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/MCMachOSymbolFlags.h"
19 #include "llvm/MC/MCObjectWriter.h"
20 #include "llvm/MC/MCSectionMachO.h"
21 #include "llvm/MC/MCSymbol.h"
22 #include "llvm/MC/MCValue.h"
23 #include "llvm/Support/Debug.h"
24 #include "llvm/Support/ErrorHandling.h"
25 #include "llvm/Support/MachO.h"
26 #include "llvm/Support/raw_ostream.h"
30 #define DEBUG_TYPE "mc"
32 void MachObjectWriter::reset() {
34 IndirectSymBase.clear();
36 LocalSymbolData.clear();
37 ExternalSymbolData.clear();
38 UndefinedSymbolData.clear();
39 MCObjectWriter::reset();
42 bool MachObjectWriter::doesSymbolRequireExternRelocation(const MCSymbol &S) {
43 // Undefined symbols are always extern.
47 // References to weak definitions require external relocation entries; the
48 // definition may not always be the one in the same object file.
49 if (S.getFlags() & SF_WeakDefinition)
52 // Otherwise, we can use an internal relocation.
56 bool MachObjectWriter::
57 MachSymbolData::operator<(const MachSymbolData &RHS) const {
58 return Symbol->getName() < RHS.Symbol->getName();
61 bool MachObjectWriter::isFixupKindPCRel(const MCAssembler &Asm, unsigned Kind) {
62 const MCFixupKindInfo &FKI = Asm.getBackend().getFixupKindInfo(
65 return FKI.Flags & MCFixupKindInfo::FKF_IsPCRel;
68 uint64_t MachObjectWriter::getFragmentAddress(const MCFragment *Fragment,
69 const MCAsmLayout &Layout) const {
70 return getSectionAddress(Fragment->getParent()) +
71 Layout.getFragmentOffset(Fragment);
74 uint64_t MachObjectWriter::getSymbolAddress(const MCSymbol &S,
75 const MCAsmLayout &Layout) const {
76 // If this is a variable, then recursively evaluate now.
78 if (const MCConstantExpr *C =
79 dyn_cast<const MCConstantExpr>(S.getVariableValue()))
84 if (!S.getVariableValue()->evaluateAsRelocatable(Target, &Layout, nullptr))
85 report_fatal_error("unable to evaluate offset for variable '" +
88 // Verify that any used symbols are defined.
89 if (Target.getSymA() && Target.getSymA()->getSymbol().isUndefined())
90 report_fatal_error("unable to evaluate offset to undefined symbol '" +
91 Target.getSymA()->getSymbol().getName() + "'");
92 if (Target.getSymB() && Target.getSymB()->getSymbol().isUndefined())
93 report_fatal_error("unable to evaluate offset to undefined symbol '" +
94 Target.getSymB()->getSymbol().getName() + "'");
96 uint64_t Address = Target.getConstant();
98 Address += getSymbolAddress(Target.getSymA()->getSymbol(), Layout);
100 Address += getSymbolAddress(Target.getSymB()->getSymbol(), Layout);
104 return getSectionAddress(S.getFragment()->getParent()) +
105 Layout.getSymbolOffset(S);
108 uint64_t MachObjectWriter::getPaddingSize(const MCSection *Sec,
109 const MCAsmLayout &Layout) const {
110 uint64_t EndAddr = getSectionAddress(Sec) + Layout.getSectionAddressSize(Sec);
111 unsigned Next = Sec->getLayoutOrder() + 1;
112 if (Next >= Layout.getSectionOrder().size())
115 const MCSection &NextSec = *Layout.getSectionOrder()[Next];
116 if (NextSec.isVirtualSection())
118 return OffsetToAlignment(EndAddr, NextSec.getAlignment());
121 void MachObjectWriter::WriteHeader(unsigned NumLoadCommands,
122 unsigned LoadCommandsSize,
123 bool SubsectionsViaSymbols) {
126 if (SubsectionsViaSymbols)
127 Flags |= MachO::MH_SUBSECTIONS_VIA_SYMBOLS;
129 // struct mach_header (28 bytes) or
130 // struct mach_header_64 (32 bytes)
132 uint64_t Start = OS.tell();
135 write32(is64Bit() ? MachO::MH_MAGIC_64 : MachO::MH_MAGIC);
137 write32(TargetObjectWriter->getCPUType());
138 write32(TargetObjectWriter->getCPUSubtype());
140 write32(MachO::MH_OBJECT);
141 write32(NumLoadCommands);
142 write32(LoadCommandsSize);
145 write32(0); // reserved
147 assert(OS.tell() - Start ==
148 (is64Bit()?sizeof(MachO::mach_header_64): sizeof(MachO::mach_header)));
151 /// WriteSegmentLoadCommand - Write a segment load command.
