diff options
| author | Jakub Konka <kubkon@jakubkonka.com> | 2024-04-16 14:25:42 +0200 |
|---|---|---|
| committer | Jakub Konka <kubkon@jakubkonka.com> | 2024-04-20 23:36:41 +0200 |
| commit | 9e0bca73e2ae8c29873c7280f28ce58502e6abab (patch) | |
| tree | 3fd87d1fd8eca202b357d134327f90dbd1aa177e /src | |
| parent | 09820a96b691f79403291d9b2179adbae50fc33b (diff) | |
| download | zig-9e0bca73e2ae8c29873c7280f28ce58502e6abab.tar.gz zig-9e0bca73e2ae8c29873c7280f28ce58502e6abab.zip | |
link/elf: implement string merging
Diffstat (limited to 'src')
| -rw-r--r-- | src/link/Elf.zig | 274 | ||||
| -rw-r--r-- | src/link/Elf/Atom.zig | 199 | ||||
| -rw-r--r-- | src/link/Elf/Object.zig | 225 | ||||
| -rw-r--r-- | src/link/Elf/SharedObject.zig | 2 | ||||
| -rw-r--r-- | src/link/Elf/Symbol.zig | 64 | ||||
| -rw-r--r-- | src/link/Elf/ZigObject.zig | 22 | ||||
| -rw-r--r-- | src/link/Elf/gc.zig | 8 | ||||
| -rw-r--r-- | src/link/Elf/relocatable.zig | 19 | ||||
| -rw-r--r-- | src/link/Elf/synthetic_sections.zig | 77 | ||||
| -rw-r--r-- | src/link/Elf/thunks.zig | 31 | ||||
| -rw-r--r-- | src/link/MachO/Atom.zig | 2 | ||||
| -rw-r--r-- | src/link/MachO/synthetic.zig | 10 | ||||
| -rw-r--r-- | src/link/MachO/thunks.zig | 2 | ||||
| -rw-r--r-- | src/link/aarch64.zig | 2 |
14 files changed, 711 insertions, 226 deletions
diff --git a/src/link/Elf.zig b/src/link/Elf.zig index b8829bebe9..c6d6d74436 100644 --- a/src/link/Elf.zig +++ b/src/link/Elf.zig @@ -210,6 +210,14 @@ atoms_extra: std.ArrayListUnmanaged(u32) = .{}, /// List of range extension thunks. thunks: std.ArrayListUnmanaged(Thunk) = .{}, +/// List of output merge sections with deduped contents. +merge_sections: std.ArrayListUnmanaged(MergeSection) = .{}, +/// List of output merge subsections. +/// Each subsection is akin to Atom but belongs to a MergeSection. +merge_subsections: std.ArrayListUnmanaged(MergeSubsection) = .{}, +/// List of input merge sections as parsed from input relocatables. +merge_input_sections: std.ArrayListUnmanaged(InputMergeSection) = .{}, + /// Table of last atom index in a section and matching atom free list if any. last_atom_and_free_list_table: LastAtomAndFreeListTable = .{}, @@ -380,6 +388,8 @@ pub fn createEmpty( _ = try self.addSection(.{ .name = "" }); // Append null symbol in output symtab try self.symtab.append(gpa, null_sym); + // Append null input merge section. + try self.merge_input_sections.append(gpa, .{}); if (!is_obj_or_ar) { try self.dynstrtab.append(gpa, 0); @@ -498,6 +508,15 @@ pub fn deinit(self: *Elf) void { th.deinit(gpa); } self.thunks.deinit(gpa); + for (self.merge_sections.items) |*sect| { + sect.deinit(gpa); + } + self.merge_sections.deinit(gpa); + self.merge_subsections.deinit(gpa); + for (self.merge_input_sections.items) |*sect| { + sect.deinit(gpa); + } + self.merge_input_sections.deinit(gpa); for (self.last_atom_and_free_list_table.values()) |*value| { value.free_list.deinit(gpa); } @@ -1295,6 +1314,7 @@ pub fn flushModule(self: *Elf, arena: Allocator, prog_node: *std.Progress.Node) // symbol for potential resolution at load-time. self.resolveSymbols(); self.markEhFrameAtomsDead(); + try self.resolveMergeSections(); try self.convertCommonSymbols(); self.markImportsExports(); @@ -1319,6 +1339,8 @@ pub fn flushModule(self: *Elf, arena: Allocator, prog_node: *std.Progress.Node) else => |e| return e, }; + try self.addCommentString(); + try self.sortMergeSections(); try self.initOutputSections(); try self.addLinkerDefinedSymbols(); self.claimUnresolved(); @@ -1338,6 +1360,7 @@ pub fn flushModule(self: *Elf, arena: Allocator, prog_node: *std.Progress.Node) self.sortDynamicSymtab(); try self.setHashSections(); try self.setVersionSymtab(); + try self.updateMergeSectionSizes(); try self.updateSectionSizes(); try self.allocatePhdrTable(); @@ -1365,7 +1388,7 @@ pub fn flushModule(self: *Elf, arena: Allocator, prog_node: *std.Progress.Node) if (shdr.sh_type == elf.SHT_NOBITS) continue; const code = try zig_object.codeAlloc(self, atom_index); defer gpa.free(code); - const file_offset = shdr.sh_offset + atom_ptr.value; + const file_offset = shdr.sh_offset + @as(u64, @intCast(atom_ptr.value)); atom_ptr.resolveRelocsAlloc(self, code) catch |err| switch (err) { error.RelocFailure, error.RelaxFailure => has_reloc_errors = true, error.UnsupportedCpuArch => { @@ -1383,6 +1406,7 @@ pub fn flushModule(self: *Elf, arena: Allocator, prog_node: *std.Progress.Node) try self.writePhdrTable(); try self.writeShdrTable(); try self.writeAtoms(); + try self.writeMergeSections(); self.writeSyntheticSections() catch |err| switch (err) { error.RelocFailure => return error.FlushFailure, error.UnsupportedCpuArch => { @@ -2952,7 +2976,10 @@ pub fn writeElfHeader(self: *Elf) !void { mem.writeInt(u32, hdr_buf[index..][0..4], 1, endian); index += 4; - const e_entry = if (self.entry_index) |entry_index| self.symbol(entry_index).address(.{}, self) else 0; + const e_entry = if (self.entry_index) |entry_index| + @as(u64, @intCast(self.symbol(entry_index).address(.{}, self))) + else + 0; const phdr_table_offset = if (self.phdr_table_index) |phndx| self.phdrs.items[phndx].p_offset else 0; switch (self.ptr_width) { .p32 => { @@ -3138,14 +3165,14 @@ fn allocateLinkerDefinedSymbols(self: *Elf) void { if (self.dynamic_section_index) |shndx| { const shdr = &self.shdrs.items[shndx]; const symbol_ptr = self.symbol(self.dynamic_index.?); - symbol_ptr.value = shdr.sh_addr; + symbol_ptr.value = @intCast(shdr.sh_addr); symbol_ptr.output_section_index = shndx; } // __ehdr_start { const symbol_ptr = self.symbol(self.ehdr_start_index.?); - symbol_ptr.value = self.image_base; + symbol_ptr.value = @intCast(self.image_base); symbol_ptr.output_section_index = 1; } @@ -3155,9 +3182,9 @@ fn allocateLinkerDefinedSymbols(self: *Elf) void { const end_sym = self.symbol(self.init_array_end_index.?); const shdr = &self.shdrs.items[shndx]; start_sym.output_section_index = shndx; - start_sym.value = shdr.sh_addr; + start_sym.value = @intCast(shdr.sh_addr); end_sym.output_section_index = shndx; - end_sym.value = shdr.sh_addr + shdr.sh_size; + end_sym.value = @intCast(shdr.sh_addr + shdr.sh_size); } // __fini_array_start, __fini_array_end @@ -3166,9 +3193,9 @@ fn allocateLinkerDefinedSymbols(self: *Elf) void { const end_sym = self.symbol(self.fini_array_end_index.?); const shdr = &self.shdrs.items[shndx]; start_sym.output_section_index = shndx; - start_sym.value = shdr.sh_addr; + start_sym.value = @intCast(shdr.sh_addr); end_sym.output_section_index = shndx; - end_sym.value = shdr.sh_addr + shdr.sh_size; + end_sym.value = @intCast(shdr.sh_addr + shdr.sh_size); } // __preinit_array_start, __preinit_array_end @@ -3177,9 +3204,9 @@ fn allocateLinkerDefinedSymbols(self: *Elf) void { const end_sym = self.symbol(self.preinit_array_end_index.?); const shdr = &self.shdrs.items[shndx]; start_sym.output_section_index = shndx; - start_sym.value = shdr.sh_addr; + start_sym.value = @intCast(shdr.sh_addr); end_sym.output_section_index = shndx; - end_sym.value = shdr.sh_addr + shdr.sh_size; + end_sym.value = @intCast(shdr.sh_addr + shdr.sh_size); } // _GLOBAL_OFFSET_TABLE_ @@ -3187,14 +3214,14 @@ fn allocateLinkerDefinedSymbols(self: *Elf) void { if (self.got_plt_section_index) |shndx| { const shdr = self.shdrs.items[shndx]; const sym = self.symbol(self.got_index.?); - sym.value = shdr.sh_addr; + sym.value = @intCast(shdr.sh_addr); sym.output_section_index = shndx; } } else { if (self.got_section_index) |shndx| { const shdr = self.shdrs.items[shndx]; const sym = self.symbol(self.got_index.?); - sym.value = shdr.sh_addr; + sym.value = @intCast(shdr.sh_addr); sym.output_section_index = shndx; } } @@ -3203,7 +3230,7 @@ fn allocateLinkerDefinedSymbols(self: *Elf) void { if (self.plt_section_index) |shndx| { const shdr = &self.shdrs.items[shndx]; const symbol_ptr = self.symbol(self.plt_index.?); - symbol_ptr.value = shdr.sh_addr; + symbol_ptr.value = @intCast(shdr.sh_addr); symbol_ptr.output_section_index = shndx; } @@ -3211,7 +3238,7 @@ fn allocateLinkerDefinedSymbols(self: *Elf) void { if (self.dso_handle_index) |index| { const shdr = &self.shdrs.items[1]; const symbol_ptr = self.symbol(index); - symbol_ptr.value = shdr.sh_addr; + symbol_ptr.value = @intCast(shdr.sh_addr); symbol_ptr.output_section_index = 0; } @@ -3219,7 +3246,7 @@ fn allocateLinkerDefinedSymbols(self: *Elf) void { if (self.eh_frame_hdr_section_index) |shndx| { const shdr = &self.shdrs.items[shndx]; const symbol_ptr = self.symbol(self.gnu_eh_frame_hdr_index.?); - symbol_ptr.value = shdr.sh_addr; + symbol_ptr.value = @intCast(shdr.sh_addr); symbol_ptr.output_section_index = shndx; } @@ -3231,9 +3258,9 @@ fn allocateLinkerDefinedSymbols(self: *Elf) void { const start_addr = end_addr - self.calcNumIRelativeRelocs() * @sizeOf(elf.Elf64_Rela); const start_sym = self.symbol(self.rela_iplt_start_index.?); const end_sym = self.symbol(self.rela_iplt_end_index.?); - start_sym.value = start_addr; + start_sym.value = @intCast(start_addr); start_sym.output_section_index = shndx; - end_sym.value = end_addr; + end_sym.value = @intCast(end_addr); end_sym.output_section_index = shndx; } @@ -3242,7 +3269,7 @@ fn allocateLinkerDefinedSymbols(self: *Elf) void { const end_symbol = self.symbol(self.end_index.?); for (self.shdrs.items, 0..) |shdr, shndx| { if (shdr.sh_flags & elf.SHF_ALLOC != 0) { - end_symbol.value = shdr.sh_addr + shdr.sh_size; + end_symbol.value = @intCast(shdr.sh_addr + shdr.sh_size); end_symbol.output_section_index = @intCast(shndx); } } @@ -3257,9 +3284,9 @@ fn allocateLinkerDefinedSymbols(self: *Elf) void { const stop = self.symbol(self.start_stop_indexes.items[index + 1]); const shndx = self.sectionByName(name["__start_".len..]).?; const shdr = &self.shdrs.items[shndx]; - start.value = shdr.sh_addr; + start.value = @intCast(shdr.sh_addr); start.output_section_index = shndx; - stop.value = shdr.sh_addr + shdr.sh_size; + stop.value = @intCast(shdr.sh_addr + shdr.sh_size); stop.output_section_index = shndx; } } @@ -3269,7 +3296,7 @@ fn allocateLinkerDefinedSymbols(self: *Elf) void { const sym = self.symbol(index); if (self.sectionByName(".sdata")) |shndx| { const shdr = self.shdrs.items[shndx]; - sym.value = shdr.sh_addr + 0x800; + sym.value = @intCast(shdr.sh_addr + 0x800); sym.output_section_index = shndx; } else { sym.value = 0; @@ -3299,10 +3326,109 @@ fn checkDuplicates(self: *Elf) !void { try self.reportDuplicates(dupes); } +pub fn addCommentString(self: *Elf) !void { + const msec_index = try self.getOrCreateMergeSection(".comment", elf.SHF_MERGE | elf.SHF_STRINGS, elf.SHT_PROGBITS); + const msec = self.mergeSection(msec_index); + const res = try msec.insertZ(self.base.comp.gpa, "zig version x.x.x"); // TODO get actual version + if (res.found_existing) return; + const msub_index = try self.addMergeSubsection(); + const msub = self.mergeSubsection(msub_index); + msub.merge_section_index = msec_index; + msub.string_index = res.key.pos; + msub.alignment = .@"1"; + msub.size = res.key.len; + msub.alive = true; + res.sub.