Commit 8fb4a4efbabbe3fe98521201eac41feee5a9a50a | zig

author	Jakub Konka <kubkon@jakubkonka.com> 2023-09-14 01:45:23+02:00
committer	GitHub <noreply@github.com> 2023-09-14 01:45:23+02:00
log	8fb4a4efbabbe3fe98521201eac41feee5a9a50a
tree	398abef45faa8065836c6a5f8cad167ae8adb4ed
parent	223f62acbd32c04db3169906f33869f43e5258f7
parent	59a586a8785b8aea417824e5b6b8d6b8a1b2b695
signature	Signed by PGP key 4AEE18F83AFDEB23

Merge pull request #17146 from ziglang/elf-linker

elf: upstream zld/ELF functionality, part 2

15 files changed, 954 insertions(+), 47 deletions(-)

CMakeLists.txt
src
- arch
  - aarch64
    CodeGen.zig
  - arm
    CodeGen.zig
  - riscv64
    CodeGen.zig
  - sparc64
    CodeGen.zig
  - x86_64
    CodeGen.zig
    Disassembler.zig
- codegen.zig
- link
  - Elf.zig
  - Elf
    Archive.zig
    Atom.zig
    Object.zig
    Symbol.zig
    ZigModule.zig
    file.zig

CMakeLists.txt+2

...	...	@@ -555,6 +555,7 @@ set(ZIG_STAGE2_SOURCES
555	555	"${CMAKE_SOURCE_DIR}/src/arch/wasm/Emit.zig"
556	556	"${CMAKE_SOURCE_DIR}/src/arch/wasm/Mir.zig"
557	557	"${CMAKE_SOURCE_DIR}/src/arch/x86_64/CodeGen.zig"
	558	"${CMAKE_SOURCE_DIR}/src/arch/x86_64/Disassembler.zig"
558	559	"${CMAKE_SOURCE_DIR}/src/arch/x86_64/Emit.zig"
559	560	"${CMAKE_SOURCE_DIR}/src/arch/x86_64/Encoding.zig"
560	561	"${CMAKE_SOURCE_DIR}/src/arch/x86_64/Mir.zig"
...	...	@@ -584,6 +585,7 @@ set(ZIG_STAGE2_SOURCES
584	585	"${CMAKE_SOURCE_DIR}/src/link/Coff/Object.zig"
585	586	"${CMAKE_SOURCE_DIR}/src/link/Coff/lld.zig"
586	587	"${CMAKE_SOURCE_DIR}/src/link/Elf.zig"
	588	"${CMAKE_SOURCE_DIR}/src/link/Elf/Archive.zig"
587	589	"${CMAKE_SOURCE_DIR}/src/link/Elf/Atom.zig"
588	590	"${CMAKE_SOURCE_DIR}/src/link/Elf/LinkerDefined.zig"
589	591	"${CMAKE_SOURCE_DIR}/src/link/Elf/Object.zig"

src/arch/aarch64/CodeGen.zig+1-1

...	...	@@ -4316,7 +4316,7 @@ fn airCall(self: *Self, inst: Air.Inst.Index, modifier: std.builtin.CallModifier
4316	4316	if (self.bin_file.cast(link.File.Elf)) \|elf_file\| {
4317	4317	const sym_index = try elf_file.getOrCreateMetadataForDecl(func.owner_decl);
4318	4318	const sym = elf_file.symbol(sym_index);
4319		_ = try sym.getOrCreateGotEntry(elf_file);
	4319	_ = try sym.getOrCreateGotEntry(sym_index, elf_file);
4320	4320	const got_addr = @as(u32, @intCast(sym.gotAddress(elf_file)));
4321	4321	try self.genSetReg(Type.usize, .x30, .{ .memory = got_addr });
4322	4322	} else if (self.bin_file.cast(link.File.MachO)) \|macho_file\| {

src/arch/arm/CodeGen.zig+1-1

...	...	@@ -4296,7 +4296,7 @@ fn airCall(self: *Self, inst: Air.Inst.Index, modifier: std.builtin.CallModifier
4296	4296	if (self.bin_file.cast(link.File.Elf)) \|elf_file\| {
4297	4297	const sym_index = try elf_file.getOrCreateMetadataForDecl(func.owner_decl);
4298	4298	const sym = elf_file.symbol(sym_index);
4299		_ = try sym.getOrCreateGotEntry(elf_file);
	4299	_ = try sym.getOrCreateGotEntry(sym_index, elf_file);
4300	4300	const got_addr = @as(u32, @intCast(sym.gotAddress(elf_file)));
4301	4301	try self.genSetReg(Type.usize, .lr, .{ .memory = got_addr });
4302	4302	} else if (self.bin_file.cast(link.File.MachO)) \|_\| {

src/arch/riscv64/CodeGen.zig+1-1

...	...	@@ -1749,7 +1749,7 @@ fn airCall(self: *Self, inst: Air.Inst.Index, modifier: std.builtin.CallModifier
1749	1749	.func => \|func\| {
1750	1750	const sym_index = try elf_file.getOrCreateMetadataForDecl(func.owner_decl);
1751	1751	const sym = elf_file.symbol(sym_index);
1752		_ = try sym.getOrCreateGotEntry(elf_file);
	1752	_ = try sym.getOrCreateGotEntry(sym_index, elf_file);
1753	1753	const got_addr = @as(u32, @intCast(sym.gotAddress(elf_file)));
1754	1754	try self.genSetReg(Type.usize, .ra, .{ .memory = got_addr });
1755	1755	_ = try self.addInst(.{

src/arch/sparc64/CodeGen.zig+1-1

...	...	@@ -1351,7 +1351,7 @@ fn airCall(self: *Self, inst: Air.Inst.Index, modifier: std.builtin.CallModifier
1351	1351	const got_addr = if (self.bin_file.cast(link.File.Elf)) \|elf_file\| blk: {
1352	1352	const sym_index = try elf_file.getOrCreateMetadataForDecl(func.owner_decl);
1353	1353	const sym = elf_file.symbol(sym_index);
1354		_ = try sym.getOrCreateGotEntry(elf_file);
	1354	_ = try sym.getOrCreateGotEntry(sym_index, elf_file);
1355	1355	break :blk @as(u32, @intCast(sym.gotAddress(elf_file)));
1356	1356	} else unreachable;
1357	1357

src/arch/x86_64/CodeGen.zig+2-2

...	...	@@ -8157,7 +8157,7 @@ fn airCall(self: *Self, inst: Air.Inst.Index, modifier: std.builtin.CallModifier
8157	8157	const sym_index = try elf_file.getOrCreateMetadataForDecl(owner_decl);
8158	8158	const sym = elf_file.symbol(sym_index);
8159	8159	sym.flags.needs_got = true;
8160		_ = try sym.getOrCreateGotEntry(elf_file);
	8160	_ = try sym.getOrCreateGotEntry(sym_index, elf_file);
8161	8161	const got_addr = sym.gotAddress(elf_file);
8162	8162	try self.asmMemory(.{ ._, .call }, Memory.sib(.qword, .{
8163	8163	.base = .{ .reg = .ds },
...	...	@@ -10236,7 +10236,7 @@ fn genLazySymbolRef(
10236	10236	return self.fail("{s} creating lazy symbol", .{@errorName(err)});
10237	10237	const sym = elf_file.symbol(sym_index);
10238	10238	sym.flags.needs_got = true;
10239		_ = try sym.getOrCreateGotEntry(elf_file);
	10239	_ = try sym.getOrCreateGotEntry(sym_index, elf_file);
10240	10240	const got_addr = sym.gotAddress(elf_file);
10241	10241	const got_mem =
10242	10242	Memory.sib(.qword, .{ .base = .{ .reg = .ds }, .disp = @intCast(got_addr) });

