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/ADT/DenseMap.h"
11 #include "llvm/ADT/Twine.h"
12 #include "llvm/ADT/iterator_range.h"
13 #include "llvm/BinaryFormat/MachO.h"
14 #include "llvm/MC/MCAsmBackend.h"
15 #include "llvm/MC/MCAsmLayout.h"
16 #include "llvm/MC/MCAssembler.h"
17 #include "llvm/MC/MCDirectives.h"
18 #include "llvm/MC/MCExpr.h"
19 #include "llvm/MC/MCFixupKindInfo.h"
20 #include "llvm/MC/MCFragment.h"
21 #include "llvm/MC/MCMachObjectWriter.h"
22 #include "llvm/MC/MCObjectWriter.h"
23 #include "llvm/MC/MCSection.h"
24 #include "llvm/MC/MCSectionMachO.h"
25 #include "llvm/MC/MCSymbol.h"
26 #include "llvm/MC/MCSymbolMachO.h"
27 #include "llvm/MC/MCValue.h"
28 #include "llvm/Support/Casting.h"
29 #include "llvm/Support/Debug.h"
30 #include "llvm/Support/ErrorHandling.h"
31 #include "llvm/Support/MathExtras.h"
32 #include "llvm/Support/raw_ostream.h"
42 #define DEBUG_TYPE "mc"
44 void MachObjectWriter::reset() {
46 IndirectSymBase.clear();
48 LocalSymbolData.clear();
49 ExternalSymbolData.clear();
50 UndefinedSymbolData.clear();
51 MCObjectWriter::reset();
54 bool MachObjectWriter::doesSymbolRequireExternRelocation(const MCSymbol &S) {
55 // Undefined symbols are always extern.
59 // References to weak definitions require external relocation entries; the
60 // definition may not always be the one in the same object file.
61 if (cast<MCSymbolMachO>(S).isWeakDefinition())
64 // Otherwise, we can use an internal relocation.
68 bool MachObjectWriter::
69 MachSymbolData::operator<(const MachSymbolData &RHS) const {
70 return Symbol->getName() < RHS.Symbol->getName();
73 bool MachObjectWriter::isFixupKindPCRel(const MCAssembler &Asm, unsigned Kind) {
74 const MCFixupKindInfo &FKI = Asm.getBackend().getFixupKindInfo(
77 return FKI.Flags & MCFixupKindInfo::FKF_IsPCRel;
80 uint64_t MachObjectWriter::getFragmentAddress(const MCFragment *Fragment,
81 const MCAsmLayout &Layout) const {
82 return getSectionAddress(Fragment->getParent()) +
83 Layout.getFragmentOffset(Fragment);
86 uint64_t MachObjectWriter::getSymbolAddress(const MCSymbol &S,
87 const MCAsmLayout &Layout) const {
88 // If this is a variable, then recursively evaluate now.
90 if (const MCConstantExpr *C =
91 dyn_cast<const MCConstantExpr>(S.getVariableValue()))
95 if (!S.getVariableValue()->evaluateAsRelocatable(Target, &Layout, nullptr))
96 report_fatal_error("unable to evaluate offset for variable '" +
99 // Verify that any used symbols are defined.
