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 Log2Size = getFixupKindLog2Size(Fixup.Kind);
443 unsigned Type = RIT_Vanilla;
446 const MCSymbol *A = Target.getSymA();
447 MCSymbolData *SD = SymbolMap.lookup(A);
448 uint32_t Value = SD->getFragment()->getAddress() + SD->getOffset();
451 if (const MCSymbol *B = Target.getSymB()) {
452 Type = RIT_LocalDifference;
454 MCSymbolData *SD = SymbolMap.lookup(B);
455 Value2 = SD->getFragment()->getAddress() + SD->getOffset();
458 // The value which goes in the fixup is current value of the expression.
459 Fixup.FixedValue = Value - Value2 + Target.getConstant();
460 if (isFixupKindPCRel(Fixup.Kind)) {
461 Fixup.FixedValue -= Address;
465 MachRelocationEntry MRE;
466 MRE.Word0 = ((Address << 0) |
472 Relocs.push_back(MRE);
474 if (Type == RIT_LocalDifference) {
477 MachRelocationEntry MRE;
478 MRE.Word0 = ((0 << 0) |
484 Relocs.push_back(MRE);
488 void ComputeRelocationInfo(MCAssembler &Asm, MCDataFragment &Fragment,
490 DenseMap<const MCSymbol*,MCSymbolData*> &SymbolMap,
491 std::vector<MachRelocationEntry> &Relocs) {
493 if (!Fixup.Value->EvaluateAsRelocatable(Target))
494 llvm_report_error("expected relocatable expression");
496 // If this is a difference or a local symbol plus an offset, then we need a
497 // scattered relocation entry.
498 if (Target.getSymB() ||
499 (Target.getSymA() && !Target.getSymA()->isUndefined() &&
500 Target.getConstant()))
501 return ComputeScatteredRelocationInfo(Asm, Fragment, Fixup, Target,
505 uint32_t Address = Fragment.getOffset() + Fixup.Offset;
508 unsigned IsPCRel = 0;
509 unsigned Log2Size = getFixupKindLog2Size(Fixup.Kind);
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 if (isFixupKindPCRel(Fixup.Kind)) {
548 Fixup.FixedValue -= Address;
552 // struct relocation_info (8 bytes)
553 MachRelocationEntry MRE;
555 MRE.Word1 = ((Index << 0) |
560 Relocs.push_back(MRE);
563 void BindIndirectSymbols(MCAssembler &Asm,
564 DenseMap<const MCSymbol*,MCSymbolData*> &SymbolMap) {
565 // This is the point where 'as' creates actual symbols for indirect symbols
566 // (in the following two passes). It would be easier for us to do this
567 // sooner when we see the attribute, but that makes getting the order in the
568 // symbol table much more complicated than it is worth.
570 // FIXME: Revisit this when the dust settles.
572 // Bind non lazy symbol pointers first.
573 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
574 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
575 // FIXME: cast<> support!
576 const MCSectionMachO &Section =
577 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
580 Section.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
581 if (Type != MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS)
584 MCSymbolData *&Entry = SymbolMap[it->Symbol];
586 Entry = new MCSymbolData(*it->Symbol, 0, 0, &Asm);
589 // Then lazy symbol pointers and symbol stubs.
590 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
591 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
592 // FIXME: cast<> support!
593 const MCSectionMachO &Section =
594 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
597 Section.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
598 if (Type != MCSectionMachO::S_LAZY_SYMBOL_POINTERS &&
599 Type != MCSectionMachO::S_SYMBOL_STUBS)
602 MCSymbolData *&Entry = SymbolMap[it->Symbol];
604 Entry = new MCSymbolData(*it->Symbol, 0, 0, &Asm);
606 // Set the symbol type to undefined lazy, but only on construction.
608 // FIXME: Do not hardcode.
609 Entry->setFlags(Entry->getFlags() | 0x0001);
614 /// ComputeSymbolTable - Compute the symbol table data
616 /// \param StringTable [out] - The string table data.