* = msub_index; +} + +fn resolveMergeSections(self: *Elf) !void { + const tracy = trace(@src()); + defer tracy.end(); + + var has_errors = false; + for (self.objects.items) |index| { + const file_ptr = self.file(index).?; + if (!file_ptr.isAlive()) continue; + file_ptr.object.initMergeSections(self) catch |err| switch (err) { + error.MalformedObject => has_errors = true, + else => |e| return e, + }; + } + + if (has_errors) return error.FlushFailure; + + for (self.objects.items) |index| { + const file_ptr = self.file(index).?; + if (!file_ptr.isAlive()) continue; + file_ptr.object.resolveMergeSubsections(self) catch |err| switch (err) { + error.MalformedObject => has_errors = true, + else => |e| return e, + }; + } + + if (has_errors) return error.FlushFailure; +} + +pub fn sortMergeSections(self: *Elf) !void { + for (self.merge_sections.items) |*msec| { + try msec.sort(self); + } +} + +pub fn updateMergeSectionSizes(self: *Elf) !void { + for (self.merge_sections.items) |*msec| { + const shdr = &self.shdrs.items[msec.output_section_index]; + for (msec.subsections.items) |msub_index| { + const msub = self.mergeSubsection(msub_index); + assert(msub.alive); + const offset = msub.alignment.forward(shdr.sh_size); + const padding = offset - shdr.sh_size; + msub.value = @intCast(offset); + shdr.sh_size += padding + msub.size; + shdr.sh_addralign = @max(shdr.sh_addralign, msub.alignment.toByteUnits() orelse 1); + } + } +} + +pub fn writeMergeSections(self: *Elf) !void { + const gpa = self.base.comp.gpa; + var buffer = std.ArrayList(u8).init(gpa); + defer buffer.deinit(); + + for (self.merge_sections.items) |msec| { + const shdr = self.shdrs.items[msec.output_section_index]; + + try buffer.ensureTotalCapacity(shdr.sh_size); + buffer.appendNTimesAssumeCapacity(0, shdr.sh_size); + + for (msec.subsections.items) |msub_index| { + const msub = self.mergeSubsection(msub_index); + assert(msub.alive); + const string = msub.getString(self); + const off: u64 = @intCast(msub.value); + @memcpy(buffer.items[off..][0..string.len], string); + } + + try self.base.file.?.pwriteAll(buffer.items, shdr.sh_offset); + buffer.clearRetainingCapacity(); + } +} + fn initOutputSections(self: *Elf) !void { for (self.objects.items) |index| { try self.file(index).?.object.initOutputSections(self); } + + for (self.merge_sections.items) |*msec| { + if (msec.subsections.items.len == 0) continue; + const name = msec.name(self); + const shndx = self.sectionByName(name) orelse try self.addSection(.{ + .name = name, + .type = msec.type, + .flags = msec.flags, + }); + msec.output_section_index = shndx; + } } fn initSyntheticSections(self: *Elf) !void { @@ -3971,6 +4097,10 @@ fn resetShdrIndexes(self: *Elf, backlinks: []const u32) !void { } } + for (self.merge_sections.items) |*msec| { + msec.output_section_index = backlinks[msec.output_section_index]; + } + { var output_rela_sections = try self.output_rela_sections.clone(gpa); defer output_rela_sections.deinit(gpa); @@ -4058,7 +4188,7 @@ fn updateSectionSizes(self: *Elf) !void { if (!atom_ptr.flags.alive) continue; const offset = atom_ptr.alignment.forward(shdr.sh_size); const padding = offset - shdr.sh_size; - atom_ptr.value = offset; + atom_ptr.value = @intCast(offset); shdr.sh_size += padding + atom_ptr.size; shdr.sh_addralign = @max(shdr.sh_addralign, atom_ptr.alignment.toByteUnits() orelse 1); } @@ -4541,7 +4671,7 @@ fn writeAtoms(self: *Elf) !void { const atom_ptr = self.atom(atom_index).?; assert(atom_ptr.flags.alive); - const offset = math.cast(usize, atom_ptr.value - base_offset) orelse + const offset = math.cast(usize, atom_ptr.value - @as(i64, @intCast(base_offset))) orelse return error.Overflow; const size = math.cast(usize, atom_ptr.size) orelse return error.Overflow; @@ -4582,7 +4712,7 @@ fn writeAtoms(self: *Elf) !void { const thunk_size = th.size(self); try buffer.ensureUnusedCapacity(thunk_size); const shdr = self.shdrs.items[th.output_section_index]; - const offset = th.value + shdr.sh_offset; + const offset = @as(u64, @intCast(th.value)) + shdr.sh_offset; try th.write(self, buffer.writer()); assert(buffer.items.len == thunk_size); try self.base.file.?.pwriteAll(buffer.items, offset); @@ -4617,6 +4747,7 @@ pub fn updateSymtabSize(self: *Elf) !void { if (self.eh_frame_section_index) |_| { nlocals += 1; } + nlocals += @intCast(self.merge_sections.items.len); if (self.requiresThunks()) for (self.thunks.items) |*th| { th.output_symtab_ctx.ilocal = nlocals + 1; @@ -4953,12 +5084,29 @@ fn writeSectionSymbols(self: *Elf) void { }; ilocal += 1; } + + for (self.merge_sections.items) |msec| { + const shdr = self.shdrs.items[msec.output_section_index]; + const out_sym = &self.symtab.items[ilocal]; + out_sym.* = .{ + .st_name = 0, + .st_value = shdr.sh_addr, + .st_info = elf.STT_SECTION, + .st_shndx = @intCast(msec.output_section_index), + .st_size = 0, + .st_other = 0, + }; + ilocal += 1; + } } pub fn sectionSymbolOutputSymtabIndex(self: Elf, shndx: u32) u32 { if (self.eh_frame_section_index) |index| { if (index == shndx) return @intCast(self.output_sections.keys().len + 1); } + for (self.merge_sections.items, 1..) |msec, index| { + if (msec.output_section_index == shndx) return @intCast(self.output_sections.keys().len + 1 + index); + } return @intCast(self.output_sections.getIndex(shndx).? + 1); } @@ -5687,35 +5835,88 @@ pub fn comdatGroupOwner(self: *Elf, index: ComdatGroupOwner.Index) *ComdatGroupO return &self.comdat_groups_owners.items[index]; } -pub fn gotAddress(self: *Elf) u64 { +pub fn addInputMergeSection(self: *Elf) !InputMergeSection.Index { + const index: InputMergeSection.Index = @intCast(self.merge_input_sections.items.len); + const msec = try self.merge_input_sections.addOne(self.base.comp.gpa); + msec.* = .{}; + return index; +} + +pub fn inputMergeSection(self: *Elf, index: InputMergeSection.Index) ?*InputMergeSection { + if (index == 0) return null; + return &self.merge_input_sections.items[index]; +} + +pub fn addMergeSubsection(self: *Elf) !MergeSubsection.Index { + const index: MergeSubsection.Index = @intCast(self.merge_subsections.items.len); + const msec = try self.merge_subsections.addOne(self.base.comp.gpa); + msec.* = .{}; + return index; +} + +pub fn mergeSubsection(self: *Elf, index: MergeSubsection.Index) *MergeSubsection { + assert(index < self.merge_subsections.items.len); + return &self.merge_subsections.items[index]; +} + +pub fn getOrCreateMergeSection(self: *Elf, name: []const u8, flags: u64, @"type": u32) !MergeSection.Index { + const gpa = self.base.comp.gpa; + const out_name = name: { + if (self.base.isRelocatable()) break :name name; + if (mem.eql(u8, name, ".rodata") or mem.startsWith(u8, name, ".rodata")) + break :name if (flags & elf.SHF_STRINGS != 0) ".rodata.str" else ".rodata.cst"; + break :name name; + }; + const out_off = try self.strings.insert(gpa, out_name); + const out_flags = flags & ~@as(u64, elf.SHF_COMPRESSED | elf.SHF_GROUP); + for (self.merge_sections.items, 0..) |msec, index| { + if (msec.name_offset == out_off) return @intCast(index); + } + const index = @as(MergeSection.Index, @intCast(self.merge_sections.items.len)); + const msec = try self.merge_sections.addOne(gpa); + msec.* = .{ + .name_offset = out_off, + .flags = out_flags, + .type = @"type", + }; + return index; +} + +pub fn mergeSection(self: *Elf, index: MergeSection.Index) *MergeSection { + assert(index < self.merge_sections.items.len); + return &self.merge_sections.items[index]; +} + +pub fn gotAddress(self: *Elf) i64 { const shndx = blk: { if (self.getTarget().cpu.arch == .x86_64 and self.got_plt_section_index != null) break :blk self.got_plt_section_index.?; break :blk if (self.got_section_index) |shndx| shndx else null; }; - return if (shndx) |index| self.shdrs.items[index].sh_addr else 0; + return if (shndx) |index| @intCast(self.shdrs.items[index].sh_addr) else 0; } -pub fn tpAddress(self: *Elf) u64 { +pub fn tpAddress(self: *Elf) i64 { const index = self.phdr_tls_index orelse return 0; const phdr = self.phdrs.items[index]; - return switch (self.getTarget().cpu.arch) { + const addr = switch (self.getTarget().cpu.arch) { .x86_64 => mem.alignForward(u64, phdr.p_vaddr + phdr.p_memsz, phdr.p_align), .aarch64 => mem.alignBackward(u64, phdr.p_vaddr - 16, phdr.p_align), else => @panic("TODO implement getTpAddress for this arch"), }; + return @intCast(addr); } -pub fn dtpAddress(self: *Elf) u64 { +pub fn dtpAddress(self: *Elf) i64 { const index = self.phdr_tls_index orelse return 0; const phdr = self.phdrs.items[index]; - return phdr.p_vaddr; + return @intCast(phdr.p_vaddr); } -pub fn tlsAddress(self: *Elf) u64 { +pub fn tlsAddress(self: *Elf) i64 { const index = self.phdr_tls_index orelse return 0; const phdr = self.phdrs.items[index]; - return phdr.p_vaddr; + return @intCast(phdr.p_vaddr); } const ErrorWithNotes = struct { @@ -6093,6 +6294,11 @@ fn fmtDumpState( try writer.print(" shdr({d}) : COMDAT({d})\n", .{ cg.shndx, cg.cg_index }); } + try writer.writeAll("\nOutput merge sections\n"); + for (self.merge_sections.items) |msec| { + try writer.print(" shdr({d}) : {}\n", .{ msec.output_section_index, msec.fmt(self) }); + } + try writer.writeAll("\nOutput shdrs\n"); for (self.shdrs.items, 0..) |shdr, shndx| { try writer.print(" shdr({d}) : phdr({?d}) : {}\n", .{ @@ -6285,6 +6491,7 @@ const gc = @import("Elf/gc.zig"); const glibc = @import("../glibc.zig"); const link = @import("../link.zig"); const lldMain = @import("../main.zig").lldMain; +const merge_section = @import("Elf/merge_section.zig"); const musl = @import("../musl.zig"); const relocatable = @import("Elf/relocatable.zig"); const relocation = @import("Elf/relocation.zig"); @@ -6310,10 +6517,13 @@ const GnuHashSection = synthetic_sections.GnuHashSection; const GotSection = synthetic_sections.GotSection; const GotPltSection = synthetic_sections.GotPltSection; const HashSection = synthetic_sections.HashSection; +const InputMergeSection = merge_section.InputMergeSection; const LdScript = @import("Elf/LdScript.zig"); const LinkerDefined = @import("Elf/LinkerDefined.zig"); const Liveness = @import("../Liveness.zig"); const LlvmObject = @import("../codegen/llvm.zig").Object; +const MergeSection = merge_section.MergeSection; +const MergeSubsection = merge_section.MergeSubsection; const Module = @import("../Module.zig"); const Object = @import("Elf/Object.zig"); const InternPool = @import("../InternPool.zig"); diff --git a/src/link/Elf/Atom.zig b/src/link/Elf/Atom.zig index 68b1f701dc..d01f4c8427 100644 --- a/src/link/Elf/Atom.zig +++ b/src/link/Elf/Atom.zig @@ -1,5 +1,5 @@ /// Address allocated for this Atom. -value: u64 = 0, +value: i64 = 0, /// Name of this Atom. name_offset: u32 = 0, @@ -44,10 +44,22 @@ pub fn name(self: Atom, elf_file: *Elf) []const u8 { }; } -pub fn address(self: Atom, elf_file: *Elf) u64 { +pub fn address(self: Atom, elf_file: *Elf) i64 { const shndx = self.outputShndx() orelse return self.value; const shdr = elf_file.shdrs.items[shndx]; - return shdr.sh_addr + self.value; + return @as(i64, @intCast(shdr.sh_addr)) + self.value; +} + +pub fn debugTombstoneValue(self: Atom, target: Symbol, elf_file: *Elf) ?u64 { + if (target.mergeSubsection(elf_file)) |msub| { + if (msub.alive) return null; + } + if (target.atom(elf_file)) |atom_ptr| { + if (atom_ptr.flags.alive) return null; + } + const atom_name = self.name(elf_file); + if (!mem.startsWith(u8, atom_name, ".debug")) return null; + return if (mem.eql(u8, atom_name, ".debug_loc") or mem.eql(u8, atom_name, ".debug_ranges")) 1 else 0; } pub fn file(self: Atom, elf_file: *Elf) ?File { @@ -91,13 +103,13 @@ pub fn capacity(self: Atom, elf_file: *Elf) u64 { next.address(elf_file) else std.math.maxInt(u32); - return next_addr - self.address(elf_file); + return @intCast(next_addr - self.address(elf_file)); } pub fn freeListEligible(self: Atom, elf_file: *Elf) bool { // No need to keep a free list node for the last block. const next = elf_file.atom(self.next_index) orelse return false; - const cap = next.address(elf_file) - self.address(elf_file); + const cap: u64 = @intCast(next.address(elf_file) - self.address(elf_file)); const ideal_cap = Elf.padToIdeal(self.size); if (cap <= ideal_cap) return false; const surplus = cap - ideal_cap; @@ -130,8 +142,8 @@ pub fn allocate(self: *Atom, elf_file: *Elf) !void { // Is it enough that we could fit this new atom? const cap = big_atom.capacity(elf_file); const ideal_capacity = Elf.padToIdeal(cap); - const ideal_capacity_end_vaddr = std.math.add(u64, big_atom.value, ideal_capacity) catch ideal_capacity; - const capacity_end_vaddr = big_atom.value + cap; + const ideal_capacity_end_vaddr = std.math.add(u64, @intCast(big_atom.value), ideal_capacity) catch ideal_capacity; + const capacity_end_vaddr = @as(u64, @intCast(big_atom.value)) + cap; const new_start_vaddr_unaligned = capacity_end_vaddr - new_atom_ideal_capacity; const new_start_vaddr = self.alignment.backward(new_start_vaddr_unaligned); if (new_start_vaddr < ideal_capacity_end_vaddr) { @@ -156,14 +168,14 @@ pub fn allocate(self: *Atom, elf_file: *Elf) !void { if (!keep_free_list_node) { free_list_removal = i; } - break :blk new_start_vaddr; + break :blk @intCast(new_start_vaddr); } else if (elf_file.atom(last_atom_index.