1 //===- lib/MC/MCAssembler.cpp - Assembler Backend Implementation ----------===//
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 #define DEBUG_TYPE "assembler"
11 #include "llvm/MC/MCAssembler.h"
12 #include "llvm/MC/MCExpr.h"
13 #include "llvm/MC/MCSectionMachO.h"
14 #include "llvm/MC/MCSymbol.h"
15 #include "llvm/MC/MCValue.h"
16 #include "llvm/ADT/DenseMap.h"
17 #include "llvm/ADT/SmallString.h"
18 #include "llvm/ADT/Statistic.h"
19 #include "llvm/ADT/StringExtras.h"
20 #include "llvm/ADT/StringMap.h"
21 #include "llvm/ADT/Twine.h"
22 #include "llvm/Support/ErrorHandling.h"
23 #include "llvm/Support/MachO.h"
24 #include "llvm/Support/raw_ostream.h"
25 #include "llvm/Support/Debug.h"
28 #include "../Target/X86/X86FixupKinds.h"
33 class MachObjectWriter;
35 STATISTIC(EmittedFragments, "Number of emitted assembler fragments");
37 // FIXME FIXME FIXME: There are number of places in this file where we convert
38 // what is a 64-bit assembler value used for computation into a value in the
39 // object file, which may truncate it. We should detect that truncation where
40 // invalid and report errors back.
42 static void WriteFileData(raw_ostream &OS, const MCSectionData &SD,
43 MachObjectWriter &MOW);
45 /// isVirtualSection - Check if this is a section which does not actually exist
46 /// in the object file.
47 static bool isVirtualSection(const MCSection &Section) {
49 const MCSectionMachO &SMO = static_cast<const MCSectionMachO&>(Section);
50 unsigned Type = SMO.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
51 return (Type == MCSectionMachO::S_ZEROFILL);
54 static unsigned getFixupKindLog2Size(unsigned Kind) {
56 default: llvm_unreachable("invalid fixup kind!");
57 case X86::reloc_pcrel_1byte:
58 case FK_Data_1: return 0;
59 case FK_Data_2: return 1;
60 case X86::reloc_pcrel_4byte:
61 case X86::reloc_riprel_4byte:
62 case FK_Data_4: return 2;
63 case FK_Data_8: return 3;
67 static bool isFixupKindPCRel(unsigned Kind) {
71 case X86::reloc_pcrel_1byte:
72 case X86::reloc_pcrel_4byte:
73 case X86::reloc_riprel_4byte:
78 class MachObjectWriter {
79 // See <mach-o/loader.h>.
81 Header_Magic32 = 0xFEEDFACE,
82 Header_Magic64 = 0xFEEDFACF
85 static const unsigned Header32Size = 28;
86 static const unsigned Header64Size = 32;
87 static const unsigned SegmentLoadCommand32Size = 56;
88 static const unsigned Section32Size = 68;
89 static const unsigned SymtabLoadCommandSize = 24;
90 static const unsigned DysymtabLoadCommandSize = 80;
91 static const unsigned Nlist32Size = 12;
92 static const unsigned RelocationInfoSize = 8;
99 HF_SubsectionsViaSymbols = 0x2000
102 enum LoadCommandType {
108 // See <mach-o/nlist.h>.
109 enum SymbolTypeType {
110 STT_Undefined = 0x00,
115 enum SymbolTypeFlags {
116 // If any of these bits are set, then the entry is a stab entry number (see
117 // <mach-o/stab.h>. Otherwise the other masks apply.
118 STF_StabsEntryMask = 0xe0,
122 STF_PrivateExtern = 0x10
125 /// IndirectSymbolFlags - Flags for encoding special values in the indirect
127 enum IndirectSymbolFlags {
128 ISF_Local = 0x80000000,
129 ISF_Absolute = 0x40000000
132 /// RelocationFlags - Special flags for addresses.
133 enum RelocationFlags {
134 RF_Scattered = 0x80000000
137 enum RelocationInfoType {
141 RIT_PreboundLazyPointer = 3,
142 RIT_LocalDifference = 4
145 /// MachSymbolData - Helper struct for containing some precomputed information
147 struct MachSymbolData {
148 MCSymbolData *SymbolData;
149 uint64_t StringIndex;
150 uint8_t SectionIndex;
152 // Support lexicographic sorting.
