[clang] Document the return value of __builtin_COLUMN (#118360)
[llvm-project.git] / llvm / lib / MC / MachObjectWriter.cpp
bloba1842f133ef60524b1237c3f2896e81b0a2b1444
1 //===- lib/MC/MachObjectWriter.cpp - Mach-O File Writer -------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
9 #include "llvm/ADT/DenseMap.h"
10 #include "llvm/ADT/Twine.h"
11 #include "llvm/BinaryFormat/MachO.h"
12 #include "llvm/MC/MCAsmBackend.h"
13 #include "llvm/MC/MCAsmInfoDarwin.h"
14 #include "llvm/MC/MCAssembler.h"
15 #include "llvm/MC/MCContext.h"
16 #include "llvm/MC/MCDirectives.h"
17 #include "llvm/MC/MCExpr.h"
18 #include "llvm/MC/MCFixupKindInfo.h"
19 #include "llvm/MC/MCFragment.h"
20 #include "llvm/MC/MCMachObjectWriter.h"
21 #include "llvm/MC/MCObjectFileInfo.h"
22 #include "llvm/MC/MCObjectWriter.h"
23 #include "llvm/MC/MCSection.h"
24 #include "llvm/MC/MCSectionMachO.h"
25 #include "llvm/MC/MCSymbol.h"
26 #include "llvm/MC/MCSymbolMachO.h"
27 #include "llvm/MC/MCValue.h"
28 #include "llvm/Support/Alignment.h"
29 #include "llvm/Support/Casting.h"
30 #include "llvm/Support/Debug.h"
31 #include "llvm/Support/ErrorHandling.h"
32 #include "llvm/Support/MathExtras.h"
33 #include "llvm/Support/raw_ostream.h"
34 #include <algorithm>
35 #include <cassert>
36 #include <cstdint>
37 #include <string>
38 #include <utility>
39 #include <vector>
41 using namespace llvm;
43 #define DEBUG_TYPE "mc"
45 void MachObjectWriter::reset() {
46 Relocations.clear();
47 IndirectSymBase.clear();
48 IndirectSymbols.clear();
49 DataRegions.clear();
50 SectionAddress.clear();
51 SectionOrder.clear();
52 StringTable.clear();
53 LocalSymbolData.clear();
54 ExternalSymbolData.clear();
55 UndefinedSymbolData.clear();
56 LOHContainer.reset();
57 VersionInfo.Major = 0;
58 VersionInfo.SDKVersion = VersionTuple();
59 TargetVariantVersionInfo.Major = 0;
60 TargetVariantVersionInfo.SDKVersion = VersionTuple();
61 LinkerOptions.clear();
62 MCObjectWriter::reset();
65 bool MachObjectWriter::doesSymbolRequireExternRelocation(const MCSymbol &S) {
66 // Undefined symbols are always extern.
67 if (S.isUndefined())
68 return true;
70 // References to weak definitions require external relocation entries; the
71 // definition may not always be the one in the same object file.
72 if (cast<MCSymbolMachO>(S).isWeakDefinition())
73 return true;
75 // Otherwise, we can use an internal relocation.
76 return false;
79 bool MachObjectWriter::
80 MachSymbolData::operator<(const MachSymbolData &RHS) const {
81 return Symbol->getName() < RHS.Symbol->getName();
84 bool MachObjectWriter::isFixupKindPCRel(const MCAssembler &Asm, unsigned Kind) {
85 const MCFixupKindInfo &FKI = Asm.getBackend().getFixupKindInfo(
86 (MCFixupKind) Kind);
88 return FKI.Flags & MCFixupKindInfo::FKF_IsPCRel;
91 uint64_t
92 MachObjectWriter::getFragmentAddress(const MCAssembler &Asm,
93 const MCFragment *Fragment) const {
94 return getSectionAddress(Fragment->getParent()) +
95 Asm.getFragmentOffset(*Fragment);
98 uint64_t MachObjectWriter::getSymbolAddress(const MCSymbol &S,
99 const MCAssembler &Asm) const {
100 // If this is a variable, then recursively evaluate now.
101 if (S.isVariable()) {
102 if (const MCConstantExpr *C =
103 dyn_cast<const MCConstantExpr>(S.getVariableValue()))
104 return C->getValue();
106 MCValue Target;
107 if (!S.getVariableValue()->evaluateAsRelocatable(Target, &Asm, nullptr))
108 report_fatal_error("unable to evaluate offset for variable '" +
109 S.getName() + "'");
111 // Verify that any used symbols are defined.
112 if (Target.getSymA() && Target.getSymA()->getSymbol().isUndefined())
113 report_fatal_error("unable to evaluate offset to undefined symbol '" +
114 Target.getSymA()->getSymbol().getName() + "'");
115 if (Target.getSymB() && Target.getSymB()->getSymbol().isUndefined())
116 report_fatal_error("unable to evaluate offset to undefined symbol '" +
117 Target.getSymB()->getSymbol().getName() + "'");
119 uint64_t Address = Target.getConstant();
120 if (Target.getSymA())
121 Address += getSymbolAddress(Target.getSymA()->getSymbol(), Asm);
122 if (Target.getSymB())
123 Address += getSymbolAddress(Target.getSymB()->getSymbol(), Asm);
124 return Address;
127 return getSectionAddress(S.getFragment()->getParent()) +
128 Asm.getSymbolOffset(S);
131 uint64_t MachObjectWriter::getPaddingSize(const MCAssembler &Asm,
132 const MCSection *Sec) const {
133 uint64_t EndAddr = getSectionAddress(Sec) + Asm.getSectionAddressSize(*Sec);
134 unsigned Next = cast<MCSectionMachO>(Sec)->getLayoutOrder() + 1;
135 if (Next >= SectionOrder.size())
136 return 0;
138 const MCSection &NextSec = *SectionOrder[Next];
139 if (NextSec.isVirtualSection())
140 return 0;
141 return offsetToAlignment(EndAddr, NextSec.getAlign());
144 static bool isSymbolLinkerVisible(const MCSymbol &Symbol) {
145 // Non-temporary labels should always be visible to the linker.
