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"
29 class MachObjectWriter;
31 STATISTIC(EmittedFragments, "Number of emitted assembler fragments");
33 // FIXME FIXME FIXME: There are number of places in this file where we convert
34 // what is a 64-bit assembler value used for computation into a value in the
35 // object file, which may truncate it. We should detect that truncation where
36 // invalid and report errors back.
38 static void WriteFileData(raw_ostream &OS, const MCSectionData &SD,
39 MachObjectWriter &MOW);
41 /// isVirtualSection - Check if this is a section which does not actually exist
42 /// in the object file.
43 static bool isVirtualSection(const MCSection &Section) {
45 const MCSectionMachO &SMO = static_cast<const MCSectionMachO&>(Section);
46 unsigned Type = SMO.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
47 return (Type == MCSectionMachO::S_ZEROFILL);
50 class MachObjectWriter {
51 // See <mach-o/loader.h>.
53 Header_Magic32 = 0xFEEDFACE,
54 Header_Magic64 = 0xFEEDFACF
57 static const unsigned Header32Size = 28;
58 static const unsigned Header64Size = 32;
59 static const unsigned SegmentLoadCommand32Size = 56;
60 static const unsigned Section32Size = 68;
61 static const unsigned SymtabLoadCommandSize = 24;
62 static const unsigned DysymtabLoadCommandSize = 80;
63 static const unsigned Nlist32Size = 12;
64 static const unsigned RelocationInfoSize = 8;
71 HF_SubsectionsViaSymbols = 0x2000
74 enum LoadCommandType {
80 // See <mach-o/nlist.h>.
87 enum SymbolTypeFlags {
88 // If any of these bits are set, then the entry is a stab entry number (see
89 // <mach-o/stab.h>. Otherwise the other masks apply.
90 STF_StabsEntryMask = 0xe0,
94 STF_PrivateExtern = 0x10
97 /// IndirectSymbolFlags - Flags for encoding special values in the indirect
99 enum IndirectSymbolFlags {
100 ISF_Local = 0x80000000,
101 ISF_Absolute = 0x40000000
104 /// RelocationFlags - Special flags for addresses.
105 enum RelocationFlags {
106 RF_Scattered = 0x80000000
109 enum RelocationInfoType {
113 RIT_PreboundLazyPointer = 3,
114 RIT_LocalDifference = 4
117 /// MachSymbolData - Helper struct for containing some precomputed information
119 struct MachSymbolData {
120 MCSymbolData *SymbolData;
121 uint64_t StringIndex;
122 uint8_t SectionIndex;
124 // Support lexicographic sorting.
125 bool operator<(const MachSymbolData &RHS) const {
126 const std::string &Name = SymbolData->getSymbol().getName();
127 return Name < RHS.SymbolData->getSymbol().getName();
135 MachObjectWriter(raw_ostream &_OS, bool _IsLSB = true)
136 : OS(_OS), IsLSB(_IsLSB) {
139 /// @name Helper Methods
142 void Write8(uint8_t Value) {
146 void Write16(uint16_t Value) {
148 Write8(uint8_t(Value >> 0));
149 Write8(uint8_t(Value >> 8));
151 Write8(uint8_t(Value >> 8));
152 Write8(uint8_t(Value >> 0));
156 void Write32(uint32_t Value) {
158 Write16(uint16_t(Value >> 0));
159 Write16(uint16_t(Value >> 16));
161 Write16(uint16_t(Value >> 16));
162 Write16(uint16_t(Value >> 0));
166 void Write64(uint64_t Value) {
168 Write32(uint32_t(Value >> 0));
169 Write32(uint32_t(Value >> 32));
171 Write32(uint32_t(Value >> 32));
172 Write32(uint32_t(Value >> 0));
176 void WriteZeros(unsigned N) {
177 const char Zeros[16] = { 0 };
179 for (unsigned i = 0, e = N / 16; i != e; ++i)
180 OS << StringRef(Zeros, 16);
182 OS << StringRef(Zeros, N % 16);
185 void WriteString(StringRef Str, unsigned ZeroFillSize = 0) {
188 WriteZeros(ZeroFillSize - Str.size());
193 void WriteHeader32(unsigned NumLoadCommands, unsigned LoadCommandsSize,
194 bool SubsectionsViaSymbols) {
197 if (SubsectionsViaSymbols)
198 Flags |= HF_SubsectionsViaSymbols;
200 // struct mach_header (28 bytes)
202 uint64_t Start = OS.tell();
205 Write32(Header_Magic32);
207 // FIXME: Support cputype.
208 Write32(MachO::CPUTypeI386);
209 // FIXME: Support cpusubtype.
210 Write32(MachO::CPUSubType_I386_ALL);
212 Write32(NumLoadCommands); // Object files have a single load command, the
214 Write32(LoadCommandsSize);
217 assert(OS.tell() - Start == Header32Size);
220 /// WriteSegmentLoadCommand32 - Write a 32-bit segment load command.
