1 //===- lib/MC/MachObjectWriter.cpp - Mach-O File Writer -------------------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 10 #include "llvm/MC/MCMachObjectWriter.h" 11 #include "llvm/ADT/StringMap.h" 12 #include "llvm/ADT/Twine.h" 13 #include "llvm/MC/MCAsmBackend.h" 14 #include "llvm/MC/MCAsmLayout.h" 15 #include "llvm/MC/MCAssembler.h" 16 #include "llvm/MC/MCExpr.h" 17 #include "llvm/MC/MCFixupKindInfo.h" 18 #include "llvm/MC/MCObjectWriter.h" 19 #include "llvm/MC/MCSectionMachO.h" 20 #include "llvm/MC/MCSymbolMachO.h" 21 #include "llvm/MC/MCValue.h" 22 #include "llvm/Support/Debug.h" 23 #include "llvm/Support/ErrorHandling.h" 24 #include "llvm/Support/MachO.h" 25 #include "llvm/Support/raw_ostream.h" 26 #include <vector> 27 using namespace llvm; 28 29 #define DEBUG_TYPE "mc" 30 31 void MachObjectWriter::reset() { 32 Relocations.clear(); 33 IndirectSymBase.clear(); 34 StringTable.clear(); 35 LocalSymbolData.clear(); 36 ExternalSymbolData.clear(); 37 UndefinedSymbolData.clear(); 38 MCObjectWriter::reset(); 39 } 40 41 bool MachObjectWriter::doesSymbolRequireExternRelocation(const MCSymbol &S) { 42 // Undefined symbols are always extern. 43 if (S.isUndefined()) 44 return true; 45 46 // References to weak definitions require external relocation entries; the 47 // definition may not always be the one in the same object file. 48 if (cast<MCSymbolMachO>(S).isWeakDefinition()) 49 return true; 50 51 // Otherwise, we can use an internal relocation. 52 return false; 53 } 54 55 bool MachObjectWriter:: 56 MachSymbolData::operator<(const MachSymbolData &RHS) const { 57 return Symbol->getName() < RHS.Symbol->getName(); 58 } 59 60 bool MachObjectWriter::isFixupKindPCRel(const MCAssembler &Asm, unsigned Kind) { 61 const MCFixupKindInfo &FKI = Asm.getBackend().getFixupKindInfo( 62 (MCFixupKind) Kind); 63 64 return FKI.Flags & MCFixupKindInfo::FKF_IsPCRel; 65 } 66 67 uint64_t MachObjectWriter::getFragmentAddress(const MCFragment *Fragment, 68 const MCAsmLayout &Layout) const { 69 return getSectionAddress(Fragment->getParent()) + 70 Layout.getFragmentOffset(Fragment); 71 } 72 73 uint64_t MachObjectWriter::getSymbolAddress(const MCSymbol &S, 74 const MCAsmLayout &Layout) const { 75 // If this is a variable, then recursively evaluate now. 76 if (S.isVariable()) { 77 if (const MCConstantExpr *C = 78 dyn_cast<const MCConstantExpr>(S.getVariableValue())) 79 return C->getValue(); 80 81 MCValue Target; 82 if (!S.getVariableValue()->evaluateAsRelocatable(Target, &Layout, nullptr)) 83 report_fatal_error("unable to evaluate offset for variable '" + 84 S.getName() + "'"); 85 86 // Verify that any used symbols are defined. 87 if (Target.getSymA() && Target.getSymA()->getSymbol().isUndefined()) 88 report_fatal_error("unable to evaluate offset to undefined symbol '" + 89 Target.getSymA()->getSymbol().getName() + "'"); 90 if (Target.getSymB() && Target.getSymB()->getSymbol().isUndefined()) 91 report_fatal_error("unable to evaluate offset to undefined symbol '" + 92 Target.getSymB()->getSymbol().getName() + "'"); 93 94 uint64_t Address = Target.getConstant(); 95 if (Target.getSymA()) 96 Address += getSymbolAddress(Target.getSymA()->getSymbol(), Layout); 97 if (Target.getSymB()) 98 Address += getSymbolAddress(Target.getSymB()->getSymbol(), Layout); 99 return Address; 100 } 101 102 return getSectionAddress(S.getFragment()->getParent()) + 103 Layout.getSymbolOffset(S); 104 } 105 106 uint64_t MachObjectWriter::getPaddingSize(const MCSection *Sec, 107 const MCAsmLayout &Layout) const { 108 uint64_t EndAddr = getSectionAddress(Sec) + Layout.getSectionAddressSize(Sec); 109 unsigned Next = Sec->getLayoutOrder() + 1; 110 if (Next >= Layout.getSectionOrder().size()) 111 return 0; 112 113 const MCSection &NextSec = *Layout.getSectionOrder()[Next]; 114 if (NextSec.isVirtualSection()) 115 return 0; 116 return OffsetToAlignment(EndAddr, NextSec.getAlignment()); 