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 MCSymbol &S,
77 const MCAsmLayout &Layout) const {
78 // If this is a variable, then recursively evaluate now.
80 if (const MCConstantExpr *C =
81 dyn_cast<const MCConstantExpr>(S.getVariableValue()))
86 if (!S.getVariableValue()->EvaluateAsRelocatable(Target, &Layout, nullptr))
87 report_fatal_error("unable to evaluate offset for variable '" +
90 // Verify that any used symbols are defined.
91 if (Target.getSymA() && Target.getSymA()->getSymbol().isUndefined())
92 report_fatal_error("unable to evaluate offset to undefined symbol '" +
93 Target.getSymA()->getSymbol().getName() + "'");
94 if (Target.getSymB() && Target.getSymB()->getSymbol().isUndefined())
95 report_fatal_error("unable to evaluate offset to undefined symbol '" +
96 Target.getSymB()->getSymbol().getName() + "'");
98 uint64_t Address = Target.getConstant();
100 Address += getSymbolAddress(Target.getSymA()->getSymbol(), Layout);
101 if (Target.getSymB())
102 Address += getSymbolAddress(Target.getSymB()->getSymbol(), Layout);
106 return getSectionAddress(S.getData().getFragment()->getParent()) +
107 Layout.getSymbolOffset(S);
110 uint64_t MachObjectWriter::getPaddingSize(const MCSectionData *SD,
111 const MCAsmLayout &Layout) const {
112 uint64_t EndAddr = getSectionAddress(SD) + Layout.getSectionAddressSize(SD);
113 unsigned Next = SD->getLayoutOrder() + 1;
114 if (Next >= Layout.getSectionOrder().size())
117 const MCSectionData &NextSD = *Layout.getSectionOrder()[Next];
118 if (NextSD.getSection().isVirtualSection())
120 return OffsetToAlignment(EndAddr, NextSD.getAlignment());
123 void MachObjectWriter::WriteHeader(unsigned NumLoadCommands,
124 unsigned LoadCommandsSize,
125 bool SubsectionsViaSymbols) {
128 if (SubsectionsViaSymbols)
129 Flags |= MachO::MH_SUBSECTIONS_VIA_SYMBOLS;
131 // struct mach_header (28 bytes) or
132 // struct mach_header_64 (32 bytes)
134 uint64_t Start = OS.tell();
137 Write32(is64Bit() ? MachO::MH_MAGIC_64 : MachO::MH_MAGIC);
139 Write32(TargetObjectWriter->getCPUType());
140 Write32(TargetObjectWriter->getCPUSubtype());
142 Write32(MachO::MH_OBJECT);
143 Write32(NumLoadCommands);
144 Write32(LoadCommandsSize);
147 Write32(0); // reserved
149 assert(OS.tell() - Start ==
150 (is64Bit()?sizeof(MachO::mach_header_64): sizeof(MachO::mach_header)));
153 /// WriteSegmentLoadCommand - Write a segment load command.
155 /// \param NumSections The number of sections in this segment.
156 /// \param SectionDataSize The total size of the sections.
