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.getData().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.getData().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 &Entry : LocalSymbolData)
303 if (Entry.Symbol == &Sym)
306 for (auto &Entry : ExternalSymbolData)
307 if (Entry.Symbol == &Sym)
310 for (auto &Entry : UndefinedSymbolData)
311 if (Entry.Symbol == &Sym)
317 const MCSymbol &MachObjectWriter::findAliasedSymbol(const MCSymbol &Sym) const {
318 const MCSymbol *S = &Sym;
319 while (S->isVariable()) {
320 const MCExpr *Value = S->getVariableValue();
321 const auto *Ref = dyn_cast<MCSymbolRefExpr>(Value);
324 S = &Ref->getSymbol();
329 void MachObjectWriter::WriteNlist(MachSymbolData &MSD,
330 const MCAsmLayout &Layout) {
331 const MCSymbol *Symbol = MSD.Symbol;
332 MCSymbolData &Data = Symbol->getData();
333 const MCSymbol *AliasedSymbol = &findAliasedSymbol(*Symbol);
334 uint8_t SectionIndex = MSD.SectionIndex;
336 uint16_t Flags = Data.getFlags();
337 uint64_t Address = 0;
338 bool IsAlias = Symbol != AliasedSymbol;
340 const MCSymbol &OrigSymbol = *Symbol;
341 MachSymbolData *AliaseeInfo;
343 AliaseeInfo = findSymbolData(*AliasedSymbol);
345 SectionIndex = AliaseeInfo->SectionIndex;
346 Symbol = AliasedSymbol;
347 // FIXME: Should this update Data as well? Do we need OrigSymbol at all?
350 // Set the N_TYPE bits. See <mach-o/nlist.h>.
352 // FIXME: Are the prebound or indirect fields possible here?
353 if (IsAlias && Symbol->isUndefined())
354 Type = MachO::N_INDR;
355 else if (Symbol->isUndefined())
356 Type = MachO::N_UNDF;
357 else if (Symbol->isAbsolute())
360 Type = MachO::N_SECT;
362 // FIXME: Set STAB bits.
364 if (Data.isPrivateExtern())
365 Type |= MachO::N_PEXT;
368 if (Data.isExternal() || (!IsAlias && Symbol->isUndefined()))
369 Type |= MachO::N_EXT;
371 // Compute the symbol address.
372 if (IsAlias && Symbol->isUndefined())
373 Address = AliaseeInfo->StringIndex;
374 else if (Symbol->isDefined())
375 Address = getSymbolAddress(OrigSymbol, Layout);
376 else if (Data.isCommon()) {
377 // Common symbols are encoded with the size in the address
378 // field, and their alignment in the flags.
379 Address = Data.getCommonSize();
381 // Common alignment is packed into the 'desc' bits.
382 if (unsigned Align = Data.getCommonAlignment()) {
383 unsigned Log2Size = Log2_32(Align);
384 assert((1U << Log2Size) == Align && "Invalid 'common' alignment!");
386 report_fatal_error("invalid 'common' alignment '" +
387 Twine(Align) + "' for '" + Symbol->getName() + "'",
389 // FIXME: Keep this mask with the SymbolFlags enumeration.
