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"
29 void MachObjectWriter::reset() {
31 IndirectSymBase.clear();
33 LocalSymbolData.clear();
34 ExternalSymbolData.clear();
35 UndefinedSymbolData.clear();
36 MCObjectWriter::reset();
39 bool MachObjectWriter::
40 doesSymbolRequireExternRelocation(const MCSymbolData *SD) {
41 // Undefined symbols are always extern.
42 if (SD->Symbol->isUndefined())
45 // References to weak definitions require external relocation entries; the
46 // definition may not always be the one in the same object file.
47 if (SD->getFlags() & SF_WeakDefinition)
50 // Otherwise, we can use an internal relocation.
54 bool MachObjectWriter::
55 MachSymbolData::operator<(const MachSymbolData &RHS) const {
56 return SymbolData->getSymbol().getName() <
57 RHS.SymbolData->getSymbol().getName();
60 bool MachObjectWriter::isFixupKindPCRel(const MCAssembler &Asm, unsigned Kind) {
61 const MCFixupKindInfo &FKI = Asm.getBackend().getFixupKindInfo(
64 return FKI.Flags & MCFixupKindInfo::FKF_IsPCRel;
67 uint64_t MachObjectWriter::getFragmentAddress(const MCFragment *Fragment,
68 const MCAsmLayout &Layout) const {
69 return getSectionAddress(Fragment->getParent()) +
70 Layout.getFragmentOffset(Fragment);
73 uint64_t MachObjectWriter::getSymbolAddress(const MCSymbolData* SD,
74 const MCAsmLayout &Layout) const {
75 const MCSymbol &S = SD->getSymbol();
77 // If this is a variable, then recursively evaluate now.
79 if (const MCConstantExpr *C =
80 dyn_cast<const MCConstantExpr>(S.getVariableValue()))
85 if (!S.getVariableValue()->EvaluateAsRelocatable(Target, &Layout))
86 report_fatal_error("unable to evaluate offset for variable '" +
89 // Verify that any used symbols are defined.
90 if (Target.getSymA() && Target.getSymA()->getSymbol().isUndefined())
91 report_fatal_error("unable to evaluate offset to undefined symbol '" +
92 Target.getSymA()->getSymbol().getName() + "'");
93 if (Target.getSymB() && Target.getSymB()->getSymbol().isUndefined())
94 report_fatal_error("unable to evaluate offset to undefined symbol '" +
95 Target.getSymB()->getSymbol().getName() + "'");
97 uint64_t Address = Target.getConstant();
99 Address += getSymbolAddress(&Layout.getAssembler().getSymbolData(
100 Target.getSymA()->getSymbol()), Layout);
101 if (Target.getSymB())
102 Address += getSymbolAddress(&Layout.getAssembler().getSymbolData(
103 Target.getSymB()->getSymbol()), Layout);
107 return getSectionAddress(SD->getFragment()->getParent()) +
108 Layout.getSymbolOffset(SD);
111 uint64_t MachObjectWriter::getPaddingSize(const MCSectionData *SD,
112 const MCAsmLayout &Layout) const {
113 uint64_t EndAddr = getSectionAddress(SD) + Layout.getSectionAddressSize(SD);
114 unsigned Next = SD->getLayoutOrder() + 1;
115 if (Next >= Layout.getSectionOrder().size())
118 const MCSectionData &NextSD = *Layout.getSectionOrder()[Next];
119 if (NextSD.getSection().isVirtualSection())
121 return OffsetToAlignment(EndAddr, NextSD.getAlignment());
124 void MachObjectWriter::WriteHeader(unsigned NumLoadCommands,
125 unsigned LoadCommandsSize,
126 bool SubsectionsViaSymbols) {
129 if (SubsectionsViaSymbols)
130 Flags |= MachO::MH_SUBSECTIONS_VIA_SYMBOLS;
132 // struct mach_header (28 bytes) or
133 // struct mach_header_64 (32 bytes)
135 uint64_t Start = OS.tell();
138 Write32(is64Bit() ? MachO::MH_MAGIC_64 : MachO::MH_MAGIC);
140 Write32(TargetObjectWriter->getCPUType());
141 Write32(TargetObjectWriter->getCPUSubtype());
143 Write32(MachO::MH_OBJECT);
144 Write32(NumLoadCommands);
145 Write32(LoadCommandsSize);
148 Write32(0); // reserved
150 assert(OS.tell() - Start ==
151 (is64Bit()?sizeof(MachO::mach_header_64): sizeof(MachO::mach_header)));
154 /// WriteSegmentLoadCommand - Write a segment load command.