153 bool operator<(const MachSymbolData &RHS) const {
154 const std::string &Name = SymbolData->getSymbol().getName();
155 return Name < RHS.SymbolData->getSymbol().getName();
163 MachObjectWriter(raw_ostream &_OS, bool _IsLSB = true)
164 : OS(_OS), IsLSB(_IsLSB) {
167 /// @name Helper Methods
170 void Write8(uint8_t Value) {
174 void Write16(uint16_t Value) {
176 Write8(uint8_t(Value >> 0));
177 Write8(uint8_t(Value >> 8));
179 Write8(uint8_t(Value >> 8));
180 Write8(uint8_t(Value >> 0));
184 void Write32(uint32_t Value) {
186 Write16(uint16_t(Value >> 0));
187 Write16(uint16_t(Value >> 16));
189 Write16(uint16_t(Value >> 16));
190 Write16(uint16_t(Value >> 0));
194 void Write64(uint64_t Value) {
196 Write32(uint32_t(Value >> 0));
197 Write32(uint32_t(Value >> 32));
199 Write32(uint32_t(Value >> 32));
200 Write32(uint32_t(Value >> 0));
204 void WriteZeros(unsigned N) {
205 const char Zeros[16] = { 0 };
207 for (unsigned i = 0, e = N / 16; i != e; ++i)
208 OS << StringRef(Zeros, 16);
210 OS << StringRef(Zeros, N % 16);
213 void WriteString(StringRef Str, unsigned ZeroFillSize = 0) {
216 WriteZeros(ZeroFillSize - Str.size());
221 void WriteHeader32(unsigned NumLoadCommands, unsigned LoadCommandsSize,
222 bool SubsectionsViaSymbols) {
225 if (SubsectionsViaSymbols)
226 Flags |= HF_SubsectionsViaSymbols;
228 // struct mach_header (28 bytes)
230 uint64_t Start = OS.tell();
233 Write32(Header_Magic32);
235 // FIXME: Support cputype.
236 Write32(MachO::CPUTypeI386);
237 // FIXME: Support cpusubtype.
238 Write32(MachO::CPUSubType_I386_ALL);
240 Write32(NumLoadCommands); // Object files have a single load command, the
242 Write32(LoadCommandsSize);
245 assert(OS.tell() - Start == Header32Size);
248 /// WriteSegmentLoadCommand32 - Write a 32-bit segment load command.
250 /// \arg NumSections - The number of sections in this segment.
251 /// \arg SectionDataSize - The total size of the sections.
252 void WriteSegmentLoadCommand32(unsigned NumSections,
254 uint64_t SectionDataStartOffset,
255 uint64_t SectionDataSize) {
256 // struct segment_command (56 bytes)
258 uint64_t Start = OS.tell();
261 Write32(LCT_Segment);
262 Write32(SegmentLoadCommand32Size + NumSections * Section32Size);
265 Write32(0); // vmaddr
266 Write32(VMSize); // vmsize
267 Write32(SectionDataStartOffset); // file offset
268 Write32(SectionDataSize); // file size
269 Write32(0x7); // maxprot
270 Write32(0x7); // initprot
271 Write32(NumSections);
274 assert(OS.tell() - Start == SegmentLoadCommand32Size);
277 void WriteSection32(const MCSectionData &SD, uint64_t FileOffset,
278 uint64_t RelocationsStart, unsigned NumRelocations) {
279 // The offset is unused for virtual sections.
280 if (isVirtualSection(SD.getSection())) {
281 assert(SD.getFileSize() == 0 && "Invalid file size!");
285 // struct section (68 bytes)
287 uint64_t Start = OS.tell();
290 // FIXME: cast<> support!
291 const MCSectionMachO &Section =
292 static_cast<const MCSectionMachO&>(SD.getSection());
293 WriteString(Section.getSectionName(), 16);
294 WriteString(Section.getSegmentName(), 16);
295 Write32(SD.getAddress()); // address
296 Write32(SD.getSize()); // size
299 unsigned Flags = Section.getTypeAndAttributes();
300 if (SD.hasInstructions())
301 Flags |= MCSectionMachO::S_ATTR_SOME_INSTRUCTIONS;
303 assert(isPowerOf2_32(SD.getAlignment()) && "Invalid alignment!");
304 Write32(Log2_32(SD.getAlignment()));
305 Write32(NumRelocations ? RelocationsStart : 0);
306 Write32(NumRelocations);
308 Write32(0); // reserved1
309 Write32(Section.getStubSize()); // reserved2
311 assert(OS.tell() - Start == Section32Size);
314 void WriteSymtabLoadCommand(uint32_t SymbolOffset, uint32_t NumSymbols,
315 uint32_t StringTableOffset,
316 uint32_t StringTableSize) {
317 // struct symtab_command (24 bytes)
319 uint64_t Start = OS.tell();
323 Write32(SymtabLoadCommandSize);
324 Write32(SymbolOffset);
326 Write32(StringTableOffset);
327 Write32(StringTableSize);
329 assert(OS.tell() - Start == SymtabLoadCommandSize);
332 void WriteDysymtabLoadCommand(uint32_t FirstLocalSymbol,
333 uint32_t NumLocalSymbols,
334 uint32_t FirstExternalSymbol,
335 uint32_t NumExternalSymbols,
336 uint32_t FirstUndefinedSymbol,
337 uint32_t NumUndefinedSymbols,
338 uint32_t IndirectSymbolOffset,
339 uint32_t NumIndirectSymbols) {
340 // struct dysymtab_command (80 bytes)
342 uint64_t Start = OS.tell();
345 Write32(LCT_Dysymtab);
346 Write32(DysymtabLoadCommandSize);
347 Write32(FirstLocalSymbol);
348 Write32(NumLocalSymbols);
349 Write32(FirstExternalSymbol);
350 Write32(NumExternalSymbols);
351 Write32(FirstUndefinedSymbol);
352 Write32(NumUndefinedSymbols);
353 Write32(0); // tocoff
355 Write32(0); // modtaboff
356 Write32(0); // nmodtab
357 Write32(0); // extrefsymoff
358 Write32(0); // nextrefsyms
359 Write32(IndirectSymbolOffset);
360 Write32(NumIndirectSymbols);
361 Write32(0); // extreloff
362 Write32(0); // nextrel
363 Write32(0); // locreloff
364 Write32(0); // nlocrel
366 assert(OS.tell() - Start == DysymtabLoadCommandSize);
369 void WriteNlist32(MachSymbolData &MSD) {
370 MCSymbolData &Data = *MSD.SymbolData;
371 const MCSymbol &Symbol = Data.getSymbol();
373 uint16_t Flags = Data.getFlags();
374 uint32_t Address = 0;
376 // Set the N_TYPE bits. See <mach-o/nlist.h>.