157 void MachObjectWriter::WriteSegmentLoadCommand(unsigned NumSections,
159 uint64_t SectionDataStartOffset,
160 uint64_t SectionDataSize) {
161 // struct segment_command (56 bytes) or
162 // struct segment_command_64 (72 bytes)
164 uint64_t Start = OS.tell();
167 unsigned SegmentLoadCommandSize =
168 is64Bit() ? sizeof(MachO::segment_command_64):
169 sizeof(MachO::segment_command);
170 Write32(is64Bit() ? MachO::LC_SEGMENT_64 : MachO::LC_SEGMENT);
171 Write32(SegmentLoadCommandSize +
172 NumSections * (is64Bit() ? sizeof(MachO::section_64) :
173 sizeof(MachO::section)));
177 Write64(0); // vmaddr
178 Write64(VMSize); // vmsize
179 Write64(SectionDataStartOffset); // file offset
180 Write64(SectionDataSize); // file size
182 Write32(0); // vmaddr
183 Write32(VMSize); // vmsize
184 Write32(SectionDataStartOffset); // file offset
185 Write32(SectionDataSize); // file size
188 Write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE);
190 Write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE);
191 Write32(NumSections);
194 assert(OS.tell() - Start == SegmentLoadCommandSize);
197 void MachObjectWriter::WriteSection(const MCAssembler &Asm,
198 const MCAsmLayout &Layout,
199 const MCSectionData &SD,
201 uint64_t RelocationsStart,
202 unsigned NumRelocations) {
203 uint64_t SectionSize = Layout.getSectionAddressSize(&SD);
205 // The offset is unused for virtual sections.
206 if (SD.getSection().isVirtualSection()) {
207 assert(Layout.getSectionFileSize(&SD) == 0 && "Invalid file size!");
211 // struct section (68 bytes) or
212 // struct section_64 (80 bytes)
214 uint64_t Start = OS.tell();
217 const MCSectionMachO &Section = cast<MCSectionMachO>(SD.getSection());
218 WriteBytes(Section.getSectionName(), 16);
219 WriteBytes(Section.getSegmentName(), 16);
221 Write64(getSectionAddress(&SD)); // address
222 Write64(SectionSize); // size
224 Write32(getSectionAddress(&SD)); // address
225 Write32(SectionSize); // size
229 unsigned Flags = Section.getTypeAndAttributes();
230 if (SD.hasInstructions())
231 Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS;
233 assert(isPowerOf2_32(SD.getAlignment()) && "Invalid alignment!");
234 Write32(Log2_32(SD.getAlignment()));
235 Write32(NumRelocations ? RelocationsStart : 0);
236 Write32(NumRelocations);
238 Write32(IndirectSymBase.lookup(&SD)); // reserved1
239 Write32(Section.getStubSize()); // reserved2
241 Write32(0); // reserved3
243 assert(OS.tell() - Start == (is64Bit() ? sizeof(MachO::section_64) :
244 sizeof(MachO::section)));
247 void MachObjectWriter::WriteSymtabLoadCommand(uint32_t SymbolOffset,
249 uint32_t StringTableOffset,
250 uint32_t StringTableSize) {
251 // struct symtab_command (24 bytes)
253 uint64_t Start = OS.tell();
256 Write32(MachO::LC_SYMTAB);
257 Write32(sizeof(MachO::symtab_command));
258 Write32(SymbolOffset);
260 Write32(StringTableOffset);
261 Write32(StringTableSize);
263 assert(OS.tell() - Start == sizeof(MachO::symtab_command));
266 void MachObjectWriter::WriteDysymtabLoadCommand(uint32_t FirstLocalSymbol,
267 uint32_t NumLocalSymbols,
268 uint32_t FirstExternalSymbol,
269 uint32_t NumExternalSymbols,
270 uint32_t FirstUndefinedSymbol,
271 uint32_t NumUndefinedSymbols,
272 uint32_t IndirectSymbolOffset,
273 uint32_t NumIndirectSymbols) {
274 // struct dysymtab_command (80 bytes)
276 uint64_t Start = OS.tell();
279 Write32(MachO::LC_DYSYMTAB);
280 Write32(sizeof(MachO::dysymtab_command));
281 Write32(FirstLocalSymbol);
282 Write32(NumLocalSymbols);
283 Write32(FirstExternalSymbol);
284 Write32(NumExternalSymbols);
285 Write32(FirstUndefinedSymbol);
286 Write32(NumUndefinedSymbols);