617 /// \param StringIndexMap [out] - Map from symbol names to offsets in the
619 void ComputeSymbolTable(MCAssembler &Asm, SmallString<256> &StringTable,
620 std::vector<MachSymbolData> &LocalSymbolData,
621 std::vector<MachSymbolData> &ExternalSymbolData,
622 std::vector<MachSymbolData> &UndefinedSymbolData) {
623 // Build section lookup table.
624 DenseMap<const MCSection*, uint8_t> SectionIndexMap;
626 for (MCAssembler::iterator it = Asm.begin(),
627 ie = Asm.end(); it != ie; ++it, ++Index)
628 SectionIndexMap[&it->getSection()] = Index;
629 assert(Index <= 256 && "Too many sections!");
631 // Index 0 is always the empty string.
632 StringMap<uint64_t> StringIndexMap;
633 StringTable += '\x00';
635 // Build the symbol arrays and the string table, but only for non-local
638 // The particular order that we collect the symbols and create the string
639 // table, then sort the symbols is chosen to match 'as'. Even though it
640 // doesn't matter for correctness, this is important for letting us diff .o
642 for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
643 ie = Asm.symbol_end(); it != ie; ++it) {
644 const MCSymbol &Symbol = it->getSymbol();
646 // Ignore assembler temporaries.
647 if (it->getSymbol().isTemporary())
650 if (!it->isExternal() && !Symbol.isUndefined())
653 uint64_t &Entry = StringIndexMap[Symbol.getName()];
655 Entry = StringTable.size();
656 StringTable += Symbol.getName();
657 StringTable += '\x00';
662 MSD.StringIndex = Entry;
664 if (Symbol.isUndefined()) {
665 MSD.SectionIndex = 0;
666 UndefinedSymbolData.push_back(MSD);
667 } else if (Symbol.isAbsolute()) {
668 MSD.SectionIndex = 0;
669 ExternalSymbolData.push_back(MSD);
671 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
672 assert(MSD.SectionIndex && "Invalid section index!");
673 ExternalSymbolData.push_back(MSD);
677 // Now add the data for local symbols.
678 for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
679 ie = Asm.symbol_end(); it != ie; ++it) {
680 const MCSymbol &Symbol = it->getSymbol();
682 // Ignore assembler temporaries.
683 if (it->getSymbol().isTemporary())
686 if (it->isExternal() || Symbol.isUndefined())
689 uint64_t &Entry = StringIndexMap[Symbol.getName()];
691 Entry = StringTable.size();
692 StringTable += Symbol.getName();
693 StringTable += '\x00';
698 MSD.StringIndex = Entry;
700 if (Symbol.isAbsolute()) {
701 MSD.SectionIndex = 0;
702 LocalSymbolData.push_back(MSD);
704 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
705 assert(MSD.SectionIndex && "Invalid section index!");
706 LocalSymbolData.push_back(MSD);
710 // External and undefined symbols are required to be in lexicographic order.
711 std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
712 std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
714 // Set the symbol indices.
716 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
717 LocalSymbolData[i].SymbolData->setIndex(Index++);
718 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
719 ExternalSymbolData[i].SymbolData->setIndex(Index++);
720 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
721 UndefinedSymbolData[i].SymbolData->setIndex(Index++);
723 // The string table is padded to a multiple of 4.
724 while (StringTable.size() % 4)
725 StringTable += '\x00';
728 void WriteObject(MCAssembler &Asm) {
729 unsigned NumSections = Asm.size();
731 // Compute the symbol -> symbol data map.
733 // FIXME: This should not be here.
734 DenseMap<const MCSymbol*, MCSymbolData *> SymbolMap;
735 for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
736 ie = Asm.symbol_end(); it != ie; ++it)
737 SymbolMap[&it->getSymbol()] = it;
739 // Create symbol data for any indirect symbols.
740 BindIndirectSymbols(Asm, SymbolMap);
742 // Compute symbol table information.