156 /// \param NumSections The number of sections in this segment.
157 /// \param SectionDataSize The total size of the sections.
158 void MachObjectWriter::WriteSegmentLoadCommand(unsigned NumSections,
160 uint64_t SectionDataStartOffset,
161 uint64_t SectionDataSize) {
162 // struct segment_command (56 bytes) or
163 // struct segment_command_64 (72 bytes)
165 uint64_t Start = OS.tell();
168 unsigned SegmentLoadCommandSize =
169 is64Bit() ? sizeof(MachO::segment_command_64):
170 sizeof(MachO::segment_command);
171 Write32(is64Bit() ? MachO::LC_SEGMENT_64 : MachO::LC_SEGMENT);
172 Write32(SegmentLoadCommandSize +
173 NumSections * (is64Bit() ? sizeof(MachO::section_64) :
174 sizeof(MachO::section)));
178 Write64(0); // vmaddr
179 Write64(VMSize); // vmsize
180 Write64(SectionDataStartOffset); // file offset
181 Write64(SectionDataSize); // file size
183 Write32(0); // vmaddr
184 Write32(VMSize); // vmsize
185 Write32(SectionDataStartOffset); // file offset
186 Write32(SectionDataSize); // file size
189 Write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE);
191 Write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE);
192 Write32(NumSections);
195 assert(OS.tell() - Start == SegmentLoadCommandSize);
198 void MachObjectWriter::WriteSection(const MCAssembler &Asm,
199 const MCAsmLayout &Layout,
200 const MCSectionData &SD,
202 uint64_t RelocationsStart,
203 unsigned NumRelocations) {
204 uint64_t SectionSize = Layout.getSectionAddressSize(&SD);
206 // The offset is unused for virtual sections.
207 if (SD.getSection().isVirtualSection()) {
208 assert(Layout.getSectionFileSize(&SD) == 0 && "Invalid file size!");
212 // struct section (68 bytes) or
213 // struct section_64 (80 bytes)
215 uint64_t Start = OS.tell();
218 const MCSectionMachO &Section = cast<MCSectionMachO>(SD.getSection());
219 WriteBytes(Section.getSectionName(), 16);
220 WriteBytes(Section.getSegmentName(), 16);
222 Write64(getSectionAddress(&SD)); // address
223 Write64(SectionSize); // size
225 Write32(getSectionAddress(&SD)); // address
226 Write32(SectionSize); // size
230 unsigned Flags = Section.getTypeAndAttributes();
231 if (SD.hasInstructions())
232 Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS;
234 assert(isPowerOf2_32(SD.getAlignment()) && "Invalid alignment!");
235 Write32(Log2_32(SD.getAlignment()));
236 Write32(NumRelocations ? RelocationsStart : 0);
237 Write32(NumRelocations);
239 Write32(IndirectSymBase.lookup(&SD)); // reserved1
240 Write32(Section.getStubSize()); // reserved2
242 Write32(0); // reserved3
244 assert(OS.tell() - Start == (is64Bit() ? sizeof(MachO::section_64) :
245 sizeof(MachO::section)));
248 void MachObjectWriter::WriteSymtabLoadCommand(uint32_t SymbolOffset,
250 uint32_t StringTableOffset,
251 uint32_t StringTableSize) {
252 // struct symtab_command (24 bytes)
254 uint64_t Start = OS.tell();
257 Write32(MachO::LC_SYMTAB);
258 Write32(sizeof(MachO::symtab_command));
259 Write32(SymbolOffset);
261 Write32(StringTableOffset);
262 Write32(StringTableSize);
264 assert(OS.tell() - Start == sizeof(MachO::symtab_command));
267 void MachObjectWriter::WriteDysymtabLoadCommand(uint32_t FirstLocalSymbol,
268 uint32_t NumLocalSymbols,
269 uint32_t FirstExternalSymbol,
270 uint32_t NumExternalSymbols,
271 uint32_t FirstUndefinedSymbol,
272 uint32_t NumUndefinedSymbols,