153 /// \param NumSections The number of sections in this segment.
154 /// \param SectionDataSize The total size of the sections.
155 void MachObjectWriter::WriteSegmentLoadCommand(unsigned NumSections,
157 uint64_t SectionDataStartOffset,
158 uint64_t SectionDataSize) {
159 // struct segment_command (56 bytes) or
160 // struct segment_command_64 (72 bytes)
162 uint64_t Start = OS.tell();
165 unsigned SegmentLoadCommandSize =
166 is64Bit() ? sizeof(MachO::segment_command_64):
167 sizeof(MachO::segment_command);
168 write32(is64Bit() ? MachO::LC_SEGMENT_64 : MachO::LC_SEGMENT);
169 write32(SegmentLoadCommandSize +
170 NumSections * (is64Bit() ? sizeof(MachO::section_64) :
171 sizeof(MachO::section)));
175 write64(0); // vmaddr
176 write64(VMSize); // vmsize
177 write64(SectionDataStartOffset); // file offset
178 write64(SectionDataSize); // file size
180 write32(0); // vmaddr
181 write32(VMSize); // vmsize
182 write32(SectionDataStartOffset); // file offset
183 write32(SectionDataSize); // file size
186 write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE);
188 write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE);
189 write32(NumSections);
192 assert(OS.tell() - Start == SegmentLoadCommandSize);
195 void MachObjectWriter::WriteSection(const MCAssembler &Asm,
196 const MCAsmLayout &Layout,
197 const MCSection &Sec, uint64_t FileOffset,
198 uint64_t RelocationsStart,
199 unsigned NumRelocations) {
200 uint64_t SectionSize = Layout.getSectionAddressSize(&Sec);
201 const MCSectionMachO &Section = cast<MCSectionMachO>(Sec);
203 // The offset is unused for virtual sections.
204 if (Section.isVirtualSection()) {
205 assert(Layout.getSectionFileSize(&Sec) == 0 && "Invalid file size!");
209 // struct section (68 bytes) or
210 // struct section_64 (80 bytes)
212 uint64_t Start = OS.tell();
215 writeBytes(Section.getSectionName(), 16);
216 writeBytes(Section.getSegmentName(), 16);
218 write64(getSectionAddress(&Sec)); // address
219 write64(SectionSize); // size
221 write32(getSectionAddress(&Sec)); // address
222 write32(SectionSize); // size
226 unsigned Flags = Section.getTypeAndAttributes();
227 if (Section.hasInstructions())
228 Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS;
230 assert(isPowerOf2_32(Section.getAlignment()) && "Invalid alignment!");
231 write32(Log2_32(Section.getAlignment()));
232 write32(NumRelocations ? RelocationsStart : 0);
233 write32(NumRelocations);
235 write32(IndirectSymBase.lookup(&Sec)); // reserved1
236 write32(Section.getStubSize()); // reserved2
238 write32(0); // reserved3
240 assert(OS.tell() - Start == (is64Bit() ? sizeof(MachO::section_64) :
241 sizeof(MachO::section)));
244 void MachObjectWriter::WriteSymtabLoadCommand(uint32_t SymbolOffset,
246 uint32_t StringTableOffset,
247 uint32_t StringTableSize) {
248 // struct symtab_command (24 bytes)