100 if (Target.getSymA() && Target.getSymA()->getSymbol().isUndefined())
101 report_fatal_error("unable to evaluate offset to undefined symbol '" +
102 Target.getSymA()->getSymbol().getName() + "'");
103 if (Target.getSymB() && Target.getSymB()->getSymbol().isUndefined())
104 report_fatal_error("unable to evaluate offset to undefined symbol '" +
105 Target.getSymB()->getSymbol().getName() + "'");
107 uint64_t Address = Target.getConstant();
108 if (Target.getSymA())
109 Address += getSymbolAddress(Target.getSymA()->getSymbol(), Layout);
110 if (Target.getSymB())
111 Address += getSymbolAddress(Target.getSymB()->getSymbol(), Layout);
115 return getSectionAddress(S.getFragment()->getParent()) +
116 Layout.getSymbolOffset(S);
119 uint64_t MachObjectWriter::getPaddingSize(const MCSection *Sec,
120 const MCAsmLayout &Layout) const {
121 uint64_t EndAddr = getSectionAddress(Sec) + Layout.getSectionAddressSize(Sec);
122 unsigned Next = Sec->getLayoutOrder() + 1;
123 if (Next >= Layout.getSectionOrder().size())
126 const MCSection &NextSec = *Layout.getSectionOrder()[Next];
127 if (NextSec.isVirtualSection())
129 return OffsetToAlignment(EndAddr, NextSec.getAlignment());
132 void MachObjectWriter::writeHeader(MachO::HeaderFileType Type,
133 unsigned NumLoadCommands,
134 unsigned LoadCommandsSize,
135 bool SubsectionsViaSymbols) {
138 if (SubsectionsViaSymbols)
139 Flags |= MachO::MH_SUBSECTIONS_VIA_SYMBOLS;
141 // struct mach_header (28 bytes) or
142 // struct mach_header_64 (32 bytes)
144 uint64_t Start = W.OS.tell();
147 W.write<uint32_t>(is64Bit() ? MachO::MH_MAGIC_64 : MachO::MH_MAGIC);
149 W.write<uint32_t>(TargetObjectWriter->getCPUType());
150 W.write<uint32_t>(TargetObjectWriter->getCPUSubtype());
152 W.write<uint32_t>(Type);
153 W.write<uint32_t>(NumLoadCommands);
154 W.write<uint32_t>(LoadCommandsSize);
155 W.write<uint32_t>(Flags);
157 W.write<uint32_t>(0); // reserved
159 assert(W.OS.tell() - Start == (is64Bit() ? sizeof(MachO::mach_header_64)
160 : sizeof(MachO::mach_header)));
163 void MachObjectWriter::writeWithPadding(StringRef Str, uint64_t Size) {
164 assert(Size >= Str.size());
166 W.OS.write_zeros(Size - Str.size());
169 /// writeSegmentLoadCommand - Write a segment load command.
171 /// \param NumSections The number of sections in this segment.
172 /// \param SectionDataSize The total size of the sections.
173 void MachObjectWriter::writeSegmentLoadCommand(
174 StringRef Name, unsigned NumSections, uint64_t VMAddr, uint64_t VMSize,
175 uint64_t SectionDataStartOffset, uint64_t SectionDataSize, uint32_t MaxProt,
177 // struct segment_command (56 bytes) or
178 // struct segment_command_64 (72 bytes)
180 uint64_t Start = W.OS.tell();
183 unsigned SegmentLoadCommandSize =
184 is64Bit() ? sizeof(MachO::segment_command_64):
185 sizeof(MachO::segment_command);
186 W.write<uint32_t>(is64Bit() ? MachO::LC_SEGMENT_64 : MachO::LC_SEGMENT);
187 W.write<uint32_t>(SegmentLoadCommandSize +
188 NumSections * (is64Bit() ? sizeof(MachO::section_64) :
189 sizeof(MachO::section)));
191 writeWithPadding(Name, 16);
193 W.write<uint64_t>(VMAddr); // vmaddr
194 W.write<uint64_t>(VMSize); // vmsize
195 W.write<uint64_t>(SectionDataStartOffset); // file offset
196 W.write<uint64_t>(SectionDataSize); // file size
198 W.write<uint32_t>(VMAddr); // vmaddr
199 W.write<uint32_t>(VMSize); // vmsize
200 W.write<uint32_t>(SectionDataStartOffset); // file offset
201 W.write<uint32_t>(SectionDataSize); // file size
204 W.write<uint32_t>(MaxProt);
206 W.write<uint32_t>(InitProt);
207 W.write<uint32_t>(NumSections);
208 W.write<uint32_t>(0); // flags
210 assert(W.OS.tell() - Start == SegmentLoadCommandSize);
213 void MachObjectWriter::writeSection(const MCAsmLayout &Layout,
214 const MCSection &Sec, uint64_t VMAddr,
215 uint64_t FileOffset, unsigned Flags,
216 uint64_t RelocationsStart,
217 unsigned NumRelocations) {
218 uint64_t SectionSize = Layout.getSectionAddressSize(&Sec);
219 const MCSectionMachO &Section = cast<MCSectionMachO>(Sec);
221 // The offset is unused for virtual sections.