146 if (!Symbol.isTemporary())
147 return true;
149 if (Symbol.isUsedInReloc())
150 return true;
152 return false;
155 const MCSymbol *MachObjectWriter::getAtom(const MCSymbol &S) const {
156 // Linker visible symbols define atoms.
157 if (isSymbolLinkerVisible(S))
158 return &S;
160 // Absolute and undefined symbols have no defining atom.
161 if (!S.isInSection())
162 return nullptr;
164 // Non-linker visible symbols in sections which can't be atomized have no
165 // defining atom.
166 if (!MCAsmInfoDarwin::isSectionAtomizableBySymbols(
167 *S.getFragment()->getParent()))
168 return nullptr;
170 // Otherwise, return the atom for the containing fragment.
171 return S.getFragment()->getAtom();
174 void MachObjectWriter::writeHeader(MachO::HeaderFileType Type,
175 unsigned NumLoadCommands,
176 unsigned LoadCommandsSize,
177 bool SubsectionsViaSymbols) {
178 uint32_t Flags = 0;
180 if (SubsectionsViaSymbols)
181 Flags |= MachO::MH_SUBSECTIONS_VIA_SYMBOLS;
183 // struct mach_header (28 bytes) or
184 // struct mach_header_64 (32 bytes)
186 uint64_t Start = W.OS.tell();
187 (void) Start;
189 W.write<uint32_t>(is64Bit() ? MachO::MH_MAGIC_64 : MachO::MH_MAGIC);
191 W.write<uint32_t>(TargetObjectWriter->getCPUType());
193 uint32_t Cpusubtype = TargetObjectWriter->getCPUSubtype();
195 // Promote arm64e subtypes to always be ptrauth-ABI-versioned, at version 0.
196 // We never need to emit unversioned binaries.
197 // And we don't support arbitrary ABI versions (or the kernel flag) yet.
198 if (TargetObjectWriter->getCPUType() == MachO::CPU_TYPE_ARM64 &&
199 Cpusubtype == MachO::CPU_SUBTYPE_ARM64E)
200 Cpusubtype = MachO::CPU_SUBTYPE_ARM64E_WITH_PTRAUTH_VERSION(
201 /*PtrAuthABIVersion=*/0, /*PtrAuthKernelABIVersion=*/false);
203 W.write<uint32_t>(Cpusubtype);
205 W.write<uint32_t>(Type);
206 W.write<uint32_t>(NumLoadCommands);
207 W.write<uint32_t>(LoadCommandsSize);
208 W.write<uint32_t>(Flags);
209 if (is64Bit())
210 W.write<uint32_t>(0); // reserved
212 assert(W.OS.tell() - Start == (is64Bit() ? sizeof(MachO::mach_header_64)
213 : sizeof(MachO::mach_header)));
216 void MachObjectWriter::writeWithPadding(StringRef Str, uint64_t Size) {
217 assert(Size >= Str.size());
218 W.OS << Str;
219 W.OS.write_zeros(Size - Str.size());
222 /// writeSegmentLoadCommand - Write a segment load command.
224 /// \param NumSections The number of sections in this segment.
225 /// \param SectionDataSize The total size of the sections.
226 void MachObjectWriter::writeSegmentLoadCommand(
227 StringRef Name, unsigned NumSections, uint64_t VMAddr, uint64_t VMSize,
228 uint64_t SectionDataStartOffset, uint64_t SectionDataSize, uint32_t MaxProt,
229 uint32_t InitProt) {
230 // struct segment_command (56 bytes) or
231 // struct segment_command_64 (72 bytes)
233 uint64_t Start = W.OS.tell();
234 (void) Start;
236 unsigned SegmentLoadCommandSize =
237 is64Bit() ? sizeof(MachO::segment_command_64):
238 sizeof(MachO::segment_command);
239 W.write<uint32_t>(is64Bit() ? MachO::LC_SEGMENT_64 : MachO::LC_SEGMENT);
240 W.write<uint32_t>(SegmentLoadCommandSize +
241 NumSections * (is64Bit() ? sizeof(MachO::section_64) :
242 sizeof(MachO::section)));
244 writeWithPadding(Name, 16);
245 if (is64Bit()) {
246 W.write<uint64_t>(VMAddr); // vmaddr
247 W.write<uint64_t>(VMSize); // vmsize
248 W.write<uint64_t>(SectionDataStartOffset); // file offset
249 W.write<uint64_t>(SectionDataSize); // file size
250 } else {
251 W.write<uint32_t>(VMAddr); // vmaddr
252 W.write<uint32_t>(VMSize); // vmsize
253 W.write<uint32_t>(SectionDataStartOffset); // file offset
254 W.write<uint32_t>(SectionDataSize); // file size
256 // maxprot
257 W.write<uint32_t>(MaxProt);
258 // initprot
259 W.write<uint32_t>(InitProt);
260 W.write<uint32_t>(NumSections);
261 W.write<uint32_t>(0); // flags
263 assert(W.OS.tell() - Start == SegmentLoadCommandSize);
266 void MachObjectWriter::writeSection(const MCAssembler &Asm,
267 const MCSection &Sec, uint64_t VMAddr,
268 uint64_t FileOffset, unsigned Flags,
269 uint64_t RelocationsStart,
270 unsigned NumRelocations) {
271 uint64_t SectionSize = Asm.getSectionAddressSize(Sec);
272 const MCSectionMachO &Section = cast<MCSectionMachO>(Sec);
274 // The offset is unused for virtual sections.