287 Write32(0); // tocoff
289 Write32(0); // modtaboff
290 Write32(0); // nmodtab
291 Write32(0); // extrefsymoff
292 Write32(0); // nextrefsyms
293 Write32(IndirectSymbolOffset);
294 Write32(NumIndirectSymbols);
295 Write32(0); // extreloff
296 Write32(0); // nextrel
297 Write32(0); // locreloff
298 Write32(0); // nlocrel
300 assert(OS.tell() - Start == sizeof(MachO::dysymtab_command));
303 MachObjectWriter::MachSymbolData *
304 MachObjectWriter::findSymbolData(const MCSymbol &Sym) {
305 for (auto &Entry : LocalSymbolData)
306 if (&Entry.SymbolData->getSymbol() == &Sym)
309 for (auto &Entry : ExternalSymbolData)
310 if (&Entry.SymbolData->getSymbol() == &Sym)
313 for (auto &Entry : UndefinedSymbolData)
314 if (&Entry.SymbolData->getSymbol() == &Sym)
320 const MCSymbol &MachObjectWriter::findAliasedSymbol(const MCSymbol &Sym) const {
321 const MCSymbol *S = &Sym;
322 while (S->isVariable()) {
323 const MCExpr *Value = S->getVariableValue();
324 const auto *Ref = dyn_cast<MCSymbolRefExpr>(Value);
327 S = &Ref->getSymbol();
332 void MachObjectWriter::WriteNlist(MachSymbolData &MSD,
333 const MCAsmLayout &Layout) {
334 MCSymbolData &Data = *MSD.SymbolData;
335 const MCSymbol *Symbol = &Data.getSymbol();
336 const MCSymbol *AliasedSymbol = &findAliasedSymbol(*Symbol);
337 uint8_t SectionIndex = MSD.SectionIndex;
339 uint16_t Flags = Data.getFlags();
340 uint64_t Address = 0;
341 bool IsAlias = Symbol != AliasedSymbol;
343 MachSymbolData *AliaseeInfo;
345 AliaseeInfo = findSymbolData(*AliasedSymbol);
347 SectionIndex = AliaseeInfo->SectionIndex;
348 Symbol = AliasedSymbol;
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 // FIXME: Should Data.getSymbol() always be *Symbol? It doesn't look like
378 Address = getSymbolAddress(Data.getSymbol(), Layout);
379 else if (Data.isCommon()) {
380 // Common symbols are encoded with the size in the address
381 // field, and their alignment in the flags.
382 Address = Data.getCommonSize();
384 // Common alignment is packed into the 'desc' bits.
385 if (unsigned Align = Data.getCommonAlignment()) {
386 unsigned Log2Size = Log2_32(Align);
387 assert((1U << Log2Size) == Align && "Invalid 'common' alignment!");
389 report_fatal_error("invalid 'common' alignment '" +
390 Twine(Align) + "' for '" + Symbol->getName() + "'",
392 // FIXME: Keep this mask with the SymbolFlags enumeration.
393 Flags = (Flags & 0xF0FF) | (Log2Size << 8);
397 if (Layout.getAssembler().isThumbFunc(Symbol))
398 Flags |= SF_ThumbFunc;
400 // struct nlist (12 bytes)
402 Write32(MSD.StringIndex);
404 Write8(SectionIndex);
406 // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc'
415 void MachObjectWriter::WriteLinkeditLoadCommand(uint32_t Type,
418 uint64_t Start = OS.tell();
422 Write32(sizeof(MachO::linkedit_data_command));
426 assert(OS.tell() - Start == sizeof(MachO::linkedit_data_command));
429 static unsigned ComputeLinkerOptionsLoadCommandSize(
430 const std::vector<std::string> &Options, bool is64Bit)
432 unsigned Size = sizeof(MachO::linker_option_command);
433 for (unsigned i = 0, e = Options.size(); i != e; ++i)
434 Size += Options[i].size() + 1;
435 return RoundUpToAlignment(Size, is64Bit ? 8 : 4);
438 void MachObjectWriter::WriteLinkerOptionsLoadCommand(
439 const std::vector<std::string> &Options)
441 unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit());
442 uint64_t Start = OS.tell();
445 Write32(MachO::LC_LINKER_OPTION);
447 Write32(Options.size());
448 uint64_t BytesWritten = sizeof(MachO::linker_option_command);
449 for (unsigned i = 0, e = Options.size(); i != e; ++i) {
450 // Write each string, including the null byte.