222 if (Section.isVirtualSection()) {
223 assert(Layout.getSectionFileSize(&Sec) == 0 && "Invalid file size!");
227 // struct section (68 bytes) or
228 // struct section_64 (80 bytes)
230 uint64_t Start = W.OS.tell();
233 writeWithPadding(Section.getSectionName(), 16);
234 writeWithPadding(Section.getSegmentName(), 16);
236 W.write<uint64_t>(VMAddr); // address
237 W.write<uint64_t>(SectionSize); // size
239 W.write<uint32_t>(VMAddr); // address
240 W.write<uint32_t>(SectionSize); // size
242 W.write<uint32_t>(FileOffset);
244 assert(isPowerOf2_32(Section.getAlignment()) && "Invalid alignment!");
245 W.write<uint32_t>(Log2_32(Section.getAlignment()));
246 W.write<uint32_t>(NumRelocations ? RelocationsStart : 0);
247 W.write<uint32_t>(NumRelocations);
248 W.write<uint32_t>(Flags);
249 W.write<uint32_t>(IndirectSymBase.lookup(&Sec)); // reserved1
250 W.write<uint32_t>(Section.getStubSize()); // reserved2
252 W.write<uint32_t>(0); // reserved3
254 assert(W.OS.tell() - Start ==
255 (is64Bit() ? sizeof(MachO::section_64) : sizeof(MachO::section)));
258 void MachObjectWriter::writeSymtabLoadCommand(uint32_t SymbolOffset,
260 uint32_t StringTableOffset,
261 uint32_t StringTableSize) {
262 // struct symtab_command (24 bytes)
264 uint64_t Start = W.OS.tell();
267 W.write<uint32_t>(MachO::LC_SYMTAB);
268 W.write<uint32_t>(sizeof(MachO::symtab_command));
269 W.write<uint32_t>(SymbolOffset);
270 W.write<uint32_t>(NumSymbols);
271 W.write<uint32_t>(StringTableOffset);
272 W.write<uint32_t>(StringTableSize);
274 assert(W.OS.tell() - Start == sizeof(MachO::symtab_command));
277 void MachObjectWriter::writeDysymtabLoadCommand(uint32_t FirstLocalSymbol,
278 uint32_t NumLocalSymbols,
279 uint32_t FirstExternalSymbol,
280 uint32_t NumExternalSymbols,
281 uint32_t FirstUndefinedSymbol,
282 uint32_t NumUndefinedSymbols,
283 uint32_t IndirectSymbolOffset,
284 uint32_t NumIndirectSymbols) {
285 // struct dysymtab_command (80 bytes)
287 uint64_t Start = W.OS.tell();
290 W.write<uint32_t>(MachO::LC_DYSYMTAB);
291 W.write<uint32_t>(sizeof(MachO::dysymtab_command));
292 W.write<uint32_t>(FirstLocalSymbol);
293 W.write<uint32_t>(NumLocalSymbols);
294 W.write<uint32_t>(FirstExternalSymbol);
295 W.write<uint32_t>(NumExternalSymbols);
296 W.write<uint32_t>(FirstUndefinedSymbol);
297 W.write<uint32_t>(NumUndefinedSymbols);
298 W.write<uint32_t>(0); // tocoff
299 W.write<uint32_t>(0); // ntoc
300 W.write<uint32_t>(0); // modtaboff
301 W.write<uint32_t>(0); // nmodtab
302 W.write<uint32_t>(0); // extrefsymoff
303 W.write<uint32_t>(0); // nextrefsyms
304 W.write<uint32_t>(IndirectSymbolOffset);
305 W.write<uint32_t>(NumIndirectSymbols);
306 W.write<uint32_t>(0); // extreloff
307 W.write<uint32_t>(0); // nextrel
308 W.write<uint32_t>(0); // locreloff
309 W.write<uint32_t>(0); // nlocrel
311 assert(W.OS.tell() - Start == sizeof(MachO::dysymtab_command));
314 MachObjectWriter::MachSymbolData *
315 MachObjectWriter::findSymbolData(const MCSymbol &Sym) {
316 for (auto *SymbolData :
317 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
318 for (MachSymbolData &Entry : *SymbolData)
319 if (Entry.Symbol == &Sym)
325 const MCSymbol &MachObjectWriter::findAliasedSymbol(const MCSymbol &Sym) const {
326 const MCSymbol *S = &Sym;
327 while (S->isVariable()) {
328 const MCExpr *Value = S->getVariableValue();
329 const auto *Ref = dyn_cast<MCSymbolRefExpr>(Value);
332 S = &Ref->getSymbol();
337 void MachObjectWriter::writeNlist(MachSymbolData &MSD,
338 const MCAsmLayout &Layout) {
339 const MCSymbol *Symbol = MSD.Symbol;
340 const MCSymbol &Data = *Symbol;
341 const MCSymbol *AliasedSymbol = &findAliasedSymbol(*Symbol);
342 uint8_t SectionIndex = MSD.SectionIndex;
344 uint64_t Address = 0;
345 bool IsAlias = Symbol != AliasedSymbol;
347 const MCSymbol &OrigSymbol = *Symbol;
348 MachSymbolData *AliaseeInfo;
350 AliaseeInfo = findSymbolData(*AliasedSymbol);
352 SectionIndex = AliaseeInfo->SectionIndex;
353 Symbol = AliasedSymbol;
354 // FIXME: Should this update Data as well?
