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 const MCSectionData &SD = Sec.getSectionData();
201 uint64_t SectionSize = Layout.getSectionAddressSize(&Sec);
202 const MCSectionMachO &Section = cast<MCSectionMachO>(Sec);
204 // The offset is unused for virtual sections.
205 if (Section.isVirtualSection()) {
206 assert(Layout.getSectionFileSize(&Sec) == 0 && "Invalid file size!");
210 // struct section (68 bytes) or
211 // struct section_64 (80 bytes)
213 uint64_t Start = OS.tell();
216 WriteBytes(Section.getSectionName(), 16);
217 WriteBytes(Section.getSegmentName(), 16);
219 Write64(getSectionAddress(&SD.getSection())); // address
220 Write64(SectionSize); // size
222 Write32(getSectionAddress(&SD.getSection())); // address
223 Write32(SectionSize); // size
227 unsigned Flags = Section.getTypeAndAttributes();
228 if (Section.hasInstructions())
229 Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS;
231 assert(isPowerOf2_32(Section.getAlignment()) && "Invalid alignment!");
232 Write32(Log2_32(Section.getAlignment()));
233 Write32(NumRelocations ? RelocationsStart : 0);
234 Write32(NumRelocations);
236 Write32(IndirectSymBase.lookup(&SD.getSection())); // reserved1
237 Write32(Section.getStubSize()); // reserved2
239 Write32(0); // reserved3
241 assert(OS.tell() - Start == (is64Bit() ? sizeof(MachO::section_64) :
242 sizeof(MachO::section)));
245 void MachObjectWriter::WriteSymtabLoadCommand(uint32_t SymbolOffset,
247 uint32_t StringTableOffset,
248 uint32_t StringTableSize) {
249 // struct symtab_command (24 bytes)
251 uint64_t Start = OS.tell();
254 Write32(MachO::LC_SYMTAB);
255 Write32(sizeof(MachO::symtab_command));
256 Write32(SymbolOffset);
258 Write32(StringTableOffset);
259 Write32(StringTableSize);
261 assert(OS.tell() - Start == sizeof(MachO::symtab_command));
264 void MachObjectWriter::WriteDysymtabLoadCommand(uint32_t FirstLocalSymbol,
265 uint32_t NumLocalSymbols,
266 uint32_t FirstExternalSymbol,
267 uint32_t NumExternalSymbols,
268 uint32_t FirstUndefinedSymbol,
269 uint32_t NumUndefinedSymbols,
270 uint32_t IndirectSymbolOffset,
271 uint32_t NumIndirectSymbols) {
272 // struct dysymtab_command (80 bytes)
274 uint64_t Start = OS.tell();
277 Write32(MachO::LC_DYSYMTAB);
278 Write32(sizeof(MachO::dysymtab_command));
279 Write32(FirstLocalSymbol);
280 Write32(NumLocalSymbols);
281 Write32(FirstExternalSymbol);
282 Write32(NumExternalSymbols);
283 Write32(FirstUndefinedSymbol);
284 Write32(NumUndefinedSymbols);
285 Write32(0); // tocoff
287 Write32(0); // modtaboff
288 Write32(0); // nmodtab
289 Write32(0); // extrefsymoff
290 Write32(0); // nextrefsyms
291 Write32(IndirectSymbolOffset);
292 Write32(NumIndirectSymbols);
293 Write32(0); // extreloff
294 Write32(0); // nextrel
295 Write32(0); // locreloff
296 Write32(0); // nlocrel
298 assert(OS.tell() - Start == sizeof(MachO::dysymtab_command));
301 MachObjectWriter::MachSymbolData *
302 MachObjectWriter::findSymbolData(const MCSymbol &Sym) {
303 for (auto &Entry : LocalSymbolData)
304 if (Entry.Symbol == &Sym)
307 for (auto &Entry : ExternalSymbolData)
308 if (Entry.Symbol == &Sym)
311 for (auto &Entry : UndefinedSymbolData)
312 if (Entry.Symbol == &Sym)
318 const MCSymbol &MachObjectWriter::findAliasedSymbol(const MCSymbol &Sym) const {
319 const MCSymbol *S = &Sym;
320 while (S->isVariable()) {
321 const MCExpr *Value = S->getVariableValue();
322 const auto *Ref = dyn_cast<MCSymbolRefExpr>(Value);
325 S = &Ref->getSymbol();
330 void MachObjectWriter::WriteNlist(MachSymbolData &MSD,
331 const MCAsmLayout &Layout) {
332 const MCSymbol *Symbol = MSD.Symbol;
333 MCSymbolData &Data = Symbol->getData();
334 const MCSymbol *AliasedSymbol = &findAliasedSymbol(*Symbol);
335 uint8_t SectionIndex = MSD.SectionIndex;
337 uint16_t Flags = Data.getFlags();
338 uint64_t Address = 0;
339 bool IsAlias = Symbol != AliasedSymbol;
341 const MCSymbol &OrigSymbol = *Symbol;
342 MachSymbolData *AliaseeInfo;
344 AliaseeInfo = findSymbolData(*AliasedSymbol);
346 SectionIndex = AliaseeInfo->SectionIndex;
347 Symbol = AliasedSymbol;
348 // FIXME: Should this update Data as well? Do we need OrigSymbol at all?