src/arch/x86_64/Disassembler.zig created+475

...	...	@@ -0,0 +1,475 @@
	1	const Disassembler = @This();
	2
	3	const std = @import("std");
	4	const assert = std.debug.assert;
	5	const math = std.math;
	6
	7	const bits = @import("bits.zig");
	8	const encoder = @import("encoder.zig");
	9
	10	const Encoding = @import("Encoding.zig");
	11	const Immediate = bits.Immediate;
	12	const Instruction = encoder.Instruction;
	13	const LegacyPrefixes = encoder.LegacyPrefixes;
	14	const Memory = bits.Memory;
	15	const Register = bits.Register;
	16	const Rex = encoder.Rex;
	17
	18	pub const Error = error{
	19	EndOfStream,
	20	LegacyPrefixAfterRex,
	21	UnknownOpcode,
	22	Overflow,
	23	Todo,
	24	};
	25
	26	code: []const u8,
	27	pos: usize = 0,
	28
	29	pub fn init(code: []const u8) Disassembler {
	30	return .{ .code = code };
	31	}
	32
	33	pub fn next(dis: *Disassembler) Error!?Instruction {
	34	const prefixes = dis.parsePrefixes() catch \|err\| switch (err) {
	35	error.EndOfStream => return null,
	36	else => \|e\| return e,
	37	};
	38
	39	const enc = try dis.parseEncoding(prefixes) orelse return error.UnknownOpcode;
	40	switch (enc.data.op_en) {
	41	.np => return inst(enc, .{}),
	42	.d, .i => {
	43	const imm = try dis.parseImm(enc.data.ops[0]);
	44	return inst(enc, .{
	45	.op1 = .{ .imm = imm },
	46	});
	47	},
	48	.zi => {
	49	const imm = try dis.parseImm(enc.data.ops[1]);
	50	return inst(enc, .{
	51	.op1 = .{ .reg = Register.rax.toBitSize(enc.data.ops[0].regBitSize()) },
	52	.op2 = .{ .imm = imm },
	53	});
	54	},
	55	.o, .oi => {
	56	const reg_low_enc = @as(u3, @truncate(dis.code[dis.pos - 1]));
	57	const op2: Instruction.Operand = if (enc.data.op_en == .oi) .{
	58	.imm = try dis.parseImm(enc.data.ops[1]),
	59	} else .none;
	60	return inst(enc, .{
	61	.op1 = .{ .reg = parseGpRegister(reg_low_enc, prefixes.rex.b, prefixes.rex, enc.data.ops[0].regBitSize()) },
	62	.op2 = op2,
	63	});
	64	},
	65	.m, .mi, .m1, .mc => {
	66	const modrm = try dis.parseModRmByte();
	67	const act_enc = Encoding.findByOpcode(enc.opcode(), .{
	68	.legacy = prefixes.legacy,
	69	.rex = prefixes.rex,
	70	}, modrm.op1) orelse return error.UnknownOpcode;
	71	const sib = if (modrm.sib()) try dis.parseSibByte() else null;
	72
	73	if (modrm.direct()) {
	74	const op2: Instruction.Operand = switch (act_enc.data.op_en) {
	75	.mi => .{ .imm = try dis.parseImm(act_enc.data.ops[1]) },
	76	.m1 => .{ .imm = Immediate.u(1) },
	77	.mc => .{ .reg = .cl },
	78	.m => .none,
	79	else => unreachable,
	80	};
	81	return inst(act_enc, .{
	82	.op1 = .{ .reg = parseGpRegister(modrm.op2, prefixes.rex.b, prefixes.rex, act_enc.data.ops[0].regBitSize()) },
	83	.op2 = op2,
	84	});
	85	}
	86
	87	const disp = try dis.parseDisplacement(modrm, sib);
	88	const op2: Instruction.Operand = switch (act_enc.data.op_en) {
	89	.mi => .{ .imm = try dis.parseImm(act_enc.data.ops[1]) },
	90	.m1 => .{ .imm = Immediate.u(1) },
	91	.mc => .{ .reg = .cl },
	92	.m => .none,
	93	else => unreachable,
	94	};
	95
	96	if (modrm.rip()) {
	97	return inst(act_enc, .{
	98	.op1 = .{ .mem = Memory.rip(Memory.PtrSize.fromBitSize(act_enc.data.ops[0].memBitSize()), disp) },
	99	.op2 = op2,
	100	});
	101	}
	102
	103	const scale_index = if (sib) \|info\| info.scaleIndex(prefixes.rex) else null;
	104	const base = if (sib) \|info\|
	105	info.baseReg(modrm, prefixes)
	106	else
	107	parseGpRegister(modrm.op2, prefixes.rex.b, prefixes.rex, 64);
	108	return inst(act_enc, .{
	109	.op1 = .{ .mem = Memory.sib(Memory.PtrSize.fromBitSize(act_enc.data.ops[0].memBitSize()), .{
	110	.base = if (base) \|base_reg\| .{ .reg = base_reg } else .none,
	111	.scale_index = scale_index,
	112	.disp = disp,
	113	}) },
	114	.op2 = op2,
	115	});
	116	},
	117	.fd => {
	118	const seg = segmentRegister(prefixes.legacy);
	119	const offset = try dis.parseOffset();
	120	return inst(enc, .{
	121	.op1 = .{ .reg = Register.rax.toBitSize(enc.data.ops[0].regBitSize()) },
	122	.op2 = .{ .mem = Memory.moffs(seg, offset) },
	123	});
	124	},
	125	.td => {
	126	const seg = segmentRegister(prefixes.legacy);
	127	const offset = try dis.parseOffset();
	128	return inst(enc, .{
	129	.op1 = .{ .mem = Memory.moffs(seg, offset) },
	130	.op2 = .{ .reg = Register.rax.toBitSize(enc.data.ops[1].regBitSize()) },
	131	});
	132	},
	133	.mr, .mri, .mrc => {
	134	const modrm = try dis.parseModRmByte();
	135	const sib = if (modrm.sib()) try dis.parseSibByte() else null;
	136	const src_bit_size = enc.data.ops[1].regBitSize();
	137
	138	if (modrm.direct()) {
	139	return inst(enc, .{
	140	.op1 = .{ .reg = parseGpRegister(modrm.op2, prefixes.rex.b, prefixes.rex, enc.data.ops[0].regBitSize()) },
	141	.op2 = .{ .reg = parseGpRegister(modrm.op1, prefixes.rex.x, prefixes.rex, src_bit_size) },
	142	});
	143	}
	144
	145	const dst_bit_size = enc.data.ops[0].memBitSize();
	146	const disp = try dis.parseDisplacement(modrm, sib);
	147	const op3: Instruction.Operand = switch (enc.data.op_en) {
	148	.mri => .{ .imm = try dis.parseImm(enc.data.ops[2]) },
	149	.mrc => .{ .reg = .cl },
	150	.mr => .none,
	151	else => unreachable,
	152	};
	153
	154	if (modrm.rip()) {
	155	return inst(enc, .{
	156	.op1 = .{ .mem = Memory.rip(Memory.PtrSize.fromBitSize(dst_bit_size), disp) },
	157	.op2 = .{ .reg = parseGpRegister(modrm.op1, prefixes.rex.r, prefixes.rex, src_bit_size) },
	158	.op3 = op3,
	159	});
	160	}
	161
	162	const scale_index = if (sib) \|info\| info.scaleIndex(prefixes.rex) else null;
	163	const base = if (sib) \|info\|
	164	info.baseReg(modrm, prefixes)
	165	else
	166	parseGpRegister(modrm.op2, prefixes.rex.b, prefixes.rex, 64);
	167	return inst(enc, .{
	168	.op1 = .{ .mem = Memory.sib(Memory.PtrSize.fromBitSize(dst_bit_size), .{
	169	.base = if (base) \|base_reg\| .{ .reg = base_reg } else .none,
	170	.scale_index = scale_index,
	171	.disp = disp,
	172	}) },
	173	.op2 = .{ .reg = parseGpRegister(modrm.op1, prefixes.rex.r, prefixes.rex, src_bit_size) },
	174	.op3 = op3,
	175	});
	176	},
	177	.rm, .rmi => {
	178	const modrm = try dis.parseModRmByte();
	179	const sib = if (modrm.sib()) try dis.parseSibByte() else null;
	180	const dst_bit_size = enc.data.ops[0].regBitSize();
	181
	182	if (modrm.direct()) {
	183	const op3: Instruction.Operand = switch (enc.data.op_en) {
	184	.rm => .none,
	185	.rmi => .{ .imm = try dis.parseImm(enc.data.ops[2]) },
	186	else => unreachable,
	187	};
	188	return inst(enc, .{
	189	.op1 = .{ .reg = parseGpRegister(modrm.op1, prefixes.rex.x, prefixes.rex, dst_bit_size) },
	190	.op2 = .{ .reg = parseGpRegister(modrm.op2, prefixes.rex.b, prefixes.rex, enc.data.ops[1].regBitSize()) },
	191	.op3 = op3,
	192	});
	193	}
	194
	195	const src_bit_size = if (enc.data.ops[1] == .m) dst_bit_size else enc.data.ops[1].memBitSize();
	196	const disp = try dis.parseDisplacement(modrm, sib);
	197	const op3: Instruction.Operand = switch (enc.data.op_en) {
	198	.rmi => .{ .imm = try dis.parseImm(enc.data.ops[2]) },
	199	.rm => .none,
	200	else => unreachable,
	201	};
	202
	203	if (modrm.rip()) {
	204	return inst(enc, .{
	205	.op1 = .{ .reg = parseGpRegister(modrm.op1, prefixes.rex.r, prefixes.rex, dst_bit_size) },
	206	.op2 = .{ .mem = Memory.rip(Memory.PtrSize.fromBitSize(src_bit_size), disp) },
	207	.op3 = op3,
	208	});
	209	}
	210
	211	const scale_index = if (sib) \|info\| info.scaleIndex(prefixes.rex) else null;
	212	const base = if (sib) \|info\|
	213	info.baseReg(modrm, prefixes)
	214	else
	215	parseGpRegister(modrm.op2, prefixes.rex.b, prefixes.rex, 64);
	216	return inst(enc, .{
	217	.op1 = .{ .reg = parseGpRegister(modrm.op1, prefixes.rex.r, prefixes.rex, dst_bit_size) },
	218	.op2 = .{ .mem = Memory.sib(Memory.PtrSize.fromBitSize(src_bit_size), .{
	219	.base = if (base) \|base_reg\| .{ .reg = base_reg } else .none,
	220	.scale_index = scale_index,
	221	.disp = disp,
	222	}) },
	223	.op3 = op3,
	224	});
	225	},
	226	.rm0, .vmi, .rvm, .rvmr, .rvmi, .mvr => unreachable, // TODO
	227	}
	228	}