378 // FIXME: Are the prebound or indirect fields possible here?
379 if (Symbol.isUndefined())
380 Type = STT_Undefined;
381 else if (Symbol.isAbsolute())
386 // FIXME: Set STAB bits.
388 if (Data.isPrivateExtern())
389 Type |= STF_PrivateExtern;
392 if (Data.isExternal() || Symbol.isUndefined())
393 Type |= STF_External;
395 // Compute the symbol address.
396 if (Symbol.isDefined()) {
397 if (Symbol.isAbsolute()) {
398 llvm_unreachable("FIXME: Not yet implemented!");
400 Address = Data.getFragment()->getAddress() + Data.getOffset();
402 } else if (Data.isCommon()) {
403 // Common symbols are encoded with the size in the address
404 // field, and their alignment in the flags.
405 Address = Data.getCommonSize();
407 // Common alignment is packed into the 'desc' bits.
408 if (unsigned Align = Data.getCommonAlignment()) {
409 unsigned Log2Size = Log2_32(Align);
410 assert((1U << Log2Size) == Align && "Invalid 'common' alignment!");
412 llvm_report_error("invalid 'common' alignment '" +
414 // FIXME: Keep this mask with the SymbolFlags enumeration.
415 Flags = (Flags & 0xF0FF) | (Log2Size << 8);
419 // struct nlist (12 bytes)
421 Write32(MSD.StringIndex);
423 Write8(MSD.SectionIndex);
425 // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc'
431 struct MachRelocationEntry {
435 void ComputeScatteredRelocationInfo(MCAssembler &Asm, MCFragment &Fragment,
437 const MCValue &Target,
438 DenseMap<const MCSymbol*,MCSymbolData*> &SymbolMap,
439 std::vector<MachRelocationEntry> &Relocs) {
440 uint32_t Address = Fragment.getOffset() + Fixup.Offset;
441 unsigned IsPCRel = 0;
442 unsigned Type = RIT_Vanilla;
445 const MCSymbol *A = Target.getSymA();
446 MCSymbolData *SD = SymbolMap.lookup(A);
447 uint32_t Value = SD->getFragment()->getAddress() + SD->getOffset();
450 if (const MCSymbol *B = Target.getSymB()) {
451 Type = RIT_LocalDifference;
453 MCSymbolData *SD = SymbolMap.lookup(B);
454 Value2 = SD->getFragment()->getAddress() + SD->getOffset();
457 unsigned Log2Size = getFixupKindLog2Size(Fixup.Kind);
459 // The value which goes in the fixup is current value of the expression.
460 Fixup.FixedValue = Value - Value2 + Target.getConstant();
461 if (isFixupKindPCRel(Fixup.Kind)) {
462 Fixup.FixedValue -= Address + (1 << Log2Size);
466 MachRelocationEntry MRE;
467 MRE.Word0 = ((Address << 0) |
473 Relocs.push_back(MRE);
475 if (Type == RIT_LocalDifference) {
478 MachRelocationEntry MRE;
479 MRE.Word0 = ((0 << 0) |
485 Relocs.push_back(MRE);
489 void ComputeRelocationInfo(MCAssembler &Asm, MCDataFragment &Fragment,
491 DenseMap<const MCSymbol*,MCSymbolData*> &SymbolMap,
492 std::vector<MachRelocationEntry> &Relocs) {
494 if (!Fixup.Value->EvaluateAsRelocatable(Target))
495 llvm_report_error("expected relocatable expression");
497 // If this is a difference or a local symbol plus an offset, then we need a
498 // scattered relocation entry.
499 if (Target.getSymB() ||
500 (Target.getSymA() && !Target.getSymA()->isUndefined() &&
501 Target.getConstant()))
502 return ComputeScatteredRelocationInfo(Asm, Fragment, Fixup, Target,
506 uint32_t Address = Fragment.getOffset() + Fixup.Offset;
509 unsigned IsPCRel = 0;
510 unsigned IsExtern = 0;
513 if (Target.isAbsolute()) { // constant
514 // SymbolNum of 0 indicates the absolute section.
516 // FIXME: When is this generated?
519 llvm_unreachable("FIXME: Not yet implemented!");
521 const MCSymbol *Symbol = Target.getSymA();
522 MCSymbolData *SD = SymbolMap.lookup(Symbol);
524 if (Symbol->isUndefined()) {
526 Index = SD->getIndex();
529 // The index is the section ordinal.
