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::
43 doesSymbolRequireExternRelocation(const MCSymbolData *SD) {
44 // Undefined symbols are always extern.
45 if (SD->getSymbol().isUndefined())
48 // References to weak definitions require external relocation entries; the
49 // definition may not always be the one in the same object file.
50 if (SD->getFlags() & SF_WeakDefinition)
53 // Otherwise, we can use an internal relocation.
57 bool MachObjectWriter::
58 MachSymbolData::operator<(const MachSymbolData &RHS) const {
59 return SymbolData->getSymbol().getName() <
60 RHS.SymbolData->getSymbol().getName();
63 bool MachObjectWriter::isFixupKindPCRel(const MCAssembler &Asm, unsigned Kind) {
64 const MCFixupKindInfo &FKI = Asm.getBackend().getFixupKindInfo(
67 return FKI.Flags & MCFixupKindInfo::FKF_IsPCRel;
70 uint64_t MachObjectWriter::getFragmentAddress(const MCFragment *Fragment,
71 const MCAsmLayout &Layout) const {
72 return getSectionAddress(Fragment->getParent()) +
73 Layout.getFragmentOffset(Fragment);
76 uint64_t MachObjectWriter::getSymbolAddress(const MCSymbolData* SD,
77 const MCAsmLayout &Layout) const {
78 const MCSymbol &S = SD->getSymbol();
80 // If this is a variable, then recursively evaluate now.
82 if (const MCConstantExpr *C =
83 dyn_cast<const MCConstantExpr>(S.getVariableValue()))
88 if (!S.getVariableValue()->EvaluateAsRelocatable(Target, &Layout, nullptr))
89 report_fatal_error("unable to evaluate offset for variable '" +
92 // Verify that any used symbols are defined.
93 if (Target.getSymA() && Target.getSymA()->getSymbol().isUndefined())
94 report_fatal_error("unable to evaluate offset to undefined symbol '" +
95 Target.getSymA()->getSymbol().getName() + "'");
96 if (Target.getSymB() && Target.getSymB()->getSymbol().isUndefined())
97 report_fatal_error("unable to evaluate offset to undefined symbol '" +
98 Target.getSymB()->getSymbol().getName() + "'");
100 uint64_t Address = Target.getConstant();
101 if (Target.getSymA())
102 Address += getSymbolAddress(&Layout.getAssembler().getSymbolData(
103 Target.getSymA()->getSymbol()), Layout);
104 if (Target.getSymB())
105 Address += getSymbolAddress(&Layout.getAssembler().getSymbolData(
106 Target.getSymB()->getSymbol()), Layout);
110 return getSectionAddress(SD->getFragment()->getParent()) +
111 Layout.getSymbolOffset(SD);
114 uint64_t MachObjectWriter::getPaddingSize(const MCSectionData *SD,
115 const MCAsmLayout &Layout) const {
116 uint64_t EndAddr = getSectionAddress(SD) + Layout.getSectionAddressSize(SD);
117 unsigned Next = SD->getLayoutOrder() + 1;
118 if (Next >= Layout.getSectionOrder().size())
121 const MCSectionData &NextSD = *Layout.getSectionOrder()[Next];
122 if (NextSD.getSection().isVirtualSection())
124 return OffsetToAlignment(EndAddr, NextSD.getAlignment());
127 void MachObjectWriter::WriteHeader(unsigned NumLoadCommands,
128 unsigned LoadCommandsSize,
129 bool SubsectionsViaSymbols) {
132 if (SubsectionsViaSymbols)
133 Flags |= MachO::MH_SUBSECTIONS_VIA_SYMBOLS;
135 // struct mach_header (28 bytes) or
136 // struct mach_header_64 (32 bytes)
138 uint64_t Start = OS.tell();
141 Write32(is64Bit() ? MachO::MH_MAGIC_64 : MachO::MH_MAGIC);
143 Write32(TargetObjectWriter->getCPUType());
144 Write32(TargetObjectWriter->getCPUSubtype());
146 Write32(MachO::MH_OBJECT);
147 Write32(NumLoadCommands);
148 Write32(LoadCommandsSize);
151 Write32(0); // reserved
153 assert(OS.tell() - Start ==
154 (is64Bit()?sizeof(MachO::mach_header_64): sizeof(MachO::mach_header)));
157 /// WriteSegmentLoadCommand - Write a segment load command.
