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 =
482 cast<MCSectionMachO>(it->SectionData->getSection());
484 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
485 Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
486 Section.getType() != MachO::S_SYMBOL_STUBS) {
487 MCSymbol &Symbol = *it->Symbol;
488 report_fatal_error("indirect symbol '" + Symbol.getName() +
489 "' not in a symbol pointer or stub section");
493 // Bind non-lazy symbol pointers first.
494 unsigned IndirectIndex = 0;
495 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
496 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
497 const MCSectionMachO &Section =
498 cast<MCSectionMachO>(it->SectionData->getSection());
500 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS)
503 // Initialize the section indirect symbol base, if necessary.
504 IndirectSymBase.insert(
505 std::make_pair(&it->SectionData->getSection(), IndirectIndex));
507 Asm.getOrCreateSymbolData(*it->Symbol);
510 // Then lazy symbol pointers and symbol stubs.
512 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
513 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
514 const MCSectionMachO &Section =
515 cast<MCSectionMachO>(it->SectionData->getSection());
517 if (Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
518 Section.getType() != MachO::S_SYMBOL_STUBS)
521 // Initialize the section indirect symbol base, if necessary.
522 IndirectSymBase.insert(
523 std::make_pair(&it->SectionData->getSection(), 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] = 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.Symbol = &Symbol;
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.Symbol = &Symbol;
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].Symbol->setIndex(Index++);
622 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
623 ExternalSymbolData[i].Symbol->setIndex(Index++);
624 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
625 UndefinedSymbolData[i].Symbol->setIndex(Index++);
627 for (const MCSection &Section : Asm) {
628 std::vector<RelAndSymbol> &Relocs = Relocations[&Section];
629 for (RelAndSymbol &Rel : Relocs) {
633 // Set the Index and the IsExtern bit.
634 unsigned Index = Rel.Sym->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<MCSection *> &Order = Layout.getSectionOrder();
648 for (int i = 0, n = Order.size(); i != n ; ++i) {
649 const MCSection *Sec = Order[i];
650 StartAddress = RoundUpToAlignment(StartAddress, Sec->getAlignment());
651 SectionAddress[Sec] = StartAddress;
652 StartAddress += Layout.getSectionAddressSize(Sec);
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(Sec, 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();
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 MCSection &Sec = *it;
806 const MCSectionData &SD = it->getSectionData();
807 uint64_t Address = getSectionAddress(&Sec);
808 uint64_t Size = Layout.getSectionAddressSize(&Sec);
809 uint64_t FileSize = Layout.getSectionFileSize(&Sec);
810 FileSize += getPaddingSize(&Sec, Layout);
812 VMSize = std::max(VMSize, Address + Size);
814 if (SD.getSection().isVirtualSection())
817 SectionDataSize = std::max(SectionDataSize, Address + Size);
818 SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
821 // The section data is padded to 4 bytes.
823 // FIXME: Is this machine dependent?
824 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
825 SectionDataFileSize += SectionDataPadding;
827 // Write the prolog, starting with the header and load command...
828 WriteHeader(NumLoadCommands, LoadCommandsSize,
829 Asm.getSubsectionsViaSymbols());
830 WriteSegmentLoadCommand(NumSections, VMSize,
831 SectionDataStart, SectionDataSize);
833 // ... and then the section headers.
834 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
835 for (MCAssembler::const_iterator it = Asm.begin(),
836 ie = Asm.end(); it != ie; ++it) {
837 std::vector<RelAndSymbol> &Relocs = Relocations[&*it];
838 unsigned NumRelocs = Relocs.size();
839 uint64_t SectionStart = SectionDataStart + getSectionAddress(&*it);
840 WriteSection(Asm, Layout, *it, SectionStart, RelocTableEnd, NumRelocs);
841 RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
844 // Write out the deployment target information, if it's available.
845 if (VersionInfo.Major != 0) {
846 assert(VersionInfo.Update < 256 && "unencodable update target version");
847 assert(VersionInfo.Minor < 256 && "unencodable minor target version");
848 assert(VersionInfo.Major < 65536 && "unencodable major target version");
849 uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
850 (VersionInfo.Major << 16);
851 Write32(VersionInfo.Kind == MCVM_OSXVersionMin ? MachO::LC_VERSION_MIN_MACOSX :
852 MachO::LC_VERSION_MIN_IPHONEOS);
853 Write32(sizeof(MachO::version_min_command));
854 Write32(EncodedVersion);
855 Write32(0); // reserved.
858 // Write the data-in-code load command, if used.
859 uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
860 if (NumDataRegions) {
861 uint64_t DataRegionsOffset = RelocTableEnd;
862 uint64_t DataRegionsSize = NumDataRegions * 8;
863 WriteLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
867 // Write the loh load command, if used.
