zld: use aarch64 for opcodes

This commit is contained in:
Jakub Konka 2021-03-07 08:18:35 +01:00
parent dc34ac2b9e
commit d484b3b3cb
3 changed files with 113 additions and 239 deletions

View File

@ -221,7 +221,8 @@ pub const Instruction = union(enum) {
offset: u12,
opc: u2,
op1: u2,
fixed: u4 = 0b111_0,
v: u1,
fixed: u3 = 0b111,
size: u2,
},
LoadStorePairOfRegisters: packed struct {
@ -505,6 +506,7 @@ pub const Instruction = union(enum) {
.offset = offset.toU12(),
.opc = opc,
.op1 = op1,
.v = 0,
.size = 0b10,
},
};
@ -517,6 +519,7 @@ pub const Instruction = union(enum) {
.offset = offset.toU12(),
.opc = opc,
.op1 = op1,
.v = 0,
.size = 0b11,
},
};

View File

@ -10,6 +10,7 @@ const fs = std.fs;
const macho = std.macho;
const math = std.math;
const log = std.log.scoped(.zld);
const aarch64 = @import("../../codegen/aarch64.zig");
const Allocator = mem.Allocator;
const CodeSignature = @import("CodeSignature.zig");
@ -19,7 +20,6 @@ const Trie = @import("Trie.zig");
usingnamespace @import("commands.zig");
usingnamespace @import("bind.zig");
usingnamespace @import("reloc.zig");
allocator: *Allocator,
@ -968,27 +968,27 @@ fn writeStubHelperCommon(self: *Zld) !void {
data_blk: {
const displacement = math.cast(i21, target_addr - this_addr) catch |_| break :data_blk;
// adr x17, disp
mem.writeIntLittle(u32, code[0..4], Arm64.adr(17, @bitCast(u21, displacement)).toU32());
mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.adr(.x17, displacement).toU32());
// nop
mem.writeIntLittle(u32, code[4..8], Arm64.nop().toU32());
mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.nop().toU32());
break :data_blk_outer;
}
data_blk: {
const new_this_addr = this_addr + @sizeOf(u32);
const displacement = math.cast(i21, target_addr - new_this_addr) catch |_| break :data_blk;
// nop
mem.writeIntLittle(u32, code[0..4], Arm64.nop().toU32());
mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.nop().toU32());
// adr x17, disp
mem.writeIntLittle(u32, code[4..8], Arm64.adr(17, @bitCast(u21, displacement)).toU32());
mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.adr(.x17, displacement).toU32());
break :data_blk_outer;
}
// Jump is too big, replace adr with adrp and add.
const this_page = @intCast(i32, this_addr >> 12);
const target_page = @intCast(i32, target_addr >> 12);
const pages = @bitCast(u21, @intCast(i21, target_page - this_page));
mem.writeIntLittle(u32, code[0..4], Arm64.adrp(17, pages).toU32());
const pages = @intCast(i21, target_page - this_page);
mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.adrp(.x17, pages).toU32());
const narrowed = @truncate(u12, target_addr);
mem.writeIntLittle(u32, code[4..8], Arm64.add(17, 17, narrowed, 1).toU32());
mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.add(.x17, .x17, narrowed, false).toU32());
}
// stp x16, x17, [sp, #-16]!
code[8] = 0xf0;
@ -1003,9 +1003,11 @@ fn writeStubHelperCommon(self: *Zld) !void {
const displacement = math.divExact(u64, target_addr - this_addr, 4) catch |_| break :binder_blk;
const literal = math.cast(u18, displacement) catch |_| break :binder_blk;
// ldr x16, label
mem.writeIntLittle(u32, code[12..16], Arm64.ldr(16, literal, 1).toU32());
mem.writeIntLittle(u32, code[12..16], aarch64.Instruction.ldr(.x16, .{
.literal = literal,
}).toU32());
// nop
mem.writeIntLittle(u32, code[16..20], Arm64.nop().toU32());
mem.writeIntLittle(u32, code[16..20], aarch64.Instruction.nop().toU32());
break :binder_blk_outer;
}
binder_blk: {
@ -1015,19 +1017,26 @@ fn writeStubHelperCommon(self: *Zld) !void {
log.debug("2: disp=0x{x}, literal=0x{x}", .{ displacement, literal });
// Pad with nop to please division.