357 // Set the N_TYPE bits. See <mach-o/nlist.h>.
359 // FIXME: Are the prebound or indirect fields possible here?
360 if (IsAlias && Symbol->isUndefined())
361 Type = MachO::N_INDR;
362 else if (Symbol->isUndefined())
363 Type = MachO::N_UNDF;
364 else if (Symbol->isAbsolute())
367 Type = MachO::N_SECT;
369 // FIXME: Set STAB bits.
371 if (Data.isPrivateExtern())
372 Type |= MachO::N_PEXT;
375 if (Data.isExternal() || (!IsAlias && Symbol->isUndefined()))
376 Type |= MachO::N_EXT;
378 // Compute the symbol address.
379 if (IsAlias && Symbol->isUndefined())
380 Address = AliaseeInfo->StringIndex;
381 else if (Symbol->isDefined())
382 Address = getSymbolAddress(OrigSymbol, Layout);
383 else if (Symbol->isCommon()) {
384 // Common symbols are encoded with the size in the address
385 // field, and their alignment in the flags.
386 Address = Symbol->getCommonSize();
389 // struct nlist (12 bytes)
391 W.write<uint32_t>(MSD.StringIndex);
393 W.OS << char(SectionIndex);
395 // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc'
397 bool EncodeAsAltEntry =
398 IsAlias && cast<MCSymbolMachO>(OrigSymbol).isAltEntry();
399 W.write<uint16_t>(cast<MCSymbolMachO>(Symbol)->getEncodedFlags(EncodeAsAltEntry));
401 W.write<uint64_t>(Address);
403 W.write<uint32_t>(Address);
406 void MachObjectWriter::writeLinkeditLoadCommand(uint32_t Type,
409 uint64_t Start = W.OS.tell();
412 W.write<uint32_t>(Type);
413 W.write<uint32_t>(sizeof(MachO::linkedit_data_command));
414 W.write<uint32_t>(DataOffset);
415 W.write<uint32_t>(DataSize);
417 assert(W.OS.tell() - Start == sizeof(MachO::linkedit_data_command));
420 static unsigned ComputeLinkerOptionsLoadCommandSize(
421 const std::vector<std::string> &Options, bool is64Bit)
423 unsigned Size = sizeof(MachO::linker_option_command);
424 for (const std::string &Option : Options)
425 Size += Option.size() + 1;
426 return alignTo(Size, is64Bit ? 8 : 4);
429 void MachObjectWriter::writeLinkerOptionsLoadCommand(
430 const std::vector<std::string> &Options)
432 unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit());
433 uint64_t Start = W.OS.tell();
436 W.write<uint32_t>(MachO::LC_LINKER_OPTION);
437 W.write<uint32_t>(Size);
438 W.write<uint32_t>(Options.size());
439 uint64_t BytesWritten = sizeof(MachO::linker_option_command);
440 for (const std::string &Option : Options) {
441 // Write each string, including the null byte.
442 W.OS << Option << '\0';
443 BytesWritten += Option.size() + 1;
446 // Pad to a multiple of the pointer size.
447 W.OS.write_zeros(OffsetToAlignment(BytesWritten, is64Bit() ? 8 : 4));
449 assert(W.OS.tell() - Start == Size);
452 void MachObjectWriter::recordRelocation(MCAssembler &Asm,
453 const MCAsmLayout &Layout,
454 const MCFragment *Fragment,
455 const MCFixup &Fixup, MCValue Target,
456 uint64_t &FixedValue) {
457 TargetObjectWriter->recordRelocation(this, Asm, Layout, Fragment, Fixup,
461 void MachObjectWriter::bindIndirectSymbols(MCAssembler &Asm) {
462 // This is the point where 'as' creates actual symbols for indirect symbols
463 // (in the following two passes). It would be easier for us to do this sooner
464 // when we see the attribute, but that makes getting the order in the symbol
465 // table much more complicated than it is worth.
467 // FIXME: Revisit this when the dust settles.
469 // Report errors for use of .indirect_symbol not in a symbol pointer section
471 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
472 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
473 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
475 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
476 Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
477 Section.getType() != MachO::S_THREAD_LOCAL_VARIABLE_POINTERS &&
478 Section.getType() != MachO::S_SYMBOL_STUBS) {
479 MCSymbol &Symbol = *it->Symbol;
480 report_fatal_error("indirect symbol '" + Symbol.getName() +
481 "' not in a symbol pointer or stub section");
485 // Bind non-lazy symbol pointers first.