159 /// \param NumSections The number of sections in this segment.
160 /// \param SectionDataSize The total size of the sections.
161 void MachObjectWriter::WriteSegmentLoadCommand(unsigned NumSections,
163 uint64_t SectionDataStartOffset,
164 uint64_t SectionDataSize) {
165 // struct segment_command (56 bytes) or
166 // struct segment_command_64 (72 bytes)
168 uint64_t Start = OS.tell();
171 unsigned SegmentLoadCommandSize =
172 is64Bit() ? sizeof(MachO::segment_command_64):
173 sizeof(MachO::segment_command);
174 Write32(is64Bit() ? MachO::LC_SEGMENT_64 : MachO::LC_SEGMENT);
175 Write32(SegmentLoadCommandSize +
176 NumSections * (is64Bit() ? sizeof(MachO::section_64) :
177 sizeof(MachO::section)));
181 Write64(0); // vmaddr
182 Write64(VMSize); // vmsize
183 Write64(SectionDataStartOffset); // file offset
184 Write64(SectionDataSize); // file size
186 Write32(0); // vmaddr
187 Write32(VMSize); // vmsize
188 Write32(SectionDataStartOffset); // file offset
189 Write32(SectionDataSize); // file size
192 Write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE);
194 Write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE);
195 Write32(NumSections);
198 assert(OS.tell() - Start == SegmentLoadCommandSize);
201 void MachObjectWriter::WriteSection(const MCAssembler &Asm,
202 const MCAsmLayout &Layout,
203 const MCSectionData &SD,
205 uint64_t RelocationsStart,
206 unsigned NumRelocations) {
207 uint64_t SectionSize = Layout.getSectionAddressSize(&SD);
209 // The offset is unused for virtual sections.
210 if (SD.getSection().isVirtualSection()) {
211 assert(Layout.getSectionFileSize(&SD) == 0 && "Invalid file size!");
215 // struct section (68 bytes) or
216 // struct section_64 (80 bytes)
218 uint64_t Start = OS.tell();
221 const MCSectionMachO &Section = cast<MCSectionMachO>(SD.getSection());
222 WriteBytes(Section.getSectionName(), 16);
223 WriteBytes(Section.getSegmentName(), 16);
225 Write64(getSectionAddress(&SD)); // address
226 Write64(SectionSize); // size
228 Write32(getSectionAddress(&SD)); // address
229 Write32(SectionSize); // size
233 unsigned Flags = Section.getTypeAndAttributes();
234 if (SD.hasInstructions())
235 Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS;
237 assert(isPowerOf2_32(SD.getAlignment()) && "Invalid alignment!");
238 Write32(Log2_32(SD.getAlignment()));
239 Write32(NumRelocations ? RelocationsStart : 0);
240 Write32(NumRelocations);
242 Write32(IndirectSymBase.lookup(&SD)); // reserved1
243 Write32(Section.getStubSize()); // reserved2
245 Write32(0); // reserved3
247 assert(OS.tell() - Start == (is64Bit() ? sizeof(MachO::section_64) :
248 sizeof(MachO::section)));
251 void MachObjectWriter::WriteSymtabLoadCommand(uint32_t SymbolOffset,
253 uint32_t StringTableOffset,
254 uint32_t StringTableSize) {
255 // struct symtab_command (24 bytes)
257 uint64_t Start = OS.tell();
260 Write32(MachO::LC_SYMTAB);
261 Write32(sizeof(MachO::symtab_command));
262 Write32(SymbolOffset);
264 Write32(StringTableOffset);
265 Write32(StringTableSize);
267 assert(OS.tell() - Start == sizeof(MachO::symtab_command));
270 void MachObjectWriter::WriteDysymtabLoadCommand(uint32_t FirstLocalSymbol,
271 uint32_t NumLocalSymbols,
272 uint32_t FirstExternalSymbol,
273 uint32_t NumExternalSymbols,
274 uint32_t FirstUndefinedSymbol,