273 uint32_t IndirectSymbolOffset,
274 uint32_t NumIndirectSymbols) {
275 // struct dysymtab_command (80 bytes)
277 uint64_t Start = OS.tell();
280 Write32(MachO::LC_DYSYMTAB);
281 Write32(sizeof(MachO::dysymtab_command));
282 Write32(FirstLocalSymbol);
283 Write32(NumLocalSymbols);
284 Write32(FirstExternalSymbol);
285 Write32(NumExternalSymbols);
286 Write32(FirstUndefinedSymbol);
287 Write32(NumUndefinedSymbols);
288 Write32(0); // tocoff
290 Write32(0); // modtaboff
291 Write32(0); // nmodtab
292 Write32(0); // extrefsymoff
293 Write32(0); // nextrefsyms
294 Write32(IndirectSymbolOffset);
295 Write32(NumIndirectSymbols);
296 Write32(0); // extreloff
297 Write32(0); // nextrel
298 Write32(0); // locreloff
299 Write32(0); // nlocrel
301 assert(OS.tell() - Start == sizeof(MachO::dysymtab_command));
304 void MachObjectWriter::WriteNlist(MachSymbolData &MSD,
305 const MCAsmLayout &Layout) {
306 MCSymbolData &Data = *MSD.SymbolData;
307 const MCSymbol &Symbol = Data.getSymbol();
309 uint16_t Flags = Data.getFlags();
310 uint64_t Address = 0;
312 // Set the N_TYPE bits. See <mach-o/nlist.h>.
314 // FIXME: Are the prebound or indirect fields possible here?
315 if (Symbol.isUndefined())
316 Type = MachO::N_UNDF;
317 else if (Symbol.isAbsolute())
320 Type = MachO::N_SECT;
322 // FIXME: Set STAB bits.
324 if (Data.isPrivateExtern())
325 Type |= MachO::N_PEXT;
328 if (Data.isExternal() || Symbol.isUndefined())
329 Type |= MachO::N_EXT;
331 // Compute the symbol address.
332 if (Symbol.isDefined()) {
333 Address = getSymbolAddress(&Data, Layout);
334 } else if (Data.isCommon()) {
335 // Common symbols are encoded with the size in the address
336 // field, and their alignment in the flags.
337 Address = Data.getCommonSize();
339 // Common alignment is packed into the 'desc' bits.
340 if (unsigned Align = Data.getCommonAlignment()) {
341 unsigned Log2Size = Log2_32(Align);
342 assert((1U << Log2Size) == Align && "Invalid 'common' alignment!");
344 report_fatal_error("invalid 'common' alignment '" +
345 Twine(Align) + "' for '" + Symbol.getName() + "'",
347 // FIXME: Keep this mask with the SymbolFlags enumeration.
348 Flags = (Flags & 0xF0FF) | (Log2Size << 8);
352 // struct nlist (12 bytes)
354 Write32(MSD.StringIndex);
356 Write8(MSD.SectionIndex);
358 // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc'
367 void MachObjectWriter::WriteLinkeditLoadCommand(uint32_t Type,
370 uint64_t Start = OS.tell();
374 Write32(sizeof(MachO::linkedit_data_command));
378 assert(OS.tell() - Start == sizeof(MachO::linkedit_data_command));
381 static unsigned ComputeLinkerOptionsLoadCommandSize(
382 const std::vector<std::string> &Options, bool is64Bit)
384 unsigned Size = sizeof(MachO::linker_options_command);
385 for (unsigned i = 0, e = Options.size(); i != e; ++i)
386 Size += Options[i].size() + 1;
387 return RoundUpToAlignment(Size, is64Bit ? 8 : 4);
390 void MachObjectWriter::WriteLinkerOptionsLoadCommand(
391 const std::vector<std::string> &Options)
393 unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit());
394 uint64_t Start = OS.tell();
397 Write32(MachO::LC_LINKER_OPTIONS);
399 Write32(Options.size());
400 uint64_t BytesWritten = sizeof(MachO::linker_options_command);
401 for (unsigned i = 0, e = Options.size(); i != e; ++i) {
402 // Write each string, including the null byte.