// nop
mem.writeIntLittle(u32, code[12..16], Arm64.nop().toU32());
mem.writeIntLittle(u32, code[12..16], aarch64.Instruction.nop().toU32());
// ldr x16, label
mem.writeIntLittle(u32, code[16..20], Arm64.ldr(16, literal, 1).toU32());
mem.writeIntLittle(u32, code[16..20], aarch64.Instruction.ldr(.x16, .{
.literal = literal,
}).toU32());
break :binder_blk_outer;
}
// Use adrp followed by ldr(immediate).
const this_page = @intCast(i32, this_addr >> 12);
const target_page = @intCast(i32, target_addr >> 12);
const pages = @bitCast(u21, @intCast(i21, target_page - this_page));
mem.writeIntLittle(u32, code[12..16], Arm64.adrp(16, pages).toU32());
const pages = @intCast(i21, target_page - this_page);
mem.writeIntLittle(u32, code[12..16], aarch64.Instruction.adrp(.x16, pages).toU32());
const narrowed = @truncate(u12, target_addr);
const offset = try math.divExact(u12, narrowed, 8);
mem.writeIntLittle(u32, code[16..20], Arm64.ldrq(16, 16, offset).toU32());
mem.writeIntLittle(u32, code[16..20], aarch64.Instruction.ldr(.x16, .{
.register = .{
.rn = .x16,
.offset = aarch64.Instruction.LoadStoreOffset.imm(offset),
},
}).toU32());
}
// br x16
code[20] = 0x00;
@ -1099,9 +1108,11 @@ fn writeStub(self: *Zld, index: u32) !void {
const displacement = math.divExact(u64, target_addr - this_addr, 4) catch |_| break :inner;
const literal = math.cast(u18, displacement) catch |_| break :inner;
// ldr x16, literal
mem.writeIntLittle(u32, code[0..4], Arm64.ldr(16, literal, 1).toU32());
mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.ldr(.x16, .{
.literal = literal,
}).toU32());
// nop
mem.writeIntLittle(u32, code[4..8], Arm64.nop().toU32());
mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.nop().toU32());
break :outer;
}
inner: {
@ -1109,22 +1120,29 @@ fn writeStub(self: *Zld, index: u32) !void {
const displacement = math.divExact(u64, target_addr - new_this_addr, 4) catch |_| break :inner;
const literal = math.cast(u18, displacement) catch |_| break :inner;
// nop
mem.writeIntLittle(u32, code[0..4], Arm64.nop().toU32());
mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.nop().toU32());
// ldr x16, literal
mem.writeIntLittle(u32, code[4..8], Arm64.ldr(16, literal, 1).toU32());
mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.ldr(.x16, .{
.literal = literal,
}).toU32());
break :outer;
}
// Use adrp followed by ldr(immediate).