117 } 118 119 void MachObjectWriter::writeHeader(MachO::HeaderFileType Type, 120 unsigned NumLoadCommands, 121 unsigned LoadCommandsSize, 122 bool SubsectionsViaSymbols) { 123 uint32_t Flags = 0; 124 125 if (SubsectionsViaSymbols) 126 Flags |= MachO::MH_SUBSECTIONS_VIA_SYMBOLS; 127 128 // struct mach_header (28 bytes) or 129 // struct mach_header_64 (32 bytes) 130 131 uint64_t Start = getStream().tell(); 132 (void) Start; 133 134 write32(is64Bit() ? MachO::MH_MAGIC_64 : MachO::MH_MAGIC); 135 136 write32(TargetObjectWriter->getCPUType()); 137 write32(TargetObjectWriter->getCPUSubtype()); 138 139 write32(Type); 140 write32(NumLoadCommands); 141 write32(LoadCommandsSize); 142 write32(Flags); 143 if (is64Bit()) 144 write32(0); // reserved 145 146 assert( 147 getStream().tell() - Start == 148 (is64Bit() ? sizeof(MachO::mach_header_64) : sizeof(MachO::mach_header))); 149 } 150 151 /// writeSegmentLoadCommand - Write a segment load command. 152 /// 153 /// \param NumSections The number of sections in this segment. 154 /// \param SectionDataSize The total size of the sections. 155 void MachObjectWriter::writeSegmentLoadCommand( 156 StringRef Name, unsigned NumSections, uint64_t VMAddr, uint64_t VMSize, 157 uint64_t SectionDataStartOffset, uint64_t SectionDataSize, uint32_t MaxProt, 158 uint32_t InitProt) { 159 // struct segment_command (56 bytes) or 160 // struct segment_command_64 (72 bytes) 161 162 uint64_t Start = getStream().tell(); 163 (void) Start; 164 165 unsigned SegmentLoadCommandSize = 166 is64Bit() ? sizeof(MachO::segment_command_64): 167 sizeof(MachO::segment_command); 168 write32(is64Bit() ? MachO::LC_SEGMENT_64 : MachO::LC_SEGMENT); 169 write32(SegmentLoadCommandSize + 170 NumSections * (is64Bit() ? sizeof(MachO::section_64) : 171 sizeof(MachO::section))); 172 173 assert(Name.size() <= 16); 174 writeBytes(Name, 16); 175 if (is64Bit()) { 176 write64(VMAddr); // vmaddr 177 write64(VMSize); // vmsize 178 write64(SectionDataStartOffset); // file offset 179 write64(SectionDataSize); // file size 180 } else { 181 write32(VMAddr); // vmaddr 182 write32(VMSize); // vmsize 183 write32(SectionDataStartOffset); // file offset 184 write32(SectionDataSize); // file size 185 } 186 // maxprot 187 write32(MaxProt); 188 // initprot 189 write32(InitProt); 190 write32(NumSections); 191 write32(0); // flags 192 193 assert(getStream().tell() - Start == SegmentLoadCommandSize); 194 } 195 196 void MachObjectWriter::writeSection(const MCAsmLayout &Layout, 197 const MCSection &Sec, uint64_t VMAddr, 198 uint64_t FileOffset, unsigned Flags, 199 uint64_t RelocationsStart, 200 unsigned NumRelocations) { 201 uint64_t SectionSize = Layout.getSectionAddressSize(&Sec); 202 const MCSectionMachO &Section = cast<MCSectionMachO>(Sec); 203 204 // The offset is unused for virtual sections. 205 if (Section.isVirtualSection()) { 206 assert(Layout.getSectionFileSize(&Sec) == 0 && "Invalid file size!"); 207 FileOffset = 0; 208 } 209 210 // struct section (68 bytes) or 211 // struct section_64 (80 bytes) 212 213 uint64_t Start = getStream().tell(); 214 (void) Start; 215 216 writeBytes(Section.getSectionName(), 16); 217 writeBytes(Section.getSegmentName(), 16); 218 if (is64Bit()) { 219 write64(VMAddr); // address 220 write64(SectionSize); // size 221 } else { 222 write32(VMAddr); // address 223 write32(SectionSize); // size 224 } 225 write32(FileOffset); 226 227 assert(isPowerOf2_32(Section.getAlignment()) && "Invalid alignment!"); 228 write32(Log2_32(Section.getAlignment())); 229 write32(NumRelocations ? RelocationsStart : 0); 230 write32(NumRelocations); 231 write32(Flags); 232 write32(IndirectSymBase.lookup(&Sec)); // reserved1 233 write32(Section.getStubSize()); // reserved2 234 if (is64Bit()) 235 write32(0); // reserved3 236 237 assert(getStream().tell() - Start == 238 (is64Bit() ? sizeof(MachO::section_64) : sizeof(MachO::section))); 