533 MCAssembler::iterator it = Asm.begin(), ie = Asm.end();
534 for (; it != ie; ++it, ++Index)
535 if (&*it == SD->getFragment()->getParent())
537 assert(it != ie && "Unable to find section index!");
538 Value = SD->getFragment()->getAddress() + SD->getOffset();
544 // The value which goes in the fixup is current value of the expression.
545 Fixup.FixedValue = Value + Target.getConstant();
547 unsigned Log2Size = getFixupKindLog2Size(Fixup.Kind);
549 if (isFixupKindPCRel(Fixup.Kind)) {
550 Fixup.FixedValue -= Address + (1<<Log2Size);
554 // struct relocation_info (8 bytes)
555 MachRelocationEntry MRE;
557 MRE.Word1 = ((Index << 0) |
562 Relocs.push_back(MRE);
565 void BindIndirectSymbols(MCAssembler &Asm,
566 DenseMap<const MCSymbol*,MCSymbolData*> &SymbolMap) {
567 // This is the point where 'as' creates actual symbols for indirect symbols
568 // (in the following two passes). It would be easier for us to do this
569 // sooner when we see the attribute, but that makes getting the order in the
570 // symbol table much more complicated than it is worth.
572 // FIXME: Revisit this when the dust settles.
574 // Bind non lazy symbol pointers first.
575 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
576 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
577 // FIXME: cast<> support!
578 const MCSectionMachO &Section =
579 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
582 Section.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
583 if (Type != MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS)
586 MCSymbolData *&Entry = SymbolMap[it->Symbol];
588 Entry = new MCSymbolData(*it->Symbol, 0, 0, &Asm);
591 // Then lazy symbol pointers and symbol stubs.
592 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
593 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
594 // FIXME: cast<> support!
595 const MCSectionMachO &Section =
596 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
599 Section.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
600 if (Type != MCSectionMachO::S_LAZY_SYMBOL_POINTERS &&
601 Type != MCSectionMachO::S_SYMBOL_STUBS)
604 MCSymbolData *&Entry = SymbolMap[it->Symbol];
606 Entry = new MCSymbolData(*it->Symbol, 0, 0, &Asm);
608 // Set the symbol type to undefined lazy, but only on construction.
610 // FIXME: Do not hardcode.
611 Entry->setFlags(Entry->getFlags() | 0x0001);
616 /// ComputeSymbolTable - Compute the symbol table data
618 /// \param StringTable [out] - The string table data.
619 /// \param StringIndexMap [out] - Map from symbol names to offsets in the
621 void ComputeSymbolTable(MCAssembler &Asm, SmallString<256> &StringTable,
622 std::vector<MachSymbolData> &LocalSymbolData,
623 std::vector<MachSymbolData> &ExternalSymbolData,
624 std::vector<MachSymbolData> &UndefinedSymbolData) {
625 // Build section lookup table.
626 DenseMap<const MCSection*, uint8_t> SectionIndexMap;
628 for (MCAssembler::iterator it = Asm.begin(),
629 ie = Asm.end(); it != ie; ++it, ++Index)
630 SectionIndexMap[&it->getSection()] = Index;
631 assert(Index <= 256 && "Too many sections!");
633 // Index 0 is always the empty string.
634 StringMap<uint64_t> StringIndexMap;
635 StringTable += '\x00';
637 // Build the symbol arrays and the string table, but only for non-local
640 // The particular order that we collect the symbols and create the string
641 // table, then sort the symbols is chosen to match 'as'. Even though it
642 // doesn't matter for correctness, this is important for letting us diff .o
644 for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
645 ie = Asm.symbol_end(); it != ie; ++it) {
646 const MCSymbol &Symbol = it->getSymbol();
648 // Ignore assembler temporaries.
649 if (it->getSymbol().isTemporary())
652 if (!it->isExternal() && !Symbol.isUndefined())
655 uint64_t &Entry = StringIndexMap[Symbol.getName()];
657 Entry = StringTable.size();
658 StringTable += Symbol.getName();
659 StringTable += '\x00';
664 MSD.StringIndex = Entry;
666 if (Symbol.isUndefined()) {
667 MSD.SectionIndex = 0;
668 UndefinedSymbolData.push_back(MSD);
669 } else if (Symbol.isAbsolute()) {
670 MSD.SectionIndex = 0;
671 ExternalSymbolData.push_back(MSD);
673 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
674 assert(MSD.SectionIndex && "Invalid section index!");
675 ExternalSymbolData.push_back(MSD);
679 // Now add the data for local symbols.
680 for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
681 ie = Asm.symbol_end(); it != ie; ++it) {
682 const MCSymbol &Symbol = it->getSymbol();
684 // Ignore assembler temporaries.
685 if (it->getSymbol().isTemporary())
688 if (it->isExternal() || Symbol.isUndefined())
691 uint64_t &Entry = StringIndexMap[Symbol.getName()];
693 Entry = StringTable.size();
694 StringTable += Symbol.getName();
695 StringTable += '\x00';
700 MSD.StringIndex = Entry;
702 if (Symbol.isAbsolute()) {
703 MSD.SectionIndex = 0;
704 LocalSymbolData.push_back(MSD);
706 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
707 assert(MSD.SectionIndex && "Invalid section index!");
708 LocalSymbolData.push_back(MSD);
712 // External and undefined symbols are required to be in lexicographic order.