451 const std::string &Option = Options[i];
452 WriteBytes(Option.c_str(), Option.size() + 1);
453 BytesWritten += Option.size() + 1;
456 // Pad to a multiple of the pointer size.
457 WriteBytes("", OffsetToAlignment(BytesWritten, is64Bit() ? 8 : 4));
459 assert(OS.tell() - Start == Size);
462 void MachObjectWriter::RecordRelocation(MCAssembler &Asm,
463 const MCAsmLayout &Layout,
464 const MCFragment *Fragment,
465 const MCFixup &Fixup, MCValue Target,
466 bool &IsPCRel, uint64_t &FixedValue) {
467 TargetObjectWriter->RecordRelocation(this, Asm, Layout, Fragment, Fixup,
471 void MachObjectWriter::BindIndirectSymbols(MCAssembler &Asm) {
472 // This is the point where 'as' creates actual symbols for indirect symbols
473 // (in the following two passes). It would be easier for us to do this sooner
474 // when we see the attribute, but that makes getting the order in the symbol
475 // table much more complicated than it is worth.
477 // FIXME: Revisit this when the dust settles.
479 // Report errors for use of .indirect_symbol not in a symbol pointer section
481 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
482 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
483 const MCSectionMachO &Section =
484 cast<MCSectionMachO>(it->SectionData->getSection());
486 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
487 Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
488 Section.getType() != MachO::S_SYMBOL_STUBS) {
489 MCSymbol &Symbol = *it->Symbol;
490 report_fatal_error("indirect symbol '" + Symbol.getName() +
491 "' not in a symbol pointer or stub section");
495 // Bind non-lazy symbol pointers first.
496 unsigned IndirectIndex = 0;
497 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
498 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
499 const MCSectionMachO &Section =
500 cast<MCSectionMachO>(it->SectionData->getSection());
502 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS)
505 // Initialize the section indirect symbol base, if necessary.
506 IndirectSymBase.insert(std::make_pair(it->SectionData, IndirectIndex));
508 Asm.getOrCreateSymbolData(*it->Symbol);
511 // Then lazy symbol pointers and symbol stubs.
513 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
514 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
515 const MCSectionMachO &Section =
516 cast<MCSectionMachO>(it->SectionData->getSection());
518 if (Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
519 Section.getType() != MachO::S_SYMBOL_STUBS)
522 // Initialize the section indirect symbol base, if necessary.
523 IndirectSymBase.insert(std::make_pair(it->SectionData, IndirectIndex));
525 // Set the symbol type to undefined lazy, but only on construction.
527 // FIXME: Do not hardcode.
529 MCSymbolData &Entry = Asm.getOrCreateSymbolData(*it->Symbol, &Created);
531 Entry.setFlags(Entry.getFlags() | 0x0001);
535 /// ComputeSymbolTable - Compute the symbol table data
536 void MachObjectWriter::ComputeSymbolTable(
537 MCAssembler &Asm, std::vector<MachSymbolData> &LocalSymbolData,
538 std::vector<MachSymbolData> &ExternalSymbolData,
539 std::vector<MachSymbolData> &UndefinedSymbolData) {
540 // Build section lookup table.
541 DenseMap<const MCSection*, uint8_t> SectionIndexMap;
543 for (MCAssembler::iterator it = Asm.begin(),
544 ie = Asm.end(); it != ie; ++it, ++Index)
545 SectionIndexMap[&it->getSection()] = Index;
546 assert(Index <= 256 && "Too many sections!");
548 // Build the string table.