275 if (Section.isVirtualSection()) {
276 assert(Asm.getSectionFileSize(Sec) == 0 && "Invalid file size!");
277 FileOffset = 0;
280 // struct section (68 bytes) or
281 // struct section_64 (80 bytes)
283 uint64_t Start = W.OS.tell();
284 (void) Start;
286 writeWithPadding(Section.getName(), 16);
287 writeWithPadding(Section.getSegmentName(), 16);
288 if (is64Bit()) {
289 W.write<uint64_t>(VMAddr); // address
290 W.write<uint64_t>(SectionSize); // size
291 } else {
292 W.write<uint32_t>(VMAddr); // address
293 W.write<uint32_t>(SectionSize); // size
295 assert(isUInt<32>(FileOffset) && "Cannot encode offset of section");
296 W.write<uint32_t>(FileOffset);
298 W.write<uint32_t>(Log2(Section.getAlign()));
299 assert((!NumRelocations || isUInt<32>(RelocationsStart)) &&
300 "Cannot encode offset of relocations");
301 W.write<uint32_t>(NumRelocations ? RelocationsStart : 0);
302 W.write<uint32_t>(NumRelocations);
303 W.write<uint32_t>(Flags);
304 W.write<uint32_t>(IndirectSymBase.lookup(&Sec)); // reserved1
305 W.write<uint32_t>(Section.getStubSize()); // reserved2
306 if (is64Bit())
307 W.write<uint32_t>(0); // reserved3
309 assert(W.OS.tell() - Start ==
310 (is64Bit() ? sizeof(MachO::section_64) : sizeof(MachO::section)));
313 void MachObjectWriter::writeSymtabLoadCommand(uint32_t SymbolOffset,
314 uint32_t NumSymbols,
315 uint32_t StringTableOffset,
316 uint32_t StringTableSize) {
317 // struct symtab_command (24 bytes)
319 uint64_t Start = W.OS.tell();
320 (void) Start;
322 W.write<uint32_t>(MachO::LC_SYMTAB);
323 W.write<uint32_t>(sizeof(MachO::symtab_command));
324 W.write<uint32_t>(SymbolOffset);
325 W.write<uint32_t>(NumSymbols);
326 W.write<uint32_t>(StringTableOffset);
327 W.write<uint32_t>(StringTableSize);
329 assert(W.OS.tell() - Start == sizeof(MachO::symtab_command));
332 void MachObjectWriter::writeDysymtabLoadCommand(uint32_t FirstLocalSymbol,
333 uint32_t NumLocalSymbols,
334 uint32_t FirstExternalSymbol,
335 uint32_t NumExternalSymbols,
336 uint32_t FirstUndefinedSymbol,
337 uint32_t NumUndefinedSymbols,
338 uint32_t IndirectSymbolOffset,
339 uint32_t NumIndirectSymbols) {
340 // struct dysymtab_command (80 bytes)
342 uint64_t Start = W.OS.tell();
343 (void) Start;
345 W.write<uint32_t>(MachO::LC_DYSYMTAB);
346 W.write<uint32_t>(sizeof(MachO::dysymtab_command));
347 W.write<uint32_t>(FirstLocalSymbol);
348 W.write<uint32_t>(NumLocalSymbols);
349 W.write<uint32_t>(FirstExternalSymbol);
350 W.write<uint32_t>(NumExternalSymbols);
351 W.write<uint32_t>(FirstUndefinedSymbol);
352 W.write<uint32_t>(NumUndefinedSymbols);
353 W.write<uint32_t>(0); // tocoff
354 W.write<uint32_t>(0); // ntoc
355 W.write<uint32_t>(0); // modtaboff
356 W.write<uint32_t>(0); // nmodtab
357 W.write<uint32_t>(0); // extrefsymoff
358 W.write<uint32_t>(0); // nextrefsyms
359 W.write<uint32_t>(IndirectSymbolOffset);
360 W.write<uint32_t>(NumIndirectSymbols);
361 W.write<uint32_t>(0); // extreloff
362 W.write<uint32_t>(0); // nextrel
363 W.write<uint32_t>(0); // locreloff
364 W.write<uint32_t>(0); // nlocrel
366 assert(W.OS.tell() - Start == sizeof(MachO::dysymtab_command));
369 MachObjectWriter::MachSymbolData *
370 MachObjectWriter::findSymbolData(const MCSymbol &Sym) {
371 for (auto *SymbolData :
372 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
373 for (MachSymbolData &Entry : *SymbolData)
374 if (Entry.Symbol == &Sym)
375 return &Entry;
377 return nullptr;
380 const MCSymbol &MachObjectWriter::findAliasedSymbol(const MCSymbol &Sym) const {
381 const MCSymbol *S = &Sym;
382 while (S->isVariable()) {
383 const MCExpr *Value = S->getVariableValue();
384 const auto *Ref = dyn_cast<MCSymbolRefExpr>(Value);
385 if (!Ref)
386 return *S;
387 S = &Ref->getSymbol();
389 return *S;
392 void MachObjectWriter::writeNlist(MachSymbolData &MSD, const MCAssembler &Asm) {
393 const MCSymbol *Symbol = MSD.Symbol;
394 const auto &Data = cast<MCSymbolMachO>(*Symbol);
395 const MCSymbol *AliasedSymbol = &findAliasedSymbol(*Symbol);
396 uint8_t SectionIndex = MSD.SectionIndex;
397 uint8_t Type = 0;
398 uint64_t Address = 0;
399 bool IsAlias = Symbol != AliasedSymbol;
401 const MCSymbol &OrigSymbol = *Symbol;
402 MachSymbolData *AliaseeInfo;
403 if (IsAlias) {
404 AliaseeInfo = findSymbolData(*AliasedSymbol);
405 if (AliaseeInfo)
406 SectionIndex = AliaseeInfo->SectionIndex;
407 Symbol = AliasedSymbol;
408 // FIXME: Should this update Data as well?