222 /// \arg NumSections - The number of sections in this segment.
223 /// \arg SectionDataSize - The total size of the sections.
224 void WriteSegmentLoadCommand32(unsigned NumSections,
226 uint64_t SectionDataStartOffset,
227 uint64_t SectionDataSize) {
228 // struct segment_command (56 bytes)
230 uint64_t Start = OS.tell();
233 Write32(LCT_Segment);
234 Write32(SegmentLoadCommand32Size + NumSections * Section32Size);
237 Write32(0); // vmaddr
238 Write32(VMSize); // vmsize
239 Write32(SectionDataStartOffset); // file offset
240 Write32(SectionDataSize); // file size
241 Write32(0x7); // maxprot
242 Write32(0x7); // initprot
243 Write32(NumSections);
246 assert(OS.tell() - Start == SegmentLoadCommand32Size);
249 void WriteSection32(const MCSectionData &SD, uint64_t FileOffset,
250 uint64_t RelocationsStart, unsigned NumRelocations) {
251 // The offset is unused for virtual sections.
252 if (isVirtualSection(SD.getSection())) {
253 assert(SD.getFileSize() == 0 && "Invalid file size!");
257 // struct section (68 bytes)
259 uint64_t Start = OS.tell();
262 // FIXME: cast<> support!
263 const MCSectionMachO &Section =
264 static_cast<const MCSectionMachO&>(SD.getSection());
265 WriteString(Section.getSectionName(), 16);
266 WriteString(Section.getSegmentName(), 16);
267 Write32(SD.getAddress()); // address
268 Write32(SD.getSize()); // size
271 unsigned Flags = Section.getTypeAndAttributes();
272 if (SD.hasInstructions())
273 Flags |= MCSectionMachO::S_ATTR_SOME_INSTRUCTIONS;
275 assert(isPowerOf2_32(SD.getAlignment()) && "Invalid alignment!");
276 Write32(Log2_32(SD.getAlignment()));
277 Write32(NumRelocations ? RelocationsStart : 0);
278 Write32(NumRelocations);
280 Write32(0); // reserved1
281 Write32(Section.getStubSize()); // reserved2
283 assert(OS.tell() - Start == Section32Size);
286 void WriteSymtabLoadCommand(uint32_t SymbolOffset, uint32_t NumSymbols,
287 uint32_t StringTableOffset,
288 uint32_t StringTableSize) {
289 // struct symtab_command (24 bytes)
291 uint64_t Start = OS.tell();
295 Write32(SymtabLoadCommandSize);
296 Write32(SymbolOffset);
298 Write32(StringTableOffset);
299 Write32(StringTableSize);
301 assert(OS.tell() - Start == SymtabLoadCommandSize);
304 void WriteDysymtabLoadCommand(uint32_t FirstLocalSymbol,
305 uint32_t NumLocalSymbols,
306 uint32_t FirstExternalSymbol,
307 uint32_t NumExternalSymbols,
308 uint32_t FirstUndefinedSymbol,
309 uint32_t NumUndefinedSymbols,
310 uint32_t IndirectSymbolOffset,
311 uint32_t NumIndirectSymbols) {
312 // struct dysymtab_command (80 bytes)
314 uint64_t Start = OS.tell();
317 Write32(LCT_Dysymtab);
318 Write32(DysymtabLoadCommandSize);
319 Write32(FirstLocalSymbol);
320 Write32(NumLocalSymbols);
321 Write32(FirstExternalSymbol);
322 Write32(NumExternalSymbols);
323 Write32(FirstUndefinedSymbol);
324 Write32(NumUndefinedSymbols);
325 Write32(0); // tocoff
327 Write32(0); // modtaboff
328 Write32(0); // nmodtab
329 Write32(0); // extrefsymoff
330 Write32(0); // nextrefsyms
331 Write32(IndirectSymbolOffset);
332 Write32(NumIndirectSymbols);
333 Write32(0); // extreloff
334 Write32(0); // nextrel
335 Write32(0); // locreloff
336 Write32(0); // nlocrel
338 assert(OS.tell() - Start == DysymtabLoadCommandSize);
341 void WriteNlist32(MachSymbolData &MSD) {
342 MCSymbolData &Data = *MSD.SymbolData;
343 const MCSymbol &Symbol = Data.getSymbol();
345 uint16_t Flags = Data.getFlags();
346 uint32_t Address = 0;
348 // Set the N_TYPE bits. See <mach-o/nlist.h>.
350 // FIXME: Are the prebound or indirect fields possible here?
351 if (Symbol.isUndefined())
352 Type = STT_Undefined;
353 else if (Symbol.isAbsolute())
358 // FIXME: Set STAB bits.
360 if (Data.isPrivateExtern())
361 Type |= STF_PrivateExtern;
364 if (Data.isExternal() || Symbol.isUndefined())
365 Type |= STF_External;
367 // Compute the symbol address.