250 uint64_t Start = OS.tell();
253 write32(MachO::LC_SYMTAB);
254 write32(sizeof(MachO::symtab_command));
255 write32(SymbolOffset);
257 write32(StringTableOffset);
258 write32(StringTableSize);
260 assert(OS.tell() - Start == sizeof(MachO::symtab_command));
263 void MachObjectWriter::WriteDysymtabLoadCommand(uint32_t FirstLocalSymbol,
264 uint32_t NumLocalSymbols,
265 uint32_t FirstExternalSymbol,
266 uint32_t NumExternalSymbols,
267 uint32_t FirstUndefinedSymbol,
268 uint32_t NumUndefinedSymbols,
269 uint32_t IndirectSymbolOffset,
270 uint32_t NumIndirectSymbols) {
271 // struct dysymtab_command (80 bytes)
273 uint64_t Start = OS.tell();
276 write32(MachO::LC_DYSYMTAB);
277 write32(sizeof(MachO::dysymtab_command));
278 write32(FirstLocalSymbol);
279 write32(NumLocalSymbols);
280 write32(FirstExternalSymbol);
281 write32(NumExternalSymbols);
282 write32(FirstUndefinedSymbol);
283 write32(NumUndefinedSymbols);
284 write32(0); // tocoff
286 write32(0); // modtaboff
287 write32(0); // nmodtab
288 write32(0); // extrefsymoff
289 write32(0); // nextrefsyms
290 write32(IndirectSymbolOffset);
291 write32(NumIndirectSymbols);
292 write32(0); // extreloff
293 write32(0); // nextrel
294 write32(0); // locreloff
295 write32(0); // nlocrel
297 assert(OS.tell() - Start == sizeof(MachO::dysymtab_command));
300 MachObjectWriter::MachSymbolData *
301 MachObjectWriter::findSymbolData(const MCSymbol &Sym) {
302 for (auto *SymbolData :
303 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
304 for (MachSymbolData &Entry : *SymbolData)
305 if (Entry.Symbol == &Sym)
311 const MCSymbol &MachObjectWriter::findAliasedSymbol(const MCSymbol &Sym) const {
312 const MCSymbol *S = &Sym;
313 while (S->isVariable()) {
314 const MCExpr *Value = S->getVariableValue();
315 const auto *Ref = dyn_cast<MCSymbolRefExpr>(Value);
318 S = &Ref->getSymbol();
323 void MachObjectWriter::WriteNlist(MachSymbolData &MSD,
324 const MCAsmLayout &Layout) {
325 const MCSymbol *Symbol = MSD.Symbol;
326 const MCSymbol &Data = *Symbol;
327 const MCSymbol *AliasedSymbol = &findAliasedSymbol(*Symbol);
328 uint8_t SectionIndex = MSD.SectionIndex;
330 uint16_t Flags = Symbol->getFlags();
331 uint64_t Address = 0;
332 bool IsAlias = Symbol != AliasedSymbol;
334 const MCSymbol &OrigSymbol = *Symbol;
335 MachSymbolData *AliaseeInfo;
337 AliaseeInfo = findSymbolData(*AliasedSymbol);
339 SectionIndex = AliaseeInfo->SectionIndex;
340 Symbol = AliasedSymbol;
341 // FIXME: Should this update Data as well? Do we need OrigSymbol at all?
344 // Set the N_TYPE bits. See <mach-o/nlist.h>.
346 // FIXME: Are the prebound or indirect fields possible here?
347 if (IsAlias && Symbol->isUndefined())
348 Type = MachO::N_INDR;
349 else if (Symbol->isUndefined())
350 Type = MachO::N_UNDF;
351 else if (Symbol->isAbsolute())
354 Type = MachO::N_SECT;
356 // FIXME: Set STAB bits.