351 // Set the N_TYPE bits. See <mach-o/nlist.h>.
353 // FIXME: Are the prebound or indirect fields possible here?
354 if (IsAlias && Symbol->isUndefined())
355 Type = MachO::N_INDR;
356 else if (Symbol->isUndefined())
357 Type = MachO::N_UNDF;
358 else if (Symbol->isAbsolute())
361 Type = MachO::N_SECT;
363 // FIXME: Set STAB bits.
365 if (Data.isPrivateExtern())
366 Type |= MachO::N_PEXT;
369 if (Data.isExternal() || (!IsAlias && Symbol->isUndefined()))
370 Type |= MachO::N_EXT;
372 // Compute the symbol address.
373 if (IsAlias && Symbol->isUndefined())
374 Address = AliaseeInfo->StringIndex;
375 else if (Symbol->isDefined())
376 Address = getSymbolAddress(OrigSymbol, Layout);
377 else if (Data.isCommon()) {
378 // Common symbols are encoded with the size in the address
379 // field, and their alignment in the flags.
380 Address = Data.getCommonSize();
382 // Common alignment is packed into the 'desc' bits.
383 if (unsigned Align = Data.getCommonAlignment()) {
384 unsigned Log2Size = Log2_32(Align);
385 assert((1U << Log2Size) == Align && "Invalid 'common' alignment!");
387 report_fatal_error("invalid 'common' alignment '" +
388 Twine(Align) + "' for '" + Symbol->getName() + "'",
390 // FIXME: Keep this mask with the SymbolFlags enumeration.
391 Flags = (Flags & 0xF0FF) | (Log2Size << 8);
395 if (Layout.getAssembler().isThumbFunc(Symbol))
396 Flags |= SF_ThumbFunc;
398 // struct nlist (12 bytes)
400 Write32(MSD.StringIndex);
402 Write8(SectionIndex);
404 // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc'
413 void MachObjectWriter::WriteLinkeditLoadCommand(uint32_t Type,
416 uint64_t Start = OS.tell();
420 Write32(sizeof(MachO::linkedit_data_command));
424 assert(OS.tell() - Start == sizeof(MachO::linkedit_data_command));
427 static unsigned ComputeLinkerOptionsLoadCommandSize(
428 const std::vector<std::string> &Options, bool is64Bit)
430 unsigned Size = sizeof(MachO::linker_option_command);
431 for (unsigned i = 0, e = Options.size(); i != e; ++i)
432 Size += Options[i].size() + 1;
433 return RoundUpToAlignment(Size, is64Bit ? 8 : 4);
436 void MachObjectWriter::WriteLinkerOptionsLoadCommand(
437 const std::vector<std::string> &Options)
439 unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit());
440 uint64_t Start = OS.tell();
443 Write32(MachO::LC_LINKER_OPTION);
445 Write32(Options.size());
446 uint64_t BytesWritten = sizeof(MachO::linker_option_command);
447 for (unsigned i = 0, e = Options.size(); i != e; ++i) {
448 // Write each string, including the null byte.
449 const std::string &Option = Options[i];
450 WriteBytes(Option.c_str(), Option.size() + 1);
451 BytesWritten += Option.size() + 1;
454 // Pad to a multiple of the pointer size.
455 WriteBytes("", OffsetToAlignment(BytesWritten, is64Bit() ? 8 : 4));
457 assert(OS.tell() - Start == Size);
460 void MachObjectWriter::RecordRelocation(MCAssembler &Asm,
461 const MCAsmLayout &Layout,
462 const MCFragment *Fragment,
463 const MCFixup &Fixup, MCValue Target,
464 bool &IsPCRel, uint64_t &FixedValue) {
465 TargetObjectWriter->RecordRelocation(this, Asm, Layout, Fragment, Fixup,
469 void MachObjectWriter::BindIndirectSymbols(MCAssembler &Asm) {
470 // This is the point where 'as' creates actual symbols for indirect symbols
471 // (in the following two passes). It would be easier for us to do this sooner
472 // when we see the attribute, but that makes getting the order in the symbol
473 // table much more complicated than it is worth.