275 uint32_t NumUndefinedSymbols,
276 uint32_t IndirectSymbolOffset,
277 uint32_t NumIndirectSymbols) {
278 // struct dysymtab_command (80 bytes)
280 uint64_t Start = OS.tell();
283 Write32(MachO::LC_DYSYMTAB);
284 Write32(sizeof(MachO::dysymtab_command));
285 Write32(FirstLocalSymbol);
286 Write32(NumLocalSymbols);
287 Write32(FirstExternalSymbol);
288 Write32(NumExternalSymbols);
289 Write32(FirstUndefinedSymbol);
290 Write32(NumUndefinedSymbols);
291 Write32(0); // tocoff
293 Write32(0); // modtaboff
294 Write32(0); // nmodtab
295 Write32(0); // extrefsymoff
296 Write32(0); // nextrefsyms
297 Write32(IndirectSymbolOffset);
298 Write32(NumIndirectSymbols);
299 Write32(0); // extreloff
300 Write32(0); // nextrel
301 Write32(0); // locreloff
302 Write32(0); // nlocrel
304 assert(OS.tell() - Start == sizeof(MachO::dysymtab_command));
307 MachObjectWriter::MachSymbolData *
308 MachObjectWriter::findSymbolData(const MCSymbol &Sym) {
309 for (auto &Entry : LocalSymbolData)
310 if (&Entry.SymbolData->getSymbol() == &Sym)
313 for (auto &Entry : ExternalSymbolData)
314 if (&Entry.SymbolData->getSymbol() == &Sym)
317 for (auto &Entry : UndefinedSymbolData)
318 if (&Entry.SymbolData->getSymbol() == &Sym)
324 const MCSymbol &MachObjectWriter::findAliasedSymbol(const MCSymbol &Sym) const {
325 const MCSymbol *S = &Sym;
326 while (S->isVariable()) {
327 const MCExpr *Value = S->getVariableValue();
328 const auto *Ref = dyn_cast<MCSymbolRefExpr>(Value);
331 S = &Ref->getSymbol();
336 void MachObjectWriter::WriteNlist(MachSymbolData &MSD,
337 const MCAsmLayout &Layout) {
338 MCSymbolData &Data = *MSD.SymbolData;
339 const MCSymbol *Symbol = &Data.getSymbol();
340 const MCSymbol *AliasedSymbol = &findAliasedSymbol(*Symbol);
341 uint8_t SectionIndex = MSD.SectionIndex;
343 uint16_t Flags = Data.getFlags();
344 uint64_t Address = 0;
345 bool IsAlias = Symbol != AliasedSymbol;
347 MachSymbolData *AliaseeInfo;
349 AliaseeInfo = findSymbolData(*AliasedSymbol);
351 SectionIndex = AliaseeInfo->SectionIndex;
352 Symbol = AliasedSymbol;
355 // Set the N_TYPE bits. See <mach-o/nlist.h>.
357 // FIXME: Are the prebound or indirect fields possible here?
358 if (IsAlias && Symbol->isUndefined())
359 Type = MachO::N_INDR;
360 else if (Symbol->isUndefined())
361 Type = MachO::N_UNDF;
362 else if (Symbol->isAbsolute())
365 Type = MachO::N_SECT;
367 // FIXME: Set STAB bits.
369 if (Data.isPrivateExtern())
370 Type |= MachO::N_PEXT;
373 if (Data.isExternal() || (!IsAlias && Symbol->isUndefined()))
374 Type |= MachO::N_EXT;
376 // Compute the symbol address.
377 if (IsAlias && Symbol->isUndefined())
378 Address = AliaseeInfo->StringIndex;
379 else if (Symbol->isDefined())
380 Address = getSymbolAddress(&Data, Layout);
381 else if (Data.isCommon()) {
382 // Common symbols are encoded with the size in the address
383 // field, and their alignment in the flags.
384 Address = Data.getCommonSize();
386 // Common alignment is packed into the 'desc' bits.
387 if (unsigned Align = Data.getCommonAlignment()) {
388 unsigned Log2Size = Log2_32(Align);
389 assert((1U << Log2Size) == Align && "Invalid 'common' alignment!");
391 report_fatal_error("invalid 'common' alignment '" +
392 Twine(Align) + "' for '" + Symbol->getName() + "'",
394 // FIXME: Keep this mask with the SymbolFlags enumeration.