358 if (Data.isPrivateExtern())
359 Type |= MachO::N_PEXT;
362 if (Data.isExternal() || (!IsAlias && Symbol->isUndefined()))
363 Type |= MachO::N_EXT;
365 // Compute the symbol address.
366 if (IsAlias && Symbol->isUndefined())
367 Address = AliaseeInfo->StringIndex;
368 else if (Symbol->isDefined())
369 Address = getSymbolAddress(OrigSymbol, Layout);
370 else if (Symbol->isCommon()) {
371 // Common symbols are encoded with the size in the address
372 // field, and their alignment in the flags.
373 Address = Symbol->getCommonSize();
375 // Common alignment is packed into the 'desc' bits.
376 if (unsigned Align = Symbol->getCommonAlignment()) {
377 unsigned Log2Size = Log2_32(Align);
378 assert((1U << Log2Size) == Align && "Invalid 'common' alignment!");
380 report_fatal_error("invalid 'common' alignment '" +
381 Twine(Align) + "' for '" + Symbol->getName() + "'",
383 // FIXME: Keep this mask with the SymbolFlags enumeration.
384 Flags = (Flags & 0xF0FF) | (Log2Size << 8);
388 if (Layout.getAssembler().isThumbFunc(Symbol))
389 Flags |= SF_ThumbFunc;
391 // struct nlist (12 bytes)
393 write32(MSD.StringIndex);
395 write8(SectionIndex);
397 // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc'
406 void MachObjectWriter::WriteLinkeditLoadCommand(uint32_t Type,
409 uint64_t Start = OS.tell();
413 write32(sizeof(MachO::linkedit_data_command));
417 assert(OS.tell() - Start == sizeof(MachO::linkedit_data_command));
420 static unsigned ComputeLinkerOptionsLoadCommandSize(
421 const std::vector<std::string> &Options, bool is64Bit)
423 unsigned Size = sizeof(MachO::linker_option_command);
424 for (const std::string &Option : Options)
425 Size += Option.size() + 1;
426 return RoundUpToAlignment(Size, is64Bit ? 8 : 4);
429 void MachObjectWriter::WriteLinkerOptionsLoadCommand(
430 const std::vector<std::string> &Options)
432 unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit());
433 uint64_t Start = OS.tell();
436 write32(MachO::LC_LINKER_OPTION);
438 write32(Options.size());
439 uint64_t BytesWritten = sizeof(MachO::linker_option_command);
440 for (const std::string &Option : Options) {
441 // Write each string, including the null byte.
442 writeBytes(Option.c_str(), Option.size() + 1);
443 BytesWritten += Option.size() + 1;
446 // Pad to a multiple of the pointer size.
447 writeBytes("", OffsetToAlignment(BytesWritten, is64Bit() ? 8 : 4));
449 assert(OS.tell() - Start == Size);
452 void MachObjectWriter::recordRelocation(MCAssembler &Asm,
453 const MCAsmLayout &Layout,
454 const MCFragment *Fragment,
455 const MCFixup &Fixup, MCValue Target,
456 bool &IsPCRel, uint64_t &FixedValue) {
457 TargetObjectWriter->recordRelocation(this, Asm, Layout, Fragment, Fixup,
461 void MachObjectWriter::BindIndirectSymbols(MCAssembler &Asm) {
462 // This is the point where 'as' creates actual symbols for indirect symbols
463 // (in the following two passes). It would be easier for us to do this sooner
464 // when we see the attribute, but that makes getting the order in the symbol
465 // table much more complicated than it is worth.
467 // FIXME: Revisit this when the dust settles.
469 // Report errors for use of .indirect_symbol not in a symbol pointer section
471 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
472 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
473 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
475 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
476 Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
477 Section.getType() != MachO::S_SYMBOL_STUBS) {
478 MCSymbol &Symbol = *it->Symbol;
479 report_fatal_error("indirect symbol '" + Symbol.getName() +
480 "' not in a symbol pointer or stub section");
484 // Bind non-lazy symbol pointers first.
