1 //===- lib/MC/MachObjectWriter.cpp - Mach-O File Writer -------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 #include "llvm/MC/MCMachObjectWriter.h"
11 #include "llvm/ADT/StringMap.h"
12 #include "llvm/ADT/Twine.h"
13 #include "llvm/MC/MCAsmBackend.h"
14 #include "llvm/MC/MCAsmLayout.h"
15 #include "llvm/MC/MCAssembler.h"
16 #include "llvm/MC/MCExpr.h"
17 #include "llvm/MC/MCFixupKindInfo.h"
18 #include "llvm/MC/MCObjectWriter.h"
19 #include "llvm/MC/MCSectionMachO.h"
20 #include "llvm/MC/MCSymbolMachO.h"
21 #include "llvm/MC/MCValue.h"
22 #include "llvm/Support/Debug.h"
23 #include "llvm/Support/ErrorHandling.h"
24 #include "llvm/Support/MachO.h"
25 #include "llvm/Support/raw_ostream.h"
29 #define DEBUG_TYPE "mc"
31 void MachObjectWriter::reset() {
33 IndirectSymBase.clear();
35 LocalSymbolData.clear();
36 ExternalSymbolData.clear();
37 UndefinedSymbolData.clear();
38 MCObjectWriter::reset();
41 bool MachObjectWriter::doesSymbolRequireExternRelocation(const MCSymbol &S) {
42 // Undefined symbols are always extern.
46 // References to weak definitions require external relocation entries; the
47 // definition may not always be the one in the same object file.
48 if (cast<MCSymbolMachO>(S).isWeakDefinition())
51 // Otherwise, we can use an internal relocation.
55 bool MachObjectWriter::
56 MachSymbolData::operator<(const MachSymbolData &RHS) const {
57 return Symbol->getName() < RHS.Symbol->getName();
60 bool MachObjectWriter::isFixupKindPCRel(const MCAssembler &Asm, unsigned Kind) {
61 const MCFixupKindInfo &FKI = Asm.getBackend().getFixupKindInfo(
64 return FKI.Flags & MCFixupKindInfo::FKF_IsPCRel;
67 uint64_t MachObjectWriter::getFragmentAddress(const MCFragment *Fragment,
68 const MCAsmLayout &Layout) const {
69 return getSectionAddress(Fragment->getParent()) +
70 Layout.getFragmentOffset(Fragment);
73 uint64_t MachObjectWriter::getSymbolAddress(const MCSymbol &S,
74 const MCAsmLayout &Layout) const {
75 // If this is a variable, then recursively evaluate now.
77 if (const MCConstantExpr *C =
78 dyn_cast<const MCConstantExpr>(S.getVariableValue()))
83 if (!S.getVariableValue()->evaluateAsRelocatable(Target, &Layout, nullptr))
84 report_fatal_error("unable to evaluate offset for variable '" +
87 // Verify that any used symbols are defined.