	229
	230	fn inst(encoding: Encoding, args: struct {
	231	prefix: Instruction.Prefix = .none,
	232	op1: Instruction.Operand = .none,
	233	op2: Instruction.Operand = .none,
	234	op3: Instruction.Operand = .none,
	235	op4: Instruction.Operand = .none,
	236	}) Instruction {
	237	var i = Instruction{ .encoding = encoding, .prefix = args.prefix, .ops = .{
	238	args.op1,
	239	args.op2,
	240	args.op3,
	241	args.op4,
	242	} };
	243	return i;
	244	}
	245
	246	const Prefixes = struct {
	247	legacy: LegacyPrefixes = .{},
	248	rex: Rex = .{},
	249	// TODO add support for VEX prefix
	250	};
	251
	252	fn parsePrefixes(dis: *Disassembler) !Prefixes {
	253	const rex_prefix_mask: u4 = 0b0100;
	254	var stream = std.io.fixedBufferStream(dis.code[dis.pos..]);
	255	const reader = stream.reader();
	256
	257	var res: Prefixes = .{};
	258
	259	while (true) {
	260	const next_byte = try reader.readByte();
	261	dis.pos += 1;
	262
	263	switch (next_byte) {
	264	0xf0, 0xf2, 0xf3, 0x2e, 0x36, 0x26, 0x64, 0x65, 0x3e, 0x66, 0x67 => {
	265	// Legacy prefix
	266	if (res.rex.present) return error.LegacyPrefixAfterRex;
	267	switch (next_byte) {
	268	0xf0 => res.legacy.prefix_f0 = true,
	269	0xf2 => res.legacy.prefix_f2 = true,
	270	0xf3 => res.legacy.prefix_f3 = true,
	271	0x2e => res.legacy.prefix_2e = true,
	272	0x36 => res.legacy.prefix_36 = true,
	273	0x26 => res.legacy.prefix_26 = true,
	274	0x64 => res.legacy.prefix_64 = true,
	275	0x65 => res.legacy.prefix_65 = true,
	276	0x3e => res.legacy.prefix_3e = true,
	277	0x66 => res.legacy.prefix_66 = true,
	278	0x67 => res.legacy.prefix_67 = true,
	279	else => unreachable,
	280	}
	281	},
	282	else => {
	283	if (rex_prefix_mask == @as(u4, @truncate(next_byte >> 4))) {
	284	// REX prefix
	285	res.rex.w = next_byte & 0b1000 != 0;
	286	res.rex.r = next_byte & 0b100 != 0;
	287	res.rex.x = next_byte & 0b10 != 0;
	288	res.rex.b = next_byte & 0b1 != 0;
	289	res.rex.present = true;
	290	continue;
	291	}
	292
	293	// TODO VEX prefix
	294
	295	dis.pos -= 1;
	296	break;
	297	},
	298	}
	299	}
	300
	301	return res;
	302	}
	303
	304	fn parseEncoding(dis: *Disassembler, prefixes: Prefixes) !?Encoding {
	305	const o_mask: u8 = 0b1111_1000;
	306
	307	var opcode: [3]u8 = .{ 0, 0, 0 };
	308	var stream = std.io.fixedBufferStream(dis.code[dis.pos..]);
	309	const reader = stream.reader();
	310
	311	comptime var opc_count = 0;
	312	inline while (opc_count < 3) : (opc_count += 1) {
	313	const byte = try reader.readByte();
	314	opcode[opc_count] = byte;
	315	dis.pos += 1;
	316
	317	if (byte == 0x0f) {
	318	// Multi-byte opcode
	319	} else if (opc_count > 0) {
	320	// Multi-byte opcode
	321	if (Encoding.findByOpcode(opcode[0 .. opc_count + 1], .{
	322	.legacy = prefixes.legacy,
	323	.rex = prefixes.rex,
	324	}, null)) \|mnemonic\| {
	325	return mnemonic;
	326	}
	327	} else {
	328	// Single-byte opcode
	329	if (Encoding.findByOpcode(opcode[0..1], .{
	330	.legacy = prefixes.legacy,
	331	.rex = prefixes.rex,
	332	}, null)) \|mnemonic\| {
	333	return mnemonic;
	334	} else {
	335	// Try O* encoding
	336	return Encoding.findByOpcode(&.{opcode[0] & o_mask}, .{
	337	.legacy = prefixes.legacy,
	338	.rex = prefixes.rex,
	339	}, null);
	340	}
	341	}
	342	}
	343	return null;
	344	}
	345
	346	fn parseGpRegister(low_enc: u3, is_extended: bool, rex: Rex, bit_size: u64) Register {
	347	const reg_id: u4 = @as(u4, @intCast(@intFromBool(is_extended))) << 3 \| low_enc;
	348	const reg = @as(Register, @enumFromInt(reg_id)).toBitSize(bit_size);
	349	return switch (reg) {
	350	.spl => if (rex.present or rex.isSet()) .spl else .ah,
	351	.dil => if (rex.present or rex.isSet()) .dil else .bh,
	352	.bpl => if (rex.present or rex.isSet()) .bpl else .ch,
	353	.sil => if (rex.present or rex.isSet()) .sil else .dh,
	354	else => reg,
	355	};
	356	}
	357
	358	fn parseImm(dis: *Disassembler, kind: Encoding.Op) !Immediate {
	359	var stream = std.io.fixedBufferStream(dis.code[dis.pos..]);
	360	var creader = std.io.countingReader(stream.reader());
	361	const reader = creader.reader();
	362	const imm = switch (kind) {
	363	.imm8s, .rel8 => Immediate.s(try reader.readInt(i8, .Little)),
	364	.imm16s, .rel16 => Immediate.s(try reader.readInt(i16, .Little)),
	365	.imm32s, .rel32 => Immediate.s(try reader.readInt(i32, .Little)),
	366	.imm8 => Immediate.u(try reader.readInt(u8, .Little)),
	367	.imm16 => Immediate.u(try reader.readInt(u16, .Little)),
	368	.imm32 => Immediate.u(try reader.readInt(u32, .Little)),
	369	.imm64 => Immediate.u(try reader.readInt(u64, .Little)),
	370	else => unreachable,
	371	};
	372	dis.pos += std.math.cast(usize, creader.bytes_read) orelse return error.Overflow;
	373	return imm;
	374	}
	375
	376	fn parseOffset(dis: *Disassembler) !u64 {
	377	var stream = std.io.fixedBufferStream(dis.code[dis.pos..]);
	378	const reader = stream.reader();
	379	const offset = try reader.readInt(u64, .Little);
	380	dis.pos += 8;
	381	return offset;
	382	}
	383
	384	const ModRm = packed struct {
	385	mod: u2,
	386	op1: u3,
	387	op2: u3,
	388
	389	inline fn direct(self: ModRm) bool {
	390	return self.mod == 0b11;
	391	}
	392
	393	inline fn rip(self: ModRm) bool {
	394	return self.mod == 0 and self.op2 == 0b101;
	395	}
	396
	397	inline fn sib(self: ModRm) bool {
	398	return !self.direct() and self.op2 == 0b100;
	399	}
	400	};
	401
	402	fn parseModRmByte(dis: *Disassembler) !ModRm {
	403	if (dis.code[dis.pos..].len == 0) return error.EndOfStream;
	404	const modrm_byte = dis.code[dis.pos];
	405	dis.pos += 1;
	406	const mod: u2 = @as(u2, @truncate(modrm_byte >> 6));
	407	const op1: u3 = @as(u3, @truncate(modrm_byte >> 3));
	408	const op2: u3 = @as(u3, @truncate(modrm_byte));
	409	return ModRm{ .mod = mod, .op1 = op1, .op2 = op2 };
	410	}
	411
	412	fn segmentRegister(prefixes: LegacyPrefixes) Register {
	413	if (prefixes.prefix_2e) return .cs;
	414	if (prefixes.prefix_36) return .ss;
	415	if (prefixes.prefix_26) return .es;
	416	if (prefixes.prefix_64) return .fs;
	417	if (prefixes.prefix_65) return .gs;
	418	return .ds;
	419	}
	420
	421	const Sib = packed struct {
	422	scale: u2,
	423	index: u3,
	424	base: u3,
	425
	426	fn scaleIndex(self: Sib, rex: Rex) ?Memory.ScaleIndex {
	427	if (self.index == 0b100 and !rex.x) return null;
	428	return .{
	429	.scale = @as(u4, 1) << self.scale,
	430	.index = parseGpRegister(self.index, rex.x, rex, 64),
	431	};
	432	}
	433
	434	fn baseReg(self: Sib, modrm: ModRm, prefixes: Prefixes) ?Register {
	435	if (self.base == 0b101 and modrm.mod == 0) {
	436	if (self.scaleIndex(prefixes.rex)) \|_\| return null;
	437	return segmentRegister(prefixes.legacy);
	438	}
	439	return parseGpRegister(self.base, prefixes.rex.b, prefixes.rex, 64);
	440	}
	441	};
	442
	443	fn parseSibByte(dis: *Disassembler) !Sib {
	444	if (dis.code[dis.pos..].len == 0) return error.EndOfStream;
	445	const sib_byte = dis.code[dis.pos];
	446	dis.pos += 1;
	447	const scale: u2 = @as(u2, @truncate(sib_byte >> 6));
	448	const index: u3 = @as(u3, @truncate(sib_byte >> 3));
	449	const base: u3 = @as(u3, @truncate(sib_byte));
	450	return Sib{ .scale = scale, .index = index, .base = base };
	451	}
	452
	453	fn parseDisplacement(dis: *Disassembler, modrm: ModRm, sib: ?Sib) !i32 {
	454	var stream = std.io.fixedBufferStream(dis.code[dis.pos..]);
	455	var creader = std.io.countingReader(stream.reader());
	456	const reader = creader.reader();
	457	const disp = disp: {
	458	if (sib) \|info\| {
	459	if (info.base == 0b101 and modrm.mod == 0) {
	460	break :disp try reader.readInt(i32, .Little);
	461	}
	462	}
	463	if (modrm.rip()) {
	464	break :disp try reader.readInt(i32, .Little);
	465	}
	466	break :disp switch (modrm.mod) {
	467	0b00 => 0,
	468	0b01 => try reader.readInt(i8, .Little),
	469	0b10 => try reader.readInt(i32, .Little),
	470	0b11 => unreachable,
	471	};
	472	};
	473	dis.pos += std.math.cast(usize, creader.bytes_read) orelse return error.Overflow;
	474	return disp;
	475	}