390 Flags = (Flags & 0xF0FF) | (Log2Size << 8);
394 if (Layout.getAssembler().isThumbFunc(Symbol))
395 Flags |= SF_ThumbFunc;
397 // struct nlist (12 bytes)
399 Write32(MSD.StringIndex);
401 Write8(SectionIndex);
403 // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc'
412 void MachObjectWriter::WriteLinkeditLoadCommand(uint32_t Type,
415 uint64_t Start = OS.tell();
419 Write32(sizeof(MachO::linkedit_data_command));
423 assert(OS.tell() - Start == sizeof(MachO::linkedit_data_command));
426 static unsigned ComputeLinkerOptionsLoadCommandSize(
427 const std::vector<std::string> &Options, bool is64Bit)
429 unsigned Size = sizeof(MachO::linker_option_command);
430 for (unsigned i = 0, e = Options.size(); i != e; ++i)
431 Size += Options[i].size() + 1;
432 return RoundUpToAlignment(Size, is64Bit ? 8 : 4);
435 void MachObjectWriter::WriteLinkerOptionsLoadCommand(
436 const std::vector<std::string> &Options)
438 unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit());
439 uint64_t Start = OS.tell();
442 Write32(MachO::LC_LINKER_OPTION);
444 Write32(Options.size());
445 uint64_t BytesWritten = sizeof(MachO::linker_option_command);
446 for (unsigned i = 0, e = Options.size(); i != e; ++i) {
447 // Write each string, including the null byte.
448 const std::string &Option = Options[i];
449 WriteBytes(Option.c_str(), Option.size() + 1);
450 BytesWritten += Option.size() + 1;
453 // Pad to a multiple of the pointer size.
454 WriteBytes("", OffsetToAlignment(BytesWritten, is64Bit() ? 8 : 4));
456 assert(OS.tell() - Start == Size);
459 void MachObjectWriter::RecordRelocation(MCAssembler &Asm,
460 const MCAsmLayout &Layout,
461 const MCFragment *Fragment,
462 const MCFixup &Fixup, MCValue Target,
463 bool &IsPCRel, uint64_t &FixedValue) {
464 TargetObjectWriter->RecordRelocation(this, Asm, Layout, Fragment, Fixup,
468 void MachObjectWriter::BindIndirectSymbols(MCAssembler &Asm) {
469 // This is the point where 'as' creates actual symbols for indirect symbols
470 // (in the following two passes). It would be easier for us to do this sooner
471 // when we see the attribute, but that makes getting the order in the symbol
472 // table much more complicated than it is worth.
474 // FIXME: Revisit this when the dust settles.
476 // Report errors for use of .indirect_symbol not in a symbol pointer section
478 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
479 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
480 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
482 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
483 Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
484 Section.getType() != MachO::S_SYMBOL_STUBS) {
485 MCSymbol &Symbol = *it->Symbol;
486 report_fatal_error("indirect symbol '" + Symbol.getName() +
487 "' not in a symbol pointer or stub section");
491 // Bind non-lazy symbol pointers first.
492 unsigned IndirectIndex = 0;
493 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
494 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
495 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
497 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS)
500 // Initialize the section indirect symbol base, if necessary.
501 IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex));
503 Asm.getOrCreateSymbolData(*it->Symbol);
506 // Then lazy symbol pointers and symbol stubs.
508 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
509 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
510 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
512 if (Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
513 Section.getType() != MachO::S_SYMBOL_STUBS)
516 // Initialize the section indirect symbol base, if necessary.
517 IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex));
519 // Set the symbol type to undefined lazy, but only on construction.
521 // FIXME: Do not hardcode.
523 MCSymbolData &Entry = Asm.getOrCreateSymbolData(*it->Symbol, &Created);
525 Entry.setFlags(Entry.getFlags() | 0x0001);
529 /// ComputeSymbolTable - Compute the symbol table data
530 void MachObjectWriter::ComputeSymbolTable(
531 MCAssembler &Asm, std::vector<MachSymbolData> &LocalSymbolData,
532 std::vector<MachSymbolData> &ExternalSymbolData,
533 std::vector<MachSymbolData> &UndefinedSymbolData) {
534 // Build section lookup table.
535 DenseMap<const MCSection*, uint8_t> SectionIndexMap;
537 for (MCAssembler::iterator it = Asm.begin(),
538 ie = Asm.end(); it != ie; ++it, ++Index)
539 SectionIndexMap[&*it] = Index;
540 assert(Index <= 256 && "Too many sections!");
542 // Build the string table.