743 SmallString<256> StringTable;
744 std::vector<MachSymbolData> LocalSymbolData;
745 std::vector<MachSymbolData> ExternalSymbolData;
746 std::vector<MachSymbolData> UndefinedSymbolData;
747 unsigned NumSymbols = Asm.symbol_size();
749 // No symbol table command is written if there are no symbols.
751 ComputeSymbolTable(Asm, StringTable, LocalSymbolData, ExternalSymbolData,
752 UndefinedSymbolData);
754 // The section data starts after the header, the segment load command (and
755 // section headers) and the symbol table.
756 unsigned NumLoadCommands = 1;
757 uint64_t LoadCommandsSize =
758 SegmentLoadCommand32Size + NumSections * Section32Size;
760 // Add the symbol table load command sizes, if used.
762 NumLoadCommands += 2;
763 LoadCommandsSize += SymtabLoadCommandSize + DysymtabLoadCommandSize;
766 // Compute the total size of the section data, as well as its file size and
768 uint64_t SectionDataStart = Header32Size + LoadCommandsSize;
769 uint64_t SectionDataSize = 0;
770 uint64_t SectionDataFileSize = 0;
772 for (MCAssembler::iterator it = Asm.begin(),
773 ie = Asm.end(); it != ie; ++it) {
774 MCSectionData &SD = *it;
776 VMSize = std::max(VMSize, SD.getAddress() + SD.getSize());
778 if (isVirtualSection(SD.getSection()))
781 SectionDataSize = std::max(SectionDataSize,
782 SD.getAddress() + SD.getSize());
783 SectionDataFileSize = std::max(SectionDataFileSize,
784 SD.getAddress() + SD.getFileSize());
787 // The section data is padded to 4 bytes.
789 // FIXME: Is this machine dependent?
790 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
791 SectionDataFileSize += SectionDataPadding;
793 // Write the prolog, starting with the header and load command...
794 WriteHeader32(NumLoadCommands, LoadCommandsSize,
795 Asm.getSubsectionsViaSymbols());
796 WriteSegmentLoadCommand32(NumSections, VMSize,
797 SectionDataStart, SectionDataSize);
799 // ... and then the section headers.
801 // We also compute the section relocations while we do this. Note that
802 // computing relocation info will also update the fixup to have the correct
803 // value; this will overwrite the appropriate data in the fragment when it
805 std::vector<MachRelocationEntry> RelocInfos;
806 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
807 for (MCAssembler::iterator it = Asm.begin(),
808 ie = Asm.end(); it != ie; ++it) {
809 MCSectionData &SD = *it;
811 // The assembler writes relocations in the reverse order they were seen.
813 // FIXME: It is probably more complicated than this.
814 unsigned NumRelocsStart = RelocInfos.size();
815 for (MCSectionData::reverse_iterator it2 = SD.rbegin(),
816 ie2 = SD.rend(); it2 != ie2; ++it2)
817 if (MCDataFragment *DF = dyn_cast<MCDataFragment>(&*it2))
818 for (unsigned i = 0, e = DF->fixup_size(); i != e; ++i)
819 ComputeRelocationInfo(Asm, *DF, DF->getFixups()[e - i - 1],
820 SymbolMap, RelocInfos);
822 unsigned NumRelocs = RelocInfos.size() - NumRelocsStart;
823 uint64_t SectionStart = SectionDataStart + SD.getAddress();
824 WriteSection32(SD, SectionStart, RelocTableEnd, NumRelocs);
825 RelocTableEnd += NumRelocs * RelocationInfoSize;
828 // Write the symbol table load command, if used.
830 unsigned FirstLocalSymbol = 0;
831 unsigned NumLocalSymbols = LocalSymbolData.size();
832 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
833 unsigned NumExternalSymbols = ExternalSymbolData.size();
834 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
835 unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
836 unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
837 unsigned NumSymTabSymbols =
838 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
839 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
840 uint64_t IndirectSymbolOffset = 0;
842 // If used, the indirect symbols are written after the section data.