395 Flags = (Flags & 0xF0FF) | (Log2Size << 8);
399 if (Layout.getAssembler().isThumbFunc(Symbol))
400 Flags |= SF_ThumbFunc;
402 // struct nlist (12 bytes)
404 Write32(MSD.StringIndex);
406 Write8(SectionIndex);
408 // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc'
417 void MachObjectWriter::WriteLinkeditLoadCommand(uint32_t Type,
420 uint64_t Start = OS.tell();
424 Write32(sizeof(MachO::linkedit_data_command));
428 assert(OS.tell() - Start == sizeof(MachO::linkedit_data_command));
431 static unsigned ComputeLinkerOptionsLoadCommandSize(
432 const std::vector<std::string> &Options, bool is64Bit)
434 unsigned Size = sizeof(MachO::linker_option_command);
435 for (unsigned i = 0, e = Options.size(); i != e; ++i)
436 Size += Options[i].size() + 1;
437 return RoundUpToAlignment(Size, is64Bit ? 8 : 4);
440 void MachObjectWriter::WriteLinkerOptionsLoadCommand(
441 const std::vector<std::string> &Options)
443 unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit());
444 uint64_t Start = OS.tell();
447 Write32(MachO::LC_LINKER_OPTION);
449 Write32(Options.size());
450 uint64_t BytesWritten = sizeof(MachO::linker_option_command);
451 for (unsigned i = 0, e = Options.size(); i != e; ++i) {
452 // Write each string, including the null byte.
453 const std::string &Option = Options[i];
454 WriteBytes(Option.c_str(), Option.size() + 1);
455 BytesWritten += Option.size() + 1;
458 // Pad to a multiple of the pointer size.
459 WriteBytes("", OffsetToAlignment(BytesWritten, is64Bit() ? 8 : 4));
461 assert(OS.tell() - Start == Size);
464 void MachObjectWriter::RecordRelocation(MCAssembler &Asm,
465 const MCAsmLayout &Layout,
466 const MCFragment *Fragment,
467 const MCFixup &Fixup, MCValue Target,
468 bool &IsPCRel, uint64_t &FixedValue) {
469 TargetObjectWriter->RecordRelocation(this, Asm, Layout, Fragment, Fixup,
473 void MachObjectWriter::BindIndirectSymbols(MCAssembler &Asm) {
474 // This is the point where 'as' creates actual symbols for indirect symbols
475 // (in the following two passes). It would be easier for us to do this sooner
476 // when we see the attribute, but that makes getting the order in the symbol
477 // table much more complicated than it is worth.
479 // FIXME: Revisit this when the dust settles.
481 // Report errors for use of .indirect_symbol not in a symbol pointer section
483 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
484 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
485 const MCSectionMachO &Section =
486 cast<MCSectionMachO>(it->SectionData->getSection());
488 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
489 Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
490 Section.getType() != MachO::S_SYMBOL_STUBS) {
491 MCSymbol &Symbol = *it->Symbol;
492 report_fatal_error("indirect symbol '" + Symbol.getName() +
493 "' not in a symbol pointer or stub section");
497 // Bind non-lazy symbol pointers first.
498 unsigned IndirectIndex = 0;
499 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
500 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
501 const MCSectionMachO &Section =
502 cast<MCSectionMachO>(it->SectionData->getSection());
504 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS)
507 // Initialize the section indirect symbol base, if necessary.
508 IndirectSymBase.insert(std::make_pair(it->SectionData, IndirectIndex));
510 Asm.getOrCreateSymbolData(*it->Symbol);
513 // Then lazy symbol pointers and symbol stubs.
515 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
516 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
517 const MCSectionMachO &Section =
518 cast<MCSectionMachO>(it->SectionData->getSection());
520 if (Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
521 Section.getType() != MachO::S_SYMBOL_STUBS)
524 // Initialize the section indirect symbol base, if necessary.
525 IndirectSymBase.insert(std::make_pair(it->SectionData, IndirectIndex));
527 // Set the symbol type to undefined lazy, but only on construction.
