 22) - 1;
+		}
+
+		wdog_kern_pat(WD_LASTVAL);
+
+		if (ptr) {
+			error = dump_append(di, ptr, 0, len);
+			if (error != 0)
+				return (error);
+			ptr += len;
+			sz -= len;
+		} else {
+			dump_va = (void *)PHYS_TO_DMAP(pa);
+			fragsz += len;
+			pa += len;
+			sz -= len;
+			error = blk_flush(di);
+			if (error != 0)
+				return (error);
+		}
+
+		/* Check for user abort */
+		c = cncheckc();
+		if (c == 0x03)
+			return (ECANCELED);
+		if (c != -1)
+			printf(" (CTRL-C to abort) ");
+	}
+
+	return (0);
+}
+
 int
 minidumpsys(struct dumperinfo *di)
 {
+	pd_entry_t *l1, *l2;
+	pt_entry_t *l3;
+	struct minidumphdr mdhdr;
+	uint32_t pmapsize;
+	vm_offset_t va;
+	vm_paddr_t pa;
+	int error;
+	uint64_t bits;
+	int i, bit;
+	int retry_count;
 
-	panic("minidumpsys");
+	retry_count = 0;
+retry:
+	retry_count++;
+	error = 0;
+	pmapsize = 0;
+
+	/* Build set of dumpable pages from kernel pmap */
+	for (va = VM_MIN_KERNEL_ADDRESS; va < kernel_vm_end; va += L2_SIZE) {
+		pmapsize += PAGE_SIZE;
+		if (!pmap_get_tables(pmap_kernel(), va, &l1, &l2, &l3))
+			continue;
+
+		/* We should always be using the l2 table for kvm */
+		if (l2 == NULL)
+			continue;
+
+		/* l2 may be a superpage */
+		if ((*l2 & PTE_RWX) != 0) {
+			pa = (*l2 >> PTE_PPN1_S) << L2_SHIFT;
+			for (i = 0; i < Ln_ENTRIES; i++, pa += PAGE_SIZE) {
+				if (is_dumpable(pa))
+					dump_add_page(pa);
+			}
+		} else {
+			for (i = 0; i < Ln_ENTRIES; i++) {
+				if ((l3[i] & PTE_V) == 0)
+					continue;
+				pa = (l3[i] >> PTE_PPN0_S) * PAGE_SIZE;
+				if (is_dumpable(pa))
+					dump_add_page(pa);
+			}
+		}
+	}
+
+	/* Calculate dump size */
+	dumpsize = pmapsize;
+	dumpsize += round_page(msgbufp->msg_size);
+	dumpsize += round_page(vm_page_dump_size);
+	for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
+		bits = vm_page_dump[i];
+		while (bits) {
+			bit = ffsl(bits) - 1;
+			pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) +
+			    bit) * PAGE_SIZE;
+			/* Clear out undumpable pages now if needed */
+			if (is_dumpable(pa))
+				dumpsize += PAGE_SIZE;
+			else
+				dump_drop_page(pa);
+			bits &= ~(1ul << bit);
+		}
+	}
+	dumpsize += PAGE_SIZE;
+
+	progress = dumpsize;
+
+	/* Initialize mdhdr */
+	bzero(&mdhdr, sizeof(mdhdr));
+	strcpy(mdhdr.magic, MINIDUMP_MAGIC);
+	mdhdr.version = MINIDUMP_VERSION;
+	mdhdr.msgbufsize = msgbufp->msg_size;
+	mdhdr.bitmapsize = vm_page_dump_size;
+	mdhdr.pmapsize = pmapsize;
+	mdhdr.kernbase = KERNBASE;
+	mdhdr.dmapphys = DMAP_MIN_PHYSADDR;
+	mdhdr.dmapbase = DMAP_MIN_ADDRESS;
+	mdhdr.dmapend = DMAP_MAX_ADDRESS;
+
+	dump_init_header(di, &kdh, KERNELDUMPMAGIC, KERNELDUMP_RISCV_VERSION,
+	    dumpsize);
+
+	error = dump_start(di, &kdh);
+	if (error != 0)
+		goto fail;
+
+	printf("Dumping %llu out of %ju MB:", (long long)dumpsize >> 20,
+	    ptoa((uintmax_t)physmem) / 1048576);
+
+	/* Dump minidump header */
+	bzero(&tmpbuffer, sizeof(tmpbuffer));
+	bcopy(&mdhdr, &tmpbuffer, sizeof(mdhdr));
+	error = blk_write(di, (char *)&tmpbuffer, 0, PAGE_SIZE);