843 if (NumIndirectSymbols)
844 IndirectSymbolOffset = RelocTableEnd;
846 // The symbol table is written after the indirect symbol data.
847 uint64_t SymbolTableOffset = RelocTableEnd + IndirectSymbolSize;
849 // The string table is written after symbol table.
850 uint64_t StringTableOffset =
851 SymbolTableOffset + NumSymTabSymbols * Nlist32Size;
852 WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
853 StringTableOffset, StringTable.size());
855 WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
856 FirstExternalSymbol, NumExternalSymbols,
857 FirstUndefinedSymbol, NumUndefinedSymbols,
858 IndirectSymbolOffset, NumIndirectSymbols);
861 // Write the actual section data.
862 for (MCAssembler::iterator it = Asm.begin(), ie = Asm.end(); it != ie; ++it)
863 WriteFileData(OS, *it, *this);
865 // Write the extra padding.
866 WriteZeros(SectionDataPadding);
868 // Write the relocation entries.
869 for (unsigned i = 0, e = RelocInfos.size(); i != e; ++i) {
870 Write32(RelocInfos[i].Word0);
871 Write32(RelocInfos[i].Word1);
874 // Write the symbol table data, if used.
876 // Write the indirect symbol entries.
877 for (MCAssembler::indirect_symbol_iterator
878 it = Asm.indirect_symbol_begin(),
879 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
880 // Indirect symbols in the non lazy symbol pointer section have some
882 const MCSectionMachO &Section =
883 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
885 Section.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
886 if (Type == MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS) {
887 // If this symbol is defined and internal, mark it as such.
888 if (it->Symbol->isDefined() &&
889 !SymbolMap.lookup(it->Symbol)->isExternal()) {
890 uint32_t Flags = ISF_Local;
891 if (it->Symbol->isAbsolute())
892 Flags |= ISF_Absolute;
898 Write32(SymbolMap[it->Symbol]->getIndex());
901 // FIXME: Check that offsets match computed ones.
903 // Write the symbol table entries.
904 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
905 WriteNlist32(LocalSymbolData[i]);
906 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
907 WriteNlist32(ExternalSymbolData[i]);
908 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
909 WriteNlist32(UndefinedSymbolData[i]);
911 // Write the string table.
912 OS << StringTable.str();
916 void ApplyFixup(const MCAsmFixup &Fixup, MCDataFragment &DF) {
917 unsigned Size = 1 << getFixupKindLog2Size(Fixup.Kind);
919 // FIXME: Endianness assumption.
920 assert(Fixup.Offset + Size <= DF.getContents().size() &&
921 "Invalid fixup offset!");
922 for (unsigned i = 0; i != Size; ++i)
923 DF.getContents()[Fixup.Offset + i] = uint8_t(Fixup.FixedValue >> (i * 8));
929 MCFragment::MCFragment() : Kind(FragmentType(~0)) {
932 MCFragment::MCFragment(FragmentType _Kind, MCSectionData *_Parent)
935 FileSize(~UINT64_C(0))
938 Parent->getFragmentList().push_back(this);
941 MCFragment::~MCFragment() {
944 uint64_t MCFragment::getAddress() const {
945 assert(getParent() && "Missing Section!");
946 return getParent()->getAddress() + Offset;
951 MCSectionData::MCSectionData() : Section(0) {}
953 MCSectionData::MCSectionData(const MCSection &_Section, MCAssembler *A)
954 : Section(&_Section),
956 Address(~UINT64_C(0)),
958 FileSize(~UINT64_C(0)),
959 HasInstructions(false)
962 A->getSectionList().push_back(this);
967 MCSymbolData::MCSymbolData() : Symbol(0) {}
969 MCSymbolData::MCSymbolData(const MCSymbol &_Symbol, MCFragment *_Fragment,
970 uint64_t _Offset, MCAssembler *A)
971 : Symbol(&_Symbol), Fragment(_Fragment), Offset(_Offset),
972 IsExternal(false), IsPrivateExtern(false),
973 CommonSize(0), CommonAlign(0), Flags(0), Index(0)
976 A->getSymbolList().push_back(this);
981 MCAssembler::MCAssembler(MCContext &_Context, raw_ostream &_OS)
982 : Context(_Context), OS(_OS), SubsectionsViaSymbols(false)
986 MCAssembler::~MCAssembler() {
989 void MCAssembler::LayoutSection(MCSectionData &SD) {
990 uint64_t Address = SD.getAddress();
992 for (MCSectionData::iterator it = SD.begin(), ie = SD.end(); it != ie; ++it) {
995 F.setOffset(Address - SD.getAddress());
997 // Evaluate fragment size.