543 for (const MCSymbol &Symbol : Asm.symbols()) {
544 if (!Asm.isSymbolLinkerVisible(Symbol))
547 StringTable.add(Symbol.getName());
549 StringTable.finalize(StringTableBuilder::MachO);
551 // Build the symbol arrays but only for non-local symbols.
553 // The particular order that we collect and then sort the symbols is chosen to
554 // match 'as'. Even though it doesn't matter for correctness, this is
555 // important for letting us diff .o files.
556 for (const MCSymbol &Symbol : Asm.symbols()) {
557 MCSymbolData &SD = Symbol.getData();
559 // Ignore non-linker visible symbols.
560 if (!Asm.isSymbolLinkerVisible(Symbol))
563 if (!SD.isExternal() && !Symbol.isUndefined())
567 MSD.Symbol = &Symbol;
568 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
570 if (Symbol.isUndefined()) {
571 MSD.SectionIndex = 0;
572 UndefinedSymbolData.push_back(MSD);
573 } else if (Symbol.isAbsolute()) {
574 MSD.SectionIndex = 0;
575 ExternalSymbolData.push_back(MSD);
577 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
578 assert(MSD.SectionIndex && "Invalid section index!");
579 ExternalSymbolData.push_back(MSD);
583 // Now add the data for local symbols.
584 for (const MCSymbol &Symbol : Asm.symbols()) {
585 MCSymbolData &SD = Symbol.getData();
587 // Ignore non-linker visible symbols.
588 if (!Asm.isSymbolLinkerVisible(Symbol))
591 if (SD.isExternal() || Symbol.isUndefined())
595 MSD.Symbol = &Symbol;
596 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
598 if (Symbol.isAbsolute()) {
599 MSD.SectionIndex = 0;
600 LocalSymbolData.push_back(MSD);
602 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
603 assert(MSD.SectionIndex && "Invalid section index!");
604 LocalSymbolData.push_back(MSD);
608 // External and undefined symbols are required to be in lexicographic order.
609 std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
610 std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
612 // Set the symbol indices.
614 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
615 LocalSymbolData[i].Symbol->setIndex(Index++);
616 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
617 ExternalSymbolData[i].Symbol->setIndex(Index++);
618 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
619 UndefinedSymbolData[i].Symbol->setIndex(Index++);
621 for (const MCSection &Section : Asm) {
622 std::vector<RelAndSymbol> &Relocs = Relocations[&Section];
623 for (RelAndSymbol &Rel : Relocs) {
627 // Set the Index and the IsExtern bit.
628 unsigned Index = Rel.Sym->getIndex();
629 assert(isInt<24>(Index));
631 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & (~0U << 24)) | Index | (1 << 27);
633 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & 0xff) | Index << 8 | (1 << 4);
638 void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm,
639 const MCAsmLayout &Layout) {
640 uint64_t StartAddress = 0;
641 const SmallVectorImpl<MCSection *> &Order = Layout.getSectionOrder();
642 for (int i = 0, n = Order.size(); i != n ; ++i) {
643 const MCSection *Sec = Order[i];
644 StartAddress = RoundUpToAlignment(StartAddress, Sec->getAlignment());
645 SectionAddress[Sec] = StartAddress;
646 StartAddress += Layout.getSectionAddressSize(Sec);
648 // Explicitly pad the section to match the alignment requirements of the
649 // following one. This is for 'gas' compatibility, it shouldn't
650 /// strictly be necessary.
651 StartAddress += getPaddingSize(Sec, Layout);
655 void MachObjectWriter::ExecutePostLayoutBinding(MCAssembler &Asm,
656 const MCAsmLayout &Layout) {
657 computeSectionAddresses(Asm, Layout);
659 // Create symbol data for any indirect symbols.