485 unsigned IndirectIndex = 0;
486 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
487 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
488 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
490 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS)
493 // Initialize the section indirect symbol base, if necessary.
494 IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex));
496 Asm.registerSymbol(*it->Symbol);
499 // Then lazy symbol pointers and symbol stubs.
501 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
502 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
503 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
505 if (Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
506 Section.getType() != MachO::S_SYMBOL_STUBS)
509 // Initialize the section indirect symbol base, if necessary.
510 IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex));
512 // Set the symbol type to undefined lazy, but only on construction.
514 // FIXME: Do not hardcode.
516 Asm.registerSymbol(*it->Symbol, &Created);
518 it->Symbol->setFlags(it->Symbol->getFlags() | 0x0001);
522 /// ComputeSymbolTable - Compute the symbol table data
523 void MachObjectWriter::ComputeSymbolTable(
524 MCAssembler &Asm, std::vector<MachSymbolData> &LocalSymbolData,
525 std::vector<MachSymbolData> &ExternalSymbolData,
526 std::vector<MachSymbolData> &UndefinedSymbolData) {
527 // Build section lookup table.
528 DenseMap<const MCSection*, uint8_t> SectionIndexMap;
530 for (MCAssembler::iterator it = Asm.begin(),
531 ie = Asm.end(); it != ie; ++it, ++Index)
532 SectionIndexMap[&*it] = Index;
533 assert(Index <= 256 && "Too many sections!");
535 // Build the string table.
536 for (const MCSymbol &Symbol : Asm.symbols()) {
537 if (!Asm.isSymbolLinkerVisible(Symbol))
540 StringTable.add(Symbol.getName());
542 StringTable.finalize(StringTableBuilder::MachO);
544 // Build the symbol arrays but only for non-local symbols.
546 // The particular order that we collect and then sort the symbols is chosen to
547 // match 'as'. Even though it doesn't matter for correctness, this is
548 // important for letting us diff .o files.
549 for (const MCSymbol &Symbol : Asm.symbols()) {
550 // Ignore non-linker visible symbols.
551 if (!Asm.isSymbolLinkerVisible(Symbol))
554 if (!Symbol.isExternal() && !Symbol.isUndefined())
558 MSD.Symbol = &Symbol;
559 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
561 if (Symbol.isUndefined()) {
562 MSD.SectionIndex = 0;
563 UndefinedSymbolData.push_back(MSD);
564 } else if (Symbol.isAbsolute()) {
565 MSD.SectionIndex = 0;
566 ExternalSymbolData.push_back(MSD);
568 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
569 assert(MSD.SectionIndex && "Invalid section index!");
570 ExternalSymbolData.push_back(MSD);
574 // Now add the data for local symbols.
575 for (const MCSymbol &Symbol : Asm.symbols()) {
576 // Ignore non-linker visible symbols.
577 if (!Asm.isSymbolLinkerVisible(Symbol))
580 if (Symbol.isExternal() || Symbol.isUndefined())
584 MSD.Symbol = &Symbol;
585 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
587 if (Symbol.isAbsolute()) {
588 MSD.SectionIndex = 0;
589 LocalSymbolData.push_back(MSD);
591 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
592 assert(MSD.SectionIndex && "Invalid section index!");
593 LocalSymbolData.push_back(MSD);
597 // External and undefined symbols are required to be in lexicographic order.
598 std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
599 std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
601 // Set the symbol indices.
603 for (auto *SymbolData :
604 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
605 for (MachSymbolData &Entry : *SymbolData)
606 Entry.Symbol->setIndex(Index++);
608 for (const MCSection &Section : Asm) {
609 for (RelAndSymbol &Rel : Relocations[&Section]) {
613 // Set the Index and the IsExtern bit.