486 unsigned IndirectIndex = 0;
487 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
488 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
489 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
491 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
492 Section.getType() != MachO::S_THREAD_LOCAL_VARIABLE_POINTERS)
495 // Initialize the section indirect symbol base, if necessary.
496 IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex));
498 Asm.registerSymbol(*it->Symbol);
501 // Then lazy symbol pointers and symbol stubs.
503 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
504 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
505 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section);
507 if (Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
508 Section.getType() != MachO::S_SYMBOL_STUBS)
511 // Initialize the section indirect symbol base, if necessary.
512 IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex));
514 // Set the symbol type to undefined lazy, but only on construction.
516 // FIXME: Do not hardcode.
518 Asm.registerSymbol(*it->Symbol, &Created);
520 cast<MCSymbolMachO>(it->Symbol)->setReferenceTypeUndefinedLazy(true);
524 /// computeSymbolTable - Compute the symbol table data
525 void MachObjectWriter::computeSymbolTable(
526 MCAssembler &Asm, std::vector<MachSymbolData> &LocalSymbolData,
527 std::vector<MachSymbolData> &ExternalSymbolData,
528 std::vector<MachSymbolData> &UndefinedSymbolData) {
529 // Build section lookup table.
530 DenseMap<const MCSection*, uint8_t> SectionIndexMap;
532 for (MCAssembler::iterator it = Asm.begin(),
533 ie = Asm.end(); it != ie; ++it, ++Index)
534 SectionIndexMap[&*it] = Index;
535 assert(Index <= 256 && "Too many sections!");
537 // Build the string table.
538 for (const MCSymbol &Symbol : Asm.symbols()) {
539 if (!Asm.isSymbolLinkerVisible(Symbol))
542 StringTable.add(Symbol.getName());
544 StringTable.finalize();
546 // Build the symbol arrays but only for non-local symbols.
548 // The particular order that we collect and then sort the symbols is chosen to
549 // match 'as'. Even though it doesn't matter for correctness, this is
550 // important for letting us diff .o files.
551 for (const MCSymbol &Symbol : Asm.symbols()) {
552 // Ignore non-linker visible symbols.
553 if (!Asm.isSymbolLinkerVisible(Symbol))
556 if (!Symbol.isExternal() && !Symbol.isUndefined())
560 MSD.Symbol = &Symbol;
561 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
563 if (Symbol.isUndefined()) {
564 MSD.SectionIndex = 0;
565 UndefinedSymbolData.push_back(MSD);
566 } else if (Symbol.isAbsolute()) {
567 MSD.SectionIndex = 0;
568 ExternalSymbolData.push_back(MSD);
570 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
571 assert(MSD.SectionIndex && "Invalid section index!");
572 ExternalSymbolData.push_back(MSD);
576 // Now add the data for local symbols.
577 for (const MCSymbol &Symbol : Asm.symbols()) {
578 // Ignore non-linker visible symbols.
579 if (!Asm.isSymbolLinkerVisible(Symbol))
582 if (Symbol.isExternal() || Symbol.isUndefined())
586 MSD.Symbol = &Symbol;
587 MSD.StringIndex = StringTable.getOffset(Symbol.getName());
589 if (Symbol.isAbsolute()) {
590 MSD.SectionIndex = 0;
591 LocalSymbolData.push_back(MSD);
593 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
594 assert(MSD.SectionIndex && "Invalid section index!");
595 LocalSymbolData.push_back(MSD);
599 // External and undefined symbols are required to be in lexicographic order.
600 llvm::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
601 llvm::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
603 // Set the symbol indices.
605 for (auto *SymbolData :
606 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
607 for (MachSymbolData &Entry : *SymbolData)
608 Entry.Symbol->setIndex(Index++);
610 for (const MCSection &Section : Asm) {
611 for (RelAndSymbol &Rel : Relocations[&Section]) {
615 // Set the Index and the IsExtern bit.