*)) |last| { const ideal_capacity = Elf.padToIdeal(last.size); - const ideal_capacity_end_vaddr = last.value + ideal_capacity; + const ideal_capacity_end_vaddr = @as(u64, @intCast(last.value)) + ideal_capacity; const new_start_vaddr = self.alignment.forward(ideal_capacity_end_vaddr); // Set up the metadata to be updated, after errors are no longer possible. atom_placement = last.atom_index; - break :blk new_start_vaddr; + break :blk @intCast(new_start_vaddr); } else { break :blk 0; } @@ -173,7 +185,7 @@ pub fn allocate(self: *Atom, elf_file: *Elf) !void { self.atom_index, self.name(elf_file), self.address(elf_file), - self.address(elf_file) + self.size, + self.address(elf_file) + @as(i64, @intCast(self.size)), }); const expand_section = if (atom_placement) |placement_index| @@ -181,7 +193,7 @@ pub fn allocate(self: *Atom, elf_file: *Elf) !void { else true; if (expand_section) { - const needed_size = self.value + self.size; + const needed_size: u64 = @intCast(self.value + @as(i64, @intCast(self.size))); try elf_file.growAllocSection(self.outputShndx().?, needed_size); last_atom_index.* = self.atom_index; @@ -231,7 +243,7 @@ pub fn shrink(self: *Atom, elf_file: *Elf) void { } pub fn grow(self: *Atom, elf_file: *Elf) !void { - if (!self.alignment.check(self.value) or self.size > self.capacity(elf_file)) + if (!self.alignment.check(@intCast(self.value)) or self.size > self.capacity(elf_file)) try self.allocate(elf_file); } @@ -321,11 +333,14 @@ pub fn writeRelocs(self: Atom, elf_file: *Elf, out_relocs: *std.ArrayList(elf.El }; const target = elf_file.symbol(target_index); const r_type = rel.r_type(); - const r_offset = self.value + rel.r_offset; + const r_offset: u64 = @intCast(self.value + @as(i64, @intCast(rel.r_offset))); var r_addend = rel.r_addend; var r_sym: u32 = 0; switch (target.type(elf_file)) { - elf.STT_SECTION => { + elf.STT_SECTION => if (target.mergeSubsection(elf_file)) |msub| { + r_addend += @intCast(target.address(.{}, elf_file)); + r_sym = elf_file.sectionSymbolOutputSymtabIndex(msub.mergeSection(elf_file).output_section_index); + } else { r_addend += @intCast(target.address(.{}, elf_file)); r_sym = elf_file.sectionSymbolOutputSymtabIndex(target.outputShndx().?); }, @@ -412,6 +427,12 @@ pub fn scanRelocs(self: Atom, elf_file: *Elf, code: ?[]const u8, undefs: anytype }; const symbol = elf_file.symbol(symbol_index); + const is_synthetic_symbol = switch (file_ptr) { + .zig_object => false, // TODO: implement this once we support merge sections in ZigObject + .object => |x| rel.r_sym() >= x.symtab.items.len, + else => unreachable, + }; + // Check for violation of One Definition Rule for COMDATs. if (symbol.file(elf_file) == null) { // TODO convert into an error @@ -424,7 +445,8 @@ pub fn scanRelocs(self: Atom, elf_file: *Elf, code: ?[]const u8, undefs: anytype } // Report an undefined symbol. - if (try self.reportUndefined(elf_file, symbol, symbol_index, rel, undefs)) continue; + if (!is_synthetic_symbol and (try self.reportUndefined(elf_file, symbol, symbol_index, rel, undefs))) + continue; if (symbol.isIFunc(elf_file)) { symbol.flags.needs_got = true; @@ -733,21 +755,21 @@ pub fn resolveRelocsAlloc(self: Atom, elf_file: *Elf, code: []u8) RelocError!voi // https://intezer.com/blog/malware-analysis/executable-and-linkable-format-101-part-3-relocations/ // // Address of the source atom. - const P = @as(i64, @intCast(self.address(elf_file) + rel.r_offset)); + const P = self.address(elf_file) + @as(i64, @intCast(rel.r_offset)); // Addend from the relocation. const A = rel.r_addend; // Address of the target symbol - can be address of the symbol within an atom or address of PLT stub. - const S = @as(i64, @intCast(target.address(.{}, elf_file))); + const S = target.address(.{}, elf_file); // Address of the global offset table. - const GOT = @as(i64, @intCast(elf_file.gotAddress())); + const GOT = elf_file.gotAddress(); // Address of the .zig.got table entry if any. - const ZIG_GOT = @as(i64, @intCast(target.zigGotAddress(elf_file))); + const ZIG_GOT = target.zigGotAddress(elf_file); // Relative offset to the start of the global offset table. - const G = @as(i64, @intCast(target.gotAddress(elf_file))) - GOT; + const G = target.gotAddress(elf_file) - GOT; // // Address of the thread pointer. - const TP = @as(i64, @intCast(elf_file.tpAddress())); + const TP = elf_file.tpAddress(); // Address of the dynamic thread pointer. - const DTP = @as(i64, @intCast(elf_file.dtpAddress())); + const DTP = elf_file.dtpAddress(); relocs_log.debug(" {s}: {x}: [{x} => {x}] G({x}) ZG({x}) ({s})", .{ relocation.fmtRelocType(rel.r_type(), cpu_arch), @@ -804,9 +826,9 @@ fn resolveDynAbsReloc( const comp = elf_file.base.comp; const gpa = comp.gpa; const cpu_arch = elf_file.getTarget().cpu.arch; - const P = self.address(elf_file) + rel.r_offset; + const P: u64 = @intCast(self.address(elf_file) + @as(i64, @intCast(rel.r_offset))); const A = rel.r_addend; - const S = @as(i64, @intCast(target.address(.{}, elf_file))); + const S = target.address(.{}, elf_file); const is_writeable = self.inputShdr(elf_file).sh_flags & elf.SHF_WRITE != 0; const num_dynrelocs = switch (self.file(elf_file).?) { @@ -874,7 +896,7 @@ fn resolveDynAbsReloc( }, .ifunc => { - const S_ = @as(i64, @intCast(target.address(.{ .plt = false }, elf_file))); + const S_ = target.address(.{ .plt = false }, elf_file); elf_file.addRelaDynAssumeCapacity(.{ .offset = P, .type = relocation.encode(.irel, cpu_arch), @@ -914,6 +936,11 @@ pub fn resolveRelocsNonAlloc(self: Atom, elf_file: *Elf, code: []u8, undefs: any else => unreachable, }; const target = elf_file.symbol(target_index); + const is_synthetic_symbol = switch (file_ptr) { + .zig_object => false, // TODO: implement this once we support merge sections in ZigObject + .object => |x| rel.r_sym() >= x.symtab.items.len, + else => unreachable, + }; // Check for violation of One Definition Rule for COMDATs. if (target.file(elf_file) == null) { @@ -927,20 +954,21 @@ pub fn resolveRelocsNonAlloc(self: Atom, elf_file: *Elf, code: []u8, undefs: any } // Report an undefined symbol. - if (try self.reportUndefined(elf_file, target, target_index, rel, undefs)) continue; + if (!is_synthetic_symbol and (try self.reportUndefined(elf_file, target, target_index, rel, undefs))) + continue; // We will use equation format to resolve relocations: // https://intezer.com/blog/malware-analysis/executable-and-linkable-format-101-part-3-relocations/ // - const P = @as(i64, @intCast(self.address(elf_file) + rel.r_offset)); + const P = self.address(elf_file) + @as(i64, @intCast(rel.r_offset)); // Addend from the relocation. const A = rel.r_addend; // Address of the target symbol - can be address of the symbol within an atom or address of PLT stub. - const S = @as(i64, @intCast(target.address(.{}, elf_file))); + const S = target.address(.{}, elf_file); // Address of the global offset table. - const GOT = @as(i64, @intCast(elf_file.gotAddress())); + const GOT = elf_file.gotAddress(); // Address of the dynamic thread pointer. - const DTP = @as(i64, @intCast(elf_file.dtpAddress())); + const DTP = elf_file.dtpAddress(); const args = ResolveArgs{ P, A, S, GOT, 0, 0, DTP, 0 }; @@ -1261,10 +1289,10 @@ const x86_64 = struct { .TLSGD => { if (target.flags.has_tlsgd) { - const S_ = @as(i64, @intCast(target.tlsGdAddress(elf_file))); + const S_ = target.tlsGdAddress(elf_file); try cwriter.writeInt(i32, @as(i32, @intCast(S_ + A - P)), .little); } else if (target.flags.has_gottp) { - const S_ = @as(i64, @intCast(target.gotTpAddress(elf_file))); + const S_ = target.gotTpAddress(elf_file); try x86_64.relaxTlsGdToIe(atom, &.{ rel, it.next().? }, @intCast(S_ - P), elf_file, stream); } else { try x86_64.relaxTlsGdToLe( @@ -1280,13 +1308,13 @@ const x86_64 = struct { .TLSLD => { if (elf_file.got.tlsld_index) |entry_index| { const tlsld_entry = elf_file.got.entries.items[entry_index]; - const S_ = @as(i64, @intCast(tlsld_entry.address(elf_file))); + const S_ = tlsld_entry.address(elf_file); try cwriter.writeInt(i32, @as(i32, @intCast(S_ + A - P)), .little); } else { try x86_64.relaxTlsLdToLe( atom, &.{ rel, it.next().? }, - @as(i32, @intCast(TP - @as(i64, @intCast(elf_file.tlsAddress())))), + @as(i32, @intCast(TP - elf_file.tlsAddress())), elf_file, stream, ); @@ -1295,7 +1323,7 @@ const x86_64 = struct { .GOTPC32_TLSDESC => { if (target.flags.has_tlsdesc) { - const S_ = @as(i64, @intCast(target.tlsDescAddress(elf_file))); + const S_ = target.tlsDescAddress(elf_file); try cwriter.writeInt(i32, @as(i32, @intCast(S_ + A - P)), .little); } else { x86_64.relaxGotPcTlsDesc(code[r_offset - 3 ..]) catch { @@ -1319,7 +1347,7 @@ const x86_64 = struct { .GOTTPOFF => { if (target.flags.has_gottp) { - const S_ = @as(i64, @intCast(target.gotTpAddress(elf_file))); + const S_ = target.gotTpAddress(elf_file); try cwriter.writeInt(i32, @as(i32, @intCast(S_ + A - P)), .little); } else { x86_64.relaxGotTpOff(code[r_offset - 3 ..]); @@ -1362,9 +1390,18 @@ const x86_64 = struct { .@"16" => try cwriter.writeInt(u16, @as(u16, @bitCast(@as(i16, @intCast(S + A)))), .little), .@"32" => try cwriter.writeInt(u32, @as(u32, @bitCast(@as(i32, @intCast(S + A)))), .little), .@"32S" => try cwriter.writeInt(i32, @as(i32, @intCast(S + A)), .little), - .@"64" => try cwriter.writeInt(i64, S + A, .little), - .DTPOFF32 => try cwriter.writeInt(i32, @as(i32, @intCast(S + A - DTP)), .little), - .DTPOFF64 => try cwriter.writeInt(i64, S + A - DTP, .little), + .@"64" => if (atom.debugTombstoneValue(target.*, elf_file)) |value| + try cwriter.writeInt(u64, value, .little) + else + try cwriter.writeInt(i64, S + A, .little), + .DTPOFF32 => if (atom.debugTombstoneValue(target.*, elf_file)) |value| + try cwriter.writeInt(u64, value, .little) + else + try cwriter.writeInt(i32, @as(i32, @intCast(S + A - DTP)), .little), + .DTPOFF64 => if (atom.debugTombstoneValue(target.