660 BindIndirectSymbols(Asm);
663 bool MachObjectWriter::IsSymbolRefDifferenceFullyResolvedImpl(
664 const MCAssembler &Asm, const MCSymbol &SymA, const MCFragment &FB,
665 bool InSet, bool IsPCRel) const {
669 // The effective address is
670 // addr(atom(A)) + offset(A)
671 // - addr(atom(B)) - offset(B)
672 // and the offsets are not relocatable, so the fixup is fully resolved when
673 // addr(atom(A)) - addr(atom(B)) == 0.
674 const MCSymbol &SA = findAliasedSymbol(SymA);
675 const MCSection &SecA = SA.getSection();
676 const MCSection &SecB = *FB.getParent();
679 // The simple (Darwin, except on x86_64) way of dealing with this was to
680 // assume that any reference to a temporary symbol *must* be a temporary
681 // symbol in the same atom, unless the sections differ. Therefore, any PCrel
682 // relocation to a temporary symbol (in the same section) is fully
683 // resolved. This also works in conjunction with absolutized .set, which
684 // requires the compiler to use .set to absolutize the differences between
685 // symbols which the compiler knows to be assembly time constants, so we
686 // don't need to worry about considering symbol differences fully resolved.
688 // If the file isn't using sub-sections-via-symbols, we can make the
689 // same assumptions about any symbol that we normally make about
692 bool hasReliableSymbolDifference = isX86_64();
693 if (!hasReliableSymbolDifference) {
694 if (!SA.isInSection() || &SecA != &SecB ||
695 (!SA.isTemporary() &&
696 FB.getAtom() != Asm.getSymbolData(SA).getFragment()->getAtom() &&
697 Asm.getSubsectionsViaSymbols()))
701 // For Darwin x86_64, there is one special case when the reference IsPCRel.
702 // If the fragment with the reference does not have a base symbol but meets
703 // the simple way of dealing with this, in that it is a temporary symbol in
704 // the same atom then it is assumed to be fully resolved. This is needed so
705 // a relocation entry is not created and so the static linker does not
706 // mess up the reference later.
707 else if(!FB.getAtom() &&
708 SA.isTemporary() && SA.isInSection() && &SecA == &SecB){
712 if (!TargetObjectWriter->useAggressiveSymbolFolding())
716 // If they are not in the same section, we can't compute the diff.
720 const MCFragment *FA = Asm.getSymbolData(SA).getFragment();
722 // Bail if the symbol has no fragment.
726 // If the atoms are the same, they are guaranteed to have the same address.
727 if (FA->getAtom() == FB.getAtom())
730 // Otherwise, we can't prove this is fully resolved.
734 void MachObjectWriter::WriteObject(MCAssembler &Asm,
735 const MCAsmLayout &Layout) {
736 // Compute symbol table information and bind symbol indices.
737 ComputeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData,
738 UndefinedSymbolData);
740 unsigned NumSections = Asm.size();
741 const MCAssembler::VersionMinInfoType &VersionInfo =
742 Layout.getAssembler().getVersionMinInfo();
744 // The section data starts after the header, the segment load command (and
745 // section headers) and the symbol table.
746 unsigned NumLoadCommands = 1;
747 uint64_t LoadCommandsSize = is64Bit() ?
748 sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64):
749 sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section);
751 // Add the deployment target version info load command size, if used.
752 if (VersionInfo.Major != 0) {
754 LoadCommandsSize += sizeof(MachO::version_min_command);
757 // Add the data-in-code load command size, if used.
758 unsigned NumDataRegions = Asm.getDataRegions().size();
759 if (NumDataRegions) {
761 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
764 // Add the loh load command size, if used.
765 uint64_t LOHRawSize = Asm.getLOHContainer().getEmitSize(*this, Layout);
766 uint64_t LOHSize = RoundUpToAlignment(LOHRawSize, is64Bit() ? 8 : 4);
769 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
772 // Add the symbol table load command sizes, if used.