const this_page = @intCast(i32, this_addr >> 12);
const target_page = @intCast(i32, target_addr >> 12);
const pages = @bitCast(u21, @intCast(i21, target_page - this_page));
mem.writeIntLittle(u32, code[0..4], Arm64.adrp(16, pages).toU32());
const pages = @intCast(i21, target_page - this_page);
mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.adrp(.x16, pages).toU32());
const narrowed = @truncate(u12, target_addr);
const offset = try math.divExact(u12, narrowed, 8);
mem.writeIntLittle(u32, code[4..8], Arm64.ldrq(16, 16, offset).toU32());
mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.ldr(.x16, .{
.register = .{
.rn = .x16,
.offset = aarch64.Instruction.LoadStoreOffset.imm(offset),
},
}).toU32());
}
// br x16
mem.writeIntLittle(u32, code[8..12], Arm64.br(16).toU32());
mem.writeIntLittle(u32, code[8..12], aarch64.Instruction.br(.x16).toU32());
},
else => unreachable,
}
@ -1160,9 +1178,11 @@ fn writeStubInStubHelper(self: *Zld, index: u32) !void {
const displacement = try math.cast(i28, @intCast(i64, stub_helper.offset) - @intCast(i64, stub_off) - 4);
const literal = @divExact(stub_size - @sizeOf(u32), 4);
// ldr w16, literal
mem.writeIntLittle(u32, code[0..4], Arm64.ldr(16, literal, 0).toU32());
mem.writeIntLittle(u32, code[0..4], aarch64.Instruction.ldr(.w16, .{
.literal = literal,
}).toU32());
// b disp
mem.writeIntLittle(u32, code[4..8], Arm64.b(displacement).toU32());
mem.writeIntLittle(u32, code[4..8], aarch64.Instruction.b(displacement).toU32());
mem.writeIntLittle(u32, code[8..12], 0x0); // Just a placeholder populated in `populateLazyBindOffsetsInStubHelper`.
},
else => unreachable,
@ -1486,9 +1506,18 @@ fn doRelocs(self: *Zld) !void {
.ARM64_RELOC_BRANCH26 => {
assert(rel.r_length == 2);
const inst = code[off..][0..4];
const displacement = @intCast(i28, @intCast(i64, target_addr) - @intCast(i64, this_addr));
var parsed = mem.bytesAsValue(meta.TagPayload(Arm64, Arm64.Branch), inst);
parsed.disp = @truncate(u26, @bitCast(u28, displacement) >> 2);
const displacement = @intCast(
i28,
@intCast(i64, target_addr) - @intCast(i64, this_addr),
);
var parsed = mem.bytesAsValue(
meta.TagPayload(
aarch64.Instruction,
aarch64.Instruction.UnconditionalBranchImmediate,
),
inst,
);
parsed.imm26 = @truncate(u26, @bitCast(u28, displacement) >> 2);
},
.ARM64_RELOC_PAGE21,
.ARM64_RELOC_GOT_LOAD_PAGE21,
@ -1501,7 +1530,13 @@ fn doRelocs(self: *Zld) !void {
const target_page = @intCast(i32, ta >> 12);
const pages = @bitCast(u21, @intCast(i21, target_page - this_page));
log.debug(" | moving by {} pages", .{pages});
var parsed = mem.bytesAsValue(meta.TagPayload(Arm64, Arm64.Address), inst);
var parsed = mem.bytesAsValue(
meta.TagPayload(
aarch64.Instruction,
aarch64.Instruction.PCRelativeAddress,
),
inst,
);
parsed.immhi = @truncate(u19, pages >> 2);
parsed.immlo = @truncate(u2, pages);
addend = null;
@ -1510,17 +1545,29 @@ fn doRelocs(self: *Zld) !void {
.ARM64_RELOC_GOT_LOAD_PAGEOFF12,
=> {
const inst = code[off..][0..4];
if (Arm64.isArithmetic(inst)) {
if (aarch64IsArithmetic(inst)) {
log.debug(" | detected ADD opcode", .{});
// add
var parsed = mem.bytesAsValue(meta.TagPayload(Arm64, Arm64.Add), inst);
var parsed = mem.bytesAsValue(
meta.TagPayload(
aarch64.Instruction,
aarch64.Instruction.AddSubtractImmediate,
),
inst,
);
const ta = if (addend) |a| target_addr + a else target_addr;
const narrowed = @truncate(u12, ta);
parsed.offset = narrowed;
parsed.imm12 = narrowed;
} else {
log.debug(" | detected LDR/STR opcode", .{});
// ldr/str
var parsed = mem.bytesAsValue(meta.TagPayload(Arm64, Arm64.LoadRegister), inst);
var parsed = mem.bytesAsValue(
meta.TagPayload(
aarch64.Instruction,
aarch64.Instruction.LoadStoreRegister,
),
inst,
);
const ta = if (addend) |a| target_addr + a else target_addr;
const narrowed = @truncate(u12, ta);
const offset: u12 = blk: {
@ -1541,27 +1588,43 @@ fn doRelocs(self: *Zld) !void {
addend = null;
},
.ARM64_RELOC_TLVP_LOAD_PAGEOFF12 => {
// TODO why is this necessary?