368 if (Symbol.isDefined()) {
369 if (Symbol.isAbsolute()) {
370 llvm_unreachable("FIXME: Not yet implemented!");
372 Address = Data.getFragment()->getAddress() + Data.getOffset();
374 } else if (Data.isCommon()) {
375 // Common symbols are encoded with the size in the address
376 // field, and their alignment in the flags.
377 Address = Data.getCommonSize();
379 // Common alignment is packed into the 'desc' bits.
380 if (unsigned Align = Data.getCommonAlignment()) {
381 unsigned Log2Size = Log2_32(Align);
382 assert((1U << Log2Size) == Align && "Invalid 'common' alignment!");
384 llvm_report_error("invalid 'common' alignment '" +
386 // FIXME: Keep this mask with the SymbolFlags enumeration.
387 Flags = (Flags & 0xF0FF) | (Log2Size << 8);
391 // struct nlist (12 bytes)
393 Write32(MSD.StringIndex);
395 Write8(MSD.SectionIndex);
397 // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc'
403 struct MachRelocationEntry {
407 void ComputeScatteredRelocationInfo(MCAssembler &Asm, MCFragment &Fragment,
409 const MCValue &Target,
410 DenseMap<const MCSymbol*,MCSymbolData*> &SymbolMap,
411 std::vector<MachRelocationEntry> &Relocs) {
412 uint32_t Address = Fragment.getOffset() + Fixup.Offset;
413 unsigned IsPCRel = 0;
414 unsigned Type = RIT_Vanilla;
417 const MCSymbol *A = Target.getSymA();
418 MCSymbolData *SD = SymbolMap.lookup(A);
419 uint32_t Value = SD->getFragment()->getAddress() + SD->getOffset();
422 if (const MCSymbol *B = Target.getSymB()) {
423 Type = RIT_LocalDifference;
425 MCSymbolData *SD = SymbolMap.lookup(B);
426 Value2 = SD->getFragment()->getAddress() + SD->getOffset();
429 unsigned Log2Size = Log2_32(Fixup.Size);
430 assert((1U << Log2Size) == Fixup.Size && "Invalid fixup size!");
432 // The value which goes in the fixup is current value of the expression.
433 Fixup.FixedValue = Value - Value2 + Target.getConstant();
435 MachRelocationEntry MRE;
436 MRE.Word0 = ((Address << 0) |
442 Relocs.push_back(MRE);
444 if (Type == RIT_LocalDifference) {
447 MachRelocationEntry MRE;
448 MRE.Word0 = ((0 << 0) |
454 Relocs.push_back(MRE);
458 void ComputeRelocationInfo(MCAssembler &Asm, MCFragment &Fragment,
460 DenseMap<const MCSymbol*,MCSymbolData*> &SymbolMap,
461 std::vector<MachRelocationEntry> &Relocs) {
463 if (!Fixup.Value->EvaluateAsRelocatable(Target))
464 llvm_report_error("expected relocatable expression");
466 // If this is a difference or a local symbol plus an offset, then we need a
467 // scattered relocation entry.
468 if (Target.getSymB() ||
469 (Target.getSymA() && !Target.getSymA()->isUndefined() &&
470 Target.getConstant()))
471 return ComputeScatteredRelocationInfo(Asm, Fragment, Fixup, Target,
475 uint32_t Address = Fragment.getOffset() + Fixup.Offset;
478 unsigned IsPCRel = 0;
479 unsigned IsExtern = 0;
482 if (Target.isAbsolute()) { // constant
483 // SymbolNum of 0 indicates the absolute section.
486 llvm_unreachable("FIXME: Not yet implemented!");
488 const MCSymbol *Symbol = Target.getSymA();
489 MCSymbolData *SD = SymbolMap.lookup(Symbol);
491 if (Symbol->isUndefined()) {
493 Index = SD->getIndex();
496 // The index is the section ordinal.
500 for (MCAssembler::iterator it = Asm.begin(),
501 ie = Asm.end(); it != ie; ++it, ++Index)
502 if (&*it == SD->getFragment()->getParent())
504 Value = SD->getFragment()->getAddress() + SD->getOffset();
510 // The value which goes in the fixup is current value of the expression.
511 Fixup.FixedValue = Value + Target.getConstant();
513 unsigned Log2Size = Log2_32(Fixup.Size);
514 assert((1U << Log2Size) == Fixup.Size && "Invalid fixup size!");
516 // struct relocation_info (8 bytes)
517 MachRelocationEntry MRE;
519 MRE.Word1 = ((Index << 0) |
524 Relocs.push_back(MRE);
527 void BindIndirectSymbols(MCAssembler &Asm,
528 DenseMap<const MCSymbol*,MCSymbolData*> &SymbolMap) {
529 // This is the point where 'as' creates actual symbols for indirect symbols
530 // (in the following two passes). It would be easier for us to do this
531 // sooner when we see the attribute, but that makes getting the order in the
532 // symbol table much more complicated than it is worth.