src/codegen.zig+1-1

...	...	@@ -861,7 +861,7 @@ fn genDeclRef(
861	861	const sym_index = try elf_file.getOrCreateMetadataForDecl(decl_index);
862	862	const sym = elf_file.symbol(sym_index);
863	863	sym.flags.needs_got = true;
864		_ = try sym.getOrCreateGotEntry(elf_file);
	864	_ = try sym.getOrCreateGotEntry(sym_index, elf_file);
865	865	return GenResult.mcv(.{ .memory = sym.gotAddress(elf_file) });
866	866	} else if (bin_file.cast(link.File.MachO)) \|macho_file\| {
867	867	const atom_index = try macho_file.getOrCreateAtomForDecl(decl_index);

src/link/Elf.zig+194-23

...	...	@@ -40,6 +40,8 @@ phdr_got_index: ?u16 = null,
40	40	phdr_load_ro_index: ?u16 = null,
41	41	/// The index into the program headers of a PT_LOAD program header with Write flag
42	42	phdr_load_rw_index: ?u16 = null,
	43	/// The index into the program headers of a PT_LOAD program header with zerofill data.
	44	phdr_load_zerofill_index: ?u16 = null,
43	45
44	46	entry_addr: ?u64 = null,
45	47	page_size: u32,
...	...	@@ -56,6 +58,7 @@ got: GotSection = .{},
56	58	text_section_index: ?u16 = null,
57	59	rodata_section_index: ?u16 = null,
58	60	data_section_index: ?u16 = null,
	61	bss_section_index: ?u16 = null,
59	62	eh_frame_section_index: ?u16 = null,
60	63	eh_frame_hdr_section_index: ?u16 = null,
61	64	dynamic_section_index: ?u16 = null,
...	...	@@ -532,6 +535,26 @@ pub fn populateMissingMetadata(self: *Elf) !void {
532	535	self.phdr_table_dirty = true;
533	536	}
534	537
	538	if (self.phdr_load_zerofill_index == null) {
	539	self.phdr_load_zerofill_index = @as(u16, @intCast(self.phdrs.items.len));
	540	const p_align = if (self.base.options.target.os.tag == .linux) self.page_size else @as(u16, ptr_size);
	541	const off = self.phdrs.items[self.phdr_load_rw_index.?].p_offset;
	542	log.debug("found PT_LOAD zerofill free space 0x{x} to 0x{x}", .{ off, off });
	543	// TODO Same as for GOT
	544	const addr: u32 = if (self.base.options.target.ptrBitWidth() >= 32) 0x14000000 else 0xf000;
	545	try self.phdrs.append(gpa, .{
	546	.p_type = elf.PT_LOAD,
	547	.p_offset = off,
	548	.p_filesz = 0,
	549	.p_vaddr = addr,
	550	.p_paddr = addr,
	551	.p_memsz = 0,
	552	.p_align = p_align,
	553	.p_flags = elf.PF_R \| elf.PF_W,
	554	});
	555	self.phdr_table_dirty = true;
	556	}
	557
535	558	if (self.shstrtab_section_index == null) {
536	559	self.shstrtab_section_index = @as(u16, @intCast(self.shdrs.items.len));
537	560	assert(self.shstrtab.buffer.items.len == 0);
...	...	@@ -655,6 +678,26 @@ pub fn populateMissingMetadata(self: *Elf) !void {
655	678	self.shdr_table_dirty = true;
656	679	}
657	680
	681	if (self.bss_section_index == null) {
	682	self.bss_section_index = @as(u16, @intCast(self.shdrs.items.len));
	683	const phdr = &self.phdrs.items[self.phdr_load_zerofill_index.?];
	684	try self.shdrs.append(gpa, .{
	685	.sh_name = try self.shstrtab.insert(gpa, ".bss"),
	686	.sh_type = elf.SHT_NOBITS,
	687	.sh_flags = elf.SHF_WRITE \| elf.SHF_ALLOC,
	688	.sh_addr = phdr.p_vaddr,
	689	.sh_offset = phdr.p_offset,
	690	.sh_size = phdr.p_filesz,
	691	.sh_link = 0,
	692	.sh_info = 0,
	693	.sh_addralign = @as(u16, ptr_size),
	694	.sh_entsize = 0,
	695	});
	696	try self.phdr_to_shdr_table.putNoClobber(gpa, self.bss_section_index.?, self.phdr_load_zerofill_index.?);
	697	try self.last_atom_and_free_list_table.putNoClobber(gpa, self.bss_section_index.?, .{});
	698	self.shdr_table_dirty = true;
	699	}
	700
658	701	if (self.symtab_section_index == null) {
659	702	self.symtab_section_index = @as(u16, @intCast(self.shdrs.items.len));
660	703	const min_align: u16 = if (small_ptr) @alignOf(elf.Elf32_Sym) else @alignOf(elf.Elf64_Sym);
...	...	@@ -868,8 +911,9 @@ pub fn growAllocSection(self: *Elf, shdr_index: u16, needed_size: u64) !void {
868	911	const shdr = &self.shdrs.items[shdr_index];
869	912	const phdr_index = self.phdr_to_shdr_table.get(shdr_index).?;
870	913	const phdr = &self.phdrs.items[phdr_index];
	914	const is_zerofill = shdr.sh_type == elf.SHT_NOBITS;
871	915
872		if (needed_size > self.allocatedSize(shdr.sh_offset)) {
	916	if (needed_size > self.allocatedSize(shdr.sh_offset) and !is_zerofill) {
873	917	// Must move the entire section.
874	918	const new_offset = self.findFreeSpace(needed_size, self.page_size);
875	919	const existing_size = if (self.last_atom_and_free_list_table.get(shdr_index)) \|meta\| blk: {
...	...	@@ -893,7 +937,10 @@ pub fn growAllocSection(self: *Elf, shdr_index: u16, needed_size: u64) !void {
893	937
894	938	shdr.sh_size = needed_size;
895	939	phdr.p_memsz = needed_size;
896		phdr.p_filesz = needed_size;
	940
	941	if (!is_zerofill) {
	942	phdr.p_filesz = needed_size;
	943	}
897	944
898	945	self.markDirty(shdr_index, phdr_index);
899	946	}
...	...	@@ -1005,13 +1052,6 @@ pub fn flushModule(self: Elf, comp: Compilation, prog_node: *std.Progress.Node
1005	1052	const directory = self.base.options.emit.?.directory; // Just an alias to make it shorter to type.
1006	1053	const full_out_path = try directory.join(arena, &[_][]const u8{self.base.options.emit.?.sub_path});
1007	1054
1008		const compiler_rt_path: ?[]const u8 = blk: {
1009		if (comp.compiler_rt_lib) \|x\| break :blk x.full_object_path;
1010		if (comp.compiler_rt_obj) \|x\| break :blk x.full_object_path;
1011		break :blk null;
1012		};
1013		_ = compiler_rt_path;
1014
1015	1055	// Here we will parse input positional and library files (if referenced).
1016	1056	// This will roughly match in any linker backend we support.
1017	1057	var positionals = std.ArrayList(Compilation.LinkObject).init(arena);
...	...	@@ -1037,6 +1077,15 @@ pub fn flushModule(self: Elf, comp: Compilation, prog_node: *std.Progress.Node
1037	1077	try positionals.append(.{ .path = key.status.success.object_path });
1038	1078	}
1039	1079
	1080	const compiler_rt_path: ?[]const u8 = blk: {
	1081	if (comp.compiler_rt_lib) \|x\| break :blk x.full_object_path;
	1082	if (comp.compiler_rt_obj) \|x\| break :blk x.full_object_path;
	1083	break :blk null;
	1084	};
	1085	if (compiler_rt_path) \|path\| {
	1086	try positionals.append(.{ .path = path });
	1087	}
	1088
1040	1089	for (positionals.items) \|obj\| {
1041	1090	const in_file = try std.fs.cwd().openFile(obj.path, .{});
1042	1091	defer in_file.close();
...	...	@@ -1093,7 +1142,7 @@ pub fn flushModule(self: Elf, comp: Compilation, prog_node: *std.Progress.Node
1093	1142	// input Object files.
1094	1143	// Any qualifing unresolved symbol will be upgraded to an absolute, weak
1095	1144	// symbol for potential resolution at load-time.
1096		self.resolveSymbols();
	1145	try self.resolveSymbols();
1097	1146	self.markImportsExports();
1098	1147	self.claimUnresolved();
1099	1148
...	...	@@ -1356,6 +1405,7 @@ const ParseError = error{
1356	1405	EndOfStream,
1357	1406	FileSystem,
1358	1407	NotSupported,
	1408	InvalidCharacter,
1359	1409	} \|\| std.os.SeekError \|\| std.fs.File.OpenError \|\| std.fs.File.ReadError;
1360	1410
1361	1411	fn parsePositional(
...	...	@@ -1367,10 +1417,32 @@ fn parsePositional(
1367	1417	) ParseError!void {
1368	1418	const tracy = trace(@src());
1369	1419	defer tracy.end();
1370		_ = must_link;
1371	1420
1372	1421	if (Object.isObject(in_file)) {
1373	1422	try self.parseObject(in_file, path, ctx);
	1423	} else {
	1424	try self.parseLibrary(in_file, path, .{
	1425	.path = null,
	1426	.needed = false,
	1427	.weak = false,
	1428	}, must_link, ctx);
	1429	}
	1430	}
	1431
	1432	fn parseLibrary(
	1433	self: *Elf,
	1434	in_file: std.fs.File,
	1435	path: []const u8,
	1436	lib: link.SystemLib,
	1437	must_link: bool,
	1438	ctx: *ParseErrorCtx,
	1439	) ParseError!void {
	1440	const tracy = trace(@src());
	1441	defer tracy.end();
	1442	_ = lib;
	1443
	1444	if (Archive.isArchive(in_file)) {
	1445	try self.parseArchive(in_file, path, must_link, ctx);
1374	1446	} else return error.UnknownFileType;
1375	1447	}
1376	1448
...	...	@@ -1395,15 +1467,109 @@ fn parseObject(self: Elf, in_file: std.fs.File, path: []const u8, ctx: ParseEr
1395	1467	if (ctx.detected_cpu_arch != self.base.options.target.cpu.arch) return error.InvalidCpuArch;