411 // Set the N_TYPE bits. See <mach-o/nlist.h>.
413 // FIXME: Are the prebound or indirect fields possible here?
414 if (IsAlias && Symbol->isUndefined())
415 Type = MachO::N_INDR;
416 else if (Symbol->isUndefined())
417 Type = MachO::N_UNDF;
418 else if (Symbol->isAbsolute())
419 Type = MachO::N_ABS;
420 else
421 Type = MachO::N_SECT;
423 // FIXME: Set STAB bits.
425 if (Data.isPrivateExtern())
426 Type |= MachO::N_PEXT;
428 // Set external bit.
429 if (Data.isExternal() || (!IsAlias && Symbol->isUndefined()))
430 Type |= MachO::N_EXT;
432 // Compute the symbol address.
433 if (IsAlias && Symbol->isUndefined())
434 Address = AliaseeInfo->StringIndex;
435 else if (Symbol->isDefined())
436 Address = getSymbolAddress(OrigSymbol, Asm);
437 else if (Symbol->isCommon()) {
438 // Common symbols are encoded with the size in the address
439 // field, and their alignment in the flags.
440 Address = Symbol->getCommonSize();
443 // struct nlist (12 bytes)
445 W.write<uint32_t>(MSD.StringIndex);
446 W.OS << char(Type);
447 W.OS << char(SectionIndex);
449 // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc'
450 // value.
451 bool EncodeAsAltEntry =
452 IsAlias && cast<MCSymbolMachO>(OrigSymbol).isAltEntry();
453 W.write<uint16_t>(cast<MCSymbolMachO>(Symbol)->getEncodedFlags(EncodeAsAltEntry));
454 if (is64Bit())
455 W.write<uint64_t>(Address);
456 else
457 W.write<uint32_t>(Address);
460 void MachObjectWriter::writeLinkeditLoadCommand(uint32_t Type,
461 uint32_t DataOffset,
462 uint32_t DataSize) {
463 uint64_t Start = W.OS.tell();
464 (void) Start;
466 W.write<uint32_t>(Type);
467 W.write<uint32_t>(sizeof(MachO::linkedit_data_command));
468 W.write<uint32_t>(DataOffset);
469 W.write<uint32_t>(DataSize);
471 assert(W.OS.tell() - Start == sizeof(MachO::linkedit_data_command));
474 static unsigned ComputeLinkerOptionsLoadCommandSize(
475 const std::vector<std::string> &Options, bool is64Bit)
477 unsigned Size = sizeof(MachO::linker_option_command);
478 for (const std::string &Option : Options)
479 Size += Option.size() + 1;
480 return alignTo(Size, is64Bit ? 8 : 4);
483 void MachObjectWriter::writeLinkerOptionsLoadCommand(
484 const std::vector<std::string> &Options)
486 unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit());
487 uint64_t Start = W.OS.tell();
488 (void) Start;
490 W.write<uint32_t>(MachO::LC_LINKER_OPTION);
491 W.write<uint32_t>(Size);
492 W.write<uint32_t>(Options.size());
493 uint64_t BytesWritten = sizeof(MachO::linker_option_command);
494 for (const std::string &Option : Options) {
495 // Write each string, including the null byte.
496 W.OS << Option << '\0';
497 BytesWritten += Option.size() + 1;
500 // Pad to a multiple of the pointer size.
501 W.OS.write_zeros(
502 offsetToAlignment(BytesWritten, is64Bit() ? Align(8) : Align(4)));
504 assert(W.OS.tell() - Start == Size);
507 static bool isFixupTargetValid(const MCValue &Target) {
508 // Target is (LHS - RHS + cst).
509 // We don't support the form where LHS is null: -RHS + cst
510 if (!Target.getSymA() && Target.getSymB())
511 return false;
512 return true;
515 void MachObjectWriter::recordRelocation(MCAssembler &Asm,
516 const MCFragment *Fragment,
517 const MCFixup &Fixup, MCValue Target,
518 uint64_t &FixedValue) {
519 if (!isFixupTargetValid(Target)) {
520 Asm.getContext().reportError(Fixup.getLoc(),
521 "unsupported relocation expression");
522 return;
525 TargetObjectWriter->recordRelocation(this, Asm, Fragment, Fixup, Target,
526 FixedValue);
529 void MachObjectWriter::bindIndirectSymbols(MCAssembler &Asm) {
530 // This is the point where 'as' creates actual symbols for indirect symbols
531 // (in the following two passes). It would be easier for us to do this sooner
532 // when we see the attribute, but that makes getting the order in the symbol
533 // table much more complicated than it is worth.
535 // FIXME: Revisit this when the dust settles.
537 // Report errors for use of .indirect_symbol not in a symbol pointer section
538 // or stub section.
539 for (IndirectSymbolData &ISD : IndirectSymbols) {
540 const MCSectionMachO &Section = cast<MCSectionMachO>(*ISD.Section);
542 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
543 Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
544 Section.getType() != MachO::S_THREAD_LOCAL_VARIABLE_POINTERS &&
545 Section.getType() != MachO::S_SYMBOL_STUBS) {
546 MCSymbol &Symbol = *ISD.Symbol;
547 report_fatal_error("indirect symbol '" + Symbol.getName() +
548 "' not in a symbol pointer or stub section");
552 // Bind non-lazy symbol pointers first.