534 // FIXME: Revisit this when the dust settles.
536 // Bind non lazy symbol pointers first.
537 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
538 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
539 // FIXME: cast<> support!
540 const MCSectionMachO &Section =
541 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
544 Section.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
545 if (Type != MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS)
548 MCSymbolData *&Entry = SymbolMap[it->Symbol];
550 Entry = new MCSymbolData(*it->Symbol, 0, 0, &Asm);
553 // Then lazy symbol pointers and symbol stubs.
554 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
555 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
556 // FIXME: cast<> support!
557 const MCSectionMachO &Section =
558 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
561 Section.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
562 if (Type != MCSectionMachO::S_LAZY_SYMBOL_POINTERS &&
563 Type != MCSectionMachO::S_SYMBOL_STUBS)
566 MCSymbolData *&Entry = SymbolMap[it->Symbol];
568 Entry = new MCSymbolData(*it->Symbol, 0, 0, &Asm);
570 // Set the symbol type to undefined lazy, but only on construction.
572 // FIXME: Do not hardcode.
573 Entry->setFlags(Entry->getFlags() | 0x0001);
578 /// ComputeSymbolTable - Compute the symbol table data
580 /// \param StringTable [out] - The string table data.
581 /// \param StringIndexMap [out] - Map from symbol names to offsets in the
583 void ComputeSymbolTable(MCAssembler &Asm, SmallString<256> &StringTable,
584 std::vector<MachSymbolData> &LocalSymbolData,
585 std::vector<MachSymbolData> &ExternalSymbolData,
586 std::vector<MachSymbolData> &UndefinedSymbolData) {
587 // Build section lookup table.
588 DenseMap<const MCSection*, uint8_t> SectionIndexMap;
590 for (MCAssembler::iterator it = Asm.begin(),
591 ie = Asm.end(); it != ie; ++it, ++Index)
592 SectionIndexMap[&it->getSection()] = Index;
593 assert(Index <= 256 && "Too many sections!");
595 // Index 0 is always the empty string.
596 StringMap<uint64_t> StringIndexMap;
597 StringTable += '\x00';
599 // Build the symbol arrays and the string table, but only for non-local
602 // The particular order that we collect the symbols and create the string
603 // table, then sort the symbols is chosen to match 'as'. Even though it
604 // doesn't matter for correctness, this is important for letting us diff .o
606 for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
607 ie = Asm.symbol_end(); it != ie; ++it) {
608 const MCSymbol &Symbol = it->getSymbol();
610 // Ignore assembler temporaries.
611 if (it->getSymbol().isTemporary())
614 if (!it->isExternal() && !Symbol.isUndefined())
617 uint64_t &Entry = StringIndexMap[Symbol.getName()];
619 Entry = StringTable.size();
620 StringTable += Symbol.getName();
621 StringTable += '\x00';
626 MSD.StringIndex = Entry;
628 if (Symbol.isUndefined()) {
629 MSD.SectionIndex = 0;
630 UndefinedSymbolData.push_back(MSD);
631 } else if (Symbol.isAbsolute()) {
632 MSD.SectionIndex = 0;
633 ExternalSymbolData.push_back(MSD);
635 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
636 assert(MSD.SectionIndex && "Invalid section index!");
637 ExternalSymbolData.push_back(MSD);
641 // Now add the data for local symbols.
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.isAbsolute()) {
665 MSD.SectionIndex = 0;
666 LocalSymbolData.push_back(MSD);
668 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
669 assert(MSD.SectionIndex && "Invalid section index!");
670 LocalSymbolData.push_back(MSD);
674 // External and undefined symbols are required to be in lexicographic order.
675 std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
676 std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
678 // Set the symbol indices.
680 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
681 LocalSymbolData[i].SymbolData->setIndex(Index++);
682 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
683 ExternalSymbolData[i].SymbolData->setIndex(Index++);
684 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
685 UndefinedSymbolData[i].SymbolData->setIndex(Index++);
687 // The string table is padded to a multiple of 4.
688 while (StringTable.size() % 4)
689 StringTable += '\x00';
692 void WriteObject(MCAssembler &Asm) {
693 unsigned NumSections = Asm.size();
695 // Compute the symbol -> symbol data map.
697 // FIXME: This should not be here.
698 DenseMap<const MCSymbol*, MCSymbolData *> SymbolMap;
699 for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
700 ie = Asm.symbol_end(); it != ie; ++it)
701 SymbolMap[&it->getSymbol()] = it;
703 // Create symbol data for any indirect symbols.
704 BindIndirectSymbols(Asm, SymbolMap);
706 // Compute symbol table information.
707 SmallString<256> StringTable;
708 std::vector<MachSymbolData> LocalSymbolData;
709 std::vector<MachSymbolData> ExternalSymbolData;
710 std::vector<MachSymbolData> UndefinedSymbolData;
711 unsigned NumSymbols = Asm.symbol_size();
713 // No symbol table command is written if there are no symbols.