+	if (error)
+		goto fail;
+
+	/* Dump msgbuf up front */
+	error = blk_write(di, (char *)msgbufp->msg_ptr, 0,
+	    round_page(msgbufp->msg_size));
+	if (error)
+		goto fail;
+
+	/* Dump bitmap */
+	error = blk_write(di, (char *)vm_page_dump, 0,
+	    round_page(vm_page_dump_size));
+	if (error)
+		goto fail;
+
+	/* Dump kernel page directory pages */
+	bzero(&tmpbuffer, sizeof(tmpbuffer));
+	for (va = VM_MIN_KERNEL_ADDRESS; va < kernel_vm_end; va += L2_SIZE) {
+		if (!pmap_get_tables(pmap_kernel(), va, &l1, &l2, &l3)) {
+			/* We always write a page, even if it is zero */
+			error = blk_write(di, (char *)&tmpbuffer, 0, PAGE_SIZE);
+			if (error)
+				goto fail;
+			/* Flush, in case we reuse tmpbuffer in the same block */
+			error = blk_flush(di);
+			if (error)
+				goto fail;
+		} else if ((*l2 & PTE_RWX) != 0) {
+			/* Generate fake l3 entries based on the l2 superpage */
+			for (i = 0; i < Ln_ENTRIES; i++) {
+				tmpbuffer[i] = (*l2 | (i << PTE_PPN0_S));
+			}
+			/* We always write a page, even if it is zero */
+			error = blk_write(di, (char *)&tmpbuffer, 0, PAGE_SIZE);
+			if (error)
+				goto fail;
+			/* Flush, in case we reuse tmpbuffer in the same block */
+			error = blk_flush(di);
+			if (error)
+				goto fail;
+			bzero(&tmpbuffer, sizeof(tmpbuffer));
+		} else {
+			pa = (*l2 >> PTE_PPN0_S) * PAGE_SIZE;
+
+			/* We always write a page, even if it is zero */
+			error = blk_write(di, NULL, pa, PAGE_SIZE);
+			if (error)
+				goto fail;
+		}
+	}
+
+	/* Dump memory chunks */
+	/* XXX cluster it up and use blk_dump() */
+	for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
+		bits = vm_page_dump[i];
+		while (bits) {
+			bit = ffsl(bits) - 1;
+			pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) +
+			    bit) * PAGE_SIZE;
+			error = blk_write(di, 0, pa, PAGE_SIZE);
+			if (error)
+				goto fail;
+			bits &= ~(1ul << bit);
+		}
+	}
+
+	error = blk_flush(di);
+	if (error)
+		goto fail;
+
+	error = dump_finish(di, &kdh);
+	if (error != 0)
+		goto fail;
+
+	printf("\nDump complete\n");
+	return (0);
+
+fail:
+	if (error < 0)
+		error = -error;
+
+	printf("\n");
+	if (error == ENOSPC) {
+		printf("Dump map grown while dumping. ");
+		if (retry_count < 5) {
+			printf("Retrying...\n");
+			goto retry;
+		}
+		printf("Dump failed.\n");
+	}
+	else if (error == ECANCELED)
+		printf("Dump aborted\n");
+	else if (error == E2BIG)
+		printf("Dump failed. Partition too small.\n");
+	else
+		printf("** DUMP FAILED (ERROR %d) **\n", error);
+	return (error);
+}
+
+/*
+ * Add a page to the minidump bitmap.
+ */
+void
+dump_add_page(vm_paddr_t pa)
+{
+	int idx, bit;
+
+	pa >>= PAGE_SHIFT;
+	idx = pa >> 6;		/* 2^6 = 64 */
+	bit = pa & 63;
+	atomic_set_long(&vm_page_dump[idx], 1ul << bit);
+}
+
+/*
+ * Remove page from the minidump bitmap.
+ */
+void
+dump_drop_page(vm_paddr_t pa)
+{
+	int idx, bit;
+
+	pa >>= PAGE_SHIFT;
+	idx = pa >> 6;		/* 2^6 = 64 */
+	bit = pa & 63;
+	atomic_clear_long(&vm_page_dump[idx], 1ul << bit);
 }