713 std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
714 std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
716 // Set the symbol indices.
718 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
719 LocalSymbolData[i].SymbolData->setIndex(Index++);
720 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
721 ExternalSymbolData[i].SymbolData->setIndex(Index++);
722 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
723 UndefinedSymbolData[i].SymbolData->setIndex(Index++);
725 // The string table is padded to a multiple of 4.
726 while (StringTable.size() % 4)
727 StringTable += '\x00';
730 void WriteObject(MCAssembler &Asm) {
731 unsigned NumSections = Asm.size();
733 // Compute the symbol -> symbol data map.
735 // FIXME: This should not be here.
736 DenseMap<const MCSymbol*, MCSymbolData *> SymbolMap;
737 for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
738 ie = Asm.symbol_end(); it != ie; ++it)
739 SymbolMap[&it->getSymbol()] = it;
741 // Create symbol data for any indirect symbols.
742 BindIndirectSymbols(Asm, SymbolMap);
744 // Compute symbol table information.
745 SmallString<256> StringTable;
746 std::vector<MachSymbolData> LocalSymbolData;
747 std::vector<MachSymbolData> ExternalSymbolData;
748 std::vector<MachSymbolData> UndefinedSymbolData;
749 unsigned NumSymbols = Asm.symbol_size();
751 // No symbol table command is written if there are no symbols.
753 ComputeSymbolTable(Asm, StringTable, LocalSymbolData, ExternalSymbolData,
754 UndefinedSymbolData);
756 // The section data starts after the header, the segment load command (and
757 // section headers) and the symbol table.
758 unsigned NumLoadCommands = 1;
759 uint64_t LoadCommandsSize =
760 SegmentLoadCommand32Size + NumSections * Section32Size;
762 // Add the symbol table load command sizes, if used.
764 NumLoadCommands += 2;
765 LoadCommandsSize += SymtabLoadCommandSize + DysymtabLoadCommandSize;
768 // Compute the total size of the section data, as well as its file size and
770 uint64_t SectionDataStart = Header32Size + LoadCommandsSize;
771 uint64_t SectionDataSize = 0;
772 uint64_t SectionDataFileSize = 0;
774 for (MCAssembler::iterator it = Asm.begin(),
775 ie = Asm.end(); it != ie; ++it) {
776 MCSectionData &SD = *it;
778 VMSize = std::max(VMSize, SD.getAddress() + SD.getSize());
780 if (isVirtualSection(SD.getSection()))
783 SectionDataSize = std::max(SectionDataSize,
784 SD.getAddress() + SD.getSize());
785 SectionDataFileSize = std::max(SectionDataFileSize,
786 SD.getAddress() + SD.getFileSize());
789 // The section data is padded to 4 bytes.
791 // FIXME: Is this machine dependent?
792 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
793 SectionDataFileSize += SectionDataPadding;
795 // Write the prolog, starting with the header and load command...
796 WriteHeader32(NumLoadCommands, LoadCommandsSize,
797 Asm.getSubsectionsViaSymbols());
798 WriteSegmentLoadCommand32(NumSections, VMSize,
799 SectionDataStart, SectionDataSize);
801 // ... and then the section headers.
803 // We also compute the section relocations while we do this. Note that
804 // computing relocation info will also update the fixup to have the correct
805 // value; this will overwrite the appropriate data in the fragment when it
807 std::vector<MachRelocationEntry> RelocInfos;
808 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
809 for (MCAssembler::iterator it = Asm.begin(),
810 ie = Asm.end(); it != ie; ++it) {
811 MCSectionData &SD = *it;
813 // The assembler writes relocations in the reverse order they were seen.
815 // FIXME: It is probably more complicated than this.
816 unsigned NumRelocsStart = RelocInfos.size();
817 for (MCSectionData::reverse_iterator it2 = SD.rbegin(),
818 ie2 = SD.rend(); it2 != ie2; ++it2)
819 if (MCDataFragment *DF = dyn_cast<MCDataFragment>(&*it2))
820 for (unsigned i = 0, e = DF->fixup_size(); i != e; ++i)
821 ComputeRelocationInfo(Asm, *DF, DF->getFixups()[e - i - 1],
822 SymbolMap, RelocInfos);
824 unsigned NumRelocs = RelocInfos.size() - NumRelocsStart;
825 uint64_t SectionStart = SectionDataStart + SD.getAddress();
826 WriteSection32(SD, SectionStart, RelocTableEnd, NumRelocs);
827 RelocTableEnd += NumRelocs * RelocationInfoSize;
830 // Write the symbol table load command, if used.
832 unsigned FirstLocalSymbol = 0;
833 unsigned NumLocalSymbols = LocalSymbolData.size();
834 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
835 unsigned NumExternalSymbols = ExternalSymbolData.size();
836 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
837 unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
838 unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
839 unsigned NumSymTabSymbols =
840 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
841 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
842 uint64_t IndirectSymbolOffset = 0;
844 // If used, the indirect symbols are written after the section data.