239 } 240 241 void MachObjectWriter::writeSymtabLoadCommand(uint32_t SymbolOffset, 242 uint32_t NumSymbols, 243 uint32_t StringTableOffset, 244 uint32_t StringTableSize) { 245 // struct symtab_command (24 bytes) 246 247 uint64_t Start = getStream().tell(); 248 (void) Start; 249 250 write32(MachO::LC_SYMTAB); 251 write32(sizeof(MachO::symtab_command)); 252 write32(SymbolOffset); 253 write32(NumSymbols); 254 write32(StringTableOffset); 255 write32(StringTableSize); 256 257 assert(getStream().tell() - Start == sizeof(MachO::symtab_command)); 258 } 259 260 void MachObjectWriter::writeDysymtabLoadCommand(uint32_t FirstLocalSymbol, 261 uint32_t NumLocalSymbols, 262 uint32_t FirstExternalSymbol, 263 uint32_t NumExternalSymbols, 264 uint32_t FirstUndefinedSymbol, 265 uint32_t NumUndefinedSymbols, 266 uint32_t IndirectSymbolOffset, 267 uint32_t NumIndirectSymbols) { 268 // struct dysymtab_command (80 bytes) 269 270 uint64_t Start = getStream().tell(); 271 (void) Start; 272 273 write32(MachO::LC_DYSYMTAB); 274 write32(sizeof(MachO::dysymtab_command)); 275 write32(FirstLocalSymbol); 276 write32(NumLocalSymbols); 277 write32(FirstExternalSymbol); 278 write32(NumExternalSymbols); 279 write32(FirstUndefinedSymbol); 280 write32(NumUndefinedSymbols); 281 write32(0); // tocoff 282 write32(0); // ntoc 283 write32(0); // modtaboff 284 write32(0); // nmodtab 285 write32(0); // extrefsymoff 286 write32(0); // nextrefsyms 287 write32(IndirectSymbolOffset); 288 write32(NumIndirectSymbols); 289 write32(0); // extreloff 290 write32(0); // nextrel 291 write32(0); // locreloff 292 write32(0); // nlocrel 293 294 assert(getStream().tell() - Start == sizeof(MachO::dysymtab_command)); 295 } 296 297 MachObjectWriter::MachSymbolData * 298 MachObjectWriter::findSymbolData(const MCSymbol &Sym) { 299 for (auto *SymbolData : 300 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData}) 301 for (MachSymbolData &Entry : *SymbolData) 302 if (Entry.Symbol == &Sym) 303 return &Entry; 304 305 return nullptr; 306 } 307 308 const MCSymbol &MachObjectWriter::findAliasedSymbol(const MCSymbol &Sym) const { 309 const MCSymbol *S = &Sym; 310 while (S->isVariable()) { 311 const MCExpr *Value = S->getVariableValue(); 312 const auto *Ref = dyn_cast<MCSymbolRefExpr>(Value); 313 if (!Ref) 314 return *S; 315 S = &Ref->getSymbol(); 316 } 317 return *S; 318 } 319 320 void MachObjectWriter::writeNlist(MachSymbolData &MSD, 321 const MCAsmLayout &Layout) { 322 const MCSymbol *Symbol = MSD.Symbol; 323 const MCSymbol &Data = *Symbol; 324 const MCSymbol *AliasedSymbol = &findAliasedSymbol(*Symbol); 325 uint8_t SectionIndex = MSD.SectionIndex; 326 uint8_t Type = 0; 327 uint64_t Address = 0; 328 bool IsAlias = Symbol != AliasedSymbol; 329 330 const MCSymbol &OrigSymbol = *Symbol; 331 MachSymbolData *AliaseeInfo; 332 if (IsAlias) { 333 AliaseeInfo = findSymbolData(*AliasedSymbol); 334 if (AliaseeInfo) 335 SectionIndex = AliaseeInfo->SectionIndex; 336 Symbol = AliasedSymbol; 337 // FIXME: Should this update Data as well? 338 } 339 340 // Set the N_TYPE bits. See <mach-o/nlist.h>. 341 // 342 // FIXME: Are the prebound or indirect fields possible here? 343 if (IsAlias && Symbol->isUndefined()) 344 Type = MachO::N_INDR; 345 else if (Symbol->isUndefined()) 346 Type = MachO::N_UNDF; 347 else if (Symbol->isAbsolute()) 348 Type = MachO::N_ABS; 349 else 350 Type = MachO::N_SECT; 351 352 // FIXME: Set STAB bits. 353 354 if (Data.isPrivateExtern()) 355 Type |= MachO::N_PEXT; 356 357 // Set external bit. 358 if (Data.isExternal() || (!IsAlias && Symbol->isUndefined())) 359 Type |= MachO::N_EXT; 360 361 // Compute the symbol address. 362 if (IsAlias && Symbol->isUndefined()) 363 Address = AliaseeInfo->StringIndex; 364 else if (Symbol->isDefined()) 365 Address = getSymbolAddress(OrigSymbol, Layout); 366 else if (Symbol->isCommon()) { 367 // Common symbols are encoded with the size in the address 368 // field, and their alignment in the flags. 