773 unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
774 UndefinedSymbolData.size();
776 NumLoadCommands += 2;
777 LoadCommandsSize += (sizeof(MachO::symtab_command) +
778 sizeof(MachO::dysymtab_command));
781 // Add the linker option load commands sizes.
782 const std::vector<std::vector<std::string> > &LinkerOptions =
783 Asm.getLinkerOptions();
784 for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
786 LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(LinkerOptions[i],
790 // Compute the total size of the section data, as well as its file size and vm
792 uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) :
793 sizeof(MachO::mach_header)) + LoadCommandsSize;
794 uint64_t SectionDataSize = 0;
795 uint64_t SectionDataFileSize = 0;
797 for (MCAssembler::const_iterator it = Asm.begin(),
798 ie = Asm.end(); it != ie; ++it) {
799 const MCSection &Sec = *it;
800 uint64_t Address = getSectionAddress(&Sec);
801 uint64_t Size = Layout.getSectionAddressSize(&Sec);
802 uint64_t FileSize = Layout.getSectionFileSize(&Sec);
803 FileSize += getPaddingSize(&Sec, Layout);
805 VMSize = std::max(VMSize, Address + Size);
807 if (it->isVirtualSection())
810 SectionDataSize = std::max(SectionDataSize, Address + Size);
811 SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
814 // The section data is padded to 4 bytes.
816 // FIXME: Is this machine dependent?
817 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
818 SectionDataFileSize += SectionDataPadding;
820 // Write the prolog, starting with the header and load command...
821 WriteHeader(NumLoadCommands, LoadCommandsSize,
822 Asm.getSubsectionsViaSymbols());
823 WriteSegmentLoadCommand(NumSections, VMSize,
824 SectionDataStart, SectionDataSize);
826 // ... and then the section headers.
827 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
828 for (MCAssembler::const_iterator it = Asm.begin(),
829 ie = Asm.end(); it != ie; ++it) {
830 std::vector<RelAndSymbol> &Relocs = Relocations[&*it];
831 unsigned NumRelocs = Relocs.size();
832 uint64_t SectionStart = SectionDataStart + getSectionAddress(&*it);
833 WriteSection(Asm, Layout, *it, SectionStart, RelocTableEnd, NumRelocs);
834 RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
837 // Write out the deployment target information, if it's available.
838 if (VersionInfo.Major != 0) {
839 assert(VersionInfo.Update < 256 && "unencodable update target version");
840 assert(VersionInfo.Minor < 256 && "unencodable minor target version");
841 assert(VersionInfo.Major < 65536 && "unencodable major target version");
842 uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
843 (VersionInfo.Major << 16);
844 Write32(VersionInfo.Kind == MCVM_OSXVersionMin ? MachO::LC_VERSION_MIN_MACOSX :
845 MachO::LC_VERSION_MIN_IPHONEOS);
846 Write32(sizeof(MachO::version_min_command));
847 Write32(EncodedVersion);
848 Write32(0); // reserved.
851 // Write the data-in-code load command, if used.
852 uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
853 if (NumDataRegions) {
854 uint64_t DataRegionsOffset = RelocTableEnd;
855 uint64_t DataRegionsSize = NumDataRegions * 8;
856 WriteLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
860 // Write the loh load command, if used.
861 uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
863 WriteLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
864 DataInCodeTableEnd, LOHSize);
866 // Write the symbol table load command, if used.
868 unsigned FirstLocalSymbol = 0;
869 unsigned NumLocalSymbols = LocalSymbolData.size();
870 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
871 unsigned NumExternalSymbols = ExternalSymbolData.size();
872 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
873 unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
874 unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
875 unsigned NumSymTabSymbols =
876 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
877 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
878 uint64_t IndirectSymbolOffset = 0;
880 // If used, the indirect symbols are written after the section data.
881 if (NumIndirectSymbols)
882 IndirectSymbolOffset = LOHTableEnd;
884 // The symbol table is written after the indirect symbol data.