998 switch (F.getKind()) {
999 case MCFragment::FT_Align: {
1000 MCAlignFragment &AF = cast<MCAlignFragment>(F);
1002 uint64_t Size = OffsetToAlignment(Address, AF.getAlignment());
1003 if (Size > AF.getMaxBytesToEmit())
1006 AF.setFileSize(Size);
1010 case MCFragment::FT_Data:
1011 case MCFragment::FT_Fill:
1012 F.setFileSize(F.getMaxFileSize());
1015 case MCFragment::FT_Org: {
1016 MCOrgFragment &OF = cast<MCOrgFragment>(F);
1019 if (!OF.getOffset().EvaluateAsRelocatable(Target))
1020 llvm_report_error("expected relocatable expression");
1022 if (!Target.isAbsolute())
1023 llvm_unreachable("FIXME: Not yet implemented!");
1024 uint64_t OrgOffset = Target.getConstant();
1025 uint64_t Offset = Address - SD.getAddress();
1027 // FIXME: We need a way to communicate this error.
1028 if (OrgOffset < Offset)
1029 llvm_report_error("invalid .org offset '" + Twine(OrgOffset) +
1030 "' (at offset '" + Twine(Offset) + "'");
1032 F.setFileSize(OrgOffset - Offset);
1036 case MCFragment::FT_ZeroFill: {
1037 MCZeroFillFragment &ZFF = cast<MCZeroFillFragment>(F);
1039 // Align the fragment offset; it is safe to adjust the offset freely since
1040 // this is only in virtual sections.
1041 uint64_t Aligned = RoundUpToAlignment(Address, ZFF.getAlignment());
1042 F.setOffset(Aligned - SD.getAddress());
1044 // FIXME: This is misnamed.
1045 F.setFileSize(ZFF.getSize());
1050 Address += F.getFileSize();
1053 // Set the section sizes.
1054 SD.setSize(Address - SD.getAddress());
1055 if (isVirtualSection(SD.getSection()))
1058 SD.setFileSize(Address - SD.getAddress());
1061 /// WriteFileData - Write the \arg F data to the output file.
1062 static void WriteFileData(raw_ostream &OS, const MCFragment &F,
1063 MachObjectWriter &MOW) {
1064 uint64_t Start = OS.tell();
1069 // FIXME: Embed in fragments instead?
1070 switch (F.getKind()) {
1071 case MCFragment::FT_Align: {
1072 MCAlignFragment &AF = cast<MCAlignFragment>(F);
1073 uint64_t Count = AF.getFileSize() / AF.getValueSize();
1075 // FIXME: This error shouldn't actually occur (the front end should emit
1076 // multiple .align directives to enforce the semantics it wants), but is
1077 // severe enough that we want to report it. How to handle this?