549 for (const MCSymbol &Symbol : Asm.symbols()) {
550 if (!Asm.isSymbolLinkerVisible(Symbol))
553 StringTable.add(Symbol.getName());
555 StringTable.finalize(StringTableBuilder::MachO);
557 // Build the symbol arrays but only for non-local symbols.
559 // The particular order that we collect and then sort the symbols is chosen to
560 // match 'as'. Even though it doesn't matter for correctness, this is
561 // important for letting us diff .o files.
562 for (const MCSymbol &Symbol : Asm.symbols()) {
563 MCSymbolData &SD = Symbol.getData();
565 // Ignore non-linker visible symbols.
566 if (!Asm.isSymbolLinkerVisible(Symbol))
569 if (!SD.isExternal() && !Symbol.isUndefined())
573 MSD.SymbolData = &SD;
574 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
576 if (Symbol.isUndefined()) {
577 MSD.SectionIndex = 0;
578 UndefinedSymbolData.push_back(MSD);
579 } else if (Symbol.isAbsolute()) {
580 MSD.SectionIndex = 0;
581 ExternalSymbolData.push_back(MSD);
583 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
584 assert(MSD.SectionIndex && "Invalid section index!");
585 ExternalSymbolData.push_back(MSD);
589 // Now add the data for local symbols.
590 for (const MCSymbol &Symbol : Asm.symbols()) {
591 MCSymbolData &SD = Symbol.getData();
593 // Ignore non-linker visible symbols.
594 if (!Asm.isSymbolLinkerVisible(Symbol))
597 if (SD.isExternal() || Symbol.isUndefined())
601 MSD.SymbolData = &SD;
602 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
604 if (Symbol.isAbsolute()) {
605 MSD.SectionIndex = 0;
606 LocalSymbolData.push_back(MSD);
608 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
609 assert(MSD.SectionIndex && "Invalid section index!");
610 LocalSymbolData.push_back(MSD);
614 // External and undefined symbols are required to be in lexicographic order.
615 std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
616 std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
618 // Set the symbol indices.
620 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
621 LocalSymbolData[i].SymbolData->setIndex(Index++);
622 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
623 ExternalSymbolData[i].SymbolData->setIndex(Index++);
624 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
625 UndefinedSymbolData[i].SymbolData->setIndex(Index++);
627 for (const MCSectionData &SD : Asm) {
628 std::vector<RelAndSymbol> &Relocs = Relocations[&SD];
629 for (RelAndSymbol &Rel : Relocs) {
633 // Set the Index and the IsExtern bit.
634 unsigned Index = Rel.Sym->getData().getIndex();
635 assert(isInt<24>(Index));
637 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & (~0U << 24)) | Index | (1 << 27);
639 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & 0xff) | Index << 8 | (1 << 4);
644 void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm,
645 const MCAsmLayout &Layout) {
646 uint64_t StartAddress = 0;
647 const SmallVectorImpl<MCSectionData*> &Order = Layout.getSectionOrder();
648 for (int i = 0, n = Order.size(); i != n ; ++i) {
649 const MCSectionData *SD = Order[i];
650 StartAddress = RoundUpToAlignment(StartAddress, SD->getAlignment());
651 SectionAddress[SD] = StartAddress;
652 StartAddress += Layout.getSectionAddressSize(SD);
654 // Explicitly pad the section to match the alignment requirements of the
655 // following one. This is for 'gas' compatibility, it shouldn't
656 /// strictly be necessary.
657 StartAddress += getPaddingSize(SD, Layout);
661 void MachObjectWriter::ExecutePostLayoutBinding(MCAssembler &Asm,
662 const MCAsmLayout &Layout) {
663 computeSectionAddresses(Asm, Layout);
665 // Create symbol data for any indirect symbols.