475 // FIXME: Revisit this when the dust settles.
477 // Report errors for use of .indirect_symbol not in a symbol pointer section
479 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
480 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
481 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
483 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
484 Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
485 Section.getType() != MachO::S_SYMBOL_STUBS) {
486 MCSymbol &Symbol = *it->Symbol;
487 report_fatal_error("indirect symbol '" + Symbol.getName() +
488 "' not in a symbol pointer or stub section");
492 // Bind non-lazy symbol pointers first.
493 unsigned IndirectIndex = 0;
494 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
495 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
496 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
498 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS)
501 // Initialize the section indirect symbol base, if necessary.
502 IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex));
504 Asm.getOrCreateSymbolData(*it->Symbol);
507 // Then lazy symbol pointers and symbol stubs.
509 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
510 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
511 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
513 if (Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
514 Section.getType() != MachO::S_SYMBOL_STUBS)
517 // Initialize the section indirect symbol base, if necessary.
518 IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex));
520 // Set the symbol type to undefined lazy, but only on construction.
522 // FIXME: Do not hardcode.
524 MCSymbolData &Entry = Asm.getOrCreateSymbolData(*it->Symbol, &Created);
526 Entry.setFlags(Entry.getFlags() | 0x0001);
530 /// ComputeSymbolTable - Compute the symbol table data
531 void MachObjectWriter::ComputeSymbolTable(
532 MCAssembler &Asm, std::vector<MachSymbolData> &LocalSymbolData,
533 std::vector<MachSymbolData> &ExternalSymbolData,
534 std::vector<MachSymbolData> &UndefinedSymbolData) {
535 // Build section lookup table.
536 DenseMap<const MCSection*, uint8_t> SectionIndexMap;
538 for (MCAssembler::iterator it = Asm.begin(),
539 ie = Asm.end(); it != ie; ++it, ++Index)
540 SectionIndexMap[&*it] = Index;
541 assert(Index <= 256 && "Too many sections!");
543 // Build the string table.
544 for (const MCSymbol &Symbol : Asm.symbols()) {
545 if (!Asm.isSymbolLinkerVisible(Symbol))
548 StringTable.add(Symbol.getName());
550 StringTable.finalize(StringTableBuilder::MachO);
552 // Build the symbol arrays but only for non-local symbols.
554 // The particular order that we collect and then sort the symbols is chosen to
555 // match 'as'. Even though it doesn't matter for correctness, this is
556 // important for letting us diff .o files.
557 for (const MCSymbol &Symbol : Asm.symbols()) {
558 MCSymbolData &SD = Symbol.getData();
560 // Ignore non-linker visible symbols.
561 if (!Asm.isSymbolLinkerVisible(Symbol))
564 if (!SD.isExternal() && !Symbol.isUndefined())
568 MSD.Symbol = &Symbol;
569 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
571 if (Symbol.isUndefined()) {
572 MSD.SectionIndex = 0;
573 UndefinedSymbolData.push_back(MSD);
574 } else if (Symbol.isAbsolute()) {
575 MSD.SectionIndex = 0;
576 ExternalSymbolData.push_back(MSD);
578 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
579 assert(MSD.SectionIndex && "Invalid section index!");
580 ExternalSymbolData.push_back(MSD);
584 // Now add the data for local symbols.
585 for (const MCSymbol &Symbol : Asm.symbols()) {
586 MCSymbolData &SD = Symbol.getData();
588 // Ignore non-linker visible symbols.
589 if (!Asm.isSymbolLinkerVisible(Symbol))
592 if (SD.isExternal() || Symbol.isUndefined())
596 MSD.Symbol = &Symbol;
597 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
599 if (Symbol.isAbsolute()) {
600 MSD.SectionIndex = 0;
601 LocalSymbolData.push_back(MSD);
603 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
604 assert(MSD.SectionIndex && "Invalid section index!");
605 LocalSymbolData.push_back(MSD);
609 // External and undefined symbols are required to be in lexicographic order.
610 std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
611 std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
613 // Set the symbol indices.