1078 if (Count * AF.getValueSize() != AF.getFileSize())
1079 llvm_report_error("undefined .align directive, value size '" +
1080 Twine(AF.getValueSize()) +
1081 "' is not a divisor of padding size '" +
1082 Twine(AF.getFileSize()) + "'");
1084 for (uint64_t i = 0; i != Count; ++i) {
1085 switch (AF.getValueSize()) {
1087 assert(0 && "Invalid size!");
1088 case 1: MOW.Write8 (uint8_t (AF.getValue())); break;
1089 case 2: MOW.Write16(uint16_t(AF.getValue())); break;
1090 case 4: MOW.Write32(uint32_t(AF.getValue())); break;
1091 case 8: MOW.Write64(uint64_t(AF.getValue())); break;
1097 case MCFragment::FT_Data: {
1098 MCDataFragment &DF = cast<MCDataFragment>(F);
1100 // Apply the fixups.
1102 // FIXME: Move elsewhere.
1103 for (MCDataFragment::const_fixup_iterator it = DF.fixup_begin(),
1104 ie = DF.fixup_end(); it != ie; ++it)
1105 MOW.ApplyFixup(*it, DF);
1107 OS << cast<MCDataFragment>(F).getContents().str();
1111 case MCFragment::FT_Fill: {
1112 MCFillFragment &FF = cast<MCFillFragment>(F);
1113 for (uint64_t i = 0, e = FF.getCount(); i != e; ++i) {
1114 switch (FF.getValueSize()) {
1116 assert(0 && "Invalid size!");
1117 case 1: MOW.Write8 (uint8_t (FF.getValue())); break;
1118 case 2: MOW.Write16(uint16_t(FF.getValue())); break;
1119 case 4: MOW.Write32(uint32_t(FF.getValue())); break;
1120 case 8: MOW.Write64(uint64_t(FF.getValue())); break;
1126 case MCFragment::FT_Org: {
1127 MCOrgFragment &OF = cast<MCOrgFragment>(F);
1129 for (uint64_t i = 0, e = OF.getFileSize(); i != e; ++i)
1130 MOW.Write8(uint8_t(OF.getValue()));
1135 case MCFragment::FT_ZeroFill: {
1136 assert(0 && "Invalid zero fill fragment in concrete section!");
1141 assert(OS.tell() - Start == F.getFileSize());
1144 /// WriteFileData - Write the \arg SD data to the output file.
1145 static void WriteFileData(raw_ostream &OS, const MCSectionData &SD,
1146 MachObjectWriter &MOW) {
1147 // Ignore virtual sections.
1148 if (isVirtualSection(SD.getSection())) {
1149 assert(SD.getFileSize() == 0);
1153 uint64_t Start = OS.tell();
1156 for (MCSectionData::const_iterator it = SD.begin(),
1157 ie = SD.end(); it != ie; ++it)
1158 WriteFileData(OS, *it, MOW);
1160 // Add section padding.
1161 assert(SD.getFileSize() >= SD.getSize() && "Invalid section sizes!");
1162 MOW.WriteZeros(SD.getFileSize() - SD.getSize());
1164 assert(OS.tell() - Start == SD.getFileSize());
1167 void MCAssembler::Finish() {
1168 DEBUG_WITH_TYPE("mc-dump", {
1169 llvm::errs() << "assembler backend - pre-layout\n--\n";
1172 // Layout the concrete sections and fragments.
1173 uint64_t Address = 0;
1174 MCSectionData *Prev = 0;
1175 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1176 MCSectionData &SD = *it;
1178 // Skip virtual sections.
1179 if (isVirtualSection(SD.getSection()))
1182 // Align this section if necessary by adding padding bytes to the previous
1184 if (uint64_t Pad = OffsetToAlignment(Address, it->getAlignment())) {
1185 assert(Prev && "Missing prev section!");
1186 Prev->setFileSize(Prev->getFileSize() + Pad);
1190 // Layout the section fragments and its size.
1191 SD.setAddress(Address);
1193 Address += SD.getFileSize();
1198 // Layout the virtual sections.
1199 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1200 MCSectionData &SD = *it;
1202 if (!isVirtualSection(SD.getSection()))
1205 SD.setAddress(Address);
1207 Address += SD.getSize();
1210 DEBUG_WITH_TYPE("mc-dump", {
1211 llvm::errs() << "assembler backend - post-layout\n--\n";
1214 // Write the object file.