553 for (auto [IndirectIndex, ISD] : enumerate(IndirectSymbols)) {
554 const auto &Section = cast<MCSectionMachO>(*ISD.Section);
556 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
557 Section.getType() != MachO::S_THREAD_LOCAL_VARIABLE_POINTERS)
558 continue;
560 // Initialize the section indirect symbol base, if necessary.
561 IndirectSymBase.insert(std::make_pair(ISD.Section, IndirectIndex));
563 Asm.registerSymbol(*ISD.Symbol);
566 // Then lazy symbol pointers and symbol stubs.
567 for (auto [IndirectIndex, ISD] : enumerate(IndirectSymbols)) {
568 const auto &Section = cast<MCSectionMachO>(*ISD.Section);
570 if (Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
571 Section.getType() != MachO::S_SYMBOL_STUBS)
572 continue;
574 // Initialize the section indirect symbol base, if necessary.
575 IndirectSymBase.insert(std::make_pair(ISD.Section, IndirectIndex));
577 // Set the symbol type to undefined lazy, but only on construction.
579 // FIXME: Do not hardcode.
580 if (Asm.registerSymbol(*ISD.Symbol))
581 cast<MCSymbolMachO>(ISD.Symbol)->setReferenceTypeUndefinedLazy(true);
585 /// computeSymbolTable - Compute the symbol table data
586 void MachObjectWriter::computeSymbolTable(
587 MCAssembler &Asm, std::vector<MachSymbolData> &LocalSymbolData,
588 std::vector<MachSymbolData> &ExternalSymbolData,
589 std::vector<MachSymbolData> &UndefinedSymbolData) {
590 // Build section lookup table.
591 DenseMap<const MCSection*, uint8_t> SectionIndexMap;
592 unsigned Index = 1;
593 for (MCSection &Sec : Asm)
594 SectionIndexMap[&Sec] = Index++;
595 assert(Index <= 256 && "Too many sections!");
597 // Build the string table.
598 for (const MCSymbol &Symbol : Asm.symbols()) {
599 if (!cast<MCSymbolMachO>(Symbol).isSymbolLinkerVisible())
600 continue;
602 StringTable.add(Symbol.getName());
604 StringTable.finalize();
606 // Build the symbol arrays but only for non-local symbols.
608 // The particular order that we collect and then sort the symbols is chosen to
609 // match 'as'. Even though it doesn't matter for correctness, this is
610 // important for letting us diff .o files.
611 for (const MCSymbol &Symbol : Asm.symbols()) {
612 // Ignore non-linker visible symbols.
613 if (!cast<MCSymbolMachO>(Symbol).isSymbolLinkerVisible())
614 continue;
616 if (!Symbol.isExternal() && !Symbol.isUndefined())
617 continue;
619 MachSymbolData MSD;
620 MSD.Symbol = &Symbol;
621 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
623 if (Symbol.isUndefined()) {
624 MSD.SectionIndex = 0;
625 UndefinedSymbolData.push_back(MSD);
626 } else if (Symbol.isAbsolute()) {
627 MSD.SectionIndex = 0;
628 ExternalSymbolData.push_back(MSD);
629 } else {
630 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
631 assert(MSD.SectionIndex && "Invalid section index!");
632 ExternalSymbolData.push_back(MSD);
636 // Now add the data for local symbols.
637 for (const MCSymbol &Symbol : Asm.symbols()) {
638 // Ignore non-linker visible symbols.
639 if (!cast<MCSymbolMachO>(Symbol).isSymbolLinkerVisible())
640 continue;
642 if (Symbol.isExternal() || Symbol.isUndefined())
643 continue;
645 MachSymbolData MSD;
646 MSD.Symbol = &Symbol;
647 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
649 if (Symbol.isAbsolute()) {
650 MSD.SectionIndex = 0;
651 LocalSymbolData.push_back(MSD);
652 } else {
653 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
654 assert(MSD.SectionIndex && "Invalid section index!");
655 LocalSymbolData.push_back(MSD);
659 // External and undefined symbols are required to be in lexicographic order.
660 llvm::sort(ExternalSymbolData);
661 llvm::sort(UndefinedSymbolData);
663 // Set the symbol indices.
664 Index = 0;
665 for (auto *SymbolData :
666 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
667 for (MachSymbolData &Entry : *SymbolData)
668 Entry.Symbol->setIndex(Index++);
670 for (const MCSection &Section : Asm) {
671 for (RelAndSymbol &Rel : Relocations[&Section]) {
672 if (!Rel.Sym)
673 continue;
675 // Set the Index and the IsExtern bit.
676 unsigned Index = Rel.Sym->getIndex();
677 assert(isInt<24>(Index));
678 if (W.Endian == llvm::endianness::little)
679 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & (~0U << 24)) | Index | (1 << 27);
680 else
681 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & 0xff) | Index << 8 | (1 << 4);
686 void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm) {
687 // Assign layout order indices to sections.
688 unsigned i = 0;
689 // Compute the section layout order. Virtual sections must go last.