614 unsigned Index = Rel.Sym->getIndex();
615 assert(isInt<24>(Index));
617 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & (~0U << 24)) | Index | (1 << 27);
619 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & 0xff) | Index << 8 | (1 << 4);
624 void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm,
625 const MCAsmLayout &Layout) {
626 uint64_t StartAddress = 0;
627 for (const MCSection *Sec : Layout.getSectionOrder()) {
628 StartAddress = RoundUpToAlignment(StartAddress, Sec->getAlignment());
629 SectionAddress[Sec] = StartAddress;
630 StartAddress += Layout.getSectionAddressSize(Sec);
632 // Explicitly pad the section to match the alignment requirements of the
633 // following one. This is for 'gas' compatibility, it shouldn't
634 /// strictly be necessary.
635 StartAddress += getPaddingSize(Sec, Layout);
639 void MachObjectWriter::ExecutePostLayoutBinding(MCAssembler &Asm,
640 const MCAsmLayout &Layout) {
641 computeSectionAddresses(Asm, Layout);
643 // Create symbol data for any indirect symbols.
644 BindIndirectSymbols(Asm);
647 bool MachObjectWriter::isSymbolRefDifferenceFullyResolvedImpl(
648 const MCAssembler &Asm, const MCSymbol &SymA, const MCFragment &FB,
649 bool InSet, bool IsPCRel) const {
653 // The effective address is
654 // addr(atom(A)) + offset(A)
655 // - addr(atom(B)) - offset(B)
656 // and the offsets are not relocatable, so the fixup is fully resolved when
657 // addr(atom(A)) - addr(atom(B)) == 0.
658 const MCSymbol &SA = findAliasedSymbol(SymA);
659 const MCSection &SecA = SA.getSection();
660 const MCSection &SecB = *FB.getParent();
663 // The simple (Darwin, except on x86_64) way of dealing with this was to
664 // assume that any reference to a temporary symbol *must* be a temporary
665 // symbol in the same atom, unless the sections differ. Therefore, any PCrel
666 // relocation to a temporary symbol (in the same section) is fully
667 // resolved. This also works in conjunction with absolutized .set, which
668 // requires the compiler to use .set to absolutize the differences between
669 // symbols which the compiler knows to be assembly time constants, so we
670 // don't need to worry about considering symbol differences fully resolved.
672 // If the file isn't using sub-sections-via-symbols, we can make the
673 // same assumptions about any symbol that we normally make about
676 bool hasReliableSymbolDifference = isX86_64();
677 if (!hasReliableSymbolDifference) {
678 if (!SA.isInSection() || &SecA != &SecB ||
679 (!SA.isTemporary() && FB.getAtom() != SA.getFragment()->getAtom() &&
680 Asm.getSubsectionsViaSymbols()))
684 // For Darwin x86_64, there is one special case when the reference IsPCRel.
685 // If the fragment with the reference does not have a base symbol but meets
686 // the simple way of dealing with this, in that it is a temporary symbol in
687 // the same atom then it is assumed to be fully resolved. This is needed so
688 // a relocation entry is not created and so the static linker does not
689 // mess up the reference later.
690 else if(!FB.getAtom() &&
691 SA.isTemporary() && SA.isInSection() && &SecA == &SecB){
696 // If they are not in the same section, we can't compute the diff.
700 const MCFragment *FA = SA.getFragment();
702 // Bail if the symbol has no fragment.
706 // If the atoms are the same, they are guaranteed to have the same address.
707 if (FA->getAtom() == FB.getAtom())
710 // Otherwise, we can't prove this is fully resolved.
714 void MachObjectWriter::writeObject(MCAssembler &Asm,
715 const MCAsmLayout &Layout) {
716 // Compute symbol table information and bind symbol indices.
717 ComputeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData,
718 UndefinedSymbolData);
720 unsigned NumSections = Asm.size();
721 const MCAssembler::VersionMinInfoType &VersionInfo =
722 Layout.getAssembler().getVersionMinInfo();
724 // The section data starts after the header, the segment load command (and
725 // section headers) and the symbol table.