88 if (Target.getSymA() && Target.getSymA()->getSymbol().isUndefined())
89 report_fatal_error("unable to evaluate offset to undefined symbol '" +
90 Target.getSymA()->getSymbol().getName() + "'");
91 if (Target.getSymB() && Target.getSymB()->getSymbol().isUndefined())
92 report_fatal_error("unable to evaluate offset to undefined symbol '" +
93 Target.getSymB()->getSymbol().getName() + "'");
95 uint64_t Address = Target.getConstant();
97 Address += getSymbolAddress(Target.getSymA()->getSymbol(), Layout);
99 Address += getSymbolAddress(Target.getSymB()->getSymbol(), Layout);
103 return getSectionAddress(S.getFragment()->getParent()) +
104 Layout.getSymbolOffset(S);
107 uint64_t MachObjectWriter::getPaddingSize(const MCSection *Sec,
108 const MCAsmLayout &Layout) const {
109 uint64_t EndAddr = getSectionAddress(Sec) + Layout.getSectionAddressSize(Sec);
110 unsigned Next = Sec->getLayoutOrder() + 1;
111 if (Next >= Layout.getSectionOrder().size())
114 const MCSection &NextSec = *Layout.getSectionOrder()[Next];
115 if (NextSec.isVirtualSection())
117 return OffsetToAlignment(EndAddr, NextSec.getAlignment());
120 void MachObjectWriter::writeHeader(unsigned NumLoadCommands,
121 unsigned LoadCommandsSize,
122 bool SubsectionsViaSymbols) {
125 if (SubsectionsViaSymbols)
126 Flags |= MachO::MH_SUBSECTIONS_VIA_SYMBOLS;
128 // struct mach_header (28 bytes) or
129 // struct mach_header_64 (32 bytes)
131 uint64_t Start = OS.tell();
134 write32(is64Bit() ? MachO::MH_MAGIC_64 : MachO::MH_MAGIC);
136 write32(TargetObjectWriter->getCPUType());
137 write32(TargetObjectWriter->getCPUSubtype());
139 write32(MachO::MH_OBJECT);
140 write32(NumLoadCommands);
141 write32(LoadCommandsSize);
144 write32(0); // reserved
146 assert(OS.tell() - Start ==
147 (is64Bit()?sizeof(MachO::mach_header_64): sizeof(MachO::mach_header)));
150 /// writeSegmentLoadCommand - Write a segment load command.
152 /// \param NumSections The number of sections in this segment.
153 /// \param SectionDataSize The total size of the sections.
154 void MachObjectWriter::writeSegmentLoadCommand(unsigned NumSections,
156 uint64_t SectionDataStartOffset,
157 uint64_t SectionDataSize) {
158 // struct segment_command (56 bytes) or
159 // struct segment_command_64 (72 bytes)
161 uint64_t Start = OS.tell();
164 unsigned SegmentLoadCommandSize =
165 is64Bit() ? sizeof(MachO::segment_command_64):
166 sizeof(MachO::segment_command);
167 write32(is64Bit() ? MachO::LC_SEGMENT_64 : MachO::LC_SEGMENT);
168 write32(SegmentLoadCommandSize +
169 NumSections * (is64Bit() ? sizeof(MachO::section_64) :
170 sizeof(MachO::section)));
174 write64(0); // vmaddr
175 write64(VMSize); // vmsize
176 write64(SectionDataStartOffset); // file offset
177 write64(SectionDataSize); // file size
179 write32(0); // vmaddr
180 write32(VMSize); // vmsize
181 write32(SectionDataStartOffset); // file offset
182 write32(SectionDataSize); // file size
185 write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE);
187 write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE);
188 write32(NumSections);
191 assert(OS.tell() - Start == SegmentLoadCommandSize);
194 void MachObjectWriter::writeSection(const MCAssembler &Asm,
195 const MCAsmLayout &Layout,
196 const MCSection &Sec, uint64_t FileOffset,
197 uint64_t RelocationsStart,
198 unsigned NumRelocations) {
199 uint64_t SectionSize = Layout.getSectionAddressSize(&Sec);
200 const MCSectionMachO &Section = cast<MCSectionMachO>(Sec);
202 // The offset is unused for virtual sections.