690 for (MCSection &Sec : Asm) {
691 if (!Sec.isVirtualSection()) {
692 SectionOrder.push_back(&Sec);
693 cast<MCSectionMachO>(Sec).setLayoutOrder(i++);
696 for (MCSection &Sec : Asm) {
697 if (Sec.isVirtualSection()) {
698 SectionOrder.push_back(&Sec);
699 cast<MCSectionMachO>(Sec).setLayoutOrder(i++);
703 uint64_t StartAddress = 0;
704 for (const MCSection *Sec : SectionOrder) {
705 StartAddress = alignTo(StartAddress, Sec->getAlign());
706 SectionAddress[Sec] = StartAddress;
707 StartAddress += Asm.getSectionAddressSize(*Sec);
709 // Explicitly pad the section to match the alignment requirements of the
710 // following one. This is for 'gas' compatibility, it shouldn't
711 /// strictly be necessary.
712 StartAddress += getPaddingSize(Asm, Sec);
716 void MachObjectWriter::executePostLayoutBinding(MCAssembler &Asm) {
717 computeSectionAddresses(Asm);
719 // Create symbol data for any indirect symbols.
720 bindIndirectSymbols(Asm);
723 bool MachObjectWriter::isSymbolRefDifferenceFullyResolvedImpl(
724 const MCAssembler &Asm, const MCSymbol &SymA, const MCFragment &FB,
725 bool InSet, bool IsPCRel) const {
726 if (InSet)
727 return true;
729 // The effective address is
730 // addr(atom(A)) + offset(A)
731 // - addr(atom(B)) - offset(B)
732 // and the offsets are not relocatable, so the fixup is fully resolved when
733 // addr(atom(A)) - addr(atom(B)) == 0.
734 const MCSymbol &SA = findAliasedSymbol(SymA);
735 const MCSection &SecA = SA.getSection();
736 const MCSection &SecB = *FB.getParent();
738 if (IsPCRel) {
739 // The simple (Darwin, except on x86_64) way of dealing with this was to
740 // assume that any reference to a temporary symbol *must* be a temporary
741 // symbol in the same atom, unless the sections differ. Therefore, any PCrel
742 // relocation to a temporary symbol (in the same section) is fully
743 // resolved. This also works in conjunction with absolutized .set, which
744 // requires the compiler to use .set to absolutize the differences between
745 // symbols which the compiler knows to be assembly time constants, so we
746 // don't need to worry about considering symbol differences fully resolved.
748 // If the file isn't using sub-sections-via-symbols, we can make the
749 // same assumptions about any symbol that we normally make about
750 // assembler locals.
752 bool hasReliableSymbolDifference = isX86_64();
753 if (!hasReliableSymbolDifference) {
754 if (!SA.isInSection() || &SecA != &SecB ||
755 (!SA.isTemporary() && FB.getAtom() != SA.getFragment()->getAtom() &&
756 SubsectionsViaSymbols))
757 return false;
758 return true;
762 // If they are not in the same section, we can't compute the diff.
763 if (&SecA != &SecB)
764 return false;
766 // If the atoms are the same, they are guaranteed to have the same address.
767 return SA.getFragment()->getAtom() == FB.getAtom();
770 static MachO::LoadCommandType getLCFromMCVM(MCVersionMinType Type) {
771 switch (Type) {
772 case MCVM_OSXVersionMin: return MachO::LC_VERSION_MIN_MACOSX;
773 case MCVM_IOSVersionMin: return MachO::LC_VERSION_MIN_IPHONEOS;
774 case MCVM_TvOSVersionMin: return MachO::LC_VERSION_MIN_TVOS;
775 case MCVM_WatchOSVersionMin: return MachO::LC_VERSION_MIN_WATCHOS;
777 llvm_unreachable("Invalid mc version min type");
780 void MachObjectWriter::populateAddrSigSection(MCAssembler &Asm) {
781 MCSection *AddrSigSection =
782 Asm.getContext().getObjectFileInfo()->getAddrSigSection();
783 unsigned Log2Size = is64Bit() ? 3 : 2;
784 for (const MCSymbol *S : getAddrsigSyms()) {
785 if (!S->isRegistered())
786 continue;
787 MachO::any_relocation_info MRE;
788 MRE.r_word0 = 0;
789 MRE.r_word1 = (Log2Size << 25) | (MachO::GENERIC_RELOC_VANILLA << 28);
790 addRelocation(S, AddrSigSection, MRE);
794 uint64_t MachObjectWriter::writeObject(MCAssembler &Asm) {
795 uint64_t StartOffset = W.OS.tell();
796 auto NumBytesWritten = [&] { return W.OS.tell() - StartOffset; };
798 populateAddrSigSection(Asm);
800 // Compute symbol table information and bind symbol indices.
801 computeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData,
802 UndefinedSymbolData);
804 if (!CGProfile.empty()) {
805 MCSection *CGProfileSection = Asm.getContext().getMachOSection(
806 "__LLVM", "__cg_profile", 0, SectionKind::getMetadata());
807 auto &Frag = cast<MCDataFragment>(*CGProfileSection->begin());
808 Frag.getContents().clear();
809 raw_svector_ostream OS(Frag.getContents());
810 for (const MCObjectWriter::CGProfileEntry &CGPE : CGProfile) {
811 uint32_t FromIndex = CGPE.From->getSymbol().getIndex();
812 uint32_t ToIndex = CGPE.To->getSymbol().getIndex();
813 support::endian::write(OS, FromIndex, W.Endian);
814 support::endian::write(OS, ToIndex, W.Endian);
815 support::endian::write(OS, CGPE.Count, W.Endian);
819 unsigned NumSections = Asm.end() - Asm.begin();
821 // The section data starts after the header, the segment load command (and
822 // section headers) and the symbol table.
823 unsigned NumLoadCommands = 1;
824 uint64_t LoadCommandsSize = is64Bit() ?
825 sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64):
826 sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section);
828 // Add the deployment target version info load command size, if used.