369 Address = Symbol->getCommonSize(); 370 } 371 372 // struct nlist (12 bytes) 373 374 write32(MSD.StringIndex); 375 write8(Type); 376 write8(SectionIndex); 377 378 // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc' 379 // value. 380 bool EncodeAsAltEntry = 381 IsAlias && cast<MCSymbolMachO>(OrigSymbol).isAltEntry(); 382 write16(cast<MCSymbolMachO>(Symbol)->getEncodedFlags(EncodeAsAltEntry)); 383 if (is64Bit()) 384 write64(Address); 385 else 386 write32(Address); 387 } 388 389 void MachObjectWriter::writeLinkeditLoadCommand(uint32_t Type, 390 uint32_t DataOffset, 391 uint32_t DataSize) { 392 uint64_t Start = getStream().tell(); 393 (void) Start; 394 395 write32(Type); 396 write32(sizeof(MachO::linkedit_data_command)); 397 write32(DataOffset); 398 write32(DataSize); 399 400 assert(getStream().tell() - Start == sizeof(MachO::linkedit_data_command)); 401 } 402 403 static unsigned ComputeLinkerOptionsLoadCommandSize( 404 const std::vector<std::string> &Options, bool is64Bit) 405 { 406 unsigned Size = sizeof(MachO::linker_option_command); 407 for (const std::string &Option : Options) 408 Size += Option.size() + 1; 409 return alignTo(Size, is64Bit ? 8 : 4); 410 } 411 412 void MachObjectWriter::writeLinkerOptionsLoadCommand( 413 const std::vector<std::string> &Options) 414 { 415 unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit()); 416 uint64_t Start = getStream().tell(); 417 (void) Start; 418 419 write32(MachO::LC_LINKER_OPTION); 420 write32(Size); 421 write32(Options.size()); 422 uint64_t BytesWritten = sizeof(MachO::linker_option_command); 423 for (const std::string &Option : Options) { 424 // Write each string, including the null byte. 425 writeBytes(Option.c_str(), Option.size() + 1); 426 BytesWritten += Option.size() + 1; 427 } 428 429 // Pad to a multiple of the pointer size. 430 writeBytes("", OffsetToAlignment(BytesWritten, is64Bit() ? 8 : 4)); 431 432 assert(getStream().tell() - Start == Size); 433 } 434 435 void MachObjectWriter::recordRelocation(MCAssembler &Asm, 436 const MCAsmLayout &Layout, 437 const MCFragment *Fragment, 438 const MCFixup &Fixup, MCValue Target, 439 bool &IsPCRel, uint64_t &FixedValue) { 440 TargetObjectWriter->recordRelocation(this, Asm, Layout, Fragment, Fixup, 441 Target, FixedValue); 442 } 443 444 void MachObjectWriter::bindIndirectSymbols(MCAssembler &Asm) { 445 // This is the point where 'as' creates actual symbols for indirect symbols 446 // (in the following two passes). It would be easier for us to do this sooner 447 // when we see the attribute, but that makes getting the order in the symbol 448 // table much more complicated than it is worth. 449 // 450 // FIXME: Revisit this when the dust settles. 451 452 // Report errors for use of .indirect_symbol not in a symbol pointer section 453 // or stub section. 454 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(), 455 ie = Asm.indirect_symbol_end(); it != ie; ++it) { 456 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section); 457 458 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS && 459 Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS && 460 Section.getType() != MachO::S_SYMBOL_STUBS) { 461 MCSymbol &Symbol = *it->Symbol; 462 report_fatal_error("indirect symbol '" + Symbol.getName() + 463 "' not in a symbol pointer or stub section"); 464 } 465 } 466 467 // Bind non-lazy symbol pointers first. 468 unsigned IndirectIndex = 0; 469 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(), 470 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) { 471 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section); 472 473 if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS) 474 continue; 475 476 // Initialize the section indirect symbol base, if necessary. 