726 unsigned NumLoadCommands = 1;
727 uint64_t LoadCommandsSize = is64Bit() ?
728 sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64):
729 sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section);
731 // Add the deployment target version info load command size, if used.
732 if (VersionInfo.Major != 0) {
734 LoadCommandsSize += sizeof(MachO::version_min_command);
737 // Add the data-in-code load command size, if used.
738 unsigned NumDataRegions = Asm.getDataRegions().size();
739 if (NumDataRegions) {
741 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
744 // Add the loh load command size, if used.
745 uint64_t LOHRawSize = Asm.getLOHContainer().getEmitSize(*this, Layout);
746 uint64_t LOHSize = RoundUpToAlignment(LOHRawSize, is64Bit() ? 8 : 4);
749 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
752 // Add the symbol table load command sizes, if used.
753 unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
754 UndefinedSymbolData.size();
756 NumLoadCommands += 2;
757 LoadCommandsSize += (sizeof(MachO::symtab_command) +
758 sizeof(MachO::dysymtab_command));
761 // Add the linker option load commands sizes.
762 for (const auto &Option : Asm.getLinkerOptions()) {
764 LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(Option, is64Bit());
767 // Compute the total size of the section data, as well as its file size and vm
769 uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) :
770 sizeof(MachO::mach_header)) + LoadCommandsSize;
771 uint64_t SectionDataSize = 0;
772 uint64_t SectionDataFileSize = 0;
774 for (const MCSection &Sec : Asm) {
775 uint64_t Address = getSectionAddress(&Sec);
776 uint64_t Size = Layout.getSectionAddressSize(&Sec);
777 uint64_t FileSize = Layout.getSectionFileSize(&Sec);
778 FileSize += getPaddingSize(&Sec, Layout);
780 VMSize = std::max(VMSize, Address + Size);
782 if (Sec.isVirtualSection())
785 SectionDataSize = std::max(SectionDataSize, Address + Size);
786 SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
789 // The section data is padded to 4 bytes.
791 // FIXME: Is this machine dependent?
792 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
793 SectionDataFileSize += SectionDataPadding;
795 // Write the prolog, starting with the header and load command...
796 WriteHeader(NumLoadCommands, LoadCommandsSize,
797 Asm.getSubsectionsViaSymbols());
798 WriteSegmentLoadCommand(NumSections, VMSize,
799 SectionDataStart, SectionDataSize);
801 // ... and then the section headers.
802 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
803 for (const MCSection &Sec : Asm) {
804 std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
805 unsigned NumRelocs = Relocs.size();
806 uint64_t SectionStart = SectionDataStart + getSectionAddress(&Sec);
807 WriteSection(Asm, Layout, Sec, SectionStart, RelocTableEnd, NumRelocs);
808 RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
811 // Write out the deployment target information, if it's available.
812 if (VersionInfo.Major != 0) {
813 assert(VersionInfo.Update < 256 && "unencodable update target version");
814 assert(VersionInfo.Minor < 256 && "unencodable minor target version");
815 assert(VersionInfo.Major < 65536 && "unencodable major target version");
816 uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
817 (VersionInfo.Major << 16);
818 write32(VersionInfo.Kind == MCVM_OSXVersionMin ? MachO::LC_VERSION_MIN_MACOSX :
819 MachO::LC_VERSION_MIN_IPHONEOS);
820 write32(sizeof(MachO::version_min_command));
821 write32(EncodedVersion);
822 write32(0); // reserved.
825 // Write the data-in-code load command, if used.
826 uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
827 if (NumDataRegions) {
828 uint64_t DataRegionsOffset = RelocTableEnd;
829 uint64_t DataRegionsSize = NumDataRegions * 8;
830 WriteLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
834 // Write the loh load command, if used.
835 uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
837 WriteLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
838 DataInCodeTableEnd, LOHSize);
840 // Write the symbol table load command, if used.