616 unsigned Index = Rel.Sym->getIndex();
617 assert(isInt<24>(Index));
618 if (W.Endian == support::little)
619 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & (~0U << 24)) | Index | (1 << 27);
621 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & 0xff) | Index << 8 | (1 << 4);
626 void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm,
627 const MCAsmLayout &Layout) {
628 uint64_t StartAddress = 0;
629 for (const MCSection *Sec : Layout.getSectionOrder()) {
630 StartAddress = alignTo(StartAddress, Sec->getAlignment());
631 SectionAddress[Sec] = StartAddress;
632 StartAddress += Layout.getSectionAddressSize(Sec);
634 // Explicitly pad the section to match the alignment requirements of the
635 // following one. This is for 'gas' compatibility, it shouldn't
636 /// strictly be necessary.
637 StartAddress += getPaddingSize(Sec, Layout);
641 void MachObjectWriter::executePostLayoutBinding(MCAssembler &Asm,
642 const MCAsmLayout &Layout) {
643 computeSectionAddresses(Asm, Layout);
645 // Create symbol data for any indirect symbols.
646 bindIndirectSymbols(Asm);
649 bool MachObjectWriter::isSymbolRefDifferenceFullyResolvedImpl(
650 const MCAssembler &Asm, const MCSymbol &A, const MCSymbol &B,
652 // FIXME: We don't handle things like
655 if (A.isVariable() || B.isVariable())
657 return MCObjectWriter::isSymbolRefDifferenceFullyResolvedImpl(Asm, A, B,
661 bool MachObjectWriter::isSymbolRefDifferenceFullyResolvedImpl(
662 const MCAssembler &Asm, const MCSymbol &SymA, const MCFragment &FB,
663 bool InSet, bool IsPCRel) const {
667 // The effective address is
668 // addr(atom(A)) + offset(A)
669 // - addr(atom(B)) - offset(B)
670 // and the offsets are not relocatable, so the fixup is fully resolved when
671 // addr(atom(A)) - addr(atom(B)) == 0.
672 const MCSymbol &SA = findAliasedSymbol(SymA);
673 const MCSection &SecA = SA.getSection();
674 const MCSection &SecB = *FB.getParent();
677 // The simple (Darwin, except on x86_64) way of dealing with this was to
678 // assume that any reference to a temporary symbol *must* be a temporary
679 // symbol in the same atom, unless the sections differ. Therefore, any PCrel
680 // relocation to a temporary symbol (in the same section) is fully
681 // resolved. This also works in conjunction with absolutized .set, which
682 // requires the compiler to use .set to absolutize the differences between
683 // symbols which the compiler knows to be assembly time constants, so we
684 // don't need to worry about considering symbol differences fully resolved.
686 // If the file isn't using sub-sections-via-symbols, we can make the
687 // same assumptions about any symbol that we normally make about
690 bool hasReliableSymbolDifference = isX86_64();
691 if (!hasReliableSymbolDifference) {
692 if (!SA.isInSection() || &SecA != &SecB ||
693 (!SA.isTemporary() && FB.getAtom() != SA.getFragment()->getAtom() &&
694 Asm.getSubsectionsViaSymbols()))
698 // For Darwin x86_64, there is one special case when the reference IsPCRel.
699 // If the fragment with the reference does not have a base symbol but meets
700 // the simple way of dealing with this, in that it is a temporary symbol in
701 // the same atom then it is assumed to be fully resolved. This is needed so
702 // a relocation entry is not created and so the static linker does not
703 // mess up the reference later.
704 else if(!FB.getAtom() &&
705 SA.isTemporary() && SA.isInSection() && &SecA == &SecB){
710 // If they are not in the same section, we can't compute the diff.
714 const MCFragment *FA = SA.getFragment();
716 // Bail if the symbol has no fragment.
720 // If the atoms are the same, they are guaranteed to have the same address.
721 if (FA->getAtom() == FB.getAtom())
724 // Otherwise, we can't prove this is fully resolved.
728 static MachO::LoadCommandType getLCFromMCVM(MCVersionMinType Type) {
730 case MCVM_OSXVersionMin: return MachO::LC_VERSION_MIN_MACOSX;
731 case MCVM_IOSVersionMin: return MachO::LC_VERSION_MIN_IPHONEOS;
732 case MCVM_TvOSVersionMin: return MachO::LC_VERSION_MIN_TVOS;
733 case MCVM_WatchOSVersionMin: return MachO::LC_VERSION_MIN_WATCHOS;
735 llvm_unreachable("Invalid mc version min type");
738 uint64_t MachObjectWriter::writeObject(MCAssembler &Asm,
739 const MCAsmLayout &Layout) {
740 uint64_t StartOffset = W.OS.tell();
742 // Compute symbol table information and bind symbol indices.