203 if (Section.isVirtualSection()) {
204 assert(Layout.getSectionFileSize(&Sec) == 0 && "Invalid file size!");
208 // struct section (68 bytes) or
209 // struct section_64 (80 bytes)
211 uint64_t Start = OS.tell();
214 writeBytes(Section.getSectionName(), 16);
215 writeBytes(Section.getSegmentName(), 16);
217 write64(getSectionAddress(&Sec)); // address
218 write64(SectionSize); // size
220 write32(getSectionAddress(&Sec)); // address
221 write32(SectionSize); // size
225 unsigned Flags = Section.getTypeAndAttributes();
226 if (Section.hasInstructions())
227 Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS;
229 assert(isPowerOf2_32(Section.getAlignment()) && "Invalid alignment!");
230 write32(Log2_32(Section.getAlignment()));
231 write32(NumRelocations ? RelocationsStart : 0);
232 write32(NumRelocations);
234 write32(IndirectSymBase.lookup(&Sec)); // reserved1
235 write32(Section.getStubSize()); // reserved2
237 write32(0); // reserved3
239 assert(OS.tell() - Start == (is64Bit() ? sizeof(MachO::section_64) :
240 sizeof(MachO::section)));
243 void MachObjectWriter::writeSymtabLoadCommand(uint32_t SymbolOffset,
245 uint32_t StringTableOffset,
246 uint32_t StringTableSize) {
247 // struct symtab_command (24 bytes)
249 uint64_t Start = OS.tell();
252 write32(MachO::LC_SYMTAB);
253 write32(sizeof(MachO::symtab_command));
254 write32(SymbolOffset);
256 write32(StringTableOffset);
257 write32(StringTableSize);
259 assert(OS.tell() - Start == sizeof(MachO::symtab_command));
262 void MachObjectWriter::writeDysymtabLoadCommand(uint32_t FirstLocalSymbol,
263 uint32_t NumLocalSymbols,
264 uint32_t FirstExternalSymbol,
265 uint32_t NumExternalSymbols,
266 uint32_t FirstUndefinedSymbol,
267 uint32_t NumUndefinedSymbols,
268 uint32_t IndirectSymbolOffset,
269 uint32_t NumIndirectSymbols) {
270 // struct dysymtab_command (80 bytes)
272 uint64_t Start = OS.tell();
275 write32(MachO::LC_DYSYMTAB);
276 write32(sizeof(MachO::dysymtab_command));
277 write32(FirstLocalSymbol);
278 write32(NumLocalSymbols);
279 write32(FirstExternalSymbol);
280 write32(NumExternalSymbols);
281 write32(FirstUndefinedSymbol);
282 write32(NumUndefinedSymbols);
283 write32(0); // tocoff
285 write32(0); // modtaboff
286 write32(0); // nmodtab
287 write32(0); // extrefsymoff
288 write32(0); // nextrefsyms
289 write32(IndirectSymbolOffset);
290 write32(NumIndirectSymbols);
291 write32(0); // extreloff
292 write32(0); // nextrel
293 write32(0); // locreloff
294 write32(0); // nlocrel
296 assert(OS.tell() - Start == sizeof(MachO::dysymtab_command));
299 MachObjectWriter::MachSymbolData *
300 MachObjectWriter::findSymbolData(const MCSymbol &Sym) {
301 for (auto *SymbolData :
302 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
303 for (MachSymbolData &Entry : *SymbolData)
304 if (Entry.Symbol == &Sym)
310 const MCSymbol &MachObjectWriter::findAliasedSymbol(const MCSymbol &Sym) const {
311 const MCSymbol *S = &Sym;
312 while (S->isVariable()) {
313 const MCExpr *Value = S->getVariableValue();
314 const auto *Ref = dyn_cast<MCSymbolRefExpr>(Value);
317 S = &Ref->getSymbol();
322 void MachObjectWriter::writeNlist(MachSymbolData &MSD,
323 const MCAsmLayout &Layout) {
324 const MCSymbol *Symbol = MSD.Symbol;
325 const MCSymbol &Data = *Symbol;
326 const MCSymbol *AliasedSymbol = &findAliasedSymbol(*Symbol);
327 uint8_t SectionIndex = MSD.SectionIndex;
329 uint64_t Address = 0;
330 bool IsAlias = Symbol != AliasedSymbol;
332 const MCSymbol &OrigSymbol = *Symbol;
333 MachSymbolData *AliaseeInfo;
335 AliaseeInfo = findSymbolData(*AliasedSymbol);
337 SectionIndex = AliaseeInfo->SectionIndex;
338 Symbol = AliasedSymbol;
339 // FIXME: Should this update Data as well? Do we need OrigSymbol at all?