666 BindIndirectSymbols(Asm);
669 bool MachObjectWriter::IsSymbolRefDifferenceFullyResolvedImpl(
670 const MCAssembler &Asm, const MCSymbol &SymA, const MCFragment &FB,
671 bool InSet, bool IsPCRel) const {
675 // The effective address is
676 // addr(atom(A)) + offset(A)
677 // - addr(atom(B)) - offset(B)
678 // and the offsets are not relocatable, so the fixup is fully resolved when
679 // addr(atom(A)) - addr(atom(B)) == 0.
680 const MCSymbol &SA = findAliasedSymbol(SymA);
681 const MCSection &SecA = SA.getSection();
682 const MCSection &SecB = FB.getParent()->getSection();
685 // The simple (Darwin, except on x86_64) way of dealing with this was to
686 // assume that any reference to a temporary symbol *must* be a temporary
687 // symbol in the same atom, unless the sections differ. Therefore, any PCrel
688 // relocation to a temporary symbol (in the same section) is fully
689 // resolved. This also works in conjunction with absolutized .set, which
690 // requires the compiler to use .set to absolutize the differences between
691 // symbols which the compiler knows to be assembly time constants, so we
692 // don't need to worry about considering symbol differences fully resolved.
694 // If the file isn't using sub-sections-via-symbols, we can make the
695 // same assumptions about any symbol that we normally make about
698 bool hasReliableSymbolDifference = isX86_64();
699 if (!hasReliableSymbolDifference) {
700 if (!SA.isInSection() || &SecA != &SecB ||
701 (!SA.isTemporary() &&
702 FB.getAtom() != Asm.getSymbolData(SA).getFragment()->getAtom() &&
703 Asm.getSubsectionsViaSymbols()))
707 // For Darwin x86_64, there is one special case when the reference IsPCRel.
708 // If the fragment with the reference does not have a base symbol but meets
709 // the simple way of dealing with this, in that it is a temporary symbol in
710 // the same atom then it is assumed to be fully resolved. This is needed so
711 // a relocation entry is not created and so the static linker does not
712 // mess up the reference later.
713 else if(!FB.getAtom() &&
714 SA.isTemporary() && SA.isInSection() && &SecA == &SecB){
718 if (!TargetObjectWriter->useAggressiveSymbolFolding())
722 // If they are not in the same section, we can't compute the diff.
726 const MCFragment *FA = Asm.getSymbolData(SA).getFragment();
728 // Bail if the symbol has no fragment.
732 // If the atoms are the same, they are guaranteed to have the same address.
733 if (FA->getAtom() == FB.getAtom())
736 // Otherwise, we can't prove this is fully resolved.
740 void MachObjectWriter::WriteObject(MCAssembler &Asm,
741 const MCAsmLayout &Layout) {
742 // Compute symbol table information and bind symbol indices.
743 ComputeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData,
744 UndefinedSymbolData);
746 unsigned NumSections = Asm.size();
747 const MCAssembler::VersionMinInfoType &VersionInfo =
748 Layout.getAssembler().getVersionMinInfo();
750 // The section data starts after the header, the segment load command (and
751 // section headers) and the symbol table.
752 unsigned NumLoadCommands = 1;
753 uint64_t LoadCommandsSize = is64Bit() ?
754 sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64):
755 sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section);
757 // Add the deployment target version info load command size, if used.
758 if (VersionInfo.Major != 0) {
760 LoadCommandsSize += sizeof(MachO::version_min_command);
763 // Add the data-in-code load command size, if used.
764 unsigned NumDataRegions = Asm.getDataRegions().size();
765 if (NumDataRegions) {
767 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
770 // Add the loh load command size, if used.
771 uint64_t LOHRawSize = Asm.getLOHContainer().getEmitSize(*this, Layout);
772 uint64_t LOHSize = RoundUpToAlignment(LOHRawSize, is64Bit() ? 8 : 4);
775 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
778 // Add the symbol table load command sizes, if used.