*, elf_file)) |value| + try cwriter.writeInt(u64, value, .little) + else + try cwriter.writeInt(i64, S + A - DTP, .little), .GOTOFF64 => try cwriter.writeInt(i64, S + A - GOT, .little), .GOTPC64 => try cwriter.writeInt(i64, GOT + A, .little), .SIZE32 => { @@ -1746,7 +1783,7 @@ const aarch64 = struct { .object => |x| x.symbols.items[rel.r_sym()], else => unreachable, }; - const S_: i64 = @intCast(th.targetAddress(target_index, elf_file)); + const S_ = th.targetAddress(target_index, elf_file); break :blk math.cast(i28, S_ + A - P) orelse return error.Overflow; }; aarch64_util.writeBranchImm(disp, code); @@ -1764,16 +1801,12 @@ const aarch64 = struct { .ADR_PREL_PG_HI21 => { // TODO: check for relaxation of ADRP+ADD - const saddr = @as(u64, @intCast(P)); - const taddr = @as(u64, @intCast(S + A)); - const pages = @as(u21, @bitCast(try aarch64_util.calcNumberOfPages(saddr, taddr))); + const pages = @as(u21, @bitCast(try aarch64_util.calcNumberOfPages(P, S + A))); aarch64_util.writeAdrpInst(pages, code); }, .ADR_GOT_PAGE => if (target.flags.has_got) { - const saddr = @as(u64, @intCast(P)); - const taddr = @as(u64, @intCast(G + GOT + A)); - const pages = @as(u21, @bitCast(try aarch64_util.calcNumberOfPages(saddr, taddr))); + const pages = @as(u21, @bitCast(try aarch64_util.calcNumberOfPages(P, G + GOT + A))); aarch64_util.writeAdrpInst(pages, code); } else { // TODO: relax @@ -1828,46 +1861,38 @@ const aarch64 = struct { }, .TLSIE_ADR_GOTTPREL_PAGE21 => { - const S_: i64 = @intCast(target.gotTpAddress(elf_file)); - const saddr: u64 = @intCast(P); - const taddr: u64 = @intCast(S_ + A); - relocs_log.debug(" [{x} => {x}]", .{ P, taddr }); - const pages: u21 = @bitCast(try aarch64_util.calcNumberOfPages(saddr, taddr)); + const S_ = target.gotTpAddress(elf_file); + relocs_log.debug(" [{x} => {x}]", .{ P, S_ + A }); + const pages: u21 = @bitCast(try aarch64_util.calcNumberOfPages(P, S_ + A)); aarch64_util.writeAdrpInst(pages, code); }, .TLSIE_LD64_GOTTPREL_LO12_NC => { - const S_: i64 = @intCast(target.gotTpAddress(elf_file)); - const taddr: u64 = @intCast(S_ + A); - relocs_log.debug(" [{x} => {x}]", .{ P, taddr }); - const offset: u12 = try math.divExact(u12, @truncate(taddr), 8); + const S_ = target.gotTpAddress(elf_file); + relocs_log.debug(" [{x} => {x}]", .{ P, S_ + A }); + const offset: u12 = try math.divExact(u12, @truncate(@as(u64, @bitCast(S_ + A))), 8); aarch64_util.writeLoadStoreRegInst(offset, code); }, .TLSGD_ADR_PAGE21 => { - const S_: i64 = @intCast(target.tlsGdAddress(elf_file)); - const saddr: u64 = @intCast(P); - const taddr: u64 = @intCast(S_ + A); - relocs_log.debug(" [{x} => {x}]", .{ P, taddr }); - const pages: u21 = @bitCast(try aarch64_util.calcNumberOfPages(saddr, taddr)); + const S_ = target.tlsGdAddress(elf_file); + relocs_log.debug(" [{x} => {x}]", .{ P, S_ + A }); + const pages: u21 = @bitCast(try aarch64_util.calcNumberOfPages(P, S_ + A)); aarch64_util.writeAdrpInst(pages, code); }, .TLSGD_ADD_LO12_NC => { - const S_: i64 = @intCast(target.tlsGdAddress(elf_file)); - const taddr: u64 = @intCast(S_ + A); - relocs_log.debug(" [{x} => {x}]", .{ P, taddr }); - const offset: u12 = @truncate(taddr); + const S_ = target.tlsGdAddress(elf_file); + relocs_log.debug(" [{x} => {x}]", .{ P, S_ + A }); + const offset: u12 = @truncate(@as(u64, @bitCast(S_ + A))); aarch64_util.writeAddImmInst(offset, code); }, .TLSDESC_ADR_PAGE21 => { if (target.flags.has_tlsdesc) { - const S_: i64 = @intCast(target.tlsDescAddress(elf_file)); - const saddr: u64 = @intCast(P); - const taddr: u64 = @intCast(S_ + A); - relocs_log.debug(" [{x} => {x}]", .{ P, taddr }); - const pages: u21 = @bitCast(try aarch64_util.calcNumberOfPages(saddr, taddr)); + const S_ = target.tlsDescAddress(elf_file); + relocs_log.debug(" [{x} => {x}]", .{ P, S_ + A }); + const pages: u21 = @bitCast(try aarch64_util.calcNumberOfPages(P, S_ + A)); aarch64_util.writeAdrpInst(pages, code); } else { relocs_log.debug(" relaxing adrp => nop", .{}); @@ -1877,10 +1902,9 @@ const aarch64 = struct { .TLSDESC_LD64_LO12 => { if (target.flags.has_tlsdesc) { - const S_: i64 = @intCast(target.tlsDescAddress(elf_file)); - const taddr: u64 = @intCast(S_ + A); - relocs_log.debug(" [{x} => {x}]", .{ P, taddr }); - const offset: u12 = try math.divExact(u12, @truncate(taddr), 8); + const S_ = target.tlsDescAddress(elf_file); + relocs_log.debug(" [{x} => {x}]", .{ P, S_ + A }); + const offset: u12 = try math.divExact(u12, @truncate(@as(u64, @bitCast(S_ + A))), 8); aarch64_util.writeLoadStoreRegInst(offset, code); } else { relocs_log.debug(" relaxing ldr => nop", .{}); @@ -1890,10 +1914,9 @@ const aarch64 = struct { .TLSDESC_ADD_LO12 => { if (target.flags.has_tlsdesc) { - const S_: i64 = @intCast(target.tlsDescAddress(elf_file)); - const taddr: u64 = @intCast(S_ + A); - relocs_log.debug(" [{x} => {x}]", .{ P, taddr }); - const offset: u12 = @truncate(taddr); + const S_ = target.tlsDescAddress(elf_file); + relocs_log.debug(" [{x} => {x}]", .{ P, S_ + A }); + const offset: u12 = @truncate(@as(u64, @bitCast(S_ + A))); aarch64_util.writeAddImmInst(offset, code); } else { const old_inst = Instruction{ @@ -1938,7 +1961,6 @@ const aarch64 = struct { ) !void { _ = it; _ = code; - _ = target; const r_type: elf.R_AARCH64 = @enumFromInt(rel.r_type()); const cwriter = stream.writer(); @@ -1948,7 +1970,10 @@ const aarch64 = struct { switch (r_type) { .NONE => unreachable, .ABS32 => try cwriter.writeInt(i32, @as(i32, @intCast(S + A)), .little), - .ABS64 => try cwriter.writeInt(i64, S + A, .little), + .ABS64 => if (atom.debugTombstoneValue(target.*, elf_file)) |value| + try cwriter.writeInt(u64, value, .little) + else + try cwriter.writeInt(i64, S + A, .little), else => try atom.reportUnhandledRelocError(rel, elf_file), } } @@ -2073,7 +2098,7 @@ const riscv = struct { const atom_addr = atom.address(elf_file); const pos = it.pos; const pair = while (it.prev()) |pair| { - if (S == atom_addr + pair.r_offset) break pair; + if (S == atom_addr + @as(i64, @intCast(pair.r_offset))) break pair; } else { // TODO: implement searching forward var err = try elf_file.addErrorWithNotes(1); @@ -2091,10 +2116,10 @@ const riscv = struct { .object => |x| elf_file.symbol(x.symbols.items[pair.r_sym()]), else => unreachable, }; - const S_ = @as(i64, @intCast(target_.address(.{}, elf_file))); + const S_ = target_.address(.{}, elf_file); const A_ = pair.r_addend; - const P_ = @as(i64, @intCast(atom_addr + pair.r_offset)); - const G_ = @as(i64, @intCast(target_.gotAddress(elf_file))) - GOT; + const P_ = atom_addr + @as(i64, @intCast(pair.r_offset)); + const G_ = target_.gotAddress(elf_file) - GOT; const disp = switch (@as(elf.R_RISCV, @enumFromInt(pair.r_type()))) { .PCREL_HI20 => math.cast(i32, S_ + A_ - P_) orelse return error.Overflow, .GOT_HI20 => math.cast(i32, G_ + GOT + A_ - P_) orelse return error.Overflow, @@ -2122,7 +2147,6 @@ const riscv = struct { code: []u8, stream: anytype, ) !void { - _ = target; _ = it; const r_type: elf.R_RISCV = @enumFromInt(rel.r_type()); @@ -2137,7 +2161,10 @@ const riscv = struct { .NONE => unreachable, .@"32" => try cwriter.writeInt(i32, @as(i32, @intCast(S + A)), .little), - .@"64" => try cwriter.writeInt(i64, S + A, .little), + .@"64" => if (atom.debugTombstoneValue(target.*, elf_file)) |value| + try cwriter.writeInt(u64, value, .little) + else + try cwriter.writeInt(i64, S + A, .little), .ADD8 => riscv_util.writeAddend(i8, .add, code[r_offset..][0..1], S + A), .SUB8 => riscv_util.writeAddend(i8, .sub, code[r_offset..][0..1], S + A), diff --git a/src/link/Elf/Object.zig b/src/link/Elf/Object.zig index 9bbc503238..391ef9a55e 100644 --- a/src/link/Elf/Object.zig +++ b/src/link/Elf/Object.zig @@ -15,6 +15,8 @@ comdat_groups: std.ArrayListUnmanaged(Elf.ComdatGroup.Index) = .{}, comdat_group_data: std.ArrayListUnmanaged(u32) = .{}, relocs: std.ArrayListUnmanaged(elf.Elf64_Rela) = .{}, +merge_sections: std.ArrayListUnmanaged(InputMergeSection.Index) = .{}, + fdes: std.ArrayListUnmanaged(Fde) = .{}, cies: std.ArrayListUnmanaged(Cie) = .{}, eh_frame_data: std.ArrayListUnmanaged(u8) = .{}, @@ -51,6 +53,7 @@ pub fn deinit(self: *Object, allocator: Allocator) void { self.fdes.deinit(allocator); self.cies.deinit(allocator); self.eh_frame_data.deinit(allocator); + self.merge_sections.deinit(allocator); } pub fn parse(self: *Object, elf_file: *Elf) !void { @@ -280,8 +283,6 @@ fn initOutputSection(self: Object, elf_file: *Elf, shdr: elf.Elf64_Shdr) error{O const name = blk: { const name = self.getString(shdr.sh_name); if (elf_file.base.isRelocatable()) break :blk name; - if (shdr.sh_flags & elf.SHF_MERGE != 0 and shdr.sh_flags & elf.SHF_STRINGS == 0) - break :blk name; // TODO: consider dropping SHF_STRINGS once ICF is implemented const sh_name_prefixes: []const [:0]const u8 = &.{ ".text", ".data.rel.ro", ".data", ".rodata", ".bss.rel.ro", ".bss", ".init_array", ".fini_array", ".tbss", ".tdata", ".gcc_except_table", ".ctors", @@ -335,7 +336,6 @@ fn skipShdr(self: *Object, index: u32, elf_file: *Elf) bool { const name = self.getString(shdr.sh_name); const ignore = blk: { if (mem.startsWith(u8, name, ".note")) break :blk true; - if (mem.startsWith(u8, name, ".comment")) break :blk true; if (mem.startsWith(u8, name, ".llvm_addrsig")) break :blk true; if (mem.startsWith(u8, name, ".riscv.attributes")) break :blk true; // TODO: riscv attributes if (comp.config.debug_format == .strip and shdr.sh_flags & elf.SHF_ALLOC == 0 and @@ -354,7 +354,7 @@ fn initSymtab(self: *Object, allocator: Allocator, elf_file: *Elf) !void { const index = try elf_file.addSymbol(); self.symbols.appendAssumeCapacity(index); const sym_ptr = elf_file.symbol(index); - sym_ptr.value = sym.st_value; + sym_ptr.value = @intCast(sym.st_value); sym_ptr.name_offset = sym.st_name; sym_ptr.esym_index = @as(u32, @intCast(i)); sym_ptr.atom_index = if (sym.st_shndx == elf.SHN_ABS) 0 else self.atoms.items[sym.st_shndx]; @@ -547,7 +547,7 @@ pub fn resolveSymbols(self: *Object, elf_file: *Elf) void { elf.SHN_ABS, elf.SHN_COMMON => 0, else => self.atoms.items[esym.st_shndx], }; - global.value = esym.st_value; + global.value = @intCast(esym.st_value); global.atom_index = atom_index; global.esym_index = esym_index; global.file_index = self.index; @@ -659,6 +659,173 @@ pub fn checkDuplicates(self: *Object, dupes: anytype, elf_file: *Elf) error{OutO } } +pub fn initMergeSections(self: *Object, elf_file: *Elf) !void { + const gpa = elf_file.base.comp.gpa; + + try self.merge_sections.resize(gpa, self.shdrs.items.len); + @memset(self.merge_sections.items, 0); + + for (self.shdrs.items, 0..) |shdr, shndx| { + if (shdr.sh_flags & elf.SHF_MERGE == 0) continue; + + const atom_index = self.atoms.items[shndx]; + const atom_ptr = elf_file.atom(atom_index) orelse continue; + if (!atom_ptr.flags.alive) continue; + if (atom_ptr.relocs(elf_file).len > 0) continue; + + const imsec_idx = try elf_file.addInputMergeSection(); + const imsec = elf_file.inputMergeSection(imsec_idx).?; + self.merge_sections.items[shndx] = imsec_idx; + + imsec.merge_section_index = try elf_file.getOrCreateMergeSection(atom_ptr.name(elf_file), shdr.sh_flags, shdr.sh_type); + imsec.atom_index = atom_index; + + const data = try self.codeDecompressAlloc(elf_file, atom_index); + defer gpa.free(data); + const sh_entsize: u32 = @intCast(shdr.sh_entsize); + + if (shdr.sh_flags & elf.SHF_STRINGS != 0) { + var pos: u32 = 0; + while (pos < data.len) switch (sh_entsize) { + 0, 1 => { + // According to mold's source code, GHC emits MS sections with sh_entsize = 0. + // This actually can also happen for output created with `-r` mode. + const string = mem.sliceTo(@as([*:0]const u8, @ptrCast(data.ptr + pos)), 0); + if (pos + string.len == data.len) { + var err = try elf_file.addErrorWithNotes(1); + try err.addMsg(elf_file, "string not null terminated", .