342 // Set the N_TYPE bits. See <mach-o/nlist.h>.
344 // FIXME: Are the prebound or indirect fields possible here?
345 if (IsAlias && Symbol->isUndefined())
346 Type = MachO::N_INDR;
347 else if (Symbol->isUndefined())
348 Type = MachO::N_UNDF;
349 else if (Symbol->isAbsolute())
352 Type = MachO::N_SECT;
354 // FIXME: Set STAB bits.
356 if (Data.isPrivateExtern())
357 Type |= MachO::N_PEXT;
360 if (Data.isExternal() || (!IsAlias && Symbol->isUndefined()))
361 Type |= MachO::N_EXT;
363 // Compute the symbol address.
364 if (IsAlias && Symbol->isUndefined())
365 Address = AliaseeInfo->StringIndex;
366 else if (Symbol->isDefined())
367 Address = getSymbolAddress(OrigSymbol, Layout);
368 else if (Symbol->isCommon()) {
369 // Common symbols are encoded with the size in the address
370 // field, and their alignment in the flags.
371 Address = Symbol->getCommonSize();
374 // struct nlist (12 bytes)
376 write32(MSD.StringIndex);
378 write8(SectionIndex);
380 // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc'
382 write16(cast<MCSymbolMachO>(Symbol)->getEncodedFlags());
389 void MachObjectWriter::writeLinkeditLoadCommand(uint32_t Type,
392 uint64_t Start = OS.tell();
396 write32(sizeof(MachO::linkedit_data_command));
400 assert(OS.tell() - Start == sizeof(MachO::linkedit_data_command));
403 static unsigned ComputeLinkerOptionsLoadCommandSize(
404 const std::vector<std::string> &Options, bool is64Bit)
406 unsigned Size = sizeof(MachO::linker_option_command);
407 for (const std::string &Option : Options)
408 Size += Option.size() + 1;
409 return RoundUpToAlignment(Size, is64Bit ? 8 : 4);
412 void MachObjectWriter::writeLinkerOptionsLoadCommand(
413 const std::vector<std::string> &Options)
415 unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit());
416 uint64_t Start = OS.tell();
419 write32(MachO::LC_LINKER_OPTION);
421 write32(Options.size());
422 uint64_t BytesWritten = sizeof(MachO::linker_option_command);
423 for (const std::string &Option : Options) {
424 // Write each string, including the null byte.
425 writeBytes(Option.c_str(), Option.size() + 1);
426 BytesWritten += Option.size() + 1;
429 // Pad to a multiple of the pointer size.
430 writeBytes("", OffsetToAlignment(BytesWritten, is64Bit() ? 8 : 4));
432 assert(OS.tell() - Start == Size);
435 void MachObjectWriter::recordRelocation(MCAssembler &Asm,
436 const MCAsmLayout &Layout,
437 const MCFragment *Fragment,
438 const MCFixup &Fixup, MCValue Target,
439 bool &IsPCRel, uint64_t &FixedValue) {
440 TargetObjectWriter->recordRelocation(this, Asm, Layout, Fragment, Fixup,
444 void MachObjectWriter::bindIndirectSymbols(MCAssembler &Asm) {
445 // This is the point where 'as' creates actual symbols for indirect symbols
446 // (in the following two passes). It would be easier for us to do this sooner
447 // when we see the attribute, but that makes getting the order in the symbol
448 // table much more complicated than it is worth.
450 // FIXME: Revisit this when the dust settles.
452 // Report errors for use of .indirect_symbol not in a symbol pointer section
454 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
455 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
456 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
458 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
459 Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
460 Section.getType() != MachO::S_SYMBOL_STUBS) {
461 MCSymbol &Symbol = *it->Symbol;
462 report_fatal_error("indirect symbol '" + Symbol.getName() +
463 "' not in a symbol pointer or stub section");
467 // Bind non-lazy symbol pointers first.