779 unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
780 UndefinedSymbolData.size();
782 NumLoadCommands += 2;
783 LoadCommandsSize += (sizeof(MachO::symtab_command) +
784 sizeof(MachO::dysymtab_command));
787 // Add the linker option load commands sizes.
788 const std::vector<std::vector<std::string> > &LinkerOptions =
789 Asm.getLinkerOptions();
790 for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
792 LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(LinkerOptions[i],
796 // Compute the total size of the section data, as well as its file size and vm
798 uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) :
799 sizeof(MachO::mach_header)) + LoadCommandsSize;
800 uint64_t SectionDataSize = 0;
801 uint64_t SectionDataFileSize = 0;
803 for (MCAssembler::const_iterator it = Asm.begin(),
804 ie = Asm.end(); it != ie; ++it) {
805 const MCSectionData &SD = *it;
806 uint64_t Address = getSectionAddress(&SD);
807 uint64_t Size = Layout.getSectionAddressSize(&SD);
808 uint64_t FileSize = Layout.getSectionFileSize(&SD);
809 FileSize += getPaddingSize(&SD, Layout);
811 VMSize = std::max(VMSize, Address + Size);
813 if (SD.getSection().isVirtualSection())
816 SectionDataSize = std::max(SectionDataSize, Address + Size);
817 SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
820 // The section data is padded to 4 bytes.
822 // FIXME: Is this machine dependent?
823 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
824 SectionDataFileSize += SectionDataPadding;
826 // Write the prolog, starting with the header and load command...
827 WriteHeader(NumLoadCommands, LoadCommandsSize,
828 Asm.getSubsectionsViaSymbols());
829 WriteSegmentLoadCommand(NumSections, VMSize,
830 SectionDataStart, SectionDataSize);
832 // ... and then the section headers.
833 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
834 for (MCAssembler::const_iterator it = Asm.begin(),
835 ie = Asm.end(); it != ie; ++it) {
836 std::vector<RelAndSymbol> &Relocs = Relocations[it];
837 unsigned NumRelocs = Relocs.size();
838 uint64_t SectionStart = SectionDataStart + getSectionAddress(it);
839 WriteSection(Asm, Layout, *it, SectionStart, RelocTableEnd, NumRelocs);
840 RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
843 // Write out the deployment target information, if it's available.
844 if (VersionInfo.Major != 0) {
845 assert(VersionInfo.Update < 256 && "unencodable update target version");
846 assert(VersionInfo.Minor < 256 && "unencodable minor target version");
847 assert(VersionInfo.Major < 65536 && "unencodable major target version");
848 uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
849 (VersionInfo.Major << 16);
850 Write32(VersionInfo.Kind == MCVM_OSXVersionMin ? MachO::LC_VERSION_MIN_MACOSX :
851 MachO::LC_VERSION_MIN_IPHONEOS);
852 Write32(sizeof(MachO::version_min_command));
853 Write32(EncodedVersion);
854 Write32(0); // reserved.
857 // Write the data-in-code load command, if used.
858 uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
859 if (NumDataRegions) {
860 uint64_t DataRegionsOffset = RelocTableEnd;
861 uint64_t DataRegionsSize = NumDataRegions * 8;
862 WriteLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
866 // Write the loh load command, if used.
867 uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
869 WriteLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
870 DataInCodeTableEnd, LOHSize);
872 // Write the symbol table load command, if used.
874 unsigned FirstLocalSymbol = 0;
875 unsigned NumLocalSymbols = LocalSymbolData.size();
876 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
877 unsigned NumExternalSymbols = ExternalSymbolData.size();
878 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
879 unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
880 unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
881 unsigned NumSymTabSymbols =
882 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
883 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
884 uint64_t IndirectSymbolOffset = 0;
886 // If used, the indirect symbols are written after the section data.