743 computeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData,
744 UndefinedSymbolData);
746 unsigned NumSections = Asm.size();
747 const MCAssembler::VersionInfoType &VersionInfo =
748 Layout.getAssembler().getVersionInfo();
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 if (VersionInfo.EmitBuildVersion)
761 LoadCommandsSize += sizeof(MachO::build_version_command);
763 LoadCommandsSize += sizeof(MachO::version_min_command);
766 // Add the data-in-code load command size, if used.
767 unsigned NumDataRegions = Asm.getDataRegions().size();
768 if (NumDataRegions) {
770 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
773 // Add the loh load command size, if used.
774 uint64_t LOHRawSize = Asm.getLOHContainer().getEmitSize(*this, Layout);
775 uint64_t LOHSize = alignTo(LOHRawSize, is64Bit() ? 8 : 4);
778 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
781 // Add the symbol table load command sizes, if used.
782 unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
783 UndefinedSymbolData.size();
785 NumLoadCommands += 2;
786 LoadCommandsSize += (sizeof(MachO::symtab_command) +
787 sizeof(MachO::dysymtab_command));
790 // Add the linker option load commands sizes.
791 for (const auto &Option : Asm.getLinkerOptions()) {
793 LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(Option, is64Bit());
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 (const MCSection &Sec : Asm) {
804 uint64_t Address = getSectionAddress(&Sec);
805 uint64_t Size = Layout.getSectionAddressSize(&Sec);
806 uint64_t FileSize = Layout.getSectionFileSize(&Sec);
807 FileSize += getPaddingSize(&Sec, Layout);
809 VMSize = std::max(VMSize, Address + Size);
811 if (Sec.isVirtualSection())
814 SectionDataSize = std::max(SectionDataSize, Address + Size);
815 SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
818 // The section data is padded to 4 bytes.
820 // FIXME: Is this machine dependent?
821 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
822 SectionDataFileSize += SectionDataPadding;
824 // Write the prolog, starting with the header and load command...
825 writeHeader(MachO::MH_OBJECT, NumLoadCommands, LoadCommandsSize,
826 Asm.getSubsectionsViaSymbols());
828 MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE;
829 writeSegmentLoadCommand("", NumSections, 0, VMSize, SectionDataStart,
830 SectionDataSize, Prot, Prot);
832 // ... and then the section headers.
833 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
834 for (const MCSection &Section : Asm) {
835 const auto &Sec = cast<MCSectionMachO>(Section);
836 std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
837 unsigned NumRelocs = Relocs.size();
838 uint64_t SectionStart = SectionDataStart + getSectionAddress(&Sec);
839 unsigned Flags = Sec.getTypeAndAttributes();
840 if (Sec.hasInstructions())
841 Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS;
842 writeSection(Layout, Sec, getSectionAddress(&Sec), SectionStart, Flags,
843 RelocTableEnd, NumRelocs);
844 RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
847 // Write out the deployment target information, if it's available.
848 if (VersionInfo.Major != 0) {
849 assert(VersionInfo.Update < 256 && "unencodable update target version");
850 assert(VersionInfo.Minor < 256 && "unencodable minor target version");
851 assert(VersionInfo.Major < 65536 && "unencodable major target version");
852 uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
853 (VersionInfo.Major << 16);
854 if (VersionInfo.EmitBuildVersion) {
855 // FIXME: Currently empty tools. Add clang version in the future.
856 W.write<uint32_t>(MachO::LC_BUILD_VERSION);
857 W.write<uint32_t>(sizeof(MachO::build_version_command));
858 W.write<uint32_t>(VersionInfo.TypeOrPlatform.Platform);
859 W.write<uint32_t>(EncodedVersion);
860 W.write<uint32_t>(0); // SDK version.
861 W.write<uint32_t>(0); // Empty tools list.
863 MachO::LoadCommandType LCType
864 = getLCFromMCVM(VersionInfo.TypeOrPlatform.Type);
865 W.write<uint32_t>(LCType);
866 W.write<uint32_t>(sizeof(MachO::version_min_command));
867 W.write<uint32_t>(EncodedVersion);
868 W.write<uint32_t>(0); // reserved.
872 // Write the data-in-code load command, if used.
873 uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
874 if (NumDataRegions) {
875 uint64_t DataRegionsOffset = RelocTableEnd;
876 uint64_t DataRegionsSize = NumDataRegions * 8;
877 writeLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
881 // Write the loh load command, if used.
882 uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
884 writeLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
885 DataInCodeTableEnd, LOHSize);
887 // Write the symbol table load command, if used.