468 unsigned IndirectIndex = 0;
469 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
470 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
471 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
473 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS)
476 // Initialize the section indirect symbol base, if necessary.
477 IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex));
479 Asm.registerSymbol(*it->Symbol);
482 // Then lazy symbol pointers and symbol stubs.
484 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
485 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
486 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
488 if (Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
489 Section.getType() != MachO::S_SYMBOL_STUBS)
492 // Initialize the section indirect symbol base, if necessary.
493 IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex));
495 // Set the symbol type to undefined lazy, but only on construction.
497 // FIXME: Do not hardcode.
499 Asm.registerSymbol(*it->Symbol, &Created);
501 cast<MCSymbolMachO>(it->Symbol)->setReferenceTypeUndefinedLazy(true);
505 /// computeSymbolTable - Compute the symbol table data
506 void MachObjectWriter::computeSymbolTable(
507 MCAssembler &Asm, std::vector<MachSymbolData> &LocalSymbolData,
508 std::vector<MachSymbolData> &ExternalSymbolData,
509 std::vector<MachSymbolData> &UndefinedSymbolData) {
510 // Build section lookup table.
511 DenseMap<const MCSection*, uint8_t> SectionIndexMap;
513 for (MCAssembler::iterator it = Asm.begin(),
514 ie = Asm.end(); it != ie; ++it, ++Index)
515 SectionIndexMap[&*it] = Index;
516 assert(Index <= 256 && "Too many sections!");
518 // Build the string table.
519 for (const MCSymbol &Symbol : Asm.symbols()) {
520 if (!Asm.isSymbolLinkerVisible(Symbol))
523 StringTable.add(Symbol.getName());
525 StringTable.finalize(StringTableBuilder::MachO);
527 // Build the symbol arrays but only for non-local symbols.
529 // The particular order that we collect and then sort the symbols is chosen to
530 // match 'as'. Even though it doesn't matter for correctness, this is
531 // important for letting us diff .o files.
532 for (const MCSymbol &Symbol : Asm.symbols()) {
533 // Ignore non-linker visible symbols.
534 if (!Asm.isSymbolLinkerVisible(Symbol))
537 if (!Symbol.isExternal() && !Symbol.isUndefined())
541 MSD.Symbol = &Symbol;
542 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
544 if (Symbol.isUndefined()) {
545 MSD.SectionIndex = 0;
546 UndefinedSymbolData.push_back(MSD);
547 } else if (Symbol.isAbsolute()) {
548 MSD.SectionIndex = 0;
549 ExternalSymbolData.push_back(MSD);
551 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
552 assert(MSD.SectionIndex && "Invalid section index!");
553 ExternalSymbolData.push_back(MSD);
557 // Now add the data for local symbols.
558 for (const MCSymbol &Symbol : Asm.symbols()) {
559 // Ignore non-linker visible symbols.
560 if (!Asm.isSymbolLinkerVisible(Symbol))
563 if (Symbol.isExternal() || Symbol.isUndefined())
567 MSD.Symbol = &Symbol;
568 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
570 if (Symbol.isAbsolute()) {
571 MSD.SectionIndex = 0;
572 LocalSymbolData.push_back(MSD);
574 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
575 assert(MSD.SectionIndex && "Invalid section index!");
576 LocalSymbolData.push_back(MSD);
580 // External and undefined symbols are required to be in lexicographic order.
581 std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
582 std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
584 // Set the symbol indices.
586 for (auto *SymbolData :
587 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
588 for (MachSymbolData &Entry : *SymbolData)
589 Entry.Symbol->setIndex(Index++);
591 for (const MCSection &Section : Asm) {
592 for (RelAndSymbol &Rel : Relocations[&Section]) {
596 // Set the Index and the IsExtern bit.