842 unsigned FirstLocalSymbol = 0;
843 unsigned NumLocalSymbols = LocalSymbolData.size();
844 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
845 unsigned NumExternalSymbols = ExternalSymbolData.size();
846 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
847 unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
848 unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
849 unsigned NumSymTabSymbols =
850 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
851 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
852 uint64_t IndirectSymbolOffset = 0;
854 // If used, the indirect symbols are written after the section data.
855 if (NumIndirectSymbols)
856 IndirectSymbolOffset = LOHTableEnd;
858 // The symbol table is written after the indirect symbol data.
859 uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
861 // The string table is written after symbol table.
862 uint64_t StringTableOffset =
863 SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
864 sizeof(MachO::nlist_64) :
865 sizeof(MachO::nlist));
866 WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
867 StringTableOffset, StringTable.data().size());
869 WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
870 FirstExternalSymbol, NumExternalSymbols,
871 FirstUndefinedSymbol, NumUndefinedSymbols,
872 IndirectSymbolOffset, NumIndirectSymbols);
875 // Write the linker options load commands.
876 for (const auto &Option : Asm.getLinkerOptions())
877 WriteLinkerOptionsLoadCommand(Option);
879 // Write the actual section data.
880 for (const MCSection &Sec : Asm) {
881 Asm.writeSectionData(&Sec, Layout);
883 uint64_t Pad = getPaddingSize(&Sec, Layout);
887 // Write the extra padding.
888 WriteZeros(SectionDataPadding);
890 // Write the relocation entries.
891 for (const MCSection &Sec : Asm) {
892 // Write the section relocation entries, in reverse order to match 'as'
893 // (approximately, the exact algorithm is more complicated than this).
894 std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
895 for (const RelAndSymbol &Rel : make_range(Relocs.rbegin(), Relocs.rend())) {
896 write32(Rel.MRE.r_word0);
897 write32(Rel.MRE.r_word1);
901 // Write out the data-in-code region payload, if there is one.
902 for (MCAssembler::const_data_region_iterator
903 it = Asm.data_region_begin(), ie = Asm.data_region_end();
905 const DataRegionData *Data = &(*it);
906 uint64_t Start = getSymbolAddress(*Data->Start, Layout);
907 uint64_t End = getSymbolAddress(*Data->End, Layout);
908 DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
909 << " start: " << Start << "(" << Data->Start->getName() << ")"
910 << " end: " << End << "(" << Data->End->getName() << ")"
911 << " size: " << End - Start
914 write16(End - Start);
918 // Write out the loh commands, if there is one.
921 unsigned Start = OS.tell();
923 Asm.getLOHContainer().emit(*this, Layout);
924 // Pad to a multiple of the pointer size.
925 writeBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
926 assert(OS.tell() - Start == LOHSize);
929 // Write the symbol table data, if used.
931 // Write the indirect symbol entries.
932 for (MCAssembler::const_indirect_symbol_iterator
933 it = Asm.indirect_symbol_begin(),
934 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
935 // Indirect symbols in the non-lazy symbol pointer section have some
937 const MCSectionMachO &Section =
938 static_cast<const MCSectionMachO &>(*it->Section);
939 if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
940 // If this symbol is defined and internal, mark it as such.
941 if (it->Symbol->isDefined() && !it->Symbol->isExternal()) {
942 uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
943 if (it->Symbol->isAbsolute())
944 Flags |= MachO::INDIRECT_SYMBOL_ABS;
950 write32(it->Symbol->getIndex());
953 // FIXME: Check that offsets match computed ones.
955 // Write the symbol table entries.
956 for (auto *SymbolData :
957 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
958 for (MachSymbolData &Entry : *SymbolData)
959 WriteNlist(Entry, Layout);
961 // Write the string table.
962 OS << StringTable.data();
966 MCObjectWriter *llvm::createMachObjectWriter(MCMachObjectTargetWriter *MOTW,
967 raw_pwrite_stream &OS,
968 bool IsLittleEndian) {
969 return new MachObjectWriter(MOTW, OS, IsLittleEndian);