845 if (NumIndirectSymbols)
846 IndirectSymbolOffset = RelocTableEnd;
848 // The symbol table is written after the indirect symbol data.
849 uint64_t SymbolTableOffset = RelocTableEnd + IndirectSymbolSize;
851 // The string table is written after symbol table.
852 uint64_t StringTableOffset =
853 SymbolTableOffset + NumSymTabSymbols * Nlist32Size;
854 WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
855 StringTableOffset, StringTable.size());
857 WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
858 FirstExternalSymbol, NumExternalSymbols,
859 FirstUndefinedSymbol, NumUndefinedSymbols,
860 IndirectSymbolOffset, NumIndirectSymbols);
863 // Write the actual section data.
864 for (MCAssembler::iterator it = Asm.begin(), ie = Asm.end(); it != ie; ++it)
865 WriteFileData(OS, *it, *this);
867 // Write the extra padding.
868 WriteZeros(SectionDataPadding);
870 // Write the relocation entries.
871 for (unsigned i = 0, e = RelocInfos.size(); i != e; ++i) {
872 Write32(RelocInfos[i].Word0);
873 Write32(RelocInfos[i].Word1);
876 // Write the symbol table data, if used.
878 // Write the indirect symbol entries.
879 for (MCAssembler::indirect_symbol_iterator
880 it = Asm.indirect_symbol_begin(),
881 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
882 // Indirect symbols in the non lazy symbol pointer section have some
884 const MCSectionMachO &Section =
885 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
887 Section.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
888 if (Type == MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS) {
889 // If this symbol is defined and internal, mark it as such.
890 if (it->Symbol->isDefined() &&
891 !SymbolMap.lookup(it->Symbol)->isExternal()) {
892 uint32_t Flags = ISF_Local;
893 if (it->Symbol->isAbsolute())
894 Flags |= ISF_Absolute;
900 Write32(SymbolMap[it->Symbol]->getIndex());
903 // FIXME: Check that offsets match computed ones.
905 // Write the symbol table entries.
906 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
907 WriteNlist32(LocalSymbolData[i]);
908 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
909 WriteNlist32(ExternalSymbolData[i]);
910 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
911 WriteNlist32(UndefinedSymbolData[i]);
913 // Write the string table.
914 OS << StringTable.str();
918 void ApplyFixup(const MCAsmFixup &Fixup, MCDataFragment &DF) {
919 unsigned Size = 1 << getFixupKindLog2Size(Fixup.Kind);
921 // FIXME: Endianness assumption.
922 assert(Fixup.Offset + Size <= DF.getContents().size() &&
923 "Invalid fixup offset!");
924 for (unsigned i = 0; i != Size; ++i)
925 DF.getContents()[Fixup.Offset + i] = uint8_t(Fixup.FixedValue >> (i * 8));
931 MCFragment::MCFragment() : Kind(FragmentType(~0)) {
934 MCFragment::MCFragment(FragmentType _Kind, MCSectionData *_Parent)
937 FileSize(~UINT64_C(0))
940 Parent->getFragmentList().push_back(this);
943 MCFragment::~MCFragment() {
946 uint64_t MCFragment::getAddress() const {
947 assert(getParent() && "Missing Section!");
948 return getParent()->getAddress() + Offset;
953 MCSectionData::MCSectionData() : Section(0) {}
955 MCSectionData::MCSectionData(const MCSection &_Section, MCAssembler *A)
956 : Section(&_Section),
958 Address(~UINT64_C(0)),
960 FileSize(~UINT64_C(0)),
961 HasInstructions(false)
964 A->getSectionList().push_back(this);
969 MCSymbolData::MCSymbolData() : Symbol(0) {}
971 MCSymbolData::MCSymbolData(const MCSymbol &_Symbol, MCFragment *_Fragment,
972 uint64_t _Offset, MCAssembler *A)
973 : Symbol(&_Symbol), Fragment(_Fragment), Offset(_Offset),
974 IsExternal(false), IsPrivateExtern(false),
975 CommonSize(0), CommonAlign(0), Flags(0), Index(0)
978 A->getSymbolList().push_back(this);
983 MCAssembler::MCAssembler(MCContext &_Context, raw_ostream &_OS)
984 : Context(_Context), OS(_OS), SubsectionsViaSymbols(false)
988 MCAssembler::~MCAssembler() {
991 void MCAssembler::LayoutSection(MCSectionData &SD) {
992 uint64_t Address = SD.getAddress();
994 for (MCSectionData::iterator it = SD.begin(), ie = SD.end(); it != ie; ++it) {
997 F.setOffset(Address - SD.getAddress());
999 // Evaluate fragment size.
1000 switch (F.getKind()) {
1001 case MCFragment::FT_Align: {
1002 MCAlignFragment &AF = cast<MCAlignFragment>(F);
1004 uint64_t Size = OffsetToAlignment(Address, AF.getAlignment());
1005 if (Size > AF.getMaxBytesToEmit())
1008 AF.setFileSize(Size);
1012 case MCFragment::FT_Data:
1013 case MCFragment::FT_Fill:
1014 F.setFileSize(F.getMaxFileSize());
1017 case MCFragment::FT_Org: {
1018 MCOrgFragment &OF = cast<MCOrgFragment>(F);
1021 if (!OF.getOffset().EvaluateAsRelocatable(Target))
1022 llvm_report_error("expected relocatable expression");
1024 if (!Target.isAbsolute())
1025 llvm_unreachable("FIXME: Not yet implemented!");
1026 uint64_t OrgOffset = Target.getConstant();
1027 uint64_t Offset = Address - SD.getAddress();
1029 // FIXME: We need a way to communicate this error.