597 unsigned Index = Rel.Sym->getIndex();
598 assert(isInt<24>(Index));
600 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & (~0U << 24)) | Index | (1 << 27);
602 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & 0xff) | Index << 8 | (1 << 4);
607 void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm,
608 const MCAsmLayout &Layout) {
609 uint64_t StartAddress = 0;
610 for (const MCSection *Sec : Layout.getSectionOrder()) {
611 StartAddress = RoundUpToAlignment(StartAddress, Sec->getAlignment());
612 SectionAddress[Sec] = StartAddress;
613 StartAddress += Layout.getSectionAddressSize(Sec);
615 // Explicitly pad the section to match the alignment requirements of the
616 // following one. This is for 'gas' compatibility, it shouldn't
617 /// strictly be necessary.
618 StartAddress += getPaddingSize(Sec, Layout);
622 void MachObjectWriter::executePostLayoutBinding(MCAssembler &Asm,
623 const MCAsmLayout &Layout) {
624 computeSectionAddresses(Asm, Layout);
626 // Create symbol data for any indirect symbols.
627 bindIndirectSymbols(Asm);
630 bool MachObjectWriter::isSymbolRefDifferenceFullyResolvedImpl(
631 const MCAssembler &Asm, const MCSymbol &SymA, const MCFragment &FB,
632 bool InSet, bool IsPCRel) const {
636 // The effective address is
637 // addr(atom(A)) + offset(A)
638 // - addr(atom(B)) - offset(B)
639 // and the offsets are not relocatable, so the fixup is fully resolved when
640 // addr(atom(A)) - addr(atom(B)) == 0.
641 const MCSymbol &SA = findAliasedSymbol(SymA);
642 const MCSection &SecA = SA.getSection();
643 const MCSection &SecB = *FB.getParent();
646 // The simple (Darwin, except on x86_64) way of dealing with this was to
647 // assume that any reference to a temporary symbol *must* be a temporary
648 // symbol in the same atom, unless the sections differ. Therefore, any PCrel
649 // relocation to a temporary symbol (in the same section) is fully
650 // resolved. This also works in conjunction with absolutized .set, which
651 // requires the compiler to use .set to absolutize the differences between
652 // symbols which the compiler knows to be assembly time constants, so we
653 // don't need to worry about considering symbol differences fully resolved.
655 // If the file isn't using sub-sections-via-symbols, we can make the
656 // same assumptions about any symbol that we normally make about
659 bool hasReliableSymbolDifference = isX86_64();
660 if (!hasReliableSymbolDifference) {
661 if (!SA.isInSection() || &SecA != &SecB ||
662 (!SA.isTemporary() && FB.getAtom() != SA.getFragment()->getAtom() &&
663 Asm.getSubsectionsViaSymbols()))
667 // For Darwin x86_64, there is one special case when the reference IsPCRel.
668 // If the fragment with the reference does not have a base symbol but meets
669 // the simple way of dealing with this, in that it is a temporary symbol in
670 // the same atom then it is assumed to be fully resolved. This is needed so
671 // a relocation entry is not created and so the static linker does not
672 // mess up the reference later.
673 else if(!FB.getAtom() &&
674 SA.isTemporary() && SA.isInSection() && &SecA == &SecB){
679 // If they are not in the same section, we can't compute the diff.
683 const MCFragment *FA = SA.getFragment();
685 // Bail if the symbol has no fragment.
689 // If the atoms are the same, they are guaranteed to have the same address.
690 if (FA->getAtom() == FB.getAtom())
693 // Otherwise, we can't prove this is fully resolved.
697 void MachObjectWriter::writeObject(MCAssembler &Asm,
698 const MCAsmLayout &Layout) {
699 // Compute symbol table information and bind symbol indices.
700 computeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData,
701 UndefinedSymbolData);
703 unsigned NumSections = Asm.size();
704 const MCAssembler::VersionMinInfoType &VersionInfo =
705 Layout.getAssembler().getVersionMinInfo();
707 // The section data starts after the header, the segment load command (and
708 // section headers) and the symbol table.
709 unsigned NumLoadCommands = 1;
710 uint64_t LoadCommandsSize = is64Bit() ?