529 // FIXME: Do not hardcode.
531 MCSymbolData &Entry = Asm.getOrCreateSymbolData(*it->Symbol, &Created);
533 Entry.setFlags(Entry.getFlags() | 0x0001);
537 /// ComputeSymbolTable - Compute the symbol table data
538 void MachObjectWriter::ComputeSymbolTable(
539 MCAssembler &Asm, std::vector<MachSymbolData> &LocalSymbolData,
540 std::vector<MachSymbolData> &ExternalSymbolData,
541 std::vector<MachSymbolData> &UndefinedSymbolData) {
542 // Build section lookup table.
543 DenseMap<const MCSection*, uint8_t> SectionIndexMap;
545 for (MCAssembler::iterator it = Asm.begin(),
546 ie = Asm.end(); it != ie; ++it, ++Index)
547 SectionIndexMap[&it->getSection()] = Index;
548 assert(Index <= 256 && "Too many sections!");
550 // Build the string table.
551 for (const MCSymbol &Symbol : Asm.symbols()) {
552 if (!Asm.isSymbolLinkerVisible(Symbol))
555 StringTable.add(Symbol.getName());
557 StringTable.finalize(StringTableBuilder::MachO);
559 // Build the symbol arrays but only for non-local symbols.
561 // The particular order that we collect and then sort the symbols is chosen to
562 // match 'as'. Even though it doesn't matter for correctness, this is
563 // important for letting us diff .o files.
564 for (const MCSymbol &Symbol : Asm.symbols()) {
565 MCSymbolData &SD = Symbol.getData();
567 // Ignore non-linker visible symbols.
568 if (!Asm.isSymbolLinkerVisible(Symbol))
571 if (!SD.isExternal() && !Symbol.isUndefined())
575 MSD.SymbolData = &SD;
576 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
578 if (Symbol.isUndefined()) {
579 MSD.SectionIndex = 0;
580 UndefinedSymbolData.push_back(MSD);
581 } else if (Symbol.isAbsolute()) {
582 MSD.SectionIndex = 0;
583 ExternalSymbolData.push_back(MSD);
585 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
586 assert(MSD.SectionIndex && "Invalid section index!");
587 ExternalSymbolData.push_back(MSD);
591 // Now add the data for local symbols.
592 for (const MCSymbol &Symbol : Asm.symbols()) {
593 MCSymbolData &SD = Symbol.getData();
595 // Ignore non-linker visible symbols.
596 if (!Asm.isSymbolLinkerVisible(Symbol))
599 if (SD.isExternal() || Symbol.isUndefined())
603 MSD.SymbolData = &SD;
604 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
606 if (Symbol.isAbsolute()) {
607 MSD.SectionIndex = 0;
608 LocalSymbolData.push_back(MSD);
610 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
611 assert(MSD.SectionIndex && "Invalid section index!");
612 LocalSymbolData.push_back(MSD);
616 // External and undefined symbols are required to be in lexicographic order.
617 std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
618 std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
620 // Set the symbol indices.
622 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
623 LocalSymbolData[i].SymbolData->setIndex(Index++);
624 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
625 ExternalSymbolData[i].SymbolData->setIndex(Index++);
626 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
627 UndefinedSymbolData[i].SymbolData->setIndex(Index++);
629 for (const MCSectionData &SD : Asm) {
630 std::vector<RelAndSymbol> &Relocs = Relocations[&SD];
631 for (RelAndSymbol &Rel : Relocs) {
635 // Set the Index and the IsExtern bit.
636 unsigned Index = Rel.Sym->getIndex();
637 assert(isInt<24>(Index));
639 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & (~0U << 24)) | Index | (1 << 27);
641 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & 0xff) | Index << 8 | (1 << 4);
646 void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm,
647 const MCAsmLayout &Layout) {
648 uint64_t StartAddress = 0;
649 const SmallVectorImpl<MCSectionData*> &Order = Layout.getSectionOrder();
650 for (int i = 0, n = Order.size(); i != n ; ++i) {
651 const MCSectionData *SD = Order[i];
652 StartAddress = RoundUpToAlignment(StartAddress, SD->getAlignment());
653 SectionAddress[SD] = StartAddress;
654 StartAddress += Layout.getSectionAddressSize(SD);
656 // Explicitly pad the section to match the alignment requirements of the
657 // following one. This is for 'gas' compatibility, it shouldn't
658 /// strictly be necessary.