const RegInfo = struct {
rt: u5,
rd: u5,
rn: u5,
size: u1,
};
const inst = code[off..][0..4];
const parsed: RegInfo = blk: {
if (Arm64.isArithmetic(inst)) {
const curr = mem.bytesAsValue(meta.TagPayload(Arm64, Arm64.Add), inst);
break :blk .{ .rt = curr.rt, .rn = curr.rn, .size = curr.size };
if (aarch64IsArithmetic(inst)) {
const curr = mem.bytesAsValue(
meta.TagPayload(
aarch64.Instruction,
aarch64.Instruction.AddSubtractImmediate,
),
inst,
);
break :blk .{ .rd = curr.rd, .rn = curr.rn, .size = curr.sf };
} else {
const curr = mem.bytesAsValue(meta.TagPayload(Arm64, Arm64.LoadRegister), inst);
break :blk .{ .rt = curr.rt, .rn = curr.rn, .size = @truncate(u1, curr.size) };
const curr = mem.bytesAsValue(
meta.TagPayload(
aarch64.Instruction,
aarch64.Instruction.LoadStoreRegister,
),
inst,
);
break :blk .{ .rd = curr.rt, .rn = curr.rn, .size = @truncate(u1, curr.size) };
}
};
const ta = if (addend) |a| target_addr + a else target_addr;
const narrowed = @truncate(u12, ta);
log.debug(" | rewriting TLV access to ADD opcode", .{});
// For TLV, we always generate an add instruction.
mem.writeIntLittle(u32, inst, Arm64.add(parsed.rt, parsed.rn, narrowed, parsed.size).toU32());
mem.writeIntLittle(u32, inst, aarch64.Instruction.add(
@intToEnum(aarch64.Register, parsed.rd),
@intToEnum(aarch64.Register, parsed.rn),
narrowed,
false,
).toU32());
},
.ARM64_RELOC_SUBTRACTOR => {
sub = @intCast(i64, target_addr);
@ -2965,3 +3028,8 @@ fn isExtern(sym: *const macho.nlist_64) callconv(.Inline) bool {
fn isWeakDef(sym: *const macho.nlist_64) callconv(.Inline) bool {
return (sym.n_desc & macho.N_WEAK_DEF) != 0;
}
fn aarch64IsArithmetic(inst: *const [4]u8) callconv(.Inline) bool {
const group_decode = @truncate(u5, inst[3]);
return ((group_decode >> 2) == 4);
}

View File

@ -1,197 +0,0 @@
const std = @import("std");
const log = std.log.scoped(.reloc);
pub const Arm64 = union(enum) {
Branch: packed struct {
disp: u26,
fixed: u5 = 0b00101,
link: u1,
},
BranchRegister: packed struct {
_1: u5 = 0b0000_0,
reg: u5,
_2: u11 = 0b1111_1000_000,
link: u1,
_3: u10 = 0b1101_0110_00,
},
Address: packed struct {
reg: u5,
immhi: u19,
_1: u5 = 0b10000,
immlo: u2,
page: u1,
},
LoadRegister: packed struct {
rt: u5,
rn: u5,
offset: u12,
opc: u2,
_2: u2 = 0b01,
v: u1,
_1: u3 = 0b111,
size: u2,
},
LoadLiteral: packed struct {
reg: u5,
literal: u19,
_1: u6 = 0b011_0_00,
size: u1,
_2: u1 = 0b0,
},