403 const std::string &Option = Options[i];
404 WriteBytes(Option.c_str(), Option.size() + 1);
405 BytesWritten += Option.size() + 1;
408 // Pad to a multiple of the pointer size.
409 WriteBytes("", OffsetToAlignment(BytesWritten, is64Bit() ? 8 : 4));
411 assert(OS.tell() - Start == Size);
415 void MachObjectWriter::RecordRelocation(const MCAssembler &Asm,
416 const MCAsmLayout &Layout,
417 const MCFragment *Fragment,
418 const MCFixup &Fixup,
421 uint64_t &FixedValue) {
422 TargetObjectWriter->RecordRelocation(this, Asm, Layout, Fragment, Fixup,
426 void MachObjectWriter::BindIndirectSymbols(MCAssembler &Asm) {
427 // This is the point where 'as' creates actual symbols for indirect symbols
428 // (in the following two passes). It would be easier for us to do this sooner
429 // when we see the attribute, but that makes getting the order in the symbol
430 // table much more complicated than it is worth.
432 // FIXME: Revisit this when the dust settles.
434 // Report errors for use of .indirect_symbol not in a symbol pointer section
436 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
437 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
438 const MCSectionMachO &Section =
439 cast<MCSectionMachO>(it->SectionData->getSection());
441 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
442 Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
443 Section.getType() != MachO::S_SYMBOL_STUBS) {
444 MCSymbol &Symbol = *it->Symbol;
445 report_fatal_error("indirect symbol '" + Symbol.getName() +
446 "' not in a symbol pointer or stub section");
450 // Bind non-lazy symbol pointers first.
451 unsigned IndirectIndex = 0;
452 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
453 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
454 const MCSectionMachO &Section =
455 cast<MCSectionMachO>(it->SectionData->getSection());
457 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS)
460 // Initialize the section indirect symbol base, if necessary.
461 IndirectSymBase.insert(std::make_pair(it->SectionData, IndirectIndex));
463 Asm.getOrCreateSymbolData(*it->Symbol);
466 // Then lazy symbol pointers and symbol stubs.
468 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
469 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
470 const MCSectionMachO &Section =
471 cast<MCSectionMachO>(it->SectionData->getSection());
473 if (Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
474 Section.getType() != MachO::S_SYMBOL_STUBS)
477 // Initialize the section indirect symbol base, if necessary.
478 IndirectSymBase.insert(std::make_pair(it->SectionData, IndirectIndex));
480 // Set the symbol type to undefined lazy, but only on construction.
482 // FIXME: Do not hardcode.
484 MCSymbolData &Entry = Asm.getOrCreateSymbolData(*it->Symbol, &Created);
486 Entry.setFlags(Entry.getFlags() | 0x0001);
490 /// ComputeSymbolTable - Compute the symbol table data
492 /// \param StringTable [out] - The string table data.
493 /// \param StringIndexMap [out] - Map from symbol names to offsets in the
495 void MachObjectWriter::
496 ComputeSymbolTable(MCAssembler &Asm, SmallString<256> &StringTable,
497 std::vector<MachSymbolData> &LocalSymbolData,
498 std::vector<MachSymbolData> &ExternalSymbolData,
499 std::vector<MachSymbolData> &UndefinedSymbolData) {
500 // Build section lookup table.
501 DenseMap<const MCSection*, uint8_t> SectionIndexMap;
503 for (MCAssembler::iterator it = Asm.begin(),
504 ie = Asm.end(); it != ie; ++it, ++Index)
505 SectionIndexMap[&it->getSection()] = Index;
506 assert(Index <= 256 && "Too many sections!");
508 // Index 0 is always the empty string.