715 ComputeSymbolTable(Asm, StringTable, LocalSymbolData, ExternalSymbolData,
716 UndefinedSymbolData);
718 // The section data starts after the header, the segment load command (and
719 // section headers) and the symbol table.
720 unsigned NumLoadCommands = 1;
721 uint64_t LoadCommandsSize =
722 SegmentLoadCommand32Size + NumSections * Section32Size;
724 // Add the symbol table load command sizes, if used.
726 NumLoadCommands += 2;
727 LoadCommandsSize += SymtabLoadCommandSize + DysymtabLoadCommandSize;
730 // Compute the total size of the section data, as well as its file size and
732 uint64_t SectionDataStart = Header32Size + LoadCommandsSize;
733 uint64_t SectionDataSize = 0;
734 uint64_t SectionDataFileSize = 0;
736 for (MCAssembler::iterator it = Asm.begin(),
737 ie = Asm.end(); it != ie; ++it) {
738 MCSectionData &SD = *it;
740 VMSize = std::max(VMSize, SD.getAddress() + SD.getSize());
742 if (isVirtualSection(SD.getSection()))
745 SectionDataSize = std::max(SectionDataSize,
746 SD.getAddress() + SD.getSize());
747 SectionDataFileSize = std::max(SectionDataFileSize,
748 SD.getAddress() + SD.getFileSize());
751 // The section data is padded to 4 bytes.
753 // FIXME: Is this machine dependent?
754 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
755 SectionDataFileSize += SectionDataPadding;
757 // Write the prolog, starting with the header and load command...
758 WriteHeader32(NumLoadCommands, LoadCommandsSize,
759 Asm.getSubsectionsViaSymbols());
760 WriteSegmentLoadCommand32(NumSections, VMSize,
761 SectionDataStart, SectionDataSize);
763 // ... and then the section headers.
765 // We also compute the section relocations while we do this. Note that
766 // computing relocation info will also update the fixup to have the correct
767 // value; this will overwrite the appropriate data in the fragment when it
769 std::vector<MachRelocationEntry> RelocInfos;
770 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
771 for (MCAssembler::iterator it = Asm.begin(),
772 ie = Asm.end(); it != ie; ++it) {
773 MCSectionData &SD = *it;
775 // The assembler writes relocations in the reverse order they were seen.
777 // FIXME: It is probably more complicated than this.
778 unsigned NumRelocsStart = RelocInfos.size();
779 for (MCSectionData::reverse_iterator it2 = SD.rbegin(),
780 ie2 = SD.rend(); it2 != ie2; ++it2)
781 for (unsigned i = 0, e = it2->fixup_size(); i != e; ++i)
782 ComputeRelocationInfo(Asm, *it2, it2->getFixups()[e - i - 1],
783 SymbolMap, RelocInfos);
785 unsigned NumRelocs = RelocInfos.size() - NumRelocsStart;
786 uint64_t SectionStart = SectionDataStart + SD.getAddress();
787 WriteSection32(SD, SectionStart, RelocTableEnd, NumRelocs);
788 RelocTableEnd += NumRelocs * RelocationInfoSize;
791 // Write the symbol table load command, if used.
793 unsigned FirstLocalSymbol = 0;
794 unsigned NumLocalSymbols = LocalSymbolData.size();
795 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
796 unsigned NumExternalSymbols = ExternalSymbolData.size();
797 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
798 unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
799 unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
800 unsigned NumSymTabSymbols =
801 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
802 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
803 uint64_t IndirectSymbolOffset = 0;
805 // If used, the indirect symbols are written after the section data.
806 if (NumIndirectSymbols)
807 IndirectSymbolOffset = RelocTableEnd;
809 // The symbol table is written after the indirect symbol data.
810 uint64_t SymbolTableOffset = RelocTableEnd + IndirectSymbolSize;
812 // The string table is written after symbol table.
813 uint64_t StringTableOffset =
814 SymbolTableOffset + NumSymTabSymbols * Nlist32Size;
815 WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
816 StringTableOffset, StringTable.size());
818 WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
819 FirstExternalSymbol, NumExternalSymbols,
820 FirstUndefinedSymbol, NumUndefinedSymbols,
821 IndirectSymbolOffset, NumIndirectSymbols);
824 // Write the actual section data.
825 for (MCAssembler::iterator it = Asm.begin(), ie = Asm.end(); it != ie; ++it)
826 WriteFileData(OS, *it, *this);
828 // Write the extra padding.
829 WriteZeros(SectionDataPadding);
831 // Write the relocation entries.
832 for (unsigned i = 0, e = RelocInfos.size(); i != e; ++i) {
833 Write32(RelocInfos[i].Word0);
834 Write32(RelocInfos[i].Word1);
837 // Write the symbol table data, if used.