711 sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64):
712 sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section);
714 // Add the deployment target version info load command size, if used.
715 if (VersionInfo.Major != 0) {
717 LoadCommandsSize += sizeof(MachO::version_min_command);
720 // Add the data-in-code load command size, if used.
721 unsigned NumDataRegions = Asm.getDataRegions().size();
722 if (NumDataRegions) {
724 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
727 // Add the loh load command size, if used.
728 uint64_t LOHRawSize = Asm.getLOHContainer().getEmitSize(*this, Layout);
729 uint64_t LOHSize = RoundUpToAlignment(LOHRawSize, is64Bit() ? 8 : 4);
732 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
735 // Add the symbol table load command sizes, if used.
736 unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
737 UndefinedSymbolData.size();
739 NumLoadCommands += 2;
740 LoadCommandsSize += (sizeof(MachO::symtab_command) +
741 sizeof(MachO::dysymtab_command));
744 // Add the linker option load commands sizes.
745 for (const auto &Option : Asm.getLinkerOptions()) {
747 LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(Option, is64Bit());
750 // Compute the total size of the section data, as well as its file size and vm
752 uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) :
753 sizeof(MachO::mach_header)) + LoadCommandsSize;
754 uint64_t SectionDataSize = 0;
755 uint64_t SectionDataFileSize = 0;
757 for (const MCSection &Sec : Asm) {
758 uint64_t Address = getSectionAddress(&Sec);
759 uint64_t Size = Layout.getSectionAddressSize(&Sec);
760 uint64_t FileSize = Layout.getSectionFileSize(&Sec);
761 FileSize += getPaddingSize(&Sec, Layout);
763 VMSize = std::max(VMSize, Address + Size);
765 if (Sec.isVirtualSection())
768 SectionDataSize = std::max(SectionDataSize, Address + Size);
769 SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
772 // The section data is padded to 4 bytes.
774 // FIXME: Is this machine dependent?
775 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
776 SectionDataFileSize += SectionDataPadding;
778 // Write the prolog, starting with the header and load command...
779 writeHeader(NumLoadCommands, LoadCommandsSize,
780 Asm.getSubsectionsViaSymbols());
781 writeSegmentLoadCommand(NumSections, VMSize,
782 SectionDataStart, SectionDataSize);
784 // ... and then the section headers.
785 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
786 for (const MCSection &Sec : Asm) {
787 std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
788 unsigned NumRelocs = Relocs.size();
789 uint64_t SectionStart = SectionDataStart + getSectionAddress(&Sec);
790 writeSection(Asm, Layout, Sec, SectionStart, RelocTableEnd, NumRelocs);
791 RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
794 // Write out the deployment target information, if it's available.
795 if (VersionInfo.Major != 0) {
796 assert(VersionInfo.Update < 256 && "unencodable update target version");
797 assert(VersionInfo.Minor < 256 && "unencodable minor target version");
798 assert(VersionInfo.Major < 65536 && "unencodable major target version");
799 uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
800 (VersionInfo.Major << 16);
801 write32(VersionInfo.Kind == MCVM_OSXVersionMin ? MachO::LC_VERSION_MIN_MACOSX :
802 MachO::LC_VERSION_MIN_IPHONEOS);
803 write32(sizeof(MachO::version_min_command));
804 write32(EncodedVersion);
805 write32(0); // reserved.
808 // Write the data-in-code load command, if used.
809 uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
810 if (NumDataRegions) {
811 uint64_t DataRegionsOffset = RelocTableEnd;
812 uint64_t DataRegionsSize = NumDataRegions * 8;
813 writeLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
817 // Write the loh load command, if used.
818 uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
820 writeLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
821 DataInCodeTableEnd, LOHSize);
823 // Write the symbol table load command, if used.