615 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
616 LocalSymbolData[i].Symbol->setIndex(Index++);
617 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
618 ExternalSymbolData[i].Symbol->setIndex(Index++);
619 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
620 UndefinedSymbolData[i].Symbol->setIndex(Index++);
622 for (const MCSection &Section : Asm) {
623 std::vector<RelAndSymbol> &Relocs = Relocations[&Section];
624 for (RelAndSymbol &Rel : Relocs) {
628 // Set the Index and the IsExtern bit.
629 unsigned Index = Rel.Sym->getIndex();
630 assert(isInt<24>(Index));
632 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & (~0U << 24)) | Index | (1 << 27);
634 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & 0xff) | Index << 8 | (1 << 4);
639 void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm,
640 const MCAsmLayout &Layout) {
641 uint64_t StartAddress = 0;
642 const SmallVectorImpl<MCSection *> &Order = Layout.getSectionOrder();
643 for (int i = 0, n = Order.size(); i != n ; ++i) {
644 const MCSection *Sec = Order[i];
645 StartAddress = RoundUpToAlignment(StartAddress, Sec->getAlignment());
646 SectionAddress[Sec] = StartAddress;
647 StartAddress += Layout.getSectionAddressSize(Sec);
649 // Explicitly pad the section to match the alignment requirements of the
650 // following one. This is for 'gas' compatibility, it shouldn't
651 /// strictly be necessary.
652 StartAddress += getPaddingSize(Sec, Layout);
656 void MachObjectWriter::ExecutePostLayoutBinding(MCAssembler &Asm,
657 const MCAsmLayout &Layout) {
658 computeSectionAddresses(Asm, Layout);
660 // Create symbol data for any indirect symbols.
661 BindIndirectSymbols(Asm);
664 bool MachObjectWriter::IsSymbolRefDifferenceFullyResolvedImpl(
665 const MCAssembler &Asm, const MCSymbol &SymA, const MCFragment &FB,
666 bool InSet, bool IsPCRel) const {
670 // The effective address is
671 // addr(atom(A)) + offset(A)
672 // - addr(atom(B)) - offset(B)
673 // and the offsets are not relocatable, so the fixup is fully resolved when
674 // addr(atom(A)) - addr(atom(B)) == 0.
675 const MCSymbol &SA = findAliasedSymbol(SymA);
676 const MCSection &SecA = SA.getSection();
677 const MCSection &SecB = *FB.getParent();
680 // The simple (Darwin, except on x86_64) way of dealing with this was to
681 // assume that any reference to a temporary symbol *must* be a temporary
682 // symbol in the same atom, unless the sections differ. Therefore, any PCrel
683 // relocation to a temporary symbol (in the same section) is fully
684 // resolved. This also works in conjunction with absolutized .set, which
685 // requires the compiler to use .set to absolutize the differences between
686 // symbols which the compiler knows to be assembly time constants, so we
687 // don't need to worry about considering symbol differences fully resolved.
689 // If the file isn't using sub-sections-via-symbols, we can make the
690 // same assumptions about any symbol that we normally make about
693 bool hasReliableSymbolDifference = isX86_64();
694 if (!hasReliableSymbolDifference) {
695 if (!SA.isInSection() || &SecA != &SecB ||
696 (!SA.isTemporary() &&
697 FB.getAtom() != Asm.getSymbolData(SA).getFragment()->getAtom() &&
698 Asm.getSubsectionsViaSymbols()))
702 // For Darwin x86_64, there is one special case when the reference IsPCRel.
703 // If the fragment with the reference does not have a base symbol but meets
704 // the simple way of dealing with this, in that it is a temporary symbol in
705 // the same atom then it is assumed to be fully resolved. This is needed so
706 // a relocation entry is not created and so the static linker does not
707 // mess up the reference later.
708 else if(!FB.getAtom() &&
709 SA.isTemporary() && SA.isInSection() && &SecA == &SecB){
713 if (!TargetObjectWriter->useAggressiveSymbolFolding())
717 // If they are not in the same section, we can't compute the diff.
721 const MCFragment *FA = Asm.getSymbolData(SA).getFragment();
723 // Bail if the symbol has no fragment.
727 // If the atoms are the same, they are guaranteed to have the same address.
728 if (FA->getAtom() == FB.getAtom())
731 // Otherwise, we can't prove this is fully resolved.
735 void MachObjectWriter::WriteObject(MCAssembler &Asm,
736 const MCAsmLayout &Layout) {
737 // Compute symbol table information and bind symbol indices.