839 // Write the indirect symbol entries.
840 for (MCAssembler::indirect_symbol_iterator
841 it = Asm.indirect_symbol_begin(),
842 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
843 // Indirect symbols in the non lazy symbol pointer section have some
845 const MCSectionMachO &Section =
846 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
848 Section.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
849 if (Type == MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS) {
850 // If this symbol is defined and internal, mark it as such.
851 if (it->Symbol->isDefined() &&
852 !SymbolMap.lookup(it->Symbol)->isExternal()) {
853 uint32_t Flags = ISF_Local;
854 if (it->Symbol->isAbsolute())
855 Flags |= ISF_Absolute;
861 Write32(SymbolMap[it->Symbol]->getIndex());
864 // FIXME: Check that offsets match computed ones.
866 // Write the symbol table entries.
867 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
868 WriteNlist32(LocalSymbolData[i]);
869 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
870 WriteNlist32(ExternalSymbolData[i]);
871 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
872 WriteNlist32(UndefinedSymbolData[i]);
874 // Write the string table.
875 OS << StringTable.str();
882 MCFragment::MCFragment() : Kind(FragmentType(~0)) {
885 MCFragment::MCFragment(FragmentType _Kind, MCSectionData *_Parent)
888 FileSize(~UINT64_C(0))
891 Parent->getFragmentList().push_back(this);
894 MCFragment::~MCFragment() {
897 uint64_t MCFragment::getAddress() const {
898 assert(getParent() && "Missing Section!");
899 return getParent()->getAddress() + Offset;
904 MCSectionData::MCSectionData() : Section(0) {}
906 MCSectionData::MCSectionData(const MCSection &_Section, MCAssembler *A)
907 : Section(&_Section),
909 Address(~UINT64_C(0)),
911 FileSize(~UINT64_C(0)),
912 HasInstructions(false)
915 A->getSectionList().push_back(this);
920 MCSymbolData::MCSymbolData() : Symbol(0) {}
922 MCSymbolData::MCSymbolData(const MCSymbol &_Symbol, MCFragment *_Fragment,
923 uint64_t _Offset, MCAssembler *A)
924 : Symbol(&_Symbol), Fragment(_Fragment), Offset(_Offset),
925 IsExternal(false), IsPrivateExtern(false),
926 CommonSize(0), CommonAlign(0), Flags(0), Index(0)
929 A->getSymbolList().push_back(this);
934 MCAssembler::MCAssembler(MCContext &_Context, raw_ostream &_OS)
935 : Context(_Context), OS(_OS), SubsectionsViaSymbols(false)
939 MCAssembler::~MCAssembler() {
942 void MCAssembler::LayoutSection(MCSectionData &SD) {
943 uint64_t Address = SD.getAddress();
945 for (MCSectionData::iterator it = SD.begin(), ie = SD.end(); it != ie; ++it) {
948 F.setOffset(Address - SD.getAddress());
950 // Evaluate fragment size.
951 switch (F.getKind()) {
952 case MCFragment::FT_Align: {
953 MCAlignFragment &AF = cast<MCAlignFragment>(F);
955 uint64_t Size = OffsetToAlignment(Address, AF.getAlignment());
956 if (Size > AF.getMaxBytesToEmit())
959 AF.setFileSize(Size);
963 case MCFragment::FT_Data:
964 F.setFileSize(F.getMaxFileSize());
967 case MCFragment::FT_Fill: {
968 MCFillFragment &FF = cast<MCFillFragment>(F);
970 F.setFileSize(F.getMaxFileSize());
973 if (!FF.getValue().EvaluateAsRelocatable(Target))
974 llvm_report_error("expected relocatable expression");
976 // If the fill value is constant, thats it.
977 if (Target.isAbsolute())
980 // Otherwise, add fixups for the values.
982 // FIXME: What we want to do here is lower this to a data fragment once we
983 // realize it will need relocations. This means that the only place we
984 // need to worry about relocations and fixing is on data fragments.
985 for (uint64_t i = 0, e = FF.getCount(); i != e; ++i)
986 FF.getFixups().push_back(MCAsmFixup(i*FF.getValueSize(), FF.getValue(),
991 case MCFragment::FT_Org: {
992 MCOrgFragment &OF = cast<MCOrgFragment>(F);
995 if (!OF.getOffset().EvaluateAsRelocatable(Target))
996 llvm_report_error("expected relocatable expression");
998 if (!Target.isAbsolute())
999 llvm_unreachable("FIXME: Not yet implemented!");
1000 uint64_t OrgOffset = Target.getConstant();
1001 uint64_t Offset = Address - SD.getAddress();
1003 // FIXME: We need a way to communicate this error.