1215 MachObjectWriter MOW(OS);
1216 MOW.WriteObject(*this);
1222 // Debugging methods
1226 raw_ostream &operator<<(raw_ostream &OS, const MCAsmFixup &AF) {
1227 OS << "<MCAsmFixup" << " Offset:" << AF.Offset << " Value:" << *AF.Value
1228 << " Kind:" << AF.Kind << ">";
1234 void MCFragment::dump() {
1235 raw_ostream &OS = llvm::errs();
1237 OS << "<MCFragment " << (void*) this << " Offset:" << Offset
1238 << " FileSize:" << FileSize;
1243 void MCAlignFragment::dump() {
1244 raw_ostream &OS = llvm::errs();
1246 OS << "<MCAlignFragment ";
1247 this->MCFragment::dump();
1249 OS << " Alignment:" << getAlignment()
1250 << " Value:" << getValue() << " ValueSize:" << getValueSize()
1251 << " MaxBytesToEmit:" << getMaxBytesToEmit() << ">";
1254 void MCDataFragment::dump() {
1255 raw_ostream &OS = llvm::errs();
1257 OS << "<MCDataFragment ";
1258 this->MCFragment::dump();
1260 OS << " Contents:[";
1261 for (unsigned i = 0, e = getContents().size(); i != e; ++i) {
1263 OS << hexdigit((Contents[i] >> 4) & 0xF) << hexdigit(Contents[i] & 0xF);
1265 OS << "] (" << getContents().size() << " bytes)";
1267 if (!getFixups().empty()) {
1270 for (fixup_iterator it = fixup_begin(), ie = fixup_end(); it != ie; ++it) {
1271 if (it != fixup_begin()) OS << ",\n ";
1280 void MCFillFragment::dump() {
1281 raw_ostream &OS = llvm::errs();
1283 OS << "<MCFillFragment ";
1284 this->MCFragment::dump();
1286 OS << " Value:" << getValue() << " ValueSize:" << getValueSize()
1287 << " Count:" << getCount() << ">";
1290 void MCOrgFragment::dump() {
1291 raw_ostream &OS = llvm::errs();
1293 OS << "<MCOrgFragment ";
1294 this->MCFragment::dump();
1296 OS << " Offset:" << getOffset() << " Value:" << getValue() << ">";
1299 void MCZeroFillFragment::dump() {
1300 raw_ostream &OS = llvm::errs();
1302 OS << "<MCZeroFillFragment ";
1303 this->MCFragment::dump();
1305 OS << " Size:" << getSize() << " Alignment:" << getAlignment() << ">";
1308 void MCSectionData::dump() {
1309 raw_ostream &OS = llvm::errs();
1311 OS << "<MCSectionData";
1312 OS << " Alignment:" << getAlignment() << " Address:" << Address
1313 << " Size:" << Size << " FileSize:" << FileSize
1315 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1316 if (it != begin()) OS << ",\n ";
1322 void MCSymbolData::dump() {
1323 raw_ostream &OS = llvm::errs();
1325 OS << "<MCSymbolData Symbol:" << getSymbol()
1326 << " Fragment:" << getFragment() << " Offset:" << getOffset()
1327 << " Flags:" << getFlags() << " Index:" << getIndex();
1329 OS << " (common, size:" << getCommonSize()
1330 << " align: " << getCommonAlignment() << ")";
1332 OS << " (external)";
1333 if (isPrivateExtern())
1334 OS << " (private extern)";
1338 void MCAssembler::dump() {
1339 raw_ostream &OS = llvm::errs();
1341 OS << "<MCAssembler\n";
1342 OS << " Sections:[";
1343 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1344 if (it != begin()) OS << ",\n ";
1350 for (symbol_iterator it = symbol_begin(), ie = symbol_end(); it != ie; ++it) {
1351 if (it != symbol_begin()) OS << ",\n ";