477 IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex)); 478 479 Asm.registerSymbol(*it->Symbol); 480 } 481 482 // Then lazy symbol pointers and symbol stubs. 483 IndirectIndex = 0; 484 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(), 485 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) { 486 const MCSectionMachO &Section = cast<MCSectionMachO>(*it->Section); 487 488 if (Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS && 489 Section.getType() != MachO::S_SYMBOL_STUBS) 490 continue; 491 492 // Initialize the section indirect symbol base, if necessary. 493 IndirectSymBase.insert(std::make_pair(it->Section, IndirectIndex)); 494 495 // Set the symbol type to undefined lazy, but only on construction. 496 // 497 // FIXME: Do not hardcode. 498 bool Created; 499 Asm.registerSymbol(*it->Symbol, &Created); 500 if (Created) 501 cast<MCSymbolMachO>(it->Symbol)->setReferenceTypeUndefinedLazy(true); 502 } 503 } 504 505 /// computeSymbolTable - Compute the symbol table data 506 void MachObjectWriter::computeSymbolTable( 507 MCAssembler &Asm, std::vector<MachSymbolData> &LocalSymbolData, 508 std::vector<MachSymbolData> &ExternalSymbolData, 509 std::vector<MachSymbolData> &UndefinedSymbolData) { 510 // Build section lookup table. 511 DenseMap<const MCSection*, uint8_t> SectionIndexMap; 512 unsigned Index = 1; 513 for (MCAssembler::iterator it = Asm.begin(), 514 ie = Asm.end(); it != ie; ++it, ++Index) 515 SectionIndexMap[&*it] = Index; 516 assert(Index <= 256 && "Too many sections!"); 517 518 // Build the string table. 519 for (const MCSymbol &Symbol : Asm.symbols()) { 520 if (!Asm.isSymbolLinkerVisible(Symbol)) 521 continue; 522 523 StringTable.add(Symbol.getName()); 524 } 525 StringTable.finalize(); 526 527 // Build the symbol arrays but only for non-local symbols. 528 // 529 // The particular order that we collect and then sort the symbols is chosen to 530 // match 'as'. Even though it doesn't matter for correctness, this is 531 // important for letting us diff .o files. 532 for (const MCSymbol &Symbol : Asm.symbols()) { 533 // Ignore non-linker visible symbols. 534 if (!Asm.isSymbolLinkerVisible(Symbol)) 535 continue; 536 537 if (!Symbol.isExternal() && !Symbol.isUndefined()) 538 continue; 539 540 MachSymbolData MSD; 541 MSD.Symbol = &Symbol; 542 MSD.StringIndex = StringTable.getOffset(Symbol.getName()); 543 544 if (Symbol.isUndefined()) { 545 MSD.SectionIndex = 0; 546 UndefinedSymbolData.push_back(MSD); 547 } else if (Symbol.isAbsolute()) { 548 MSD.SectionIndex = 0; 549 ExternalSymbolData.push_back(MSD); 550 } else { 551 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection()); 552 assert(MSD.SectionIndex && "Invalid section index!"); 553 ExternalSymbolData.push_back(MSD); 554 } 555 } 556 557 // Now add the data for local symbols. 558 for (const MCSymbol &Symbol : Asm.symbols()) { 559 // Ignore non-linker visible symbols. 560 if (!Asm.isSymbolLinkerVisible(Symbol)) 561 continue; 562 563 if (Symbol.isExternal() || Symbol.isUndefined()) 564 continue; 565 566 MachSymbolData MSD; 567 MSD.Symbol = &Symbol; 568 MSD.StringIndex = StringTable.getOffset(Symbol.getName()); 569 570 if (Symbol.isAbsolute()) { 571 MSD.SectionIndex = 0; 572 LocalSymbolData.push_back(MSD); 573 } else { 574 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection()); 575 assert(MSD.SectionIndex && "Invalid section index!"); 576 LocalSymbolData.push_back(MSD); 577 } 578 } 579 580 // External and undefined symbols are required to be in lexicographic order. 581 std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end()); 582 std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end()); 583 584 // Set the symbol indices. 