1004 if (OrgOffset < Offset)
1005 llvm_report_error("invalid .org offset '" + Twine(OrgOffset) +
1006 "' (at offset '" + Twine(Offset) + "'");
1008 F.setFileSize(OrgOffset - Offset);
1012 case MCFragment::FT_ZeroFill: {
1013 MCZeroFillFragment &ZFF = cast<MCZeroFillFragment>(F);
1015 // Align the fragment offset; it is safe to adjust the offset freely since
1016 // this is only in virtual sections.
1017 uint64_t Aligned = RoundUpToAlignment(Address, ZFF.getAlignment());
1018 F.setOffset(Aligned - SD.getAddress());
1020 // FIXME: This is misnamed.
1021 F.setFileSize(ZFF.getSize());
1026 Address += F.getFileSize();
1029 // Set the section sizes.
1030 SD.setSize(Address - SD.getAddress());
1031 if (isVirtualSection(SD.getSection()))
1034 SD.setFileSize(Address - SD.getAddress());
1037 /// WriteFileData - Write the \arg F data to the output file.
1038 static void WriteFileData(raw_ostream &OS, const MCFragment &F,
1039 MachObjectWriter &MOW) {
1040 uint64_t Start = OS.tell();
1045 // FIXME: Embed in fragments instead?
1046 switch (F.getKind()) {
1047 case MCFragment::FT_Align: {
1048 MCAlignFragment &AF = cast<MCAlignFragment>(F);
1049 uint64_t Count = AF.getFileSize() / AF.getValueSize();
1051 // FIXME: This error shouldn't actually occur (the front end should emit
1052 // multiple .align directives to enforce the semantics it wants), but is
1053 // severe enough that we want to report it. How to handle this?
1054 if (Count * AF.getValueSize() != AF.getFileSize())
1055 llvm_report_error("undefined .align directive, value size '" +
1056 Twine(AF.getValueSize()) +
1057 "' is not a divisor of padding size '" +
1058 Twine(AF.getFileSize()) + "'");
1060 for (uint64_t i = 0; i != Count; ++i) {
1061 switch (AF.getValueSize()) {
1063 assert(0 && "Invalid size!");
1064 case 1: MOW.Write8 (uint8_t (AF.getValue())); break;
1065 case 2: MOW.Write16(uint16_t(AF.getValue())); break;
1066 case 4: MOW.Write32(uint32_t(AF.getValue())); break;
1067 case 8: MOW.Write64(uint64_t(AF.getValue())); break;
1073 case MCFragment::FT_Data:
1074 OS << cast<MCDataFragment>(F).getContents().str();
1077 case MCFragment::FT_Fill: {
1078 MCFillFragment &FF = cast<MCFillFragment>(F);
1083 if (!FF.getValue().EvaluateAsRelocatable(Target))
1084 llvm_report_error("expected relocatable expression");
1086 if (Target.isAbsolute())
1087 Value = Target.getConstant();
1088 for (uint64_t i = 0, e = FF.getCount(); i != e; ++i) {
1089 if (!Target.isAbsolute()) {
1092 // FIXME: Find a better way to write in the fixes (move to
1094 const MCAsmFixup *Fixup = FF.LookupFixup(i * FF.getValueSize());
1095 assert(Fixup && "Missing fixup for fill value!");
1096 Value = Fixup->FixedValue;
1099 switch (FF.getValueSize()) {
1101 assert(0 && "Invalid size!");
1102 case 1: MOW.Write8 (uint8_t (Value)); break;
1103 case 2: MOW.Write16(uint16_t(Value)); break;
1104 case 4: MOW.Write32(uint32_t(Value)); break;
1105 case 8: MOW.Write64(uint64_t(Value)); break;
1111 case MCFragment::FT_Org: {
1112 MCOrgFragment &OF = cast<MCOrgFragment>(F);
1114 for (uint64_t i = 0, e = OF.getFileSize(); i != e; ++i)
1115 MOW.Write8(uint8_t(OF.getValue()));
1120 case MCFragment::FT_ZeroFill: {
1121 assert(0 && "Invalid zero fill fragment in concrete section!");
1126 assert(OS.tell() - Start == F.getFileSize());
1129 /// WriteFileData - Write the \arg SD data to the output file.
1130 static void WriteFileData(raw_ostream &OS, const MCSectionData &SD,
1131 MachObjectWriter &MOW) {
1132 // Ignore virtual sections.
1133 if (isVirtualSection(SD.getSection())) {
1134 assert(SD.getFileSize() == 0);
1138 uint64_t Start = OS.tell();
1141 for (MCSectionData::const_iterator it = SD.begin(),
1142 ie = SD.end(); it != ie; ++it)
1143 WriteFileData(OS, *it, MOW);
1145 // Add section padding.
1146 assert(SD.getFileSize() >= SD.getSize() && "Invalid section sizes!");
1147 MOW.WriteZeros(SD.getFileSize() - SD.getSize());
1149 assert(OS.tell() - Start == SD.getFileSize());
1152 void MCAssembler::Finish() {
1153 DEBUG_WITH_TYPE("mc-dump", {
1154 llvm::errs() << "assembler backend - pre-layout\n--\n";
1157 // Layout the concrete sections and fragments.