1396	1468	}
1397	1469
1398		fn resolveSymbols(self: *Elf) void {
1399		if (self.zig_module_index) \|index\| {
1400		const zig_module = self.file(index).?.zig_module;
1401		zig_module.resolveSymbols(self);
	1470	fn parseArchive(
	1471	self: *Elf,
	1472	in_file: std.fs.File,
	1473	path: []const u8,
	1474	must_link: bool,
	1475	ctx: *ParseErrorCtx,
	1476	) ParseError!void {
	1477	const tracy = trace(@src());
	1478	defer tracy.end();
	1479
	1480	const gpa = self.base.allocator;
	1481	const data = try in_file.readToEndAlloc(gpa, std.math.maxInt(u32));
	1482	var archive = Archive{ .path = path, .data = data };
	1483	defer archive.deinit(gpa);
	1484	try archive.parse(self);
	1485
	1486	for (archive.objects.items) \|extracted\| {
	1487	const index = @as(File.Index, @intCast(try self.files.addOne(gpa)));
	1488	self.files.set(index, .{ .object = extracted });
	1489	const object = &self.files.items(.data)[index].object;
	1490	object.index = index;
	1491	object.alive = must_link;
	1492	try object.parse(self);
	1493	try self.objects.append(gpa, index);
	1494
	1495	ctx.detected_cpu_arch = object.header.?.e_machine.toTargetCpuArch().?;
	1496	if (ctx.detected_cpu_arch != self.base.options.target.cpu.arch) return error.InvalidCpuArch;
	1497	}
	1498	}
	1499
	1500	/// When resolving symbols, we approach the problem similarly to `mold`.
	1501	/// 1. Resolve symbols across all objects (including those preemptively extracted archives).
	1502	/// 2. Resolve symbols across all shared objects.
	1503	/// 3. Mark live objects (see `Elf.markLive`)
	1504	/// 4. Reset state of all resolved globals since we will redo this bit on the pruned set.
	1505	/// 5. Remove references to dead objects/shared objects
	1506	/// 6. Re-run symbol resolution on pruned objects and shared objects sets.
	1507	fn resolveSymbols(self: *Elf) error{Overflow}!void {
	1508	// Resolve symbols in the ZigModule. For now, we assume that it's always live.
	1509	if (self.zig_module_index) \|index\| self.file(index).?.resolveSymbols(self);
	1510	// Resolve symbols on the set of all objects and shared objects (even if some are unneeded).
	1511	for (self.objects.items) \|index\| self.file(index).?.resolveSymbols(self);
	1512
	1513	// Mark live objects.
	1514	self.markLive();
	1515
	1516	// Reset state of all globals after marking live objects.
	1517	if (self.zig_module_index) \|index\| self.file(index).?.resetGlobals(self);
	1518	for (self.objects.items) \|index\| self.file(index).?.resetGlobals(self);
	1519
	1520	// Prune dead objects and shared objects.
	1521	var i: usize = 0;
	1522	while (i < self.objects.items.len) {
	1523	const index = self.objects.items[i];
	1524	if (!self.file(index).?.isAlive()) {
	1525	_ = self.objects.orderedRemove(i);
	1526	} else i += 1;
	1527	}
	1528
	1529	// Dedup comdat groups.
	1530	for (self.objects.items) \|index\| {
	1531	const object = self.file(index).?.object;
	1532	for (object.comdat_groups.items) \|cg_index\| {
	1533	const cg = self.comdatGroup(cg_index);
	1534	const cg_owner = self.comdatGroupOwner(cg.owner);
	1535	const owner_file_index = if (self.file(cg_owner.file)) \|file_ptr\|
	1536	file_ptr.object.index
	1537	else
	1538	std.math.maxInt(File.Index);
	1539	cg_owner.file = @min(owner_file_index, index);
	1540	}
1402	1541	}
1403	1542
1404	1543	for (self.objects.items) \|index\| {
1405	1544	const object = self.file(index).?.object;
1406		object.resolveSymbols(self);
	1545	for (object.comdat_groups.items) \|cg_index\| {
	1546	const cg = self.comdatGroup(cg_index);
	1547	const cg_owner = self.comdatGroupOwner(cg.owner);
	1548	if (cg_owner.file != index) {
	1549	for (try object.comdatGroupMembers(cg.shndx)) \|shndx\| {
	1550	const atom_index = object.atoms.items[shndx];
	1551	if (self.atom(atom_index)) \|atom_ptr\| {
	1552	atom_ptr.alive = false;
	1553	// atom_ptr.markFdesDead(self);
	1554	}
	1555	}
	1556	}
	1557	}
	1558	}
	1559
	1560	// Re-resolve the symbols.
	1561	if (self.zig_module_index) \|index\| self.file(index).?.resolveSymbols(self);
	1562	for (self.objects.items) \|index\| self.file(index).?.resolveSymbols(self);
	1563	}
	1564
	1565	/// Traverses all objects and shared objects marking any object referenced by
	1566	/// a live object/shared object as alive itself.
	1567	/// This routine will prune unneeded objects extracted from archives and
	1568	/// unneeded shared objects.
	1569	fn markLive(self: *Elf) void {
	1570	for (self.objects.items) \|index\| {
	1571	const file_ptr = self.file(index).?;
	1572	if (file_ptr.isAlive()) file_ptr.markLive(self);
1407	1573	}
1408	1574	}
1409	1575
...	...	@@ -1477,11 +1643,11 @@ fn scanRelocs(self: *Elf) !void {
1477	1643
1478	1644	try self.reportUndefined(&undefs);
1479	1645
1480		for (self.symbols.items) \|*sym\| {
	1646	for (self.symbols.items, 0..) \|*sym, sym_index\| {
1481	1647	if (sym.flags.needs_got) {
1482	1648	log.debug("'{s}' needs GOT", .{sym.name(self)});
1483	1649	// TODO how can we tell we need to write it again, aka the entry is dirty?
1484		const gop = try sym.getOrCreateGotEntry(self);
	1650	const gop = try sym.getOrCreateGotEntry(@intCast(sym_index), self);
1485	1651	try self.got.writeEntry(self, gop.index);
1486	1652	}
1487	1653	}
...	...	@@ -1500,7 +1666,7 @@ fn allocateObjects(self: *Elf) !void {
1500	1666	const local = self.symbol(local_index);
1501	1667	const atom_ptr = local.atom(self) orelse continue;
1502	1668	if (!atom_ptr.alive) continue;
1503		local.value = atom_ptr.value;
	1669	local.value += atom_ptr.value;
1504	1670	}
1505	1671
1506	1672	for (object.globals()) \|global_index\| {
...	...	@@ -1508,7 +1674,7 @@ fn allocateObjects(self: *Elf) !void {
1508	1674	const atom_ptr = global.atom(self) orelse continue;
1509	1675	if (!atom_ptr.alive) continue;
1510	1676	if (global.file_index == index) {
1511		global.value = atom_ptr.value;
	1677	global.value += atom_ptr.value;
1512	1678	}
1513	1679	}
1514	1680	}
...	...	@@ -1524,7 +1690,11 @@ fn writeObjects(self: *Elf) !void {
1524	1690	if (!atom_ptr.alive) continue;
1525	1691
1526	1692	const shdr = &self.shdrs.items[atom_ptr.output_section_index];
	1693	if (shdr.sh_type == elf.SHT_NOBITS) continue;
	1694	if (shdr.sh_flags & elf.SHF_ALLOC == 0) continue; // TODO we don't yet know how to handle non-alloc sections
	1695
1527	1696	const file_offset = shdr.sh_offset + atom_ptr.value - shdr.sh_addr;
	1697	log.debug("writing atom({d}) at 0x{x}", .{ atom_ptr.atom_index, file_offset });
1528	1698	const code = try atom_ptr.codeInObjectUncompressAlloc(self);
1529	1699	defer gpa.free(code);
1530	1700
...	...	@@ -2529,7 +2699,7 @@ fn updateDeclCode(
2529	2699	esym.st_value = atom_ptr.value;
2530	2700
2531	2701	sym.flags.needs_got = true;
2532		const gop = try sym.getOrCreateGotEntry(self);
	2702	const gop = try sym.getOrCreateGotEntry(sym_index, self);
2533	2703	try self.got.writeEntry(self, gop.index);
2534	2704	}
2535	2705
...	...	@@ -2764,7 +2934,7 @@ fn updateLazySymbol(self: *Elf, sym: link.File.LazySymbol, symbol_index: Symbol.
2764	2934	local_esym.st_value = atom_ptr.value;
2765	2935
2766	2936	local_sym.flags.needs_got = true;
2767		const gop = try local_sym.getOrCreateGotEntry(self);
	2937	const gop = try local_sym.getOrCreateGotEntry(symbol_index, self);
2768	2938	try self.got.writeEntry(self, gop.index);
2769	2939
2770	2940	const section_offset = atom_ptr.value - self.phdrs.items[phdr_index].p_vaddr;
...	...	@@ -3635,7 +3805,7 @@ pub fn addSymbol(self: *Elf) !Symbol.Index {
3635	3805	break :blk index;
3636	3806	}
3637	3807	};
3638		self.symbols.items[index] = .{ .index = index };
	3808	self.symbols.items[index] = .{};
3639	3809	return index;
3640	3810	}
3641	3811
...	...	@@ -4012,6 +4182,7 @@ const synthetic_sections = @import("Elf/synthetic_sections.zig");
4012	4182
4013	4183	const Air = @import("../Air.zig");
4014	4184	const Allocator = std.mem.Allocator;
	4185	const Archive = @import("Elf/Archive.zig");
4015	4186	pub const Atom = @import("Elf/Atom.zig");
4016	4187	const Cache = std.Build.Cache;
4017	4188	const Compilation = @import("../Compilation.zig");