509 StringMap<uint64_t> StringIndexMap;
510 StringTable += '\x00';
512 // Build the symbol arrays and the string table, but only for non-local
515 // The particular order that we collect the symbols and create the string
516 // table, then sort the symbols is chosen to match 'as'. Even though it
517 // doesn't matter for correctness, this is important for letting us diff .o
519 for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
520 ie = Asm.symbol_end(); it != ie; ++it) {
521 const MCSymbol &Symbol = it->getSymbol();
523 // Ignore non-linker visible symbols.
524 if (!Asm.isSymbolLinkerVisible(it->getSymbol()))
527 if (!it->isExternal() && !Symbol.isUndefined())
530 uint64_t &Entry = StringIndexMap[Symbol.getName()];
532 Entry = StringTable.size();
533 StringTable += Symbol.getName();
534 StringTable += '\x00';
539 MSD.StringIndex = Entry;
541 if (Symbol.isUndefined()) {
542 MSD.SectionIndex = 0;
543 UndefinedSymbolData.push_back(MSD);
544 } else if (Symbol.isAbsolute()) {
545 MSD.SectionIndex = 0;
546 ExternalSymbolData.push_back(MSD);
548 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
549 assert(MSD.SectionIndex && "Invalid section index!");
550 ExternalSymbolData.push_back(MSD);
554 // Now add the data for local symbols.
555 for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
556 ie = Asm.symbol_end(); it != ie; ++it) {
557 const MCSymbol &Symbol = it->getSymbol();
559 // Ignore non-linker visible symbols.
560 if (!Asm.isSymbolLinkerVisible(it->getSymbol()))
563 if (it->isExternal() || Symbol.isUndefined())
566 uint64_t &Entry = StringIndexMap[Symbol.getName()];
568 Entry = StringTable.size();
569 StringTable += Symbol.getName();
570 StringTable += '\x00';
575 MSD.StringIndex = Entry;
577 if (Symbol.isAbsolute()) {
578 MSD.SectionIndex = 0;
579 LocalSymbolData.push_back(MSD);
581 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
582 assert(MSD.SectionIndex && "Invalid section index!");
583 LocalSymbolData.push_back(MSD);
587 // External and undefined symbols are required to be in lexicographic order.
588 std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
589 std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
591 // Set the symbol indices.
593 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
594 LocalSymbolData[i].SymbolData->setIndex(Index++);
595 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
596 ExternalSymbolData[i].SymbolData->setIndex(Index++);
597 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
598 UndefinedSymbolData[i].SymbolData->setIndex(Index++);
600 // The string table is padded to a multiple of 4.
601 while (StringTable.size() % 4)
602 StringTable += '\x00';
605 void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm,
606 const MCAsmLayout &Layout) {
607 uint64_t StartAddress = 0;
608 const SmallVectorImpl<MCSectionData*> &Order = Layout.getSectionOrder();
609 for (int i = 0, n = Order.size(); i != n ; ++i) {
610 const MCSectionData *SD = Order[i];
611 StartAddress = RoundUpToAlignment(StartAddress, SD->getAlignment());
612 SectionAddress[SD] = StartAddress;
613 StartAddress += Layout.getSectionAddressSize(SD);
615 // Explicitly pad the section to match the alignment requirements of the
616 // following one. This is for 'gas' compatibility, it shouldn't
617 /// strictly be necessary.
618 StartAddress += getPaddingSize(SD, Layout);
622 void MachObjectWriter::markAbsoluteVariableSymbols(MCAssembler &Asm,
623 const MCAsmLayout &Layout) {
624 for (MCAssembler::symbol_iterator i = Asm.symbol_begin(),
625 e = Asm.symbol_end();
627 MCSymbolData &SD = *i;
628 if (!SD.getSymbol().isVariable())
631 // Is the variable is a symbol difference (SA - SB + C) expression,
632 // and neither symbol is external, mark the variable as absolute.