829 if (VersionInfo.Major != 0) {
830 ++NumLoadCommands;
831 if (VersionInfo.EmitBuildVersion)
832 LoadCommandsSize += sizeof(MachO::build_version_command);
833 else
834 LoadCommandsSize += sizeof(MachO::version_min_command);
837 // Add the target variant version info load command size, if used.
838 if (TargetVariantVersionInfo.Major != 0) {
839 ++NumLoadCommands;
840 assert(TargetVariantVersionInfo.EmitBuildVersion &&
841 "target variant should use build version");
842 LoadCommandsSize += sizeof(MachO::build_version_command);
845 // Add the data-in-code load command size, if used.
846 unsigned NumDataRegions = DataRegions.size();
847 if (NumDataRegions) {
848 ++NumLoadCommands;
849 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
852 // Add the loh load command size, if used.
853 uint64_t LOHRawSize = LOHContainer.getEmitSize(Asm, *this);
854 uint64_t LOHSize = alignTo(LOHRawSize, is64Bit() ? 8 : 4);
855 if (LOHSize) {
856 ++NumLoadCommands;
857 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
860 // Add the symbol table load command sizes, if used.
861 unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
862 UndefinedSymbolData.size();
863 if (NumSymbols) {
864 NumLoadCommands += 2;
865 LoadCommandsSize += (sizeof(MachO::symtab_command) +
866 sizeof(MachO::dysymtab_command));
869 // Add the linker option load commands sizes.
870 for (const auto &Option : LinkerOptions) {
871 ++NumLoadCommands;
872 LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(Option, is64Bit());
875 // Compute the total size of the section data, as well as its file size and vm
876 // size.
877 uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) :
878 sizeof(MachO::mach_header)) + LoadCommandsSize;
879 uint64_t SectionDataSize = 0;
880 uint64_t SectionDataFileSize = 0;
881 uint64_t VMSize = 0;
882 for (const MCSection &Sec : Asm) {
883 uint64_t Address = getSectionAddress(&Sec);
884 uint64_t Size = Asm.getSectionAddressSize(Sec);
885 uint64_t FileSize = Asm.getSectionFileSize(Sec);
886 FileSize += getPaddingSize(Asm, &Sec);
888 VMSize = std::max(VMSize, Address + Size);
890 if (Sec.isVirtualSection())
891 continue;
893 SectionDataSize = std::max(SectionDataSize, Address + Size);
894 SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
897 // The section data is padded to pointer size bytes.
899 // FIXME: Is this machine dependent?
900 unsigned SectionDataPadding =
901 offsetToAlignment(SectionDataFileSize, is64Bit() ? Align(8) : Align(4));
902 SectionDataFileSize += SectionDataPadding;
904 // Write the prolog, starting with the header and load command...
905 writeHeader(MachO::MH_OBJECT, NumLoadCommands, LoadCommandsSize,
906 SubsectionsViaSymbols);
907 uint32_t Prot =
908 MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE;
909 writeSegmentLoadCommand("", NumSections, 0, VMSize, SectionDataStart,
910 SectionDataSize, Prot, Prot);
912 // ... and then the section headers.
913 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
914 for (const MCSection &Section : Asm) {
915 const auto &Sec = cast<MCSectionMachO>(Section);
916 std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
917 unsigned NumRelocs = Relocs.size();
918 uint64_t SectionStart = SectionDataStart + getSectionAddress(&Sec);
919 unsigned Flags = Sec.getTypeAndAttributes();
920 if (Sec.hasInstructions())
921 Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS;
922 if (!cast<MCSectionMachO>(Sec).isVirtualSection() &&
923 !isUInt<32>(SectionStart)) {
924 Asm.getContext().reportError(
925 SMLoc(), "cannot encode offset of section; object file too large");
926 return NumBytesWritten();
928 if (NumRelocs && !isUInt<32>(RelocTableEnd)) {
929 Asm.getContext().reportError(
930 SMLoc(),
931 "cannot encode offset of relocations; object file too large");
932 return NumBytesWritten();
934 writeSection(Asm, Sec, getSectionAddress(&Sec), SectionStart, Flags,
935 RelocTableEnd, NumRelocs);
936 RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
939 // Write out the deployment target information, if it's available.
940 auto EmitDeploymentTargetVersion =
941 [&](const VersionInfoType &VersionInfo) {
942 auto EncodeVersion = [](VersionTuple V) -> uint32_t {
943 assert(!V.empty() && "empty version");
944 unsigned Update = V.getSubminor().value_or(0);
945 unsigned Minor = V.getMinor().value_or(0);
946 assert(Update < 256 && "unencodable update target version");
947 assert(Minor < 256 && "unencodable minor target version");
948 assert(V.getMajor() < 65536 && "unencodable major target version");
949 return Update | (Minor << 8) | (V.getMajor() << 16);
951 uint32_t EncodedVersion = EncodeVersion(VersionTuple(
952 VersionInfo.Major, VersionInfo.Minor, VersionInfo.Update));
953 uint32_t SDKVersion = !VersionInfo.SDKVersion.empty()
954 ? EncodeVersion(VersionInfo.SDKVersion)
955 : 0;
956 if (VersionInfo.EmitBuildVersion) {
957 // FIXME: Currently empty tools. Add clang version in the future.