src/link/Elf/Archive.zig created+153

...	...	@@ -0,0 +1,153 @@
	1	path: []const u8,
	2	data: []const u8,
	3
	4	objects: std.ArrayListUnmanaged(Object) = .{},
	5	strtab: []const u8 = &[0]u8{},
	6
	7	// Archive files start with the ARMAG identifying string. Then follows a
	8	// `struct ar_hdr', and as many bytes of member file data as its `ar_size'
	9	// member indicates, for each member file.
	10	/// String that begins an archive file.
	11	pub const ARMAG: *const [SARMAG:0]u8 = "!<arch>\n";
	12	/// Size of that string.
	13	pub const SARMAG: u4 = 8;
	14
	15	/// String in ar_fmag at the end of each header.
	16	const ARFMAG: *const [2:0]u8 = "`\n";
	17
	18	const SYM64NAME: *const [7:0]u8 = "/SYM64/";
	19
	20	const ar_hdr = extern struct {
	21	/// Member file name, sometimes / terminated.
	22	ar_name: [16]u8,
	23
	24	/// File date, decimal seconds since Epoch.
	25	ar_date: [12]u8,
	26
	27	/// User ID, in ASCII format.
	28	ar_uid: [6]u8,
	29
	30	/// Group ID, in ASCII format.
	31	ar_gid: [6]u8,
	32
	33	/// File mode, in ASCII octal.
	34	ar_mode: [8]u8,
	35
	36	/// File size, in ASCII decimal.
	37	ar_size: [10]u8,
	38
	39	/// Always contains ARFMAG.
	40	ar_fmag: [2]u8,
	41
	42	fn date(self: ar_hdr) !u64 {
	43	const value = getValue(&self.ar_date);
	44	return std.fmt.parseInt(u64, value, 10);
	45	}
	46
	47	fn size(self: ar_hdr) !u32 {
	48	const value = getValue(&self.ar_size);
	49	return std.fmt.parseInt(u32, value, 10);
	50	}
	51
	52	fn getValue(raw: []const u8) []const u8 {
	53	return mem.trimRight(u8, raw, &[_]u8{@as(u8, 0x20)});
	54	}
	55
	56	fn isStrtab(self: ar_hdr) bool {
	57	return mem.eql(u8, getValue(&self.ar_name), "//");
	58	}
	59
	60	fn isSymtab(self: ar_hdr) bool {
	61	return mem.eql(u8, getValue(&self.ar_name), "/");
	62	}
	63	};
	64
	65	pub fn isArchive(file: std.fs.File) bool {
	66	const reader = file.reader();
	67	const magic = reader.readBytesNoEof(Archive.SARMAG) catch return false;
	68	defer file.seekTo(0) catch {};
	69	if (!mem.eql(u8, &magic, ARMAG)) return false;
	70	return true;
	71	}
	72
	73	pub fn deinit(self: *Archive, allocator: Allocator) void {
	74	allocator.free(self.data);
	75	self.objects.deinit(allocator);
	76	}
	77
	78	pub fn parse(self: Archive, elf_file: Elf) !void {
	79	const gpa = elf_file.base.allocator;
	80
	81	var stream = std.io.fixedBufferStream(self.data);
	82	const reader = stream.reader();
	83	_ = try reader.readBytesNoEof(SARMAG);
	84
	85	while (true) {
	86	if (stream.pos % 2 != 0) {
	87	stream.pos += 1;
	88	}
	89
	90	const hdr = reader.readStruct(ar_hdr) catch break;
	91
	92	if (!mem.eql(u8, &hdr.ar_fmag, ARFMAG)) {
	93	// TODO convert into an error
	94	log.debug(
	95	"{s}: invalid header delimiter: expected '{s}', found '{s}'",
	96	.{ self.path, std.fmt.fmtSliceEscapeLower(ARFMAG), std.fmt.fmtSliceEscapeLower(&hdr.ar_fmag) },
	97	);
	98	return;
	99	}
	100
	101	const size = try hdr.size();
	102	defer {
	103	_ = stream.seekBy(size) catch {};
	104	}
	105
	106	if (hdr.isSymtab()) continue;
	107	if (hdr.isStrtab()) {
	108	self.strtab = self.data[stream.pos..][0..size];
	109	continue;
	110	}
	111
	112	const name = ar_hdr.getValue(&hdr.ar_name);
	113
	114	if (mem.eql(u8, name, "__.SYMDEF") or mem.eql(u8, name, "__.SYMDEF SORTED")) continue;
	115
	116	const object_name = blk: {
	117	if (name[0] == '/') {
	118	const off = try std.fmt.parseInt(u32, name[1..], 10);
	119	break :blk self.getString(off);
	120	}
	121	break :blk name;
	122	};
	123
	124	const object = Object{
	125	.archive = self.path,
	126	.path = try gpa.dupe(u8, object_name[0 .. object_name.len - 1]), // To account for trailing '/'
	127	.data = try gpa.dupe(u8, self.data[stream.pos..][0..size]),
	128	.index = undefined,
	129	.alive = false,
	130	};
	131
	132	log.debug("extracting object '{s}' from archive '{s}'", .{ object.path, self.path });
	133
	134	try self.objects.append(gpa, object);
	135	}
	136	}
	137
	138	fn getString(self: Archive, off: u32) []const u8 {
	139	assert(off < self.strtab.len);
	140	return mem.sliceTo(@as([*:'\n']const u8, @ptrCast(self.strtab.ptr + off)), 0);
	141	}
	142
	143	const std = @import("std");
	144	const assert = std.debug.assert;
	145	const elf = std.elf;
	146	const fs = std.fs;
	147	const log = std.log.scoped(.link);
	148	const mem = std.mem;
	149
	150	const Allocator = mem.Allocator;
	151	const Archive = @This();
	152	const Elf = @import("../Elf.zig");
	153	const Object = @import("Object.zig");

src/link/Elf/Atom.zig+94-8

...	...	@@ -322,11 +322,12 @@ pub fn scanRelocs(self: Atom, elf_file: *Elf, undefs: anytype) !void {
322	322
323	323	if (rel.r_type() == elf.R_X86_64_NONE) continue;
324	324
325		const symbol = switch (file_ptr) {
326		.zig_module => \|x\| elf_file.symbol(x.symbol(rel.r_sym())),
327		.object => \|x\| elf_file.symbol(x.symbols.items[rel.r_sym()]),
	325	const symbol_index = switch (file_ptr) {
	326	.zig_module => \|x\| x.symbol(rel.r_sym()),
	327	.object => \|x\| x.symbols.items[rel.r_sym()],
328	328	else => unreachable,
329	329	};
	330	const symbol = elf_file.symbol(symbol_index);
330	331
331	332	// Check for violation of One Definition Rule for COMDATs.
332	333	if (symbol.file(elf_file) == null) {
...	...	@@ -340,7 +341,7 @@ pub fn scanRelocs(self: Atom, elf_file: *Elf, undefs: anytype) !void {
340	341	}
341	342
342	343	// Report an undefined symbol.
343		try self.reportUndefined(elf_file, symbol, rel, undefs);
	344	try self.reportUndefined(elf_file, symbol, symbol_index, rel, undefs);
344	345
345	346	// While traversing relocations, mark symbols that require special handling such as
346	347	// pointer indirection via GOT, or a stub trampoline via PLT.
...	...	@@ -379,7 +380,14 @@ pub fn scanRelocs(self: Atom, elf_file: *Elf, undefs: anytype) !void {