1030 if (OrgOffset < Offset)
1031 llvm_report_error("invalid .org offset '" + Twine(OrgOffset) +
1032 "' (at offset '" + Twine(Offset) + "'");
1034 F.setFileSize(OrgOffset - Offset);
1038 case MCFragment::FT_ZeroFill: {
1039 MCZeroFillFragment &ZFF = cast<MCZeroFillFragment>(F);
1041 // Align the fragment offset; it is safe to adjust the offset freely since
1042 // this is only in virtual sections.
1043 uint64_t Aligned = RoundUpToAlignment(Address, ZFF.getAlignment());
1044 F.setOffset(Aligned - SD.getAddress());
1046 // FIXME: This is misnamed.
1047 F.setFileSize(ZFF.getSize());
1052 Address += F.getFileSize();
1055 // Set the section sizes.
1056 SD.setSize(Address - SD.getAddress());
1057 if (isVirtualSection(SD.getSection()))
1060 SD.setFileSize(Address - SD.getAddress());
1063 /// WriteFileData - Write the \arg F data to the output file.
1064 static void WriteFileData(raw_ostream &OS, const MCFragment &F,
1065 MachObjectWriter &MOW) {
1066 uint64_t Start = OS.tell();
1071 // FIXME: Embed in fragments instead?
1072 switch (F.getKind()) {
1073 case MCFragment::FT_Align: {
1074 MCAlignFragment &AF = cast<MCAlignFragment>(F);
1075 uint64_t Count = AF.getFileSize() / AF.getValueSize();
1077 // FIXME: This error shouldn't actually occur (the front end should emit
1078 // multiple .align directives to enforce the semantics it wants), but is
1079 // severe enough that we want to report it. How to handle this?
1080 if (Count * AF.getValueSize() != AF.getFileSize())
1081 llvm_report_error("undefined .align directive, value size '" +
1082 Twine(AF.getValueSize()) +
1083 "' is not a divisor of padding size '" +
1084 Twine(AF.getFileSize()) + "'");
1086 for (uint64_t i = 0; i != Count; ++i) {
1087 switch (AF.getValueSize()) {
1089 assert(0 && "Invalid size!");
1090 case 1: MOW.Write8 (uint8_t (AF.getValue())); break;
1091 case 2: MOW.Write16(uint16_t(AF.getValue())); break;
1092 case 4: MOW.Write32(uint32_t(AF.getValue())); break;
1093 case 8: MOW.Write64(uint64_t(AF.getValue())); break;
1099 case MCFragment::FT_Data: {
1100 MCDataFragment &DF = cast<MCDataFragment>(F);
1102 // Apply the fixups.
1104 // FIXME: Move elsewhere.
1105 for (MCDataFragment::const_fixup_iterator it = DF.fixup_begin(),
1106 ie = DF.fixup_end(); it != ie; ++it)
1107 MOW.ApplyFixup(*it, DF);
1109 OS << cast<MCDataFragment>(F).getContents().str();
1113 case MCFragment::FT_Fill: {
1114 MCFillFragment &FF = cast<MCFillFragment>(F);
1115 for (uint64_t i = 0, e = FF.getCount(); i != e; ++i) {
1116 switch (FF.getValueSize()) {
1118 assert(0 && "Invalid size!");
1119 case 1: MOW.Write8 (uint8_t (FF.getValue())); break;
1120 case 2: MOW.Write16(uint16_t(FF.getValue())); break;
1121 case 4: MOW.Write32(uint32_t(FF.getValue())); break;
1122 case 8: MOW.Write64(uint64_t(FF.getValue())); break;
1128 case MCFragment::FT_Org: {
1129 MCOrgFragment &OF = cast<MCOrgFragment>(F);
1131 for (uint64_t i = 0, e = OF.getFileSize(); i != e; ++i)
1132 MOW.Write8(uint8_t(OF.getValue()));
1137 case MCFragment::FT_ZeroFill: {
1138 assert(0 && "Invalid zero fill fragment in concrete section!");
1143 assert(OS.tell() - Start == F.getFileSize());
1146 /// WriteFileData - Write the \arg SD data to the output file.
1147 static void WriteFileData(raw_ostream &OS, const MCSectionData &SD,
1148 MachObjectWriter &MOW) {
1149 // Ignore virtual sections.
1150 if (isVirtualSection(SD.getSection())) {
1151 assert(SD.getFileSize() == 0);
1155 uint64_t Start = OS.tell();
1158 for (MCSectionData::const_iterator it = SD.begin(),
1159 ie = SD.end(); it != ie; ++it)
1160 WriteFileData(OS, *it, MOW);
1162 // Add section padding.