sys/riscv/riscv/pmap.c

@@ -1645,9 +1645,7 @@ free_pv_chunk(struct pv_chunk *pc)
 	PV_STAT(atomic_add_int(&pc_chunk_frees, 1));
 	/* entire chunk is free, return it */
 	m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t)pc));
-#if 0 /* TODO: For minidump */
 	dump_drop_page(m->phys_addr);
-#endif
 	vm_page_unwire(m, PQ_NONE);
 	vm_page_free(m);
 }
@@ -1709,9 +1707,7 @@ retry:
 	}
 	PV_STAT(atomic_add_int(&pc_chunk_count, 1));
 	PV_STAT(atomic_add_int(&pc_chunk_allocs, 1));
-#if 0 /* TODO: This is for minidump */
 	dump_add_page(m->phys_addr);
-#endif
 	pc = (void *)PHYS_TO_DMAP(m->phys_addr);
 	pc->pc_pmap = pmap;
 	pc->pc_map[0] = PC_FREE0 & ~1ul;	/* preallocated bit 0 */

sys/riscv/riscv/uma_machdep.c

@@ -55,11 +55,8 @@ uma_small_alloc(uma_zone_t zone, vm_size_t bytes, int domain, u_int8_t *flags,
 	if (m == NULL)
 		return (NULL);
 	pa = m->phys_addr;
-#if 0
-	/* RISCVTODO: minidump */
 	if ((wait & M_NODUMP) == 0)
 		dump_add_page(pa);
-#endif
 	va = (void *)PHYS_TO_DMAP(pa);
 	if ((wait & M_ZERO) && (m->flags & PG_ZERO) == 0)
 		bzero(va, PAGE_SIZE);
@@ -73,10 +70,7 @@ uma_small_free(void *mem, vm_size_t size, u_int8_t flags)
 	vm_paddr_t pa;
 
 	pa = DMAP_TO_PHYS((vm_offset_t)mem);
-#if 0
-	/* RISCVTODO: minidump */
 	dump_drop_page(pa);
-#endif
 	m = PHYS_TO_VM_PAGE(pa);
 	vm_page_unwire_noq(m);
 	vm_page_free(m);

sys/vm/vm_page.c

@@ -633,7 +633,7 @@ vm_page_startup(vm_offset_t vaddr)
 #endif
 
 #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \
-    defined(__i386__) || defined(__mips__)
+    defined(__i386__) || defined(__mips__) || defined(__riscv)
 	/*
 	 * Allocate a bitmap to indicate that a random physical page
 	 * needs to be included in a minidump.
@@ -658,7 +658,8 @@ vm_page_startup(vm_offset_t vaddr)
 #else
 	(void)last_pa;
 #endif
-#if defined(__aarch64__) || defined(__amd64__) || defined(__mips__)
+#if defined(__aarch64__) || defined(__amd64__) || defined(__mips__) || \
+    defined(__riscv)
 	/*
 	 * Include the UMA bootstrap pages, witness pages and vm_page_dump
 	 * in a crash dump.  When pmap_map() uses the direct map, they are
@@ -773,7 +774,8 @@ vm_page_startup(vm_offset_t vaddr)
 		high_avail = new_end;
 	new_end = vm_reserv_startup(&vaddr, new_end, high_avail);
 #endif
-#if defined(__aarch64__) || defined(__amd64__) || defined(__mips__)
+#if defined(__aarch64__) || defined(__amd64__) || defined(__mips__) || \
+    defined(__riscv)
 	/*
 	 * Include vm_page_array and vm_reserv_array in a crash dump.
 	 */