{}); + try err.addNote(elf_file, "in {}:{s}", .{ self.fmtPath(), atom_ptr.name(elf_file) }); + return error.MalformedObject; + } + try imsec.insertZ(gpa, string); + try imsec.offsets.append(gpa, pos); + pos += @as(u32, @intCast(string.len)) + 1; // account for null + }, + else => |entsize| { + const string = data.ptr[pos..][0..entsize]; + if (string[string.len - 1] != 0) { + var err = try elf_file.addErrorWithNotes(1); + try err.addMsg(elf_file, "string not null terminated", .{}); + try err.addNote(elf_file, "in {}:{s}", .{ self.fmtPath(), atom_ptr.name(elf_file) }); + return error.MalformedObject; + } + try imsec.insert(gpa, string); + try imsec.offsets.append(gpa, pos); + pos += @as(u32, @intCast(string.len)); + }, + }; + } else { + if (sh_entsize == 0) continue; // Malformed, don't split but don't error out + if (shdr.sh_size % sh_entsize != 0) { + var err = try elf_file.addErrorWithNotes(1); + try err.addMsg(elf_file, "size not a multiple of sh_entsize", .{}); + try err.addNote(elf_file, "in {}:{s}", .{ self.fmtPath(), atom_ptr.name(elf_file) }); + return error.MalformedObject; + } + + var pos: u32 = 0; + while (pos < data.len) : (pos += sh_entsize) { + const string = data.ptr[pos..][0..sh_entsize]; + try imsec.insert(gpa, string); + try imsec.offsets.append(gpa, pos); + } + } + + atom_ptr.flags.alive = false; + } +} + +pub fn resolveMergeSubsections(self: *Object, elf_file: *Elf) !void { + const gpa = elf_file.base.comp.gpa; + + for (self.merge_sections.items) |index| { + const imsec = elf_file.inputMergeSection(index) orelse continue; + const msec = elf_file.mergeSection(imsec.merge_section_index); + const atom_ptr = elf_file.atom(imsec.atom_index).?; + const isec = atom_ptr.inputShdr(elf_file); + + try imsec.subsections.resize(gpa, imsec.strings.items.len); + + for (imsec.strings.items, imsec.subsections.items) |str, *imsec_msub| { + const string = imsec.bytes.items[str.pos..][0..str.len]; + const res = try msec.insert(gpa, string); + if (!res.found_existing) { + const msub_index = try elf_file.addMergeSubsection(); + const msub = elf_file.mergeSubsection(msub_index); + msub.merge_section_index = imsec.merge_section_index; + msub.string_index = res.key.pos; + msub.alignment = atom_ptr.alignment; + msub.size = res.key.len; + msub.alive = !elf_file.base.gc_sections or isec.sh_flags & elf.SHF_ALLOC == 0; + res.sub.* = msub_index; + } + imsec_msub.* = res.sub.*; + } + + imsec.clearAndFree(gpa); + } + + for (self.symtab.items, 0..) |*esym, idx| { + const sym_index = self.symbols.items[idx]; + const sym = elf_file.symbol(sym_index); + + if (esym.st_shndx == elf.SHN_COMMON or esym.st_shndx == elf.SHN_UNDEF or esym.st_shndx == elf.SHN_ABS) continue; + + const imsec_index = self.merge_sections.items[esym.st_shndx]; + const imsec = elf_file.inputMergeSection(imsec_index) orelse continue; + const msub_index, const offset = imsec.findSubsection(@intCast(esym.st_value)) orelse { + var err = try elf_file.addErrorWithNotes(2); + try err.addMsg(elf_file, "invalid symbol value: {x}", .{esym.st_value}); + try err.addNote(elf_file, "for symbol {s}", .{sym.name(elf_file)}); + try err.addNote(elf_file, "in {}", .{self.fmtPath()}); + return error.MalformedObject; + }; + + try sym.addExtra(.{ .subsection = msub_index }, elf_file); + sym.flags.merge_subsection = true; + sym.value = offset; + } + + for (self.atoms.items) |atom_index| { + const atom_ptr = elf_file.atom(atom_index) orelse continue; + if (!atom_ptr.flags.alive) continue; + const extras = atom_ptr.extra(elf_file) orelse continue; + const relocs = self.relocs.items[extras.rel_index..][0..extras.rel_count]; + for (relocs) |*rel| { + const esym = self.symtab.items[rel.r_sym()]; + if (esym.st_type() != elf.STT_SECTION) continue; + + const imsec_index = self.merge_sections.items[esym.st_shndx]; + const imsec = elf_file.inputMergeSection(imsec_index) orelse continue; + const msub_index, const offset = imsec.findSubsection(@intCast(@as(i64, @intCast(esym.st_value)) + rel.r_addend)) orelse { + var err = try elf_file.addErrorWithNotes(1); + try err.addMsg(elf_file, "invalid relocation at offset 0x{x}", .{rel.r_offset}); + try err.addNote(elf_file, "in {}:{s}", .{ self.fmtPath(), atom_ptr.name(elf_file) }); + return error.MalformedObject; + }; + const msub = elf_file.mergeSubsection(msub_index); + const msec = msub.mergeSection(elf_file); + + const out_sym_idx: u64 = @intCast(self.symbols.items.len); + try self.symbols.ensureUnusedCapacity(gpa, 1); + const name = try std.fmt.allocPrint(gpa, "{s}$subsection{d}", .{ msec.name(elf_file), msub_index }); + defer gpa.free(name); + const sym_index = try elf_file.addSymbol(); + const sym = elf_file.symbol(sym_index); + sym.* = .{ + .value = @bitCast(@as(i64, @intCast(offset)) - rel.r_addend), + .name_offset = try self.addString(gpa, name), + .esym_index = rel.r_sym(), + .file_index = self.index, + }; + try sym.addExtra(.{ .subsection = msub_index }, elf_file); + sym.flags.merge_subsection = true; + self.symbols.addOneAssumeCapacity().* = sym_index; + rel.r_info = (out_sym_idx << 32) | rel.r_type(); + } + } +} + /// We will create dummy shdrs per each resolved common symbols to make it /// play nicely with the rest of the system. pub fn convertCommonSymbols(self: *Object, elf_file: *Elf) !void { @@ -749,6 +916,11 @@ pub fn addAtomsToOutputSections(self: *Object, elf_file: *Elf) !void { for (self.locals()) |local_index| { const local = elf_file.symbol(local_index); + if (local.mergeSubsection(elf_file)) |msub| { + if (!msub.alive) continue; + local.output_section_index = msub.mergeSection(elf_file).output_section_index; + continue; + } const atom = local.atom(elf_file) orelse continue; if (!atom.flags.alive) continue; local.output_section_index = atom.output_section_index; @@ -756,11 +928,23 @@ pub fn addAtomsToOutputSections(self: *Object, elf_file: *Elf) !void { for (self.globals()) |global_index| { const global = elf_file.symbol(global_index); + if (global.file(elf_file).?.index() != self.index) continue; + if (global.mergeSubsection(elf_file)) |msub| { + if (!msub.alive) continue; + global.output_section_index = msub.mergeSection(elf_file).output_section_index; + continue; + } const atom = global.atom(elf_file) orelse continue; if (!atom.flags.alive) continue; - if (global.file(elf_file).?.index() != self.index) continue; global.output_section_index = atom.output_section_index; } + + for (self.symbols.items[self.symtab.items.len..]) |local_index| { + const local = elf_file.symbol(local_index); + const msub = local.mergeSubsection(elf_file).?; + if (!msub.alive) continue; + local.output_section_index = msub.mergeSection(elf_file).output_section_index; + } } pub fn initRelaSections(self: Object, elf_file: *Elf) !void { @@ -845,9 +1029,17 @@ pub fn writeAr(self: Object, elf_file: *Elf, writer: anytype) !void { } pub fn updateSymtabSize(self: *Object, elf_file: *Elf) !void { + const isAlive = struct { + fn isAlive(sym: *const Symbol, ctx: *Elf) bool { + if (sym.mergeSubsection(ctx)) |msub| return msub.alive; + if (sym.atom(ctx)) |atom_ptr| return atom_ptr.flags.alive; + return true; + } + }.isAlive; + for (self.locals()) |local_index| { const local = elf_file.symbol(local_index); - if (local.atom(elf_file)) |atom| if (!atom.flags.alive) continue; + if (!isAlive(local, elf_file)) continue; const esym = local.elfSym(elf_file); switch (esym.st_type()) { elf.STT_SECTION => continue, @@ -864,7 +1056,7 @@ pub fn updateSymtabSize(self: *Object, elf_file: *Elf) !void { const global = elf_file.symbol(global_index); const file_ptr = global.file(elf_file) orelse continue; if (file_ptr.index() != self.index) continue; - if (global.atom(elf_file)) |atom| if (!atom.flags.alive) continue; + if (!isAlive(global, elf_file)) continue; global.flags.output_symtab = true; if (global.isLocal(elf_file)) { try global.addExtra(.{ .symtab = self.output_symtab_ctx.nlocals }, elf_file); @@ -904,14 +1096,16 @@ pub fn writeSymtab(self: Object, elf_file: *Elf) void { pub fn locals(self: Object) []const Symbol.Index { if (self.symbols.items.len == 0) return &[0]Symbol.Index{}; - const end = self.first_global orelse self.symbols.items.len; + assert(self.symbols.items.len >= self.symtab.items.len); + const end = self.first_global orelse self.symtab.items.len; return self.symbols.items[0..end]; } pub fn globals(self: Object) []const Symbol.Index { if (self.symbols.items.len == 0) return &[0]Symbol.Index{}; - const start = self.first_global orelse self.symbols.items.len; - return self.symbols.items[start..]; + assert(self.symbols.items.len >= self.symtab.items.len); + const start = self.first_global orelse self.symtab.items.len; + return self.symbols.items[start..self.symtab.items.len]; } /// Returns atom's code and optionally uncompresses data if required (for compressed sections). @@ -956,6 +1150,14 @@ pub fn getString(self: Object, off: u32) [:0]const u8 { return mem.sliceTo(@as([*:0]const u8, @ptrCast(self.strtab.items.ptr + off)), 0); } +fn addString(self: *Object, allocator: Allocator, str: []const u8) !u32 { + const off: u32 = @intCast(self.strtab.items.len); + try self.strtab.ensureUnusedCapacity(allocator, str.len + 1); + self.strtab.appendSliceAssumeCapacity(str); + self.strtab.appendAssumeCapacity(0); + return off; +} + /// Caller owns the memory. fn preadShdrContentsAlloc(self: Object, allocator: Allocator, handle: std.fs.File, index: u32) ![]u8 { assert(index < self.shdrs.items.len); @@ -1161,5 +1363,6 @@ const Cie = eh_frame.Cie; const Elf = @import("../Elf.zig"); const Fde = eh_frame.Fde; const File = @import("file.zig").File; +const InputMergeSection = @import("merge_section.zig").InputMergeSection; const Symbol = @import("Symbol.zig"); const Alignment = Atom.Alignment; diff --git a/src/link/Elf/SharedObject.zig b/src/link/Elf/SharedObject.zig index 4491c04dc8..2ec5b211e8 100644 --- a/src/link/Elf/SharedObject.zig +++ b/src/link/Elf/SharedObject.zig @@ -231,7 +231,7 @@ pub fn resolveSymbols(self: *SharedObject, elf_file: *Elf) void { const global = elf_file.symbol(index); if (self.asFile().symbolRank(this_sym, false) < global.symbolRank(elf_file)) { - global.value = this_sym.st_value; + global.value = @intCast(this_sym.st_value); global.atom_index = 0; global.esym_index = esym_index; global.version_index = self.versyms.items[esym_index]; diff --git a/src/link/Elf/Symbol.zig b/src/link/Elf/Symbol.zig index 1cf41b459a..21061f036b 100644 --- a/src/link/Elf/Symbol.zig +++ b/src/link/Elf/Symbol.zig @@ -1,7 +1,7 @@ //! Represents a defined symbol. /// Allocated address value of this symbol. -value: u64 = 0, +value: i64 = 0, /// Offset into the linker's string table. name_offset: u32 = 0, @@ -14,7 +14,7 @@ file_index: File.Index = 0, /// Use `atom` to get the pointer to the atom. atom_index: Atom.Index = 0, -/// Assigned output section index for this atom. +/// Assigned output section index for this symbol. output_section_index: u32 = 0, /// Index of the source symbol this symbol references. @@ -33,7 +33,8 @@ extra_index: u32 = 0, pub fn isAbs(symbol: Symbol, elf_file: *Elf) bool { const file_ptr = symbol.file(elf_file).?; if (file_ptr == .shared_object) return symbol.elfSym(elf_file).st_shndx == elf.SHN_ABS; - return !symbol.flags.import and symbol.atom(elf_file) == null and symbol.outputShndx() == null and + return !symbol.flags.import and symbol.atom(elf_file) == null and + symbol.mergeSubsection(elf_file) == null and symbol.outputShndx() == null and file_ptr != .linker_defined; } @@ -70,6 +71,12 @@ pub fn atom(symbol: Symbol, elf_file: *Elf) ?