Add: packed struct {
rt: u5,
rn: u5,
offset: u12,
_1: u9 = 0b0_0_100010_0,
size: u1,
},
Nop: packed struct {
fixed: u32 = 0b1101010100_0_00_011_0010_0000_000_11111,
},
pub fn toU32(self: Arm64) u32 {
const as_u32 = switch (self) {
.Branch => |x| @bitCast(u32, x),
.BranchRegister => |x| @bitCast(u32, x),
.Address => |x| @bitCast(u32, x),
.LoadRegister => |x| @bitCast(u32, x),
.LoadLiteral => |x| @bitCast(u32, x),
.Add => |x| @bitCast(u32, x),
.Nop => |x| @bitCast(u32, x),
};
return as_u32;
}
pub fn b(disp: i28) Arm64 {
return Arm64{
.Branch = .{
.disp = @truncate(u26, @bitCast(u28, disp) >> 2),
.link = 0,
},
};
}
pub fn bl(disp: i28) Arm64 {
return Arm64{
.Branch = .{
.disp = @truncate(u26, @bitCast(u28, disp) >> 2),
.link = 1,
},
};
}
pub fn br(reg: u5) Arm64 {
return Arm64{
.BranchRegister = .{
.reg = reg,
.link = 0,
},
};
}
pub fn blr(reg: u5) Arm64 {
return Arm64{
.BranchRegister = .{
.reg = reg,
.link = 1,
},
};
}
pub fn adr(reg: u5, disp: u21) Arm64 {
return Arm64{
.Address = .{
.reg = reg,
.immhi = @truncate(u19, disp >> 2),
.immlo = @truncate(u2, disp),
.page = 0,
},
};
}
pub fn adrp(reg: u5, disp: u21) Arm64 {
return Arm64{
.Address = .{
.reg = reg,
.immhi = @truncate(u19, disp >> 2),
.immlo = @truncate(u2, disp),
.page = 1,
},
};
}
pub fn ldr(reg: u5, literal: u19, size: u1) Arm64 {
return Arm64{
.LoadLiteral = .{
.reg = reg,
.literal = literal,
.size = size,
},
};
}
pub fn add(rt: u5, rn: u5, offset: u12, size: u1) Arm64 {
return Arm64{
.Add = .{
.rt = rt,
.rn = rn,
.offset = offset,
.size = size,
},
};
}
pub fn ldrq(rt: u5, rn: u5, offset: u12) Arm64 {
return Arm64{
.LoadRegister = .{
.rt = rt,
.rn = rn,
.offset = offset,
.opc = 0b01,
.v = 0b0,
.size = 0b11,
},
};
}
pub fn ldrh(rt: u5, rn: u5, offset: u12) Arm64 {
return Arm64{
.LoadRegister = .{
.rt = rt,
.rn = rn,
.offset = offset,
.opc = 0b01,
.v = 0b0,
.size = 0b01,
},
};
}
pub fn ldrb(rt: u5, rn: u5, offset: u12) Arm64 {
return Arm64{
.LoadRegister = .{
.rt = rt,
.rn = rn,
.offset = offset,
.opc = 0b01,
.v = 0b0,
.size = 0b00,
},
};
}
pub fn nop() Arm64 {
return Arm64{
.Nop = .{},
};
}
pub fn isArithmetic(inst: *const [4]u8) bool {
const group_decode = @truncate(u5, inst[3]);
log.debug("{b}", .{group_decode});
return ((group_decode >> 2) == 4);
// if ((group_decode >> 2) == 4) {
// log.debug("Arithmetic imm", .{});
// } else if (((group_decode & 0b01010) >> 3) == 1) {
// log.debug("Load/store", .{});
// }
}
};