633 const MCExpr *Expr = SD.getSymbol().getVariableValue();
635 if (Expr->EvaluateAsRelocatable(Value, &Layout)) {
636 if (Value.getSymA() && Value.getSymB())
637 const_cast<MCSymbol*>(&SD.getSymbol())->setAbsolute();
642 void MachObjectWriter::ExecutePostLayoutBinding(MCAssembler &Asm,
643 const MCAsmLayout &Layout) {
644 computeSectionAddresses(Asm, Layout);
646 // Create symbol data for any indirect symbols.
647 BindIndirectSymbols(Asm);
649 // Mark symbol difference expressions in variables (from .set or = directives)
651 markAbsoluteVariableSymbols(Asm, Layout);
653 // Compute symbol table information and bind symbol indices.
654 ComputeSymbolTable(Asm, StringTable, LocalSymbolData, ExternalSymbolData,
655 UndefinedSymbolData);
658 bool MachObjectWriter::
659 IsSymbolRefDifferenceFullyResolvedImpl(const MCAssembler &Asm,
660 const MCSymbolData &DataA,
661 const MCFragment &FB,
663 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 MCSymbolData *A_Base = nullptr, *B_Base = nullptr;
674 const MCSymbol &SA = DataA.getSymbol().AliasedSymbol();
675 const MCSection &SecA = SA.getSection();
676 const MCSection &SecB = FB.getParent()->getSection();
679 // The simple (Darwin, except on x86_64) way of dealing with this was to
680 // assume that any reference to a temporary symbol *must* be a temporary
681 // symbol in the same atom, unless the sections differ. Therefore, any PCrel
682 // relocation to a temporary symbol (in the same section) is fully
683 // resolved. This also works in conjunction with absolutized .set, which
684 // requires the compiler to use .set to absolutize the differences between
685 // symbols which the compiler knows to be assembly time constants, so we
686 // don't need to worry about considering symbol differences fully resolved.
688 // If the file isn't using sub-sections-via-symbols, we can make the
689 // same assumptions about any symbol that we normally make about
692 bool hasReliableSymbolDifference = isX86_64();
693 if (!hasReliableSymbolDifference) {
694 if (!SA.isInSection() || &SecA != &SecB ||
695 (!SA.isTemporary() &&
696 FB.getAtom() != Asm.getSymbolData(SA).getFragment()->getAtom() &&
697 Asm.getSubsectionsViaSymbols()))
701 // For Darwin x86_64, there is one special case when the reference IsPCRel.
702 // If the fragment with the reference does not have a base symbol but meets
703 // the simple way of dealing with this, in that it is a temporary symbol in
704 // the same atom then it is assumed to be fully resolved. This is needed so
705 // a relocation entry is not created and so the static linker does not
706 // mess up the reference later.
707 else if(!FB.getAtom() &&
708 SA.isTemporary() && SA.isInSection() && &SecA == &SecB){
712 if (!TargetObjectWriter->useAggressiveSymbolFolding())
716 const MCFragment *FA = Asm.getSymbolData(SA).getFragment();
718 // Bail if the symbol has no fragment.
722 A_Base = FA->getAtom();
726 B_Base = FB.getAtom();
730 // If the atoms are the same, they are guaranteed to have the same address.
731 if (A_Base == B_Base)
734 // Otherwise, we can't prove this is fully resolved.
738 void MachObjectWriter::WriteObject(MCAssembler &Asm,
739 const MCAsmLayout &Layout) {
740 unsigned NumSections = Asm.size();
741 const MCAssembler::VersionMinInfoType &VersionInfo =
742 Layout.getAssembler().getVersionMinInfo();
744 // The section data starts after the header, the segment load command (and
745 // section headers) and the symbol table.
746 unsigned NumLoadCommands = 1;
747 uint64_t LoadCommandsSize = is64Bit() ?
748 sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64):
749 sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section);
751 // Add the deployment target version info load command size, if used.
752 if (VersionInfo.Major != 0) {
754 LoadCommandsSize += sizeof(MachO::version_min_command);
757 // Add the data-in-code load command size, if used.
758 unsigned NumDataRegions = Asm.getDataRegions().size();
759 if (NumDataRegions) {
761 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
764 // Add the loh load command size, if used.