379	380	}
380	381
381	382	// This function will report any undefined non-weak symbols that are not imports.
382		fn reportUndefined(self: Atom, elf_file: Elf, sym: const Symbol, rel: elf.Elf64_Rela, undefs: anytype) !void {
	383	fn reportUndefined(
	384	self: Atom,
	385	elf_file: *Elf,
	386	sym: *const Symbol,
	387	sym_index: Symbol.Index,
	388	rel: elf.Elf64_Rela,
	389	undefs: anytype,
	390	) !void {
383	391	const rel_esym = switch (elf_file.file(self.file_index).?) {
384	392	.zig_module => \|x\| x.elfSym(rel.r_sym()).*,
385	393	.object => \|x\| x.symtab[rel.r_sym()],
...	...	@@ -392,7 +400,7 @@ fn reportUndefined(self: Atom, elf_file: Elf, sym: const Symbol, rel: elf.Elf6
392	400	!sym.flags.import and
393	401	esym.st_shndx == elf.SHN_UNDEF)
394	402	{
395		const gop = try undefs.getOrPut(sym.index);
	403	const gop = try undefs.getOrPut(sym_index);
396	404	if (!gop.found_existing) {
397	405	gop.value_ptr.* = std.ArrayList(Atom.Index).init(elf_file.base.allocator);
398	406	}
...	...	@@ -417,6 +425,7 @@ pub fn resolveRelocs(self: Atom, elf_file: *Elf, code: []u8) !void {
417	425	.object => \|x\| elf_file.symbol(x.symbols.items[rel.r_sym()]),
418	426	else => unreachable,
419	427	};
	428	const r_offset = std.math.cast(usize, rel.r_offset) orelse return error.Overflow;
420	429
421	430	// We will use equation format to resolve relocations:
422	431	// https://intezer.com/blog/malware-analysis/executable-and-linkable-format-101-part-3-relocations/
...	...	@@ -446,24 +455,49 @@ pub fn resolveRelocs(self: Atom, elf_file: *Elf, code: []u8) !void {
446	455
447	456	relocs_log.debug(" {s}: {x}: [{x} => {x}] G({x}) ({s})", .{
448	457	fmtRelocType(r_type),
449		rel.r_offset,
	458	r_offset,
450	459	P,
451	460	S + A,
452	461	G + GOT + A,
453	462	target.name(elf_file),
454	463	});
455	464
456		try stream.seekTo(rel.r_offset);
	465	try stream.seekTo(r_offset);
457	466
458	467	switch (rel.r_type()) {
459	468	elf.R_X86_64_NONE => unreachable,
460	469
461	470	elf.R_X86_64_64 => try cwriter.writeIntLittle(i64, S + A),
462	471
	472	elf.R_X86_64_32 => try cwriter.writeIntLittle(u32, @as(u32, @truncate(@as(u64, @intCast(S + A))))),
	473	elf.R_X86_64_32S => try cwriter.writeIntLittle(i32, @as(i32, @truncate(S + A))),
	474
463	475	elf.R_X86_64_PLT32,
464	476	elf.R_X86_64_PC32,
465	477	=> try cwriter.writeIntLittle(i32, @as(i32, @intCast(S + A - P))),
466	478
	479	elf.R_X86_64_GOTPCREL => try cwriter.writeIntLittle(i32, @as(i32, @intCast(G + GOT + A - P))),
	480	elf.R_X86_64_GOTPC32 => try cwriter.writeIntLittle(i32, @as(i32, @intCast(GOT + A - P))),
	481	elf.R_X86_64_GOTPC64 => try cwriter.writeIntLittle(i64, GOT + A - P),
	482
	483	elf.R_X86_64_GOTPCRELX => {
	484	if (!target.flags.import and !target.isIFunc(elf_file) and !target.isAbs(elf_file)) blk: {
	485	x86_64.relaxGotpcrelx(code[r_offset - 2 ..]) catch break :blk;
	486	try cwriter.writeIntLittle(i32, @as(i32, @intCast(S + A - P)));
	487	continue;
	488	}
	489	try cwriter.writeIntLittle(i32, @as(i32, @intCast(G + GOT + A - P)));
	490	},
	491
	492	elf.R_X86_64_REX_GOTPCRELX => {
	493	if (!target.flags.import and !target.isIFunc(elf_file) and !target.isAbs(elf_file)) blk: {
	494	x86_64.relaxRexGotpcrelx(code[r_offset - 3 ..]) catch break :blk;
	495	try cwriter.writeIntLittle(i32, @as(i32, @intCast(S + A - P)));
	496	continue;
	497	}
	498	try cwriter.writeIntLittle(i32, @as(i32, @intCast(G + GOT + A - P)));
	499	},
	500
467	501	else => {
468	502	log.err("TODO: unhandled relocation type {}", .{fmtRelocType(rel.r_type())});
469	503	@panic("TODO unhandled relocation type");
...	...	@@ -591,6 +625,58 @@ fn format2(
591	625	// future.
592	626	pub const Index = u16;
593	627
	628	const x86_64 = struct {
	629	pub fn relaxGotpcrelx(code: []u8) !void {
	630	const old_inst = disassemble(code) orelse return error.RelaxFail;
	631	const inst = switch (old_inst.encoding.mnemonic) {
	632	.call => try Instruction.new(old_inst.prefix, .call, &.{
	633	// TODO: hack to force imm32s in the assembler
	634	.{ .imm = Immediate.s(-129) },
	635	}),
	636	.jmp => try Instruction.new(old_inst.prefix, .jmp, &.{
	637	// TODO: hack to force imm32s in the assembler
	638	.{ .imm = Immediate.s(-129) },
	639	}),
	640	else => return error.RelaxFail,
	641	};
	642	relocs_log.debug(" relaxing {} => {}", .{ old_inst.encoding, inst.encoding });
	643	const nop = try Instruction.new(.none, .nop, &.{});
	644	encode(&.{ nop, inst }, code) catch return error.RelaxFail;
	645	}
	646
	647	pub fn relaxRexGotpcrelx(code: []u8) !void {
	648	const old_inst = disassemble(code) orelse return error.RelaxFail;
	649	switch (old_inst.encoding.mnemonic) {
	650	.mov => {
	651	const inst = try Instruction.new(old_inst.prefix, .lea, &old_inst.ops);
	652	relocs_log.debug(" relaxing {} => {}", .{ old_inst.encoding, inst.encoding });
	653	encode(&.{inst}, code) catch return error.RelaxFail;
	654	},
	655	else => return error.RelaxFail,
	656	}
	657	}
	658
	659	fn disassemble(code: []const u8) ?Instruction {
	660	var disas = Disassembler.init(code);
	661	const inst = disas.next() catch return null;
	662	return inst;
	663	}
	664
	665	fn encode(insts: []const Instruction, code: []u8) !void {
	666	var stream = std.io.fixedBufferStream(code);
	667	const writer = stream.writer();
	668	for (insts) \|inst\| {
	669	try inst.encode(writer, .{});
	670	}
	671	}
	672
	673	const bits = @import("../../arch/x86_64/bits.zig");
	674	const encoder = @import("../../arch/x86_64/encoder.zig");
	675	const Disassembler = @import("../../arch/x86_64/Disassembler.zig");
	676	const Immediate = bits.Immediate;
	677	const Instruction = encoder.Instruction;
	678	};
	679
594	680	const std = @import("std");
595	681	const assert = std.debug.assert;
596	682	const elf = std.elf;