1158 uint64_t Address = 0;
1159 MCSectionData *Prev = 0;
1160 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1161 MCSectionData &SD = *it;
1163 // Skip virtual sections.
1164 if (isVirtualSection(SD.getSection()))
1167 // Align this section if necessary by adding padding bytes to the previous
1169 if (uint64_t Pad = OffsetToAlignment(Address, it->getAlignment())) {
1170 assert(Prev && "Missing prev section!");
1171 Prev->setFileSize(Prev->getFileSize() + Pad);
1175 // Layout the section fragments and its size.
1176 SD.setAddress(Address);
1178 Address += SD.getFileSize();
1183 // Layout the virtual sections.
1184 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1185 MCSectionData &SD = *it;
1187 if (!isVirtualSection(SD.getSection()))
1190 SD.setAddress(Address);
1192 Address += SD.getSize();
1195 DEBUG_WITH_TYPE("mc-dump", {
1196 llvm::errs() << "assembler backend - post-layout\n--\n";
1199 // Write the object file.
1200 MachObjectWriter MOW(OS);
1201 MOW.WriteObject(*this);
1207 // Debugging methods
1211 raw_ostream &operator<<(raw_ostream &OS, const MCAsmFixup &AF) {
1212 OS << "<MCAsmFixup" << " Offset:" << AF.Offset << " Value:" << AF.Value
1213 << " Size:" << AF.Size << ">";
1219 void MCFragment::dump() {
1220 raw_ostream &OS = llvm::errs();
1222 OS << "<MCFragment " << (void*) this << " Offset:" << Offset
1223 << " FileSize:" << FileSize;
1225 if (!Fixups.empty()) {
1228 for (fixup_iterator it = fixup_begin(), ie = fixup_end(); it != ie; ++it) {
1229 if (it != fixup_begin()) OS << ",\n ";
1238 void MCAlignFragment::dump() {
1239 raw_ostream &OS = llvm::errs();
1241 OS << "<MCAlignFragment ";
1242 this->MCFragment::dump();
1244 OS << " Alignment:" << getAlignment()
1245 << " Value:" << getValue() << " ValueSize:" << getValueSize()
1246 << " MaxBytesToEmit:" << getMaxBytesToEmit() << ">";
1249 void MCDataFragment::dump() {
1250 raw_ostream &OS = llvm::errs();
1252 OS << "<MCDataFragment ";
1253 this->MCFragment::dump();
1255 OS << " Contents:[";
1256 for (unsigned i = 0, e = getContents().size(); i != e; ++i) {
1258 OS << hexdigit((Contents[i] >> 4) & 0xF) << hexdigit(Contents[i] & 0xF);
1263 void MCFillFragment::dump() {
1264 raw_ostream &OS = llvm::errs();
1266 OS << "<MCFillFragment ";
1267 this->MCFragment::dump();
1269 OS << " Value:" << getValue() << " ValueSize:" << getValueSize()
1270 << " Count:" << getCount() << ">";
1273 void MCOrgFragment::dump() {
1274 raw_ostream &OS = llvm::errs();
1276 OS << "<MCOrgFragment ";
1277 this->MCFragment::dump();
1279 OS << " Offset:" << getOffset() << " Value:" << getValue() << ">";
1282 void MCZeroFillFragment::dump() {
1283 raw_ostream &OS = llvm::errs();
1285 OS << "<MCZeroFillFragment ";
1286 this->MCFragment::dump();
1288 OS << " Size:" << getSize() << " Alignment:" << getAlignment() << ">";
1291 void MCSectionData::dump() {
1292 raw_ostream &OS = llvm::errs();
1294 OS << "<MCSectionData";
1295 OS << " Alignment:" << getAlignment() << " Address:" << Address
1296 << " Size:" << Size << " FileSize:" << FileSize
1298 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1299 if (it != begin()) OS << ",\n ";
1305 void MCSymbolData::dump() {
1306 raw_ostream &OS = llvm::errs();
1308 OS << "<MCSymbolData Symbol:" << getSymbol()
1309 << " Fragment:" << getFragment() << " Offset:" << getOffset()
1310 << " Flags:" << getFlags() << " Index:" << getIndex();
1312 OS << " (common, size:" << getCommonSize()
1313 << " align: " << getCommonAlignment() << ")";
1315 OS << " (external)";
1316 if (isPrivateExtern())
1317 OS << " (private extern)";
1321 void MCAssembler::dump() {
1322 raw_ostream &OS = llvm::errs();
1324 OS << "<MCAssembler\n";
1325 OS << " Sections:[";
1326 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1327 if (it != begin()) OS << ",\n ";
1333 for (symbol_iterator it = symbol_begin(), ie = symbol_end(); it != ie; ++it) {
1334 if (it != symbol_begin()) OS << ",\n ";