885 uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
887 // The string table is written after symbol table.
888 uint64_t StringTableOffset =
889 SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
890 sizeof(MachO::nlist_64) :
891 sizeof(MachO::nlist));
892 WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
893 StringTableOffset, StringTable.data().size());
895 WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
896 FirstExternalSymbol, NumExternalSymbols,
897 FirstUndefinedSymbol, NumUndefinedSymbols,
898 IndirectSymbolOffset, NumIndirectSymbols);
901 // Write the linker options load commands.
902 for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
903 WriteLinkerOptionsLoadCommand(LinkerOptions[i]);
906 // Write the actual section data.
907 for (MCAssembler::const_iterator it = Asm.begin(),
908 ie = Asm.end(); it != ie; ++it) {
909 MCSection &Sec = *it;
910 Asm.writeSectionData(&Sec, Layout);
912 uint64_t Pad = getPaddingSize(&Sec, Layout);
916 // Write the extra padding.
917 WriteZeros(SectionDataPadding);
919 // Write the relocation entries.
920 for (MCAssembler::const_iterator it = Asm.begin(),
921 ie = Asm.end(); it != ie; ++it) {
922 // Write the section relocation entries, in reverse order to match 'as'
923 // (approximately, the exact algorithm is more complicated than this).
924 std::vector<RelAndSymbol> &Relocs = Relocations[&*it];
925 for (unsigned i = 0, e = Relocs.size(); i != e; ++i) {
926 Write32(Relocs[e - i - 1].MRE.r_word0);
927 Write32(Relocs[e - i - 1].MRE.r_word1);
931 // Write out the data-in-code region payload, if there is one.
932 for (MCAssembler::const_data_region_iterator
933 it = Asm.data_region_begin(), ie = Asm.data_region_end();
935 const DataRegionData *Data = &(*it);
936 uint64_t Start = getSymbolAddress(*Data->Start, Layout);
937 uint64_t End = getSymbolAddress(*Data->End, Layout);
938 DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
939 << " start: " << Start << "(" << Data->Start->getName() << ")"
940 << " end: " << End << "(" << Data->End->getName() << ")"
941 << " size: " << End - Start
944 Write16(End - Start);
948 // Write out the loh commands, if there is one.
951 unsigned Start = OS.tell();
953 Asm.getLOHContainer().Emit(*this, Layout);
954 // Pad to a multiple of the pointer size.
955 WriteBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
956 assert(OS.tell() - Start == LOHSize);
959 // Write the symbol table data, if used.
961 // Write the indirect symbol entries.
962 for (MCAssembler::const_indirect_symbol_iterator
963 it = Asm.indirect_symbol_begin(),
964 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
965 // Indirect symbols in the non-lazy symbol pointer section have some
967 const MCSectionMachO &Section =
968 static_cast<const MCSectionMachO &>(*it->Section);
969 if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
970 // If this symbol is defined and internal, mark it as such.
971 if (it->Symbol->isDefined() &&
972 !Asm.getSymbolData(*it->Symbol).isExternal()) {
973 uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
974 if (it->Symbol->isAbsolute())
975 Flags |= MachO::INDIRECT_SYMBOL_ABS;
981 Write32(it->Symbol->getIndex());
984 // FIXME: Check that offsets match computed ones.
986 // Write the symbol table entries.
987 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
988 WriteNlist(LocalSymbolData[i], Layout);
989 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
990 WriteNlist(ExternalSymbolData[i], Layout);
991 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
992 WriteNlist(UndefinedSymbolData[i], Layout);
994 // Write the string table.
995 OS << StringTable.data();
999 MCObjectWriter *llvm::createMachObjectWriter(MCMachObjectTargetWriter *MOTW,
1000 raw_pwrite_stream &OS,
1001 bool IsLittleEndian) {
1002 return new MachObjectWriter(MOTW, OS, IsLittleEndian);