738 ComputeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData,
739 UndefinedSymbolData);
741 unsigned NumSections = Asm.size();
742 const MCAssembler::VersionMinInfoType &VersionInfo =
743 Layout.getAssembler().getVersionMinInfo();
745 // The section data starts after the header, the segment load command (and
746 // section headers) and the symbol table.
747 unsigned NumLoadCommands = 1;
748 uint64_t LoadCommandsSize = is64Bit() ?
749 sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64):
750 sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section);
752 // Add the deployment target version info load command size, if used.
753 if (VersionInfo.Major != 0) {
755 LoadCommandsSize += sizeof(MachO::version_min_command);
758 // Add the data-in-code load command size, if used.
759 unsigned NumDataRegions = Asm.getDataRegions().size();
760 if (NumDataRegions) {
762 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
765 // Add the loh load command size, if used.
766 uint64_t LOHRawSize = Asm.getLOHContainer().getEmitSize(*this, Layout);
767 uint64_t LOHSize = RoundUpToAlignment(LOHRawSize, is64Bit() ? 8 : 4);
770 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
773 // Add the symbol table load command sizes, if used.
774 unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
775 UndefinedSymbolData.size();
777 NumLoadCommands += 2;
778 LoadCommandsSize += (sizeof(MachO::symtab_command) +
779 sizeof(MachO::dysymtab_command));
782 // Add the linker option load commands sizes.
783 const std::vector<std::vector<std::string> > &LinkerOptions =
784 Asm.getLinkerOptions();
785 for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
787 LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(LinkerOptions[i],
791 // Compute the total size of the section data, as well as its file size and vm
793 uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) :
794 sizeof(MachO::mach_header)) + LoadCommandsSize;
795 uint64_t SectionDataSize = 0;
796 uint64_t SectionDataFileSize = 0;
798 for (MCAssembler::const_iterator it = Asm.begin(),
799 ie = Asm.end(); it != ie; ++it) {
800 const MCSection &Sec = *it;
801 const MCSectionData &SD = it->getSectionData();
802 uint64_t Address = getSectionAddress(&Sec);
803 uint64_t Size = Layout.getSectionAddressSize(&Sec);
804 uint64_t FileSize = Layout.getSectionFileSize(&Sec);
805 FileSize += getPaddingSize(&Sec, Layout);
807 VMSize = std::max(VMSize, Address + Size);
809 if (SD.getSection().isVirtualSection())
812 SectionDataSize = std::max(SectionDataSize, Address + Size);
813 SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
816 // The section data is padded to 4 bytes.
818 // FIXME: Is this machine dependent?
819 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
820 SectionDataFileSize += SectionDataPadding;
822 // Write the prolog, starting with the header and load command...
823 WriteHeader(NumLoadCommands, LoadCommandsSize,
824 Asm.getSubsectionsViaSymbols());
825 WriteSegmentLoadCommand(NumSections, VMSize,
826 SectionDataStart, SectionDataSize);
828 // ... and then the section headers.
829 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
830 for (MCAssembler::const_iterator it = Asm.begin(),
831 ie = Asm.end(); it != ie; ++it) {
832 std::vector<RelAndSymbol> &Relocs = Relocations[&*it];
833 unsigned NumRelocs = Relocs.size();
834 uint64_t SectionStart = SectionDataStart + getSectionAddress(&*it);
835 WriteSection(Asm, Layout, *it, SectionStart, RelocTableEnd, NumRelocs);
836 RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
839 // Write out the deployment target information, if it's available.
840 if (VersionInfo.Major != 0) {
841 assert(VersionInfo.Update < 256 && "unencodable update target version");
842 assert(VersionInfo.Minor < 256 && "unencodable minor target version");
843 assert(VersionInfo.Major < 65536 && "unencodable major target version");
844 uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
845 (VersionInfo.Major << 16);
846 Write32(VersionInfo.Kind == MCVM_OSXVersionMin ? MachO::LC_VERSION_MIN_MACOSX :
847 MachO::LC_VERSION_MIN_IPHONEOS);
848 Write32(sizeof(MachO::version_min_command));
849 Write32(EncodedVersion);
850 Write32(0); // reserved.
853 // Write the data-in-code load command, if used.
854 uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
855 if (NumDataRegions) {
856 uint64_t DataRegionsOffset = RelocTableEnd;
857 uint64_t DataRegionsSize = NumDataRegions * 8;
858 WriteLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
862 // Write the loh load command, if used.
863 uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
865 WriteLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
866 DataInCodeTableEnd, LOHSize);
868 // Write the symbol table load command, if used.