659 StartAddress += getPaddingSize(SD, Layout);
663 void MachObjectWriter::ExecutePostLayoutBinding(MCAssembler &Asm,
664 const MCAsmLayout &Layout) {
665 computeSectionAddresses(Asm, Layout);
667 // Create symbol data for any indirect symbols.
668 BindIndirectSymbols(Asm);
671 bool MachObjectWriter::IsSymbolRefDifferenceFullyResolvedImpl(
672 const MCAssembler &Asm, const MCSymbolData &DataA, const MCFragment &FB,
673 bool InSet, bool IsPCRel) const {
677 // The effective address is
678 // addr(atom(A)) + offset(A)
679 // - addr(atom(B)) - offset(B)
680 // and the offsets are not relocatable, so the fixup is fully resolved when
681 // addr(atom(A)) - addr(atom(B)) == 0.
682 const MCSymbolData *A_Base = nullptr, *B_Base = nullptr;
684 const MCSymbol &SA = findAliasedSymbol(DataA.getSymbol());
685 const MCSection &SecA = SA.getSection();
686 const MCSection &SecB = FB.getParent()->getSection();
689 // The simple (Darwin, except on x86_64) way of dealing with this was to
690 // assume that any reference to a temporary symbol *must* be a temporary
691 // symbol in the same atom, unless the sections differ. Therefore, any PCrel
692 // relocation to a temporary symbol (in the same section) is fully
693 // resolved. This also works in conjunction with absolutized .set, which
694 // requires the compiler to use .set to absolutize the differences between
695 // symbols which the compiler knows to be assembly time constants, so we
696 // don't need to worry about considering symbol differences fully resolved.
698 // If the file isn't using sub-sections-via-symbols, we can make the
699 // same assumptions about any symbol that we normally make about
702 bool hasReliableSymbolDifference = isX86_64();
703 if (!hasReliableSymbolDifference) {
704 if (!SA.isInSection() || &SecA != &SecB ||
705 (!SA.isTemporary() &&
706 FB.getAtom() != Asm.getSymbolData(SA).getFragment()->getAtom() &&
707 Asm.getSubsectionsViaSymbols()))
711 // For Darwin x86_64, there is one special case when the reference IsPCRel.
712 // If the fragment with the reference does not have a base symbol but meets
713 // the simple way of dealing with this, in that it is a temporary symbol in
714 // the same atom then it is assumed to be fully resolved. This is needed so
715 // a relocation entry is not created and so the static linker does not
716 // mess up the reference later.
717 else if(!FB.getAtom() &&
718 SA.isTemporary() && SA.isInSection() && &SecA == &SecB){
722 if (!TargetObjectWriter->useAggressiveSymbolFolding())
726 // If they are not in the same section, we can't compute the diff.
730 const MCFragment *FA = Asm.getSymbolData(SA).getFragment();
732 // Bail if the symbol has no fragment.
736 A_Base = FA->getAtom();
737 B_Base = FB.getAtom();
739 // If the atoms are the same, they are guaranteed to have the same address.
740 if (A_Base == B_Base)
743 // Otherwise, we can't prove this is fully resolved.
747 void MachObjectWriter::WriteObject(MCAssembler &Asm,
748 const MCAsmLayout &Layout) {
749 // Compute symbol table information and bind symbol indices.
750 ComputeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData,
751 UndefinedSymbolData);
753 unsigned NumSections = Asm.size();
754 const MCAssembler::VersionMinInfoType &VersionInfo =
755 Layout.getAssembler().getVersionMinInfo();
757 // The section data starts after the header, the segment load command (and
758 // section headers) and the symbol table.
759 unsigned NumLoadCommands = 1;
760 uint64_t LoadCommandsSize = is64Bit() ?
761 sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64):
762 sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section);
764 // Add the deployment target version info load command size, if used.