825 unsigned FirstLocalSymbol = 0;
826 unsigned NumLocalSymbols = LocalSymbolData.size();
827 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
828 unsigned NumExternalSymbols = ExternalSymbolData.size();
829 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
830 unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
831 unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
832 unsigned NumSymTabSymbols =
833 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
834 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
835 uint64_t IndirectSymbolOffset = 0;
837 // If used, the indirect symbols are written after the section data.
838 if (NumIndirectSymbols)
839 IndirectSymbolOffset = LOHTableEnd;
841 // The symbol table is written after the indirect symbol data.
842 uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
844 // The string table is written after symbol table.
845 uint64_t StringTableOffset =
846 SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
847 sizeof(MachO::nlist_64) :
848 sizeof(MachO::nlist));
849 writeSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
850 StringTableOffset, StringTable.data().size());
852 writeDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
853 FirstExternalSymbol, NumExternalSymbols,
854 FirstUndefinedSymbol, NumUndefinedSymbols,
855 IndirectSymbolOffset, NumIndirectSymbols);
858 // Write the linker options load commands.
859 for (const auto &Option : Asm.getLinkerOptions())
860 writeLinkerOptionsLoadCommand(Option);
862 // Write the actual section data.
863 for (const MCSection &Sec : Asm) {
864 Asm.writeSectionData(&Sec, Layout);
866 uint64_t Pad = getPaddingSize(&Sec, Layout);
870 // Write the extra padding.
871 WriteZeros(SectionDataPadding);
873 // Write the relocation entries.
874 for (const MCSection &Sec : Asm) {
875 // Write the section relocation entries, in reverse order to match 'as'
876 // (approximately, the exact algorithm is more complicated than this).
877 std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
878 for (const RelAndSymbol &Rel : make_range(Relocs.rbegin(), Relocs.rend())) {
879 write32(Rel.MRE.r_word0);
880 write32(Rel.MRE.r_word1);
884 // Write out the data-in-code region payload, if there is one.
885 for (MCAssembler::const_data_region_iterator
886 it = Asm.data_region_begin(), ie = Asm.data_region_end();
888 const DataRegionData *Data = &(*it);
889 uint64_t Start = getSymbolAddress(*Data->Start, Layout);
890 uint64_t End = getSymbolAddress(*Data->End, Layout);
891 DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
892 << " start: " << Start << "(" << Data->Start->getName() << ")"
893 << " end: " << End << "(" << Data->End->getName() << ")"
894 << " size: " << End - Start
897 write16(End - Start);
901 // Write out the loh commands, if there is one.
904 unsigned Start = OS.tell();
906 Asm.getLOHContainer().emit(*this, Layout);
907 // Pad to a multiple of the pointer size.
908 writeBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
909 assert(OS.tell() - Start == LOHSize);
912 // Write the symbol table data, if used.
914 // Write the indirect symbol entries.
915 for (MCAssembler::const_indirect_symbol_iterator
916 it = Asm.indirect_symbol_begin(),
917 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
918 // Indirect symbols in the non-lazy symbol pointer section have some
920 const MCSectionMachO &Section =
921 static_cast<const MCSectionMachO &>(*it->Section);
922 if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
923 // If this symbol is defined and internal, mark it as such.
924 if (it->Symbol->isDefined() && !it->Symbol->isExternal()) {
925 uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
926 if (it->Symbol->isAbsolute())
927 Flags |= MachO::INDIRECT_SYMBOL_ABS;
933 write32(it->Symbol->getIndex());
936 // FIXME: Check that offsets match computed ones.
938 // Write the symbol table entries.
939 for (auto *SymbolData :
940 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
941 for (MachSymbolData &Entry : *SymbolData)
942 writeNlist(Entry, Layout);
944 // Write the string table.
945 OS << StringTable.data();
949 MCObjectWriter *llvm::createMachObjectWriter(MCMachObjectTargetWriter *MOTW,
950 raw_pwrite_stream &OS,
951 bool IsLittleEndian) {
952 return new MachObjectWriter(MOTW, OS, IsLittleEndian);