870 unsigned FirstLocalSymbol = 0;
871 unsigned NumLocalSymbols = LocalSymbolData.size();
872 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
873 unsigned NumExternalSymbols = ExternalSymbolData.size();
874 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
875 unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
876 unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
877 unsigned NumSymTabSymbols =
878 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
879 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
880 uint64_t IndirectSymbolOffset = 0;
882 // If used, the indirect symbols are written after the section data.
883 if (NumIndirectSymbols)
884 IndirectSymbolOffset = LOHTableEnd;
886 // The symbol table is written after the indirect symbol data.
887 uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
889 // The string table is written after symbol table.
890 uint64_t StringTableOffset =
891 SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
892 sizeof(MachO::nlist_64) :
893 sizeof(MachO::nlist));
894 WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
895 StringTableOffset, StringTable.data().size());
897 WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
898 FirstExternalSymbol, NumExternalSymbols,
899 FirstUndefinedSymbol, NumUndefinedSymbols,
900 IndirectSymbolOffset, NumIndirectSymbols);
903 // Write the linker options load commands.
904 for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
905 WriteLinkerOptionsLoadCommand(LinkerOptions[i]);
908 // Write the actual section data.
909 for (MCAssembler::const_iterator it = Asm.begin(),
910 ie = Asm.end(); it != ie; ++it) {
911 MCSection &Sec = *it;
912 Asm.writeSectionData(&Sec, Layout);
914 uint64_t Pad = getPaddingSize(&Sec, Layout);
918 // Write the extra padding.
919 WriteZeros(SectionDataPadding);
921 // Write the relocation entries.
922 for (MCAssembler::const_iterator it = Asm.begin(),
923 ie = Asm.end(); it != ie; ++it) {
924 // Write the section relocation entries, in reverse order to match 'as'
925 // (approximately, the exact algorithm is more complicated than this).
926 std::vector<RelAndSymbol> &Relocs = Relocations[&*it];
927 for (unsigned i = 0, e = Relocs.size(); i != e; ++i) {
928 Write32(Relocs[e - i - 1].MRE.r_word0);
929 Write32(Relocs[e - i - 1].MRE.r_word1);
933 // Write out the data-in-code region payload, if there is one.
934 for (MCAssembler::const_data_region_iterator
935 it = Asm.data_region_begin(), ie = Asm.data_region_end();
937 const DataRegionData *Data = &(*it);
938 uint64_t Start = getSymbolAddress(*Data->Start, Layout);
939 uint64_t End = getSymbolAddress(*Data->End, Layout);
940 DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
941 << " start: " << Start << "(" << Data->Start->getName() << ")"
942 << " end: " << End << "(" << Data->End->getName() << ")"
943 << " size: " << End - Start
946 Write16(End - Start);
950 // Write out the loh commands, if there is one.
953 unsigned Start = OS.tell();
955 Asm.getLOHContainer().Emit(*this, Layout);
956 // Pad to a multiple of the pointer size.
957 WriteBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
958 assert(OS.tell() - Start == LOHSize);
961 // Write the symbol table data, if used.
963 // Write the indirect symbol entries.
964 for (MCAssembler::const_indirect_symbol_iterator
965 it = Asm.indirect_symbol_begin(),
966 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
967 // Indirect symbols in the non-lazy symbol pointer section have some
969 const MCSectionMachO &Section =
970 static_cast<const MCSectionMachO &>(*it->Section);
971 if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
972 // If this symbol is defined and internal, mark it as such.
973 if (it->Symbol->isDefined() &&
974 !Asm.getSymbolData(*it->Symbol).isExternal()) {
975 uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
976 if (it->Symbol->isAbsolute())
977 Flags |= MachO::INDIRECT_SYMBOL_ABS;
983 Write32(it->Symbol->getIndex());
986 // FIXME: Check that offsets match computed ones.
988 // Write the symbol table entries.
989 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
990 WriteNlist(LocalSymbolData[i], Layout);
991 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
992 WriteNlist(ExternalSymbolData[i], Layout);
993 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
994 WriteNlist(UndefinedSymbolData[i], Layout);
996 // Write the string table.
997 OS << StringTable.data();
1001 MCObjectWriter *llvm::createMachObjectWriter(MCMachObjectTargetWriter *MOTW,
1002 raw_pwrite_stream &OS,
1003 bool IsLittleEndian) {
1004 return new MachObjectWriter(MOTW, OS, IsLittleEndian);