1163 assert(SD.getFileSize() >= SD.getSize() && "Invalid section sizes!");
1164 MOW.WriteZeros(SD.getFileSize() - SD.getSize());
1166 assert(OS.tell() - Start == SD.getFileSize());
1169 void MCAssembler::Finish() {
1170 DEBUG_WITH_TYPE("mc-dump", {
1171 llvm::errs() << "assembler backend - pre-layout\n--\n";
1174 // Layout the concrete sections and fragments.
1175 uint64_t Address = 0;
1176 MCSectionData *Prev = 0;
1177 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1178 MCSectionData &SD = *it;
1180 // Skip virtual sections.
1181 if (isVirtualSection(SD.getSection()))
1184 // Align this section if necessary by adding padding bytes to the previous
1186 if (uint64_t Pad = OffsetToAlignment(Address, it->getAlignment())) {
1187 assert(Prev && "Missing prev section!");
1188 Prev->setFileSize(Prev->getFileSize() + Pad);
1192 // Layout the section fragments and its size.
1193 SD.setAddress(Address);
1195 Address += SD.getFileSize();
1200 // Layout the virtual sections.
1201 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1202 MCSectionData &SD = *it;
1204 if (!isVirtualSection(SD.getSection()))
1207 SD.setAddress(Address);
1209 Address += SD.getSize();
1212 DEBUG_WITH_TYPE("mc-dump", {
1213 llvm::errs() << "assembler backend - post-layout\n--\n";
1216 // Write the object file.
1217 MachObjectWriter MOW(OS);
1218 MOW.WriteObject(*this);
1224 // Debugging methods
1228 raw_ostream &operator<<(raw_ostream &OS, const MCAsmFixup &AF) {
1229 OS << "<MCAsmFixup" << " Offset:" << AF.Offset << " Value:" << *AF.Value
1230 << " Kind:" << AF.Kind << ">";
1236 void MCFragment::dump() {
1237 raw_ostream &OS = llvm::errs();
1239 OS << "<MCFragment " << (void*) this << " Offset:" << Offset
1240 << " FileSize:" << FileSize;
1245 void MCAlignFragment::dump() {
1246 raw_ostream &OS = llvm::errs();
1248 OS << "<MCAlignFragment ";
1249 this->MCFragment::dump();
1251 OS << " Alignment:" << getAlignment()
1252 << " Value:" << getValue() << " ValueSize:" << getValueSize()
1253 << " MaxBytesToEmit:" << getMaxBytesToEmit() << ">";
1256 void MCDataFragment::dump() {
1257 raw_ostream &OS = llvm::errs();
1259 OS << "<MCDataFragment ";
1260 this->MCFragment::dump();
1262 OS << " Contents:[";
1263 for (unsigned i = 0, e = getContents().size(); i != e; ++i) {
1265 OS << hexdigit((Contents[i] >> 4) & 0xF) << hexdigit(Contents[i] & 0xF);
1267 OS << "] (" << getContents().size() << " bytes)";
1269 if (!getFixups().empty()) {
1272 for (fixup_iterator it = fixup_begin(), ie = fixup_end(); it != ie; ++it) {
1273 if (it != fixup_begin()) OS << ",\n ";
1282 void MCFillFragment::dump() {
1283 raw_ostream &OS = llvm::errs();
1285 OS << "<MCFillFragment ";
1286 this->MCFragment::dump();
1288 OS << " Value:" << getValue() << " ValueSize:" << getValueSize()
1289 << " Count:" << getCount() << ">";
1292 void MCOrgFragment::dump() {
1293 raw_ostream &OS = llvm::errs();
1295 OS << "<MCOrgFragment ";
1296 this->MCFragment::dump();
1298 OS << " Offset:" << getOffset() << " Value:" << getValue() << ">";
1301 void MCZeroFillFragment::dump() {
1302 raw_ostream &OS = llvm::errs();
1304 OS << "<MCZeroFillFragment ";
1305 this->MCFragment::dump();
1307 OS << " Size:" << getSize() << " Alignment:" << getAlignment() << ">";
1310 void MCSectionData::dump() {
1311 raw_ostream &OS = llvm::errs();
1313 OS << "<MCSectionData";
1314 OS << " Alignment:" << getAlignment() << " Address:" << Address
1315 << " Size:" << Size << " FileSize:" << FileSize
1317 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1318 if (it != begin()) OS << ",\n ";
1324 void MCSymbolData::dump() {
1325 raw_ostream &OS = llvm::errs();
1327 OS << "<MCSymbolData Symbol:" << getSymbol()
1328 << " Fragment:" << getFragment() << " Offset:" << getOffset()
1329 << " Flags:" << getFlags() << " Index:" << getIndex();
1331 OS << " (common, size:" << getCommonSize()
1332 << " align: " << getCommonAlignment() << ")";
1334 OS << " (external)";
1335 if (isPrivateExtern())
1336 OS << " (private extern)";
1340 void MCAssembler::dump() {
1341 raw_ostream &OS = llvm::errs();
1343 OS << "<MCAssembler\n";
1344 OS << " Sections:[";
1345 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1346 if (it != begin()) OS << ",\n ";
1352 for (symbol_iterator it = symbol_begin(), ie = symbol_end(); it != ie; ++it) {
1353 if (it != symbol_begin()) OS << ",\n ";