585 Index = 0; 586 for (auto *SymbolData : 587 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData}) 588 for (MachSymbolData &Entry : *SymbolData) 589 Entry.Symbol->setIndex(Index++); 590 591 for (const MCSection &Section : Asm) { 592 for (RelAndSymbol &Rel : Relocations[&Section]) { 593 if (!Rel.Sym) 594 continue; 595 596 // Set the Index and the IsExtern bit. 597 unsigned Index = Rel.Sym->getIndex(); 598 assert(isInt<24>(Index)); 599 if (IsLittleEndian) 600 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & (~0U << 24)) | Index | (1 << 27); 601 else 602 Rel.MRE.r_word1 = (Rel.MRE.r_word1 & 0xff) | Index << 8 | (1 << 4); 603 } 604 } 605 } 606 607 void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm, 608 const MCAsmLayout &Layout) { 609 uint64_t StartAddress = 0; 610 for (const MCSection *Sec : Layout.getSectionOrder()) { 611 StartAddress = alignTo(StartAddress, Sec->getAlignment()); 612 SectionAddress[Sec] = StartAddress; 613 StartAddress += Layout.getSectionAddressSize(Sec); 614 615 // Explicitly pad the section to match the alignment requirements of the 616 // following one. This is for 'gas' compatibility, it shouldn't 617 /// strictly be necessary. 618 StartAddress += getPaddingSize(Sec, Layout); 619 } 620 } 621 622 void MachObjectWriter::executePostLayoutBinding(MCAssembler &Asm, 623 const MCAsmLayout &Layout) { 624 computeSectionAddresses(Asm, Layout); 625 626 // Create symbol data for any indirect symbols. 627 bindIndirectSymbols(Asm); 628 } 629 630 bool MachObjectWriter::isSymbolRefDifferenceFullyResolvedImpl( 631 const MCAssembler &Asm, const MCSymbol &A, const MCSymbol &B, 632 bool InSet) const { 633 // FIXME: We don't handle things like 634 // foo = . 635 // creating atoms. 636 if (A.isVariable() || B.isVariable()) 637 return false; 638 return MCObjectWriter::isSymbolRefDifferenceFullyResolvedImpl(Asm, A, B, 639 InSet); 640 } 641 642 bool MachObjectWriter::isSymbolRefDifferenceFullyResolvedImpl( 643 const MCAssembler &Asm, const MCSymbol &SymA, const MCFragment &FB, 644 bool InSet, bool IsPCRel) const { 645 if (InSet) 646 return true; 647 648 // The effective address is 649 // addr(atom(A)) + offset(A) 650 // - addr(atom(B)) - offset(B) 651 // and the offsets are not relocatable, so the fixup is fully resolved when 652 // addr(atom(A)) - addr(atom(B)) == 0. 653 const MCSymbol &SA = findAliasedSymbol(SymA); 654 const MCSection &SecA = SA.getSection(); 655 const MCSection &SecB = *FB.getParent(); 656 657 if (IsPCRel) { 658 // The simple (Darwin, except on x86_64) way of dealing with this was to 659 // assume that any reference to a temporary symbol *must* be a temporary 660 // symbol in the same atom, unless the sections differ. Therefore, any PCrel 661 // relocation to a temporary symbol (in the same section) is fully 662 // resolved. This also works in conjunction with absolutized .set, which 663 // requires the compiler to use .set to absolutize the differences between 664 // symbols which the compiler knows to be assembly time constants, so we 665 // don't need to worry about considering symbol differences fully resolved. 666 // 667 // If the file isn't using sub-sections-via-symbols, we can make the 668 // same assumptions about any symbol that we normally make about 669 // assembler locals. 670 671 bool hasReliableSymbolDifference = isX86_64(); 672 if (!hasReliableSymbolDifference) { 673 if (!SA.isInSection() || &SecA != &SecB || 674 (!SA.isTemporary() && FB.getAtom() != SA.getFragment()->getAtom() && 675 Asm.getSubsectionsViaSymbols())) 676 return false; 677 return true; 678 } 679 // For Darwin x86_64, there is one special case when the reference IsPCRel. 680 // If the fragment with the reference does not have a base symbol but meets 681 // the simple way of dealing with this, in that it is a temporary symbol in 682 // the same atom then it is assumed to be fully resolved. This is needed so 683 // a relocation entry is not created and so the static linker does not 684 // mess up the reference later. 