765 if (VersionInfo.Major != 0) {
767 LoadCommandsSize += sizeof(MachO::version_min_command);
770 // Add the data-in-code load command size, if used.
771 unsigned NumDataRegions = Asm.getDataRegions().size();
772 if (NumDataRegions) {
774 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
777 // Add the loh load command size, if used.
778 uint64_t LOHRawSize = Asm.getLOHContainer().getEmitSize(*this, Layout);
779 uint64_t LOHSize = RoundUpToAlignment(LOHRawSize, is64Bit() ? 8 : 4);
782 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
785 // Add the symbol table load command sizes, if used.
786 unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
787 UndefinedSymbolData.size();
789 NumLoadCommands += 2;
790 LoadCommandsSize += (sizeof(MachO::symtab_command) +
791 sizeof(MachO::dysymtab_command));
794 // Add the linker option load commands sizes.
795 const std::vector<std::vector<std::string> > &LinkerOptions =
796 Asm.getLinkerOptions();
797 for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
799 LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(LinkerOptions[i],
803 // Compute the total size of the section data, as well as its file size and vm
805 uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) :
806 sizeof(MachO::mach_header)) + LoadCommandsSize;
807 uint64_t SectionDataSize = 0;
808 uint64_t SectionDataFileSize = 0;
810 for (MCAssembler::const_iterator it = Asm.begin(),
811 ie = Asm.end(); it != ie; ++it) {
812 const MCSectionData &SD = *it;
813 uint64_t Address = getSectionAddress(&SD);
814 uint64_t Size = Layout.getSectionAddressSize(&SD);
815 uint64_t FileSize = Layout.getSectionFileSize(&SD);
816 FileSize += getPaddingSize(&SD, Layout);
818 VMSize = std::max(VMSize, Address + Size);
820 if (SD.getSection().isVirtualSection())
823 SectionDataSize = std::max(SectionDataSize, Address + Size);
824 SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
827 // The section data is padded to 4 bytes.
829 // FIXME: Is this machine dependent?
830 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
831 SectionDataFileSize += SectionDataPadding;
833 // Write the prolog, starting with the header and load command...
834 WriteHeader(NumLoadCommands, LoadCommandsSize,
835 Asm.getSubsectionsViaSymbols());
836 WriteSegmentLoadCommand(NumSections, VMSize,
837 SectionDataStart, SectionDataSize);
839 // ... and then the section headers.
840 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
841 for (MCAssembler::const_iterator it = Asm.begin(),
842 ie = Asm.end(); it != ie; ++it) {
843 std::vector<RelAndSymbol> &Relocs = Relocations[it];
844 unsigned NumRelocs = Relocs.size();
845 uint64_t SectionStart = SectionDataStart + getSectionAddress(it);
846 WriteSection(Asm, Layout, *it, SectionStart, RelocTableEnd, NumRelocs);
847 RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
850 // Write out the deployment target information, if it's available.
851 if (VersionInfo.Major != 0) {
852 assert(VersionInfo.Update < 256 && "unencodable update target version");
853 assert(VersionInfo.Minor < 256 && "unencodable minor target version");
854 assert(VersionInfo.Major < 65536 && "unencodable major target version");
855 uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
856 (VersionInfo.Major << 16);
857 Write32(VersionInfo.Kind == MCVM_OSXVersionMin ? MachO::LC_VERSION_MIN_MACOSX :
858 MachO::LC_VERSION_MIN_IPHONEOS);
859 Write32(sizeof(MachO::version_min_command));
860 Write32(EncodedVersion);
861 Write32(0); // reserved.
864 // Write the data-in-code load command, if used.
865 uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
866 if (NumDataRegions) {
867 uint64_t DataRegionsOffset = RelocTableEnd;
868 uint64_t DataRegionsSize = NumDataRegions * 8;
869 WriteLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
873 // Write the loh load command, if used.
874 uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
876 WriteLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
877 DataInCodeTableEnd, LOHSize);
879 // Write the symbol table load command, if used.
881 unsigned FirstLocalSymbol = 0;
882 unsigned NumLocalSymbols = LocalSymbolData.size();
883 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
884 unsigned NumExternalSymbols = ExternalSymbolData.size();
885 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
886 unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
887 unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
888 unsigned NumSymTabSymbols =
889 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
890 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
891 uint64_t IndirectSymbolOffset = 0;
893 // If used, the indirect symbols are written after the section data.