889 unsigned FirstLocalSymbol = 0;
890 unsigned NumLocalSymbols = LocalSymbolData.size();
891 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
892 unsigned NumExternalSymbols = ExternalSymbolData.size();
893 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
894 unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
895 unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
896 unsigned NumSymTabSymbols =
897 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
898 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
899 uint64_t IndirectSymbolOffset = 0;
901 // If used, the indirect symbols are written after the section data.
902 if (NumIndirectSymbols)
903 IndirectSymbolOffset = LOHTableEnd;
905 // The symbol table is written after the indirect symbol data.
906 uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
908 // The string table is written after symbol table.
909 uint64_t StringTableOffset =
910 SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
911 sizeof(MachO::nlist_64) :
912 sizeof(MachO::nlist));
913 writeSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
914 StringTableOffset, StringTable.getSize());
916 writeDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
917 FirstExternalSymbol, NumExternalSymbols,
918 FirstUndefinedSymbol, NumUndefinedSymbols,
919 IndirectSymbolOffset, NumIndirectSymbols);
922 // Write the linker options load commands.
923 for (const auto &Option : Asm.getLinkerOptions())
924 writeLinkerOptionsLoadCommand(Option);
926 // Write the actual section data.
927 for (const MCSection &Sec : Asm) {
928 Asm.writeSectionData(W.OS, &Sec, Layout);
930 uint64_t Pad = getPaddingSize(&Sec, Layout);
931 W.OS.write_zeros(Pad);
934 // Write the extra padding.
935 W.OS.write_zeros(SectionDataPadding);
937 // Write the relocation entries.
938 for (const MCSection &Sec : Asm) {
939 // Write the section relocation entries, in reverse order to match 'as'
940 // (approximately, the exact algorithm is more complicated than this).
941 std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
942 for (const RelAndSymbol &Rel : make_range(Relocs.rbegin(), Relocs.rend())) {
943 W.write<uint32_t>(Rel.MRE.r_word0);
944 W.write<uint32_t>(Rel.MRE.r_word1);
948 // Write out the data-in-code region payload, if there is one.
949 for (MCAssembler::const_data_region_iterator
950 it = Asm.data_region_begin(), ie = Asm.data_region_end();
952 const DataRegionData *Data = &(*it);
953 uint64_t Start = getSymbolAddress(*Data->Start, Layout);
956 End = getSymbolAddress(*Data->End, Layout);
958 report_fatal_error("Data region not terminated");
960 LLVM_DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
961 << " start: " << Start << "(" << Data->Start->getName()
963 << " end: " << End << "(" << Data->End->getName() << ")"
964 << " size: " << End - Start << "\n");
965 W.write<uint32_t>(Start);
966 W.write<uint16_t>(End - Start);
967 W.write<uint16_t>(Data->Kind);
970 // Write out the loh commands, if there is one.
973 unsigned Start = W.OS.tell();
975 Asm.getLOHContainer().emit(*this, Layout);
976 // Pad to a multiple of the pointer size.
977 W.OS.write_zeros(OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
978 assert(W.OS.tell() - Start == LOHSize);
981 // Write the symbol table data, if used.
983 // Write the indirect symbol entries.
984 for (MCAssembler::const_indirect_symbol_iterator
985 it = Asm.indirect_symbol_begin(),
986 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
987 // Indirect symbols in the non-lazy symbol pointer section have some
989 const MCSectionMachO &Section =
990 static_cast<const MCSectionMachO &>(*it->Section);
991 if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
992 // If this symbol is defined and internal, mark it as such.
993 if (it->Symbol->isDefined() && !it->Symbol->isExternal()) {
994 uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
995 if (it->Symbol->isAbsolute())
996 Flags |= MachO::INDIRECT_SYMBOL_ABS;
997 W.write<uint32_t>(Flags);
1002 W.write<uint32_t>(it->Symbol->getIndex());
1005 // FIXME: Check that offsets match computed ones.
1007 // Write the symbol table entries.
1008 for (auto *SymbolData :
1009 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
1010 for (MachSymbolData &Entry : *SymbolData)
1011 writeNlist(Entry, Layout);
1013 // Write the string table.
1014 StringTable.write(W.OS);
1017 return W.OS.tell() - StartOffset;
1020 std::unique_ptr<MCObjectWriter>
1021 llvm::createMachObjectWriter(std::unique_ptr<MCMachObjectTargetWriter> MOTW,
1022 raw_pwrite_stream &OS, bool IsLittleEndian) {
1023 return llvm::make_unique<MachObjectWriter>(std::move(MOTW), OS,