685 else if(!FB.getAtom() && 686 SA.isTemporary() && SA.isInSection() && &SecA == &SecB){ 687 return true; 688 } 689 } 690 691 // If they are not in the same section, we can't compute the diff. 692 if (&SecA != &SecB) 693 return false; 694 695 const MCFragment *FA = SA.getFragment(); 696 697 // Bail if the symbol has no fragment. 698 if (!FA) 699 return false; 700 701 // If the atoms are the same, they are guaranteed to have the same address. 702 if (FA->getAtom() == FB.getAtom()) 703 return true; 704 705 // Otherwise, we can't prove this is fully resolved. 706 return false; 707 } 708 709 void MachObjectWriter::writeObject(MCAssembler &Asm, 710 const MCAsmLayout &Layout) { 711 // Compute symbol table information and bind symbol indices. 712 computeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData, 713 UndefinedSymbolData); 714 715 unsigned NumSections = Asm.size(); 716 const MCAssembler::VersionMinInfoType &VersionInfo = 717 Layout.getAssembler().getVersionMinInfo(); 718 719 // The section data starts after the header, the segment load command (and 720 // section headers) and the symbol table. 721 unsigned NumLoadCommands = 1; 722 uint64_t LoadCommandsSize = is64Bit() ? 723 sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64): 724 sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section); 725 726 // Add the deployment target version info load command size, if used. 727 if (VersionInfo.Major != 0) { 728 ++NumLoadCommands; 729 LoadCommandsSize += sizeof(MachO::version_min_command); 730 } 731 732 // Add the data-in-code load command size, if used. 733 unsigned NumDataRegions = Asm.getDataRegions().size(); 734 if (NumDataRegions) { 735 ++NumLoadCommands; 736 LoadCommandsSize += sizeof(MachO::linkedit_data_command); 737 } 738 739 // Add the loh load command size, if used. 740 uint64_t LOHRawSize = Asm.getLOHContainer().getEmitSize(*this, Layout); 741 uint64_t LOHSize = alignTo(LOHRawSize, is64Bit() ? 8 : 4); 742 if (LOHSize) { 743 ++NumLoadCommands; 744 LoadCommandsSize += sizeof(MachO::linkedit_data_command); 745 } 746 747 // Add the symbol table load command sizes, if used. 748 unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() + 749 UndefinedSymbolData.size(); 750 if (NumSymbols) { 751 NumLoadCommands += 2; 752 LoadCommandsSize += (sizeof(MachO::symtab_command) + 753 sizeof(MachO::dysymtab_command)); 754 } 755 756 // Add the linker option load commands sizes. 757 for (const auto &Option : Asm.getLinkerOptions()) { 758 ++NumLoadCommands; 759 LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(Option, is64Bit()); 760 } 761 762 // Compute the total size of the section data, as well as its file size and vm 763 // size. 764 uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) : 765 sizeof(MachO::mach_header)) + LoadCommandsSize; 766 uint64_t SectionDataSize = 0; 767 uint64_t SectionDataFileSize = 0; 768 uint64_t VMSize = 0; 769 for (const MCSection &Sec : Asm) { 770 uint64_t Address = getSectionAddress(&Sec); 771 uint64_t Size = Layout.getSectionAddressSize(&Sec); 772 uint64_t FileSize = Layout.getSectionFileSize(&Sec); 773 FileSize += getPaddingSize(&Sec, Layout); 774 775 VMSize = std::max(VMSize, Address + Size); 776 777 if (Sec.isVirtualSection()) 778 continue; 779 780 SectionDataSize = std::max(SectionDataSize, Address + Size); 781 SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize); 782 } 783 784 // The section data is padded to 4 bytes. 785 // 786 // FIXME: Is this machine dependent? 787 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4); 788 SectionDataFileSize += SectionDataPadding; 789 790 // Write the prolog, starting with the header and load command... 791 writeHeader(MachO::MH_OBJECT, NumLoadCommands, LoadCommandsSize, 792 Asm.getSubsectionsViaSymbols()); 793 uint32_t Prot = 794 MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE; 795 writeSegmentLoadCommand("", NumSections, 0, VMSize, SectionDataStart, 796 SectionDataSize, Prot, Prot); 797 798 // ... and then the section headers. 