894 if (NumIndirectSymbols)
895 IndirectSymbolOffset = LOHTableEnd;
897 // The symbol table is written after the indirect symbol data.
898 uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
900 // The string table is written after symbol table.
901 uint64_t StringTableOffset =
902 SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
903 sizeof(MachO::nlist_64) :
904 sizeof(MachO::nlist));
905 WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
906 StringTableOffset, StringTable.data().size());
908 WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
909 FirstExternalSymbol, NumExternalSymbols,
910 FirstUndefinedSymbol, NumUndefinedSymbols,
911 IndirectSymbolOffset, NumIndirectSymbols);
914 // Write the linker options load commands.
915 for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
916 WriteLinkerOptionsLoadCommand(LinkerOptions[i]);
919 // Write the actual section data.
920 for (MCAssembler::const_iterator it = Asm.begin(),
921 ie = Asm.end(); it != ie; ++it) {
922 Asm.writeSectionData(it, Layout);
924 uint64_t Pad = getPaddingSize(it, Layout);
928 // Write the extra padding.
929 WriteZeros(SectionDataPadding);
931 // Write the relocation entries.
932 for (MCAssembler::const_iterator it = Asm.begin(),
933 ie = Asm.end(); it != ie; ++it) {
934 // Write the section relocation entries, in reverse order to match 'as'
935 // (approximately, the exact algorithm is more complicated than this).
936 std::vector<RelAndSymbol> &Relocs = Relocations[it];
937 for (unsigned i = 0, e = Relocs.size(); i != e; ++i) {
938 Write32(Relocs[e - i - 1].MRE.r_word0);
939 Write32(Relocs[e - i - 1].MRE.r_word1);
943 // Write out the data-in-code region payload, if there is one.
944 for (MCAssembler::const_data_region_iterator
945 it = Asm.data_region_begin(), ie = Asm.data_region_end();
947 const DataRegionData *Data = &(*it);
949 getSymbolAddress(&Layout.getAssembler().getSymbolData(*Data->Start),
952 getSymbolAddress(&Layout.getAssembler().getSymbolData(*Data->End),
954 DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
955 << " start: " << Start << "(" << Data->Start->getName() << ")"
956 << " end: " << End << "(" << Data->End->getName() << ")"
957 << " size: " << End - Start
960 Write16(End - Start);
964 // Write out the loh commands, if there is one.
967 unsigned Start = OS.tell();
969 Asm.getLOHContainer().Emit(*this, Layout);
970 // Pad to a multiple of the pointer size.
971 WriteBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
972 assert(OS.tell() - Start == LOHSize);
975 // Write the symbol table data, if used.
977 // Write the indirect symbol entries.
978 for (MCAssembler::const_indirect_symbol_iterator
979 it = Asm.indirect_symbol_begin(),
980 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
981 // Indirect symbols in the non-lazy symbol pointer section have some
983 const MCSectionMachO &Section =
984 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
985 if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
986 // If this symbol is defined and internal, mark it as such.
987 if (it->Symbol->isDefined() &&
988 !Asm.getSymbolData(*it->Symbol).isExternal()) {
989 uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
990 if (it->Symbol->isAbsolute())
991 Flags |= MachO::INDIRECT_SYMBOL_ABS;
997 Write32(Asm.getSymbolData(*it->Symbol).getIndex());
1000 // FIXME: Check that offsets match computed ones.
1002 // Write the symbol table entries.
1003 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
1004 WriteNlist(LocalSymbolData[i], Layout);
1005 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
1006 WriteNlist(ExternalSymbolData[i], Layout);
1007 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
1008 WriteNlist(UndefinedSymbolData[i], Layout);
1010 // Write the string table.
1011 OS << StringTable.data();
1015 MCObjectWriter *llvm::createMachObjectWriter(MCMachObjectTargetWriter *MOTW,
1016 raw_pwrite_stream &OS,
1017 bool IsLittleEndian) {
1018 return new MachObjectWriter(MOTW, OS, IsLittleEndian);