765 uint64_t LOHRawSize = Asm.getLOHContainer().getEmitSize(*this, Layout);
766 uint64_t LOHSize = RoundUpToAlignment(LOHRawSize, is64Bit() ? 8 : 4);
769 LoadCommandsSize += sizeof(MachO::linkedit_data_command);
772 // Add the symbol table load command sizes, if used.
773 unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
774 UndefinedSymbolData.size();
776 NumLoadCommands += 2;
777 LoadCommandsSize += (sizeof(MachO::symtab_command) +
778 sizeof(MachO::dysymtab_command));
781 // Add the linker option load commands sizes.
782 const std::vector<std::vector<std::string> > &LinkerOptions =
783 Asm.getLinkerOptions();
784 for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
786 LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(LinkerOptions[i],
790 // Compute the total size of the section data, as well as its file size and vm
792 uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) :
793 sizeof(MachO::mach_header)) + LoadCommandsSize;
794 uint64_t SectionDataSize = 0;
795 uint64_t SectionDataFileSize = 0;
797 for (MCAssembler::const_iterator it = Asm.begin(),
798 ie = Asm.end(); it != ie; ++it) {
799 const MCSectionData &SD = *it;
800 uint64_t Address = getSectionAddress(&SD);
801 uint64_t Size = Layout.getSectionAddressSize(&SD);
802 uint64_t FileSize = Layout.getSectionFileSize(&SD);
803 FileSize += getPaddingSize(&SD, Layout);
805 VMSize = std::max(VMSize, Address + Size);
807 if (SD.getSection().isVirtualSection())
810 SectionDataSize = std::max(SectionDataSize, Address + Size);
811 SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
814 // The section data is padded to 4 bytes.
816 // FIXME: Is this machine dependent?
817 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
818 SectionDataFileSize += SectionDataPadding;
820 // Write the prolog, starting with the header and load command...
821 WriteHeader(NumLoadCommands, LoadCommandsSize,
822 Asm.getSubsectionsViaSymbols());
823 WriteSegmentLoadCommand(NumSections, VMSize,
824 SectionDataStart, SectionDataSize);
826 // ... and then the section headers.
827 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
828 for (MCAssembler::const_iterator it = Asm.begin(),
829 ie = Asm.end(); it != ie; ++it) {
830 std::vector<MachO::any_relocation_info> &Relocs = Relocations[it];
831 unsigned NumRelocs = Relocs.size();
832 uint64_t SectionStart = SectionDataStart + getSectionAddress(it);
833 WriteSection(Asm, Layout, *it, SectionStart, RelocTableEnd, NumRelocs);
834 RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
837 // Write out the deployment target information, if it's available.
838 if (VersionInfo.Major != 0) {
839 assert(VersionInfo.Update < 256 && "unencodable update target version");
840 assert(VersionInfo.Minor < 256 && "unencodable minor target version");
841 assert(VersionInfo.Major < 65536 && "unencodable major target version");
842 uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
843 (VersionInfo.Major << 16);
844 Write32(VersionInfo.Kind == MCVM_OSXVersionMin ? MachO::LC_VERSION_MIN_MACOSX :
845 MachO::LC_VERSION_MIN_IPHONEOS);
846 Write32(sizeof(MachO::version_min_command));
847 Write32(EncodedVersion);
848 Write32(0); // reserved.
851 // Write the data-in-code load command, if used.
852 uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
853 if (NumDataRegions) {
854 uint64_t DataRegionsOffset = RelocTableEnd;
855 uint64_t DataRegionsSize = NumDataRegions * 8;
856 WriteLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
860 // Write the loh load command, if used.
861 uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
863 WriteLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
864 DataInCodeTableEnd, LOHSize);
866 // Write the symbol table load command, if used.
868 unsigned FirstLocalSymbol = 0;
869 unsigned NumLocalSymbols = LocalSymbolData.size();
870 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
871 unsigned NumExternalSymbols = ExternalSymbolData.size();
872 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
873 unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
874 unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
875 unsigned NumSymTabSymbols =
876 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
877 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
878 uint64_t IndirectSymbolOffset = 0;
880 // If used, the indirect symbols are written after the section data.