887 if (NumIndirectSymbols)
888 IndirectSymbolOffset = LOHTableEnd;
890 // The symbol table is written after the indirect symbol data.
891 uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
893 // The string table is written after symbol table.
894 uint64_t StringTableOffset =
895 SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
896 sizeof(MachO::nlist_64) :
897 sizeof(MachO::nlist));
898 WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
899 StringTableOffset, StringTable.data().size());
901 WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
902 FirstExternalSymbol, NumExternalSymbols,
903 FirstUndefinedSymbol, NumUndefinedSymbols,
904 IndirectSymbolOffset, NumIndirectSymbols);
907 // Write the linker options load commands.
908 for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
909 WriteLinkerOptionsLoadCommand(LinkerOptions[i]);
912 // Write the actual section data.
913 for (MCAssembler::const_iterator it = Asm.begin(),
914 ie = Asm.end(); it != ie; ++it) {
915 Asm.writeSectionData(it, Layout);
917 uint64_t Pad = getPaddingSize(it, Layout);
921 // Write the extra padding.
922 WriteZeros(SectionDataPadding);
924 // Write the relocation entries.
925 for (MCAssembler::const_iterator it = Asm.begin(),
926 ie = Asm.end(); it != ie; ++it) {
927 // Write the section relocation entries, in reverse order to match 'as'
928 // (approximately, the exact algorithm is more complicated than this).
929 std::vector<RelAndSymbol> &Relocs = Relocations[it];
930 for (unsigned i = 0, e = Relocs.size(); i != e; ++i) {
931 Write32(Relocs[e - i - 1].MRE.r_word0);
932 Write32(Relocs[e - i - 1].MRE.r_word1);
936 // Write out the data-in-code region payload, if there is one.
937 for (MCAssembler::const_data_region_iterator
938 it = Asm.data_region_begin(), ie = Asm.data_region_end();
940 const DataRegionData *Data = &(*it);
941 uint64_t Start = getSymbolAddress(*Data->Start, Layout);
942 uint64_t End = getSymbolAddress(*Data->End, Layout);
943 DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
944 << " start: " << Start << "(" << Data->Start->getName() << ")"
945 << " end: " << End << "(" << Data->End->getName() << ")"
946 << " size: " << End - Start
949 Write16(End - Start);
953 // Write out the loh commands, if there is one.
956 unsigned Start = OS.tell();
958 Asm.getLOHContainer().Emit(*this, Layout);
959 // Pad to a multiple of the pointer size.
960 WriteBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
961 assert(OS.tell() - Start == LOHSize);
964 // Write the symbol table data, if used.
966 // Write the indirect symbol entries.
967 for (MCAssembler::const_indirect_symbol_iterator
968 it = Asm.indirect_symbol_begin(),
969 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
970 // Indirect symbols in the non-lazy symbol pointer section have some
972 const MCSectionMachO &Section =
973 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
974 if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
975 // If this symbol is defined and internal, mark it as such.
976 if (it->Symbol->isDefined() &&
977 !Asm.getSymbolData(*it->Symbol).isExternal()) {
978 uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
979 if (it->Symbol->isAbsolute())
980 Flags |= MachO::INDIRECT_SYMBOL_ABS;
986 Write32(Asm.getSymbolData(*it->Symbol).getIndex());
989 // FIXME: Check that offsets match computed ones.
991 // Write the symbol table entries.
992 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
993 WriteNlist(LocalSymbolData[i], Layout);
994 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
995 WriteNlist(ExternalSymbolData[i], Layout);
996 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
997 WriteNlist(UndefinedSymbolData[i], Layout);
999 // Write the string table.
1000 OS << StringTable.data();
1004 MCObjectWriter *llvm::createMachObjectWriter(MCMachObjectTargetWriter *MOTW,
1005 raw_pwrite_stream &OS,
1006 bool IsLittleEndian) {
1007 return new MachObjectWriter(MOTW, OS, IsLittleEndian);