868 uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
870 WriteLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
871 DataInCodeTableEnd, LOHSize);
873 // Write the symbol table load command, if used.
875 unsigned FirstLocalSymbol = 0;
876 unsigned NumLocalSymbols = LocalSymbolData.size();
877 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
878 unsigned NumExternalSymbols = ExternalSymbolData.size();
879 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
880 unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
881 unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
882 unsigned NumSymTabSymbols =
883 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
884 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
885 uint64_t IndirectSymbolOffset = 0;
887 // If used, the indirect symbols are written after the section data.
888 if (NumIndirectSymbols)
889 IndirectSymbolOffset = LOHTableEnd;
891 // The symbol table is written after the indirect symbol data.
892 uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
894 // The string table is written after symbol table.
895 uint64_t StringTableOffset =
896 SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
897 sizeof(MachO::nlist_64) :
898 sizeof(MachO::nlist));
899 WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
900 StringTableOffset, StringTable.data().size());
902 WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
903 FirstExternalSymbol, NumExternalSymbols,
904 FirstUndefinedSymbol, NumUndefinedSymbols,
905 IndirectSymbolOffset, NumIndirectSymbols);
908 // Write the linker options load commands.
909 for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
910 WriteLinkerOptionsLoadCommand(LinkerOptions[i]);
913 // Write the actual section data.
914 for (MCAssembler::const_iterator it = Asm.begin(),
915 ie = Asm.end(); it != ie; ++it) {
916 const MCSectionData &SD = it->getSectionData();
917 Asm.writeSectionData(&SD, Layout);
919 uint64_t Pad = getPaddingSize(&*it, Layout);
923 // Write the extra padding.
924 WriteZeros(SectionDataPadding);
926 // Write the relocation entries.
927 for (MCAssembler::const_iterator it = Asm.begin(),
928 ie = Asm.end(); it != ie; ++it) {
929 // Write the section relocation entries, in reverse order to match 'as'
930 // (approximately, the exact algorithm is more complicated than this).
931 std::vector<RelAndSymbol> &Relocs = Relocations[&*it];
932 for (unsigned i = 0, e = Relocs.size(); i != e; ++i) {
933 Write32(Relocs[e - i - 1].MRE.r_word0);
934 Write32(Relocs[e - i - 1].MRE.r_word1);
938 // Write out the data-in-code region payload, if there is one.
939 for (MCAssembler::const_data_region_iterator
940 it = Asm.data_region_begin(), ie = Asm.data_region_end();
942 const DataRegionData *Data = &(*it);
943 uint64_t Start = getSymbolAddress(*Data->Start, Layout);
944 uint64_t End = getSymbolAddress(*Data->End, Layout);
945 DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
946 << " start: " << Start << "(" << Data->Start->getName() << ")"
947 << " end: " << End << "(" << Data->End->getName() << ")"
948 << " size: " << End - Start
951 Write16(End - Start);
955 // Write out the loh commands, if there is one.
958 unsigned Start = OS.tell();
960 Asm.getLOHContainer().Emit(*this, Layout);
961 // Pad to a multiple of the pointer size.
962 WriteBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
963 assert(OS.tell() - Start == LOHSize);
966 // Write the symbol table data, if used.
968 // Write the indirect symbol entries.
969 for (MCAssembler::const_indirect_symbol_iterator
970 it = Asm.indirect_symbol_begin(),
971 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
972 // Indirect symbols in the non-lazy symbol pointer section have some
974 const MCSectionMachO &Section =
975 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
976 if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
977 // If this symbol is defined and internal, mark it as such.
978 if (it->Symbol->isDefined() &&
979 !Asm.getSymbolData(*it->Symbol).isExternal()) {
980 uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
981 if (it->Symbol->isAbsolute())
982 Flags |= MachO::INDIRECT_SYMBOL_ABS;
988 Write32(it->Symbol->getIndex());
991 // FIXME: Check that offsets match computed ones.
993 // Write the symbol table entries.
994 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
995 WriteNlist(LocalSymbolData[i], Layout);
996 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
997 WriteNlist(ExternalSymbolData[i], Layout);
998 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
999 WriteNlist(UndefinedSymbolData[i], Layout);
1001 // Write the string table.
1002 OS << StringTable.data();
1006 MCObjectWriter *llvm::createMachObjectWriter(MCMachObjectTargetWriter *MOTW,
1007 raw_pwrite_stream &OS,
1008 bool IsLittleEndian) {
1009 return new MachObjectWriter(MOTW, OS, IsLittleEndian);