958 W.write<uint32_t>(MachO::LC_BUILD_VERSION);
959 W.write<uint32_t>(sizeof(MachO::build_version_command));
960 W.write<uint32_t>(VersionInfo.TypeOrPlatform.Platform);
961 W.write<uint32_t>(EncodedVersion);
962 W.write<uint32_t>(SDKVersion);
963 W.write<uint32_t>(0); // Empty tools list.
964 } else {
965 MachO::LoadCommandType LCType =
966 getLCFromMCVM(VersionInfo.TypeOrPlatform.Type);
967 W.write<uint32_t>(LCType);
968 W.write<uint32_t>(sizeof(MachO::version_min_command));
969 W.write<uint32_t>(EncodedVersion);
970 W.write<uint32_t>(SDKVersion);
973 if (VersionInfo.Major != 0)
974 EmitDeploymentTargetVersion(VersionInfo);
975 if (TargetVariantVersionInfo.Major != 0)
976 EmitDeploymentTargetVersion(TargetVariantVersionInfo);
978 // Write the data-in-code load command, if used.
979 uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
980 if (NumDataRegions) {
981 uint64_t DataRegionsOffset = RelocTableEnd;
982 uint64_t DataRegionsSize = NumDataRegions * 8;
983 writeLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
984 DataRegionsSize);
987 // Write the loh load command, if used.
988 uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
989 if (LOHSize)
990 writeLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
991 DataInCodeTableEnd, LOHSize);
993 // Write the symbol table load command, if used.
994 if (NumSymbols) {
995 unsigned FirstLocalSymbol = 0;
996 unsigned NumLocalSymbols = LocalSymbolData.size();
997 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
998 unsigned NumExternalSymbols = ExternalSymbolData.size();
999 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
1000 unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
1001 unsigned NumIndirectSymbols = IndirectSymbols.size();
1002 unsigned NumSymTabSymbols =
1003 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
1004 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
1005 uint64_t IndirectSymbolOffset = 0;
1007 // If used, the indirect symbols are written after the section data.
1008 if (NumIndirectSymbols)
1009 IndirectSymbolOffset = LOHTableEnd;
1011 // The symbol table is written after the indirect symbol data.
1012 uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
1014 // The string table is written after symbol table.
1015 uint64_t StringTableOffset =
1016 SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
1017 sizeof(MachO::nlist_64) :
1018 sizeof(MachO::nlist));
1019 writeSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
1020 StringTableOffset, StringTable.getSize());
1022 writeDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
1023 FirstExternalSymbol, NumExternalSymbols,
1024 FirstUndefinedSymbol, NumUndefinedSymbols,
1025 IndirectSymbolOffset, NumIndirectSymbols);
1028 // Write the linker options load commands.
1029 for (const auto &Option : LinkerOptions)
1030 writeLinkerOptionsLoadCommand(Option);
1032 // Write the actual section data.
1033 for (const MCSection &Sec : Asm) {
1034 Asm.writeSectionData(W.OS, &Sec);
1036 uint64_t Pad = getPaddingSize(Asm, &Sec);
1037 W.OS.write_zeros(Pad);
1040 // Write the extra padding.
1041 W.OS.write_zeros(SectionDataPadding);
1043 // Write the relocation entries.
1044 for (const MCSection &Sec : Asm) {
1045 // Write the section relocation entries, in reverse order to match 'as'
1046 // (approximately, the exact algorithm is more complicated than this).
1047 std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
1048 for (const RelAndSymbol &Rel : llvm::reverse(Relocs)) {
1049 W.write<uint32_t>(Rel.MRE.r_word0);
1050 W.write<uint32_t>(Rel.MRE.r_word1);
1054 // Write out the data-in-code region payload, if there is one.
1055 for (DataRegionData Data : DataRegions) {
1056 uint64_t Start = getSymbolAddress(*Data.Start, Asm);
1057 uint64_t End;
1058 if (Data.End)
1059 End = getSymbolAddress(*Data.End, Asm);
1060 else
1061 report_fatal_error("Data region not terminated");
1063 LLVM_DEBUG(dbgs() << "data in code region-- kind: " << Data.Kind
1064 << " start: " << Start << "(" << Data.Start->getName()
1065 << ")" << " end: " << End << "(" << Data.End->getName()
1066 << ")" << " size: " << End - Start << "\n");
1067 W.write<uint32_t>(Start);
1068 W.write<uint16_t>(End - Start);
1069 W.write<uint16_t>(Data.Kind);
1072 // Write out the loh commands, if there is one.
1073 if (LOHSize) {
1074 #ifndef NDEBUG
1075 unsigned Start = W.OS.tell();
1076 #endif
1077 LOHContainer.emit(Asm, *this);
1078 // Pad to a multiple of the pointer size.
1079 W.OS.write_zeros(
1080 offsetToAlignment(LOHRawSize, is64Bit() ? Align(8) : Align(4)));
1081 assert(W.OS.tell() - Start == LOHSize);
1084 // Write the symbol table data, if used.
1085 if (NumSymbols) {
1086 // Write the indirect symbol entries.
1087 for (auto &ISD : IndirectSymbols) {
1088 // Indirect symbols in the non-lazy symbol pointer section have some
1089 // special handling.
1090 const MCSectionMachO &Section =
1091 static_cast<const MCSectionMachO &>(*ISD.Section);
1092 if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
1093 // If this symbol is defined and internal, mark it as such.
1094 if (ISD.Symbol->isDefined() && !ISD.Symbol->isExternal()) {
1095 uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
1096 if (ISD.Symbol->isAbsolute())
1097 Flags |= MachO::INDIRECT_SYMBOL_ABS;
1098 W.write<uint32_t>(Flags);
1099 continue;
1103 W.write<uint32_t>(ISD.Symbol->getIndex());
1106 // FIXME: Check that offsets match computed ones.
1108 // Write the symbol table entries.
1109 for (auto *SymbolData :
1110 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
1111 for (MachSymbolData &Entry : *SymbolData)
1112 writeNlist(Entry, Asm);
1114 // Write the string table.
1115 StringTable.write(W.OS);
1118 return NumBytesWritten();