*Atom { return elf_file.atom(symbol.atom_index); } +pub fn mergeSubsection(symbol: Symbol, elf_file: *Elf) ?*MergeSubsection { + if (!symbol.flags.merge_subsection) return null; + const extras = symbol.extra(elf_file).?; + return elf_file.mergeSubsection(extras.subsection); +} + pub fn file(symbol: Symbol, elf_file: *Elf) ?File { return elf_file.file(symbol.file_index); } @@ -92,7 +99,11 @@ pub fn symbolRank(symbol: Symbol, elf_file: *Elf) u32 { return file_ptr.symbolRank(sym, in_archive); } -pub fn address(symbol: Symbol, opts: struct { plt: bool = true }, elf_file: *Elf) u64 { +pub fn address(symbol: Symbol, opts: struct { plt: bool = true }, elf_file: *Elf) i64 { + if (symbol.mergeSubsection(elf_file)) |msub| { + if (!msub.alive) return 0; + return msub.address(elf_file) + symbol.value; + } if (symbol.flags.has_copy_rel) { return symbol.copyRelAddress(elf_file); } @@ -108,19 +119,23 @@ pub fn address(symbol: Symbol, opts: struct { plt: bool = true }, elf_file: *Elf if (!atom_ptr.flags.alive) { if (mem.eql(u8, atom_ptr.name(elf_file), ".eh_frame")) { const sym_name = symbol.name(elf_file); + const sh_addr, const sh_size = blk: { + const shndx = elf_file.eh_frame_section_index orelse break :blk .{ 0, 0 }; + const shdr = elf_file.shdrs.items[shndx]; + break :blk .{ shdr.sh_addr, shdr.sh_size }; + }; if (mem.startsWith(u8, sym_name, "__EH_FRAME_BEGIN__") or mem.startsWith(u8, sym_name, "__EH_FRAME_LIST__") or mem.startsWith(u8, sym_name, ".eh_frame_seg") or symbol.elfSym(elf_file).st_type() == elf.STT_SECTION) { - return elf_file.shdrs.items[elf_file.eh_frame_section_index.?].sh_addr; + return @intCast(sh_addr); } if (mem.startsWith(u8, sym_name, "__FRAME_END__") or mem.startsWith(u8, sym_name, "__EH_FRAME_LIST_END__")) { - const shdr = elf_file.shdrs.items[elf_file.eh_frame_section_index.?]; - return shdr.sh_addr + shdr.sh_size; + return @intCast(sh_addr + sh_size); } // TODO I think we potentially should error here @@ -143,57 +158,57 @@ pub fn outputSymtabIndex(symbol: Symbol, elf_file: *Elf) ?u32 { return if (symbol.isLocal(elf_file)) idx + symtab_ctx.ilocal else idx + symtab_ctx.iglobal; } -pub fn gotAddress(symbol: Symbol, elf_file: *Elf) u64 { +pub fn gotAddress(symbol: Symbol, elf_file: *Elf) i64 { if (!symbol.flags.has_got) return 0; const extras = symbol.extra(elf_file).?; const entry = elf_file.got.entries.items[extras.got]; return entry.address(elf_file); } -pub fn pltGotAddress(symbol: Symbol, elf_file: *Elf) u64 { +pub fn pltGotAddress(symbol: Symbol, elf_file: *Elf) i64 { if (!(symbol.flags.has_plt and symbol.flags.has_got)) return 0; const extras = symbol.extra(elf_file).?; const shdr = elf_file.shdrs.items[elf_file.plt_got_section_index.?]; const cpu_arch = elf_file.getTarget().cpu.arch; - return shdr.sh_addr + extras.plt_got * PltGotSection.entrySize(cpu_arch); + return @intCast(shdr.sh_addr + extras.plt_got * PltGotSection.entrySize(cpu_arch)); } -pub fn pltAddress(symbol: Symbol, elf_file: *Elf) u64 { +pub fn pltAddress(symbol: Symbol, elf_file: *Elf) i64 { if (!symbol.flags.has_plt) return 0; const extras = symbol.extra(elf_file).?; const shdr = elf_file.shdrs.items[elf_file.plt_section_index.?]; const cpu_arch = elf_file.getTarget().cpu.arch; - return shdr.sh_addr + extras.plt * PltSection.entrySize(cpu_arch) + PltSection.preambleSize(cpu_arch); + return @intCast(shdr.sh_addr + extras.plt * PltSection.entrySize(cpu_arch) + PltSection.preambleSize(cpu_arch)); } -pub fn gotPltAddress(symbol: Symbol, elf_file: *Elf) u64 { +pub fn gotPltAddress(symbol: Symbol, elf_file: *Elf) i64 { if (!symbol.flags.has_plt) return 0; const extras = symbol.extra(elf_file).?; const shdr = elf_file.shdrs.items[elf_file.got_plt_section_index.?]; - return shdr.sh_addr + extras.plt * 8 + GotPltSection.preamble_size; + return @intCast(shdr.sh_addr + extras.plt * 8 + GotPltSection.preamble_size); } -pub fn copyRelAddress(symbol: Symbol, elf_file: *Elf) u64 { +pub fn copyRelAddress(symbol: Symbol, elf_file: *Elf) i64 { if (!symbol.flags.has_copy_rel) return 0; const shdr = elf_file.shdrs.items[elf_file.copy_rel_section_index.?]; - return shdr.sh_addr + symbol.value; + return @as(i64, @intCast(shdr.sh_addr)) + symbol.value; } -pub fn tlsGdAddress(symbol: Symbol, elf_file: *Elf) u64 { +pub fn tlsGdAddress(symbol: Symbol, elf_file: *Elf) i64 { if (!symbol.flags.has_tlsgd) return 0; const extras = symbol.extra(elf_file).?; const entry = elf_file.got.entries.items[extras.tlsgd]; return entry.address(elf_file); } -pub fn gotTpAddress(symbol: Symbol, elf_file: *Elf) u64 { +pub fn gotTpAddress(symbol: Symbol, elf_file: *Elf) i64 { if (!symbol.flags.has_gottp) return 0; const extras = symbol.extra(elf_file).?; const entry = elf_file.got.entries.items[extras.gottp]; return entry.address(elf_file); } -pub fn tlsDescAddress(symbol: Symbol, elf_file: *Elf) u64 { +pub fn tlsDescAddress(symbol: Symbol, elf_file: *Elf) i64 { if (!symbol.flags.has_tlsdesc) return 0; const extras = symbol.extra(elf_file).?; const entry = elf_file.got.entries.items[extras.tlsdesc]; @@ -213,7 +228,7 @@ pub fn getOrCreateZigGotEntry(symbol: *Symbol, symbol_index: Index, elf_file: *E return .{ .found_existing = false, .index = index }; } -pub fn zigGotAddress(symbol: Symbol, elf_file: *Elf) u64 { +pub fn zigGotAddress(symbol: Symbol, elf_file: *Elf) i64 { if (!symbol.flags.has_zig_got) return 0; const extras = symbol.extra(elf_file).?; return elf_file.zig_got.entryAddress(extras.zig_got, elf_file); @@ -243,6 +258,7 @@ const AddExtraOpts = struct { gottp: ?u32 = null, tlsdesc: ?u32 = null, zig_got: ?u32 = null, + subsection: ?u32 = null, }; pub fn addExtra(symbol: *Symbol, opts: AddExtraOpts, elf_file: *Elf) !void { @@ -280,6 +296,7 @@ pub fn setOutputSym(symbol: Symbol, elf_file: *Elf, out: *elf.Elf64_Sym) void { if (symbol.flags.has_copy_rel) break :blk @intCast(elf_file.copy_rel_section_index.?); if (file_ptr == .shared_object or esym.st_shndx == elf.SHN_UNDEF) break :blk elf.SHN_UNDEF; if (elf_file.base.isRelocatable() and esym.st_shndx == elf.SHN_COMMON) break :blk elf.SHN_COMMON; + if (symbol.mergeSubsection(elf_file)) |msub| break :blk @intCast(msub.mergeSection(elf_file).output_section_index); if (symbol.atom(elf_file) == null and file_ptr != .linker_defined) break :blk elf.SHN_ABS; break :blk @intCast(symbol.outputShndx() orelse elf.SHN_UNDEF); }; @@ -298,7 +315,7 @@ pub fn setOutputSym(symbol: Symbol, elf_file: *Elf, out: *elf.Elf64_Sym) void { out.st_info = (st_bind << 4) | st_type; out.st_other = esym.st_other; out.st_shndx = st_shndx; - out.st_value = st_value; + out.st_value = @intCast(st_value); out.st_size = esym.st_size; } @@ -450,6 +467,9 @@ pub const Flags = packed struct { /// TODO this is really not needed if only we operated on esyms between /// codegen and ZigObject. is_tls: bool = false, + + /// Whether the symbol is a merge subsection. + merge_subsection: bool = false, }; pub const Extra = struct { @@ -463,6 +483,7 @@ pub const Extra = struct { gottp: u32 = 0, tlsdesc: u32 = 0, zig_got: u32 = 0, + subsection: u32 = 0, }; pub const Index = u32; @@ -479,6 +500,7 @@ const File = @import("file.zig").File; const GotSection = synthetic_sections.GotSection; const GotPltSection = synthetic_sections.GotPltSection; const LinkerDefined = @import("LinkerDefined.zig"); +const MergeSubsection = @import("merge_section.zig").MergeSubsection; const Object = @import("Object.zig"); const PltSection = synthetic_sections.PltSection; const PltGotSection = synthetic_sections.PltGotSection; diff --git a/src/link/Elf/ZigObject.zig b/src/link/Elf/ZigObject.zig index af74f2637b..98af014068 100644 --- a/src/link/Elf/ZigObject.zig +++ b/src/link/Elf/ZigObject.zig @@ -343,7 +343,7 @@ pub fn resolveSymbols(self: *ZigObject, elf_file: *Elf) void { atom.outputShndx().? else elf.SHN_UNDEF; - global.value = esym.st_value; + global.value = @intCast(esym.st_value); global.atom_index = atom_index; global.esym_index = esym_index; global.file_index = self.index; @@ -631,7 +631,7 @@ pub fn codeAlloc(self: ZigObject, elf_file: *Elf, atom_index: Atom.Index) ![]u8 return code; } - const file_offset = shdr.sh_offset + atom.value; + const file_offset = shdr.sh_offset + @as(u64, @intCast(atom.value)); const size = std.math.cast(usize, atom.size) orelse return error.Overflow; const code = try gpa.alloc(u8, size); errdefer gpa.free(code); @@ -659,7 +659,7 @@ pub fn getDeclVAddr( .r_info = (@as(u64, @intCast(this_sym.esym_index)) << 32) | r_type, .r_addend = reloc_info.addend, }); - return vaddr; + return @intCast(vaddr); } pub fn getAnonDeclVAddr( @@ -678,7 +678,7 @@ pub fn getAnonDeclVAddr( .r_info = (@as(u64, @intCast(sym.esym_index)) << 32) | r_type, .r_addend = reloc_info.addend, }); - return vaddr; + return @intCast(vaddr); } pub fn lowerAnonDecl( @@ -929,7 +929,7 @@ fn updateDeclCode( if (old_size > 0 and elf_file.base.child_pid == null) { const capacity = atom_ptr.capacity(elf_file); - const need_realloc = code.len > capacity or !required_alignment.check(atom_ptr.value); + const need_realloc = code.len > capacity or !required_alignment.check(@intCast(atom_ptr.value)); if (need_realloc) { try atom_ptr.grow(elf_file); log.debug("growing {} from 0x{x} to 0x{x}", .{ decl_name.fmt(&mod.intern_pool), old_vaddr, atom_ptr.value }); @@ -984,7 +984,7 @@ fn updateDeclCode( const shdr = elf_file.shdrs.items[shdr_index]; if (shdr.sh_type != elf.SHT_NOBITS) { - const file_offset = shdr.sh_offset + atom_ptr.value; + const file_offset = shdr.sh_offset + @as(u64, @intCast(atom_ptr.value)); try elf_file.base.file.?.pwriteAll(code, file_offset); } } @@ -1107,7 +1107,7 @@ pub fn updateFunc( try self.dwarf.?.commitDeclState( mod, decl_index, - sym.address(.{}, elf_file), + @intCast(sym.address(.{}, elf_file)), sym.atom(elf_file).?.size, ds, ); @@ -1186,7 +1186,7 @@ pub fn updateDecl( try self.dwarf.?.commitDeclState( mod, decl_index, - sym.address(.{}, elf_file), + @intCast(sym.address(.{}, elf_file)), sym.atom(elf_file).?.size, ds, ); @@ -1275,7 +1275,7 @@ fn updateLazySymbol( } const shdr = elf_file.shdrs.items[output_section_index]; - const file_offset = shdr.sh_offset + atom_ptr.value; + const file_offset = shdr.sh_offset + @as(u64, @intCast(atom_ptr.value)); try elf_file.base.file.?.pwriteAll(code, file_offset); } @@ -1373,7 +1373,7 @@ fn lowerConst( local_esym.st_value = 0; const shdr = elf_file.shdrs.items[output_section_index]; - const file_offset = shdr.sh_offset + atom_ptr.value; + const file_offset = shdr.sh_offset + @as(u64, @intCast(atom_ptr.value)); try elf_file.base.file.?.pwriteAll(code, file_offset); return .