src/link/Elf/Object.zig+3-3

...	...	@@ -485,9 +485,9 @@ pub fn claimUnresolved(self: Object, elf_file: Elf) void {
485	485	pub fn resetGlobals(self: Object, elf_file: Elf) void {
486	486	for (self.globals()) \|index\| {
487	487	const global = elf_file.symbol(index);
488		const name = global.name;
	488	const off = global.name_offset;
489	489	global.* = .{};
490		global.name = name;
	490	global.name_offset = off;
491	491	}
492	492	}
493	493
...	...	@@ -499,7 +499,7 @@ pub fn markLive(self: Object, elf_file: Elf) void {
499	499	if (sym.st_bind() == elf.STB_WEAK) continue;
500	500
501	501	const global = elf_file.symbol(index);
502		const file = global.getFile(elf_file) orelse continue;
	502	const file = global.file(elf_file) orelse continue;
503	503	const should_keep = sym.st_shndx == elf.SHN_UNDEF or
504	504	(sym.st_shndx == elf.SHN_COMMON and global.elfSym(elf_file).st_shndx != elf.SHN_COMMON);
505	505	if (should_keep and !file.isAlive()) {

src/link/Elf/Symbol.zig+3-5

...	...	@@ -1,7 +1,5 @@
1	1	//! Represents a defined symbol.
2	2
3		index: Index = 0,
4
5	3	/// Allocated address value of this symbol.
6	4	value: u64 = 0,
7	5
...	...	@@ -117,10 +115,10 @@ const GetOrCreateGotEntryResult = struct {
117	115	index: GotSection.Index,
118	116	};
119	117
120		pub fn getOrCreateGotEntry(symbol: Symbol, elf_file: Elf) !GetOrCreateGotEntryResult {
	118	pub fn getOrCreateGotEntry(symbol: Symbol, symbol_index: Index, elf_file: Elf) !GetOrCreateGotEntryResult {
121	119	assert(symbol.flags.needs_got);
122	120	if (symbol.flags.has_got) return .{ .found_existing = true, .index = symbol.extra(elf_file).?.got };
123		const index = try elf_file.got.addGotSymbol(symbol.index, elf_file);
	121	const index = try elf_file.got.addGotSymbol(symbol_index, elf_file);
124	122	symbol.flags.has_got = true;
125	123	return .{ .found_existing = false, .index = index };
126	124	}
...	...	@@ -270,7 +268,7 @@ fn format2(
270	268	_ = options;
271	269	_ = unused_fmt_string;
272	270	const symbol = ctx.symbol;
273		try writer.print("%{d} : {s} : @{x}", .{ symbol.index, symbol.fmtName(ctx.elf_file), symbol.value });
	271	try writer.print("%{d} : {s} : @{x}", .{ symbol.esym_index, symbol.fmtName(ctx.elf_file), symbol.value });
274	272	if (symbol.file(ctx.elf_file)) \|file_ptr\| {
275	273	if (symbol.isAbs(ctx.elf_file)) {
276	274	if (symbol.elfSym(ctx.elf_file).st_shndx == elf.SHN_UNDEF) {

src/link/Elf/ZigModule.zig+9

...	...	@@ -148,6 +148,15 @@ pub fn scanRelocs(self: ZigModule, elf_file: Elf, undefs: anytype) !void {
148	148	}
149	149	}
150	150
	151	pub fn resetGlobals(self: ZigModule, elf_file: Elf) void {
	152	for (self.globals()) \|index\| {
	153	const global = elf_file.symbol(index);
	154	const off = global.name_offset;
	155	global.* = .{};
	156	global.name_offset = off;
	157	}
	158	}
	159
151	160	pub fn updateSymtabSize(self: ZigModule, elf_file: Elf) void {
152	161	for (self.locals()) \|local_index\| {
153	162	const local = elf_file.symbol(local_index);

src/link/Elf/file.zig+14-1

...	...	@@ -62,6 +62,19 @@ pub const File = union(enum) {
62	62	return (@as(u32, base) << 24) + file.index();
63	63	}
64	64
	65	pub fn resolveSymbols(file: File, elf_file: *Elf) void {
	66	switch (file) {
	67	inline else => \|x\| x.resolveSymbols(elf_file),
	68	}
	69	}
	70
	71	pub fn resetGlobals(file: File, elf_file: *Elf) void {
	72	switch (file) {
	73	.linker_defined => unreachable,
	74	inline else => \|x\| x.resetGlobals(elf_file),
	75	}
	76	}
	77
65	78	pub fn setAlive(file: File) void {
66	79	switch (file) {
67	80	.zig_module, .linker_defined => {},
...	...	@@ -71,7 +84,7 @@ pub const File = union(enum) {
71	84
72	85	pub fn markLive(file: File, elf_file: *Elf) void {
73	86	switch (file) {
74		.zig_module, .linker_defined => {},
	87	.zig_module, .linker_defined => unreachable,
75	88	inline else => \|x\| x.markLive(elf_file),
76	89	}
77	90	}