881 if (NumIndirectSymbols)
882 IndirectSymbolOffset = LOHTableEnd;
884 // The symbol table is written after the indirect symbol data.
885 uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
887 // The string table is written after symbol table.
888 uint64_t StringTableOffset =
889 SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
890 sizeof(MachO::nlist_64) :
891 sizeof(MachO::nlist));
892 WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
893 StringTableOffset, StringTable.size());
895 WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
896 FirstExternalSymbol, NumExternalSymbols,
897 FirstUndefinedSymbol, NumUndefinedSymbols,
898 IndirectSymbolOffset, NumIndirectSymbols);
901 // Write the linker options load commands.
902 for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
903 WriteLinkerOptionsLoadCommand(LinkerOptions[i]);
906 // Write the actual section data.
907 for (MCAssembler::const_iterator it = Asm.begin(),
908 ie = Asm.end(); it != ie; ++it) {
909 Asm.writeSectionData(it, Layout);
911 uint64_t Pad = getPaddingSize(it, Layout);
912 for (unsigned int i = 0; i < Pad; ++i)
916 // Write the extra padding.
917 WriteZeros(SectionDataPadding);
919 // Write the relocation entries.
920 for (MCAssembler::const_iterator it = Asm.begin(),
921 ie = Asm.end(); it != ie; ++it) {
922 // Write the section relocation entries, in reverse order to match 'as'
923 // (approximately, the exact algorithm is more complicated than this).
924 std::vector<MachO::any_relocation_info> &Relocs = Relocations[it];
925 for (unsigned i = 0, e = Relocs.size(); i != e; ++i) {
926 Write32(Relocs[e - i - 1].r_word0);
927 Write32(Relocs[e - i - 1].r_word1);
931 // Write out the data-in-code region payload, if there is one.
932 for (MCAssembler::const_data_region_iterator
933 it = Asm.data_region_begin(), ie = Asm.data_region_end();
935 const DataRegionData *Data = &(*it);
937 getSymbolAddress(&Layout.getAssembler().getSymbolData(*Data->Start),
940 getSymbolAddress(&Layout.getAssembler().getSymbolData(*Data->End),
942 DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
943 << " start: " << Start << "(" << Data->Start->getName() << ")"
944 << " end: " << End << "(" << Data->End->getName() << ")"
945 << " size: " << End - Start
948 Write16(End - Start);
952 // Write out the loh commands, if there is one.
955 unsigned Start = OS.tell();
957 Asm.getLOHContainer().Emit(*this, Layout);
958 // Pad to a multiple of the pointer size.
959 WriteBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
960 assert(OS.tell() - Start == LOHSize);
963 // Write the symbol table data, if used.
965 // Write the indirect symbol entries.
966 for (MCAssembler::const_indirect_symbol_iterator
967 it = Asm.indirect_symbol_begin(),
968 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
969 // Indirect symbols in the non-lazy symbol pointer section have some
971 const MCSectionMachO &Section =
972 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
973 if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
974 // If this symbol is defined and internal, mark it as such.
975 if (it->Symbol->isDefined() &&
976 !Asm.getSymbolData(*it->Symbol).isExternal()) {
977 uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
978 if (it->Symbol->isAbsolute())
979 Flags |= MachO::INDIRECT_SYMBOL_ABS;
985 Write32(Asm.getSymbolData(*it->Symbol).getIndex());
988 // FIXME: Check that offsets match computed ones.
990 // Write the symbol table entries.
991 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
992 WriteNlist(LocalSymbolData[i], Layout);
993 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
994 WriteNlist(ExternalSymbolData[i], Layout);
995 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
996 WriteNlist(UndefinedSymbolData[i], Layout);
998 // Write the string table.
999 OS << StringTable.str();
1003 MCObjectWriter *llvm::createMachObjectWriter(MCMachObjectTargetWriter *MOTW,
1005 bool IsLittleEndian) {
1006 return new MachObjectWriter(MOTW, OS, IsLittleEndian);