{ .ok = sym_index }; @@ -1457,7 +1457,7 @@ pub fn updateExports( const actual_esym_index = global_esym_index & symbol_mask; const global_esym = &self.global_esyms.items(.elf_sym)[actual_esym_index]; - global_esym.st_value = elf_file.symbol(sym_index).value; + global_esym.st_value = @intCast(elf_file.symbol(sym_index).value); global_esym.st_shndx = esym.st_shndx; global_esym.st_info = (stb_bits << 4) | stt_bits; global_esym.st_name = name_off; diff --git a/src/link/Elf/gc.zig b/src/link/Elf/gc.zig index d95dd78b33..49b5664dec 100644 --- a/src/link/Elf/gc.zig +++ b/src/link/Elf/gc.zig @@ -68,6 +68,10 @@ fn collectRoots(roots: *std.ArrayList(*Atom), files: []const File.Index, elf_fil } fn markSymbol(sym: *Symbol, roots: *std.ArrayList(*Atom), elf_file: *Elf) !void { + if (sym.mergeSubsection(elf_file)) |msub| { + msub.alive = true; + return; + } const atom = sym.atom(elf_file) orelse return; if (markAtom(atom)) try roots.append(atom); } @@ -96,6 +100,10 @@ fn markLive(atom: *Atom, elf_file: *Elf) void { for (atom.relocs(elf_file)) |rel| { const target_sym = elf_file.symbol(file.symbol(rel.r_sym())); + if (target_sym.mergeSubsection(elf_file)) |msub| { + msub.alive = true; + continue; + } const target_atom = target_sym.atom(elf_file) orelse continue; target_atom.flags.alive = true; gc_track_live_log.debug("{}marking live atom({d})", .{ track_live_level, target_atom.atom_index }); diff --git a/src/link/Elf/relocatable.zig b/src/link/Elf/relocatable.zig index 98abed420e..e7a984bc15 100644 --- a/src/link/Elf/relocatable.zig +++ b/src/link/Elf/relocatable.zig @@ -181,6 +181,8 @@ pub fn flushObject(elf_file: *Elf, comp: *Compilation, module_obj_path: ?[]const elf_file.markEhFrameAtomsDead(); claimUnresolved(elf_file); + try elf_file.addCommentString(); + try elf_file.sortMergeSections(); try initSections(elf_file); try elf_file.sortShdrs(); if (elf_file.zigObjectPtr()) |zig_object| { @@ -191,6 +193,7 @@ pub fn flushObject(elf_file: *Elf, comp: *Compilation, module_obj_path: ?[]const try object.addAtomsToOutputSections(elf_file); try object.addAtomsToRelaSections(elf_file); } + try elf_file.updateMergeSectionSizes(); try updateSectionSizes(elf_file); try allocateAllocSections(elf_file); @@ -201,6 +204,7 @@ pub fn flushObject(elf_file: *Elf, comp: *Compilation, module_obj_path: ?[]const } try writeAtoms(elf_file); + try elf_file.writeMergeSections(); try writeSyntheticSections(elf_file); try elf_file.writeShdrTable(); try elf_file.writeElfHeader(); @@ -275,6 +279,17 @@ fn initSections(elf_file: *Elf) !void { try object.initRelaSections(elf_file); } + for (elf_file.merge_sections.items) |*msec| { + if (msec.subsections.items.len == 0) continue; + const name = msec.name(elf_file); + const shndx = elf_file.sectionByName(name) orelse try elf_file.addSection(.{ + .name = name, + .type = msec.type, + .flags = msec.flags, + }); + msec.output_section_index = shndx; + } + const needs_eh_frame = for (elf_file.objects.items) |index| { if (elf_file.file(index).?.object.cies.items.len > 0) break true; } else false; @@ -328,7 +343,7 @@ fn updateSectionSizes(elf_file: *Elf) !void { if (!atom_ptr.flags.alive) continue; const offset = atom_ptr.alignment.forward(shdr.sh_size); const padding = offset - shdr.sh_size; - atom_ptr.value = offset; + atom_ptr.value = @intCast(offset); shdr.sh_size += padding + atom_ptr.size; shdr.sh_addralign = @max(shdr.sh_addralign, atom_ptr.alignment.toByteUnits() orelse 1); } @@ -434,7 +449,7 @@ fn writeAtoms(elf_file: *Elf) !void { const atom_ptr = elf_file.atom(atom_index).?; assert(atom_ptr.flags.alive); - const offset = math.cast(usize, atom_ptr.value - shdr.sh_addr - base_offset) orelse + const offset = math.cast(usize, atom_ptr.value - @as(i64, @intCast(shdr.sh_addr - base_offset))) orelse return error.Overflow; const size = math.cast(usize, atom_ptr.size) orelse return error.Overflow; diff --git a/src/link/Elf/synthetic_sections.zig b/src/link/Elf/synthetic_sections.zig index e76809c79a..59adc7d40e 100644 --- a/src/link/Elf/synthetic_sections.zig +++ b/src/link/Elf/synthetic_sections.zig @@ -270,11 +270,11 @@ pub const ZigGotSection = struct { return shdr.sh_offset + @as(u64, entry_size) * index; } - pub fn entryAddress(zig_got: ZigGotSection, index: Index, elf_file: *Elf) u64 { + pub fn entryAddress(zig_got: ZigGotSection, index: Index, elf_file: *Elf) i64 { _ = zig_got; const entry_size = elf_file.archPtrWidthBytes(); const shdr = elf_file.shdrs.items[elf_file.zig_got_section_index.?]; - return shdr.sh_addr + @as(u64, entry_size) * index; + return @as(i64, @intCast(shdr.sh_addr)) + entry_size * index; } pub fn size(zig_got: ZigGotSection, elf_file: *Elf) usize { @@ -291,23 +291,23 @@ pub const ZigGotSection = struct { const target = elf_file.getTarget(); const endian = target.cpu.arch.endian(); const off = zig_got.entryOffset(index, elf_file); - const vaddr = zig_got.entryAddress(index, elf_file); + const vaddr: u64 = @intCast(zig_got.entryAddress(index, elf_file)); const entry = zig_got.entries.items[index]; const value = elf_file.symbol(entry).address(.{}, elf_file); switch (entry_size) { 2 => { var buf: [2]u8 = undefined; - std.mem.writeInt(u16, &buf, @as(u16, @intCast(value)), endian); + std.mem.writeInt(u16, &buf, @intCast(value), endian); try elf_file.base.file.?.pwriteAll(&buf, off); }, 4 => { var buf: [4]u8 = undefined; - std.mem.writeInt(u32, &buf, @as(u32, @intCast(value)), endian); + std.mem.writeInt(u32, &buf, @intCast(value), endian); try elf_file.base.file.?.pwriteAll(&buf, off); }, 8 => { var buf: [8]u8 = undefined; - std.mem.writeInt(u64, &buf, value, endian); + std.mem.writeInt(u64, &buf, @intCast(value), endian); try elf_file.base.file.?.pwriteAll(&buf, off); if (elf_file.base.child_pid) |pid| { @@ -356,9 +356,9 @@ pub const ZigGotSection = struct { const symbol = elf_file.symbol(entry); const offset = symbol.zigGotAddress(elf_file); elf_file.addRelaDynAssumeCapacity(.{ - .offset = offset, + .offset = @intCast(offset), .type = relocation.encode(.rel, cpu_arch), - .addend = @intCast(symbol.address(.{ .plt = false }, elf_file)), + .addend = symbol.address(.{ .plt = false }, elf_file), }); } } @@ -386,7 +386,7 @@ pub const ZigGotSection = struct { .st_info = elf.STT_OBJECT, .st_other = 0, .st_shndx = @intCast(elf_file.zig_got_section_index.?), - .st_value = st_value, + .st_value = @intCast(st_value), .st_size = st_size, }; } @@ -457,10 +457,10 @@ pub const GotSection = struct { }; } - pub fn address(entry: Entry, elf_file: *Elf) u64 { - const ptr_bytes = @as(u64, elf_file.archPtrWidthBytes()); + pub fn address(entry: Entry, elf_file: *Elf) i64 { + const ptr_bytes = elf_file.archPtrWidthBytes(); const shdr = &elf_file.shdrs.items[elf_file.got_section_index.?]; - return shdr.sh_addr + @as(u64, entry.cell_index) * ptr_bytes; + return @as(i64, @intCast(shdr.sh_addr)) + entry.cell_index * ptr_bytes; } }; @@ -608,8 +608,7 @@ pub const GotSection = struct { 0; try writeInt(offset, elf_file, writer); } else { - const offset = @as(i64, @intCast(symbol.?.address(.{}, elf_file))) - - @as(i64, @intCast(elf_file.tpAddress())); + const offset = symbol.?.address(.{}, elf_file) - elf_file.tpAddress(); try writeInt(offset, elf_file, writer); } }, @@ -620,7 +619,7 @@ pub const GotSection = struct { } else { try writeInt(0, elf_file, writer); const offset = if (apply_relocs) - @as(i64, @intCast(symbol.?.address(.{}, elf_file))) - @as(i64, @intCast(elf_file.tlsAddress())) + symbol.?.address(.{}, elf_file) - elf_file.tlsAddress() else 0; try writeInt(offset, elf_file, writer); @@ -646,7 +645,7 @@ pub const GotSection = struct { switch (entry.tag) { .got => { - const offset = symbol.?.gotAddress(elf_file); + const offset: u64 = @intCast(symbol.?.gotAddress(elf_file)); if (symbol.?.flags.import) { elf_file.addRelaDynAssumeCapacity(.{ .offset = offset, @@ -659,7 +658,7 @@ pub const GotSection = struct { elf_file.addRelaDynAssumeCapacity(.{ .offset = offset, .type = relocation.encode(.irel, cpu_arch), - .addend = @intCast(symbol.?.address(.{ .plt = false }, elf_file)), + .addend = symbol.?.address(.{ .plt = false }, elf_file), }); continue; } @@ -669,14 +668,14 @@ pub const GotSection = struct { elf_file.addRelaDynAssumeCapacity(.{ .offset = offset, .type = relocation.encode(.rel, cpu_arch), - .addend = @intCast(symbol.?.address(.{ .plt = false }, elf_file)), + .addend = symbol.?.address(.{ .plt = false }, elf_file), }); } }, .tlsld => { if (is_dyn_lib) { - const offset = entry.address(elf_file); + const offset: u64 = @intCast(entry.address(elf_file)); elf_file.addRelaDynAssumeCapacity(.{ .offset = offset, .type = relocation.encode(.dtpmod, cpu_arch), @@ -685,7 +684,7 @@ pub const GotSection = struct { }, .tlsgd => { - const offset = symbol.?.tlsGdAddress(elf_file); + const offset: u64 = @intCast(symbol.?.tlsGdAddress(elf_file)); if (symbol.?.flags.import) { elf_file.addRelaDynAssumeCapacity(.{ .offset = offset, @@ -707,7 +706,7 @@ pub const GotSection = struct { }, .gottp => { - const offset = symbol.?.gotTpAddress(elf_file); + const offset: u64 = @intCast(symbol.?.gotTpAddress(elf_file)); if (symbol.?.flags.import) { elf_file.addRelaDynAssumeCapacity(.{ .offset = offset, @@ -718,18 +717,18 @@ pub const GotSection = struct { elf_file.addRelaDynAssumeCapacity(.{ .offset = offset, .type = relocation.encode(.tpoff, cpu_arch), - .addend = @intCast(symbol.?.address(.{}, elf_file) - elf_file.tlsAddress()), + .addend = symbol.?.address(.{}, elf_file) - elf_file.tlsAddress(), }); } }, .tlsdesc => { - const offset = symbol.?.tlsDescAddress(elf_file); + const offset: u64 = @intCast(symbol.?.tlsDescAddress(elf_file)); elf_file.addRelaDynAssumeCapacity(.{ .offset = offset, .sym = if (symbol.?.flags.import) extra.?.dynamic else 0, .type = relocation.encode(.tlsdesc, cpu_arch), - .addend = if (symbol.?.flags.import) 0 else @intCast(symbol.?.address(.{}, elf_file) - elf_file.tlsAddress()), + .addend = if (symbol.?.flags.import) 0 else symbol.?.address(.{}, elf_file) - elf_file.tlsAddress(), }); }, } @@ -806,7 +805,7 @@ pub const GotSection = struct { .st_info = elf.STT_OBJECT, .st_other = 0, .st_shndx = @intCast(elf_file.got_section_index.?), - .st_value = st_value, + .st_value = @intCast(st_value), .st_size = st_size, }; } @@ -900,7 +899,7 @@ pub const PltSection = struct { const sym = elf_file.symbol(sym_index); assert(sym.flags.import); const extra = sym.extra(elf_file).?; - const r_offset = sym.gotPltAddress(elf_file); + const r_offset: u64 = @intCast(sym.gotPltAddress(elf_file)); const r_sym: u64 = extra.dynamic; const r_type = relocation.encode(.jump_slot, cpu_arch); elf_file.rela_plt.appendAssumeCapacity(.{ @@ -936,7 +935,7 @@ pub const PltSection = struct { .st_info = elf.STT_FUNC, .st_other = 0, .st_shndx = @intCast(elf_file.plt_section_index.?), - .st_value = sym.pltAddress(elf_file), + .st_value = @intCast(sym.pltAddress(elf_file)), .st_size = entrySize(cpu_arch), }; } @@ -1009,13 +1008,13 @@ pub const PltSection = struct { const aarch64 = struct { fn write(plt: PltSection, elf_file: *Elf, writer: anytype) !void { { - const plt_addr = elf_file.shdrs.items[elf_file.plt_section_index.?].sh_addr; - const got_plt_addr = elf_file.shdrs.items[elf_file.got_plt_section_index.?].sh_addr; + const plt_addr: i64 = @intCast(elf_file.shdrs.items[elf_file.plt_section_index.?].sh_addr); + const got_plt_addr: i64 = @intCast(elf_file.shdrs.items[elf_file.got_plt_section_index.?].sh_addr); // TODO: relax if possible // .got.plt[2] const pages = try aarch64_util.calcNumberOfPages(plt_addr + 4, got_plt_addr + 16); - const ldr_off = try math.divExact(u12, @truncate(got_plt_addr + 16), 8); - const add_off: u12 = @truncate(got_plt_addr + 16); + const ldr_off = try math.divExact(u12, @truncate(@as(u64, @bitCast(got_plt_addr + 16))), 8); + const add_off: u12 = @truncate(@as(u64, @bitCast(got_plt_addr + 16))); const preamble = &[_]Instruction{ Instruction.stp( @@ -1043,8 +1042,8 @@ pub const PltSection = struct { const target_addr = sym.gotPltAddress(elf_file); const source_addr = sym.pltAddress(elf_file); const pages = try aarch64_util.calcNumberOfPages(source_addr, target_addr); - const ldr_off = try math.divExact(u12, @truncate(target_addr), 8); - const add_off: u12 = @truncate(target_addr); + const ldr_off = try math.divExact(u12, @truncate(@as(u64, @bitCast(target_addr))), 8); + const add_off: u12 = @truncate(@as(u64, @bitCast(target_addr))); const insts = &[_]Instruction{ Instruction.adrp(.x16, pages), Instruction.ldr(.x17, .x16, Instruction.LoadStoreOffset.imm(ldr_off)), @@ -1077,7 +1076,7 @@ pub const GotPltSection = struct { { // [0]: _DYNAMIC const symbol = elf_file.symbol(elf_file.dynamic_index.?); - try writer.writeInt(u64, symbol.address(.{}, elf_file), .little); + try writer.writeInt(u64, @intCast(symbol.address(.{}, elf_file)), .little); } // [1]: 0x0 // [2]: 0x0 @@ -1153,7 +1152,7 @@ pub const PltGotSection = struct { .st_info = elf.STT_FUNC, .st_other = 0, .st_shndx = @intCast(elf_file.plt_got_section_index.?), - .st_value = sym.pltGotAddress(elf_file), + .st_value = @intCast(sym.pltGotAddress(elf_file)), .st_size = 16, }; } @@ -1184,7 +1183,7 @@ pub const PltGotSection = struct { const target_addr = sym.gotAddress(elf_file); const source_addr = sym.pltGotAddress(elf_file); const pages = try aarch64_util.calcNumberOfPages(source_addr, target_addr); - const off = try math.divExact(u12, @truncate(target_addr), 8); + const off = try math.divExact(u12, @truncate(@as(u64, @bitCast(target_addr))), 8); const insts = &[_]Instruction{ Instruction.adrp(.x16, pages), Instruction.ldr(.x17, .x16, Instruction.LoadStoreOffset.imm(off)), @@ -1247,9 +1246,9 @@ pub const CopyRelSection = struct { const symbol = elf_file.symbol(sym_index); const shared_object = symbol.file(elf_file).?.shared_object; const alignment = try symbol.dsoAlignment(elf_file); - symbol.value = mem.alignForward(u64, shdr.sh_size, alignment); + symbol.value = @intCast(mem.alignForward(u64, shdr.sh_size, alignment)); shdr.sh_addralign = @max(shdr.sh_addralign, alignment); - shdr.sh_size = symbol.value + symbol.elfSym(elf_file).st_size; + shdr.sh_size = @as(u64, @intCast(symbol.value)) + symbol.elfSym(elf_file).st_size; const aliases = shared_object.symbolAliases(sym_index, elf_file); for (aliases) |alias| { @@ -1270,7 +1269,7 @@ pub const CopyRelSection = struct { assert(sym.flags.import and sym.flags.has_copy_rel); const extra = sym.extra(elf_file).?; elf_file.addRelaDynAssumeCapacity(.{ - .offset = sym.address(.{}, elf_file), + .offset = @intCast(sym.address(.{}, elf_file)), .sym = extra.dynamic, .type = relocation.encode(.copy, cpu_arch), }); diff --git a/src/link/Elf/thunks.zig b/src/link/Elf/thunks.zig index 6e1f8aeee6..48694416c9 100644 --- a/src/link/Elf/thunks.zig +++ b/src/link/Elf/thunks.zig @@ -7,7 +7,7 @@ pub fn createThunks(shndx: u32, elf_file: *Elf) !void { assert(atoms.len > 0); for (atoms) |atom_index| { - elf_file.atom(atom_index).?.value = @bitCast(@as(i64, -1)); + elf_file.atom(atom_index).?.value = -1; } var i: usize = 0; @@ -22,7 +22,8 @@ pub fn createThunks(shndx: u32, elf_file: *Elf) !void { const atom_index = atoms[i]; const atom = elf_file.atom(atom_index).?; assert(atom.flags.alive); - if (atom.alignment.forward(shdr.sh_size) - start_atom.value >= max_distance) break; + if (@as(i64, @intCast(atom.alignment.forward(shdr.sh_size))) - start_atom.value >= max_distance) + break; atom.value = try advance(shdr, atom.size, atom.alignment); } @@ -60,12 +61,12 @@ pub fn createThunks(shndx: u32, elf_file: *Elf) !void { } } -fn advance(shdr: *elf.Elf64_Shdr, size: u64, alignment: Atom.Alignment) !u64 { +fn advance(shdr: *elf.Elf64_Shdr, size: u64, alignment: Atom.Alignment) !i64 { const offset = alignment.forward(shdr.sh_size); const padding = offset - shdr.sh_size; shdr.sh_size += padding + size; shdr.sh_addralign = @max(shdr.sh_addralign, alignment.toByteUnits() orelse 1); - return offset; + return @intCast(offset); } /// A branch will need an extender if its target is larger than @@ -79,7 +80,7 @@ fn maxAllowedDistance(cpu_arch: std.Target.Cpu.Arch) u32 { } pub const Thunk = struct { - value: u64 = 0, + value: i64 = 0, output_section_index: u32 = 0, symbols: std.AutoArrayHashMapUnmanaged(Symbol.Index, void) = .{}, output_symtab_ctx: Elf.SymtabCtx = .{}, @@ -93,14 +94,14 @@ pub const Thunk = struct { return thunk.symbols.keys().len * trampolineSize(cpu_arch); } - pub fn address(thunk: Thunk, elf_file: *Elf) u64 { + pub fn address(thunk: Thunk, elf_file: *Elf) i64 { const shdr = elf_file.shdrs.items[thunk.output_section_index]; - return shdr.sh_addr + thunk.value; + return @as(i64, @intCast(shdr.sh_addr)) + thunk.value; } - pub fn targetAddress(thunk: Thunk, sym_index: Symbol.Index, elf_file: *Elf) u64 { + pub fn targetAddress(thunk: Thunk, sym_index: Symbol.Index, elf_file: *Elf) i64 { const cpu_arch = elf_file.getTarget().cpu.arch; - return thunk.address(elf_file) + thunk.symbols.getIndex(sym_index).? * trampolineSize(cpu_arch); + return thunk.address(elf_file) + @as(i64, @intCast(thunk.symbols.getIndex(sym_index).? * trampolineSize(cpu_arch))); } pub fn write(thunk: Thunk, elf_file: *Elf, writer: anytype) !void { @@ -132,7 +133,7 @@ pub const Thunk = struct { .st_info = elf.STT_FUNC, .st_other = 0, .st_shndx = @intCast(thunk.output_section_index), - .st_value = thunk.targetAddress(sym_index, elf_file), + .st_value = @intCast(thunk.targetAddress(sym_index, elf_file)), .st_size = trampolineSize(cpu_arch), }; } @@ -205,9 +206,9 @@ const aarch64 = struct { if (target.flags.has_plt) return false; if (atom.output_section_index != target.output_section_index) return false; const target_atom = target.atom(elf_file).?; - if (target_atom.value == @as(u64, @bitCast(@as(i64, -1)))) return false; - const saddr = @as(i64, @intCast(atom.address(elf_file) + rel.r_offset)); - const taddr: i64 = @intCast(target.address(.{}, elf_file)); + if (target_atom.value == -1) return false; + const saddr = atom.address(elf_file) + @as(i64, @intCast(rel.r_offset)); + const taddr = target.address(.{}, elf_file); _ = math.cast(i28, taddr + rel.r_addend - saddr) orelse return false; return true; } @@ -215,11 +216,11 @@ const aarch64 = struct { fn write(thunk: Thunk, elf_file: *Elf, writer: anytype) !void { for (thunk.symbols.keys(), 0..) |sym_index, i| { const sym = elf_file.symbol(sym_index); - const saddr = thunk.address(elf_file) + i * trampoline_size; + const saddr = thunk.address(elf_file) + @as(i64, @intCast(i * trampoline_size)); const taddr = sym.address(.{}, elf_file); const pages = try util.calcNumberOfPages(saddr, taddr); try writer.writeInt(u32, Instruction.adrp(.x16, pages).toU32(), .little); - const off: u12 = @truncate(taddr); + const off: u12 = @truncate(@as(u64, @bitCast(taddr))); try writer.writeInt(u32, Instruction.add(.x16, .x16, off, false).toU32(), .little); try writer.writeInt(u32, Instruction.br(.x16).toU32(), .little); } diff --git a/src/link/MachO/Atom.zig b/src/link/MachO/Atom.zig index b4600176ce..86edceda05 100644 --- a/src/link/MachO/Atom.zig +++ b/src/link/MachO/Atom.zig @@ -770,7 +770,7 @@ fn resolveRelocInner( }; break :target math.cast(u64, target) orelse return error.Overflow; }; - const pages = @as(u21, @bitCast(try aarch64.calcNumberOfPages(source, target))); + const pages = @as(u21, @bitCast(try aarch64.calcNumberOfPages(@intCast(source), @intCast(target)))); aarch64.writeAdrpInst(pages, code[rel_offset..][0..4]); }, diff --git a/src/link/MachO/synthetic.zig b/src/link/MachO/synthetic.zig index 863b68f9b1..a7729b9137 100644 --- a/src/link/MachO/synthetic.zig +++ b/src/link/MachO/synthetic.zig @@ -267,7 +267,7 @@ pub const StubsSection = struct { }, .aarch64 => { // TODO relax if possible - const pages = try aarch64.calcNumberOfPages(source, target); + const pages = try aarch64.calcNumberOfPages(@intCast(source), @intCast(target)); try writer.writeInt(u32, aarch64.Instruction.adrp(.x16, pages).toU32(), .little); const off = try math.divExact(u12, @truncate(target), 8); try writer.writeInt( @@ -411,7 +411,7 @@ pub const StubsHelperSection = struct { .aarch64 => { { // TODO relax if possible - const pages = try aarch64.calcNumberOfPages(sect.addr, dyld_private_addr); + const pages = try aarch64.calcNumberOfPages(@intCast(sect.addr), @intCast(dyld_private_addr)); try writer.writeInt(u32, aarch64.Instruction.adrp(.x17, pages).toU32(), .little); const off: u12 = @truncate(dyld_private_addr); try writer.writeInt(u32, aarch64.Instruction.add(.x17, .x17, off, false).toU32(), .little); @@ -424,7 +424,7 @@ pub const StubsHelperSection = struct { ).toU32(), .little); { // TODO relax if possible - const pages = try aarch64.calcNumberOfPages(sect.addr + 12, dyld_stub_binder_addr); + const pages = try aarch64.calcNumberOfPages(@intCast(sect.addr + 12), @intCast(dyld_stub_binder_addr)); try writer.writeInt(u32, aarch64.Instruction.adrp(.x16, pages).toU32(), .little); const off = try math.divExact(u12, @truncate(dyld_stub_binder_addr), 8); try writer.writeInt(u32, aarch64.Instruction.ldr( @@ -679,7 +679,7 @@ pub const ObjcStubsSection = struct { { const target = sym.getObjcSelrefsAddress(macho_file); const source = addr; - const pages = try aarch64.calcNumberOfPages(source, target); + const pages = try aarch64.calcNumberOfPages(@intCast(source), @intCast(target)); try writer.writeInt(u32, aarch64.Instruction.adrp(.x1, pages).toU32(), .little); const off = try math.divExact(u12, @truncate(target), 8); try writer.writeInt( @@ -692,7 +692,7 @@ pub const ObjcStubsSection = struct { const target_sym = macho_file.getSymbol(macho_file.objc_msg_send_index.?); const target = target_sym.getGotAddress(macho_file); const source = addr + 2 * @sizeOf(u32); - const pages = try aarch64.calcNumberOfPages(source, target); + const pages = try aarch64.calcNumberOfPages(@intCast(source), @intCast(target)); try writer.writeInt(u32, aarch64.Instruction.adrp(.x16, pages).toU32(), .little); const off = try math.divExact(u12, @truncate(target), 8); try writer.writeInt( diff --git a/src/link/MachO/thunks.zig b/src/link/MachO/thunks.zig index 02dacc1aec..f66a990b1d 100644 --- a/src/link/MachO/thunks.zig +++ b/src/link/MachO/thunks.zig @@ -99,7 +99,7 @@ pub const Thunk = struct { const sym = macho_file.getSymbol(sym_index); const saddr = thunk.getAddress(macho_file) + i * trampoline_size; const taddr = sym.getAddress(.{}, macho_file); - const pages = try aarch64.calcNumberOfPages(saddr, taddr); + const pages = try aarch64.calcNumberOfPages(@intCast(saddr), @intCast(taddr)); try writer.writeInt(u32, aarch64.Instruction.adrp(.x16, pages).toU32(), .little); const off: u12 = @truncate(taddr); try writer.writeInt(u32, aarch64.Instruction.add(.x16, .x16, off, false).toU32(), .little); diff --git a/src/link/aarch64.zig b/src/link/aarch64.zig index 576249e988..8b79810c8d 100644 --- a/src/link/aarch64.zig +++ b/src/link/aarch64.zig @@ -25,7 +25,7 @@ pub fn writeLoadStoreRegInst(value: u12, code: *[4]u8) void { mem.writeInt(u32, code, inst.toU32(), .little); } -pub fn calcNumberOfPages(saddr: u64, taddr: u64) error{Overflow}!i21 { +pub fn calcNumberOfPages(saddr: i64, taddr: i64) error{Overflow}!i21 { const spage = math.cast(i32, saddr >> 12) orelse return error.Overflow; const tpage = math.cast(i32, taddr >> 12) orelse return error.Overflow; const pages = math.cast(i21, tpage - spage) orelse return error.Overflow; |