799 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize; 800 for (const MCSection &Section : Asm) { 801 const auto &Sec = cast<MCSectionMachO>(Section); 802 std::vector<RelAndSymbol> &Relocs = Relocations[&Sec]; 803 unsigned NumRelocs = Relocs.size(); 804 uint64_t SectionStart = SectionDataStart + getSectionAddress(&Sec); 805 unsigned Flags = Sec.getTypeAndAttributes(); 806 if (Sec.hasInstructions()) 807 Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS; 808 writeSection(Layout, Sec, getSectionAddress(&Sec), SectionStart, Flags, 809 RelocTableEnd, NumRelocs); 810 RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info); 811 } 812 813 // Write out the deployment target information, if it's available. 814 if (VersionInfo.Major != 0) { 815 assert(VersionInfo.Update < 256 && "unencodable update target version"); 816 assert(VersionInfo.Minor < 256 && "unencodable minor target version"); 817 assert(VersionInfo.Major < 65536 && "unencodable major target version"); 818 uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) | 819 (VersionInfo.Major << 16); 820 MachO::LoadCommandType LCType; 821 switch (VersionInfo.Kind) { 822 case MCVM_OSXVersionMin: 823 LCType = MachO::LC_VERSION_MIN_MACOSX; 824 break; 825 case MCVM_IOSVersionMin: 826 LCType = MachO::LC_VERSION_MIN_IPHONEOS; 827 break; 828 case MCVM_TvOSVersionMin: 829 LCType = MachO::LC_VERSION_MIN_TVOS; 830 break; 831 case MCVM_WatchOSVersionMin: 832 LCType = MachO::LC_VERSION_MIN_WATCHOS; 833 break; 834 } 835 write32(LCType); 836 write32(sizeof(MachO::version_min_command)); 837 write32(EncodedVersion); 838 write32(0); // reserved. 839 } 840 841 // Write the data-in-code load command, if used. 842 uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8; 843 if (NumDataRegions) { 844 uint64_t DataRegionsOffset = RelocTableEnd; 845 uint64_t DataRegionsSize = NumDataRegions * 8; 846 writeLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset, 847 DataRegionsSize); 848 } 849 850 // Write the loh load command, if used. 851 uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize; 852 if (LOHSize) 853 writeLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT, 854 DataInCodeTableEnd, LOHSize); 855 856 // Write the symbol table load command, if used. 857 if (NumSymbols) { 858 unsigned FirstLocalSymbol = 0; 859 unsigned NumLocalSymbols = LocalSymbolData.size(); 860 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols; 861 unsigned NumExternalSymbols = ExternalSymbolData.size(); 862 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols; 863 unsigned NumUndefinedSymbols = UndefinedSymbolData.size(); 864 unsigned NumIndirectSymbols = Asm.indirect_symbol_size(); 865 unsigned NumSymTabSymbols = 866 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols; 867 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4; 868 uint64_t IndirectSymbolOffset = 0; 869 870 // If used, the indirect symbols are written after the section data. 871 if (NumIndirectSymbols) 872 IndirectSymbolOffset = LOHTableEnd; 873 874 // The symbol table is written after the indirect symbol data. 875 uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize; 876 877 // The string table is written after symbol table. 878 uint64_t StringTableOffset = 879 SymbolTableOffset + NumSymTabSymbols * (is64Bit() ? 880 sizeof(MachO::nlist_64) : 881 sizeof(MachO::nlist)); 882 writeSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols, 883 StringTableOffset, StringTable.data().size()); 884 885 writeDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols, 886 FirstExternalSymbol, NumExternalSymbols, 887 FirstUndefinedSymbol, NumUndefinedSymbols, 888 IndirectSymbolOffset, NumIndirectSymbols); 889 } 890 891 // Write the linker options load commands. 892 for (const auto &Option : Asm.getLinkerOptions()) 893 writeLinkerOptionsLoadCommand(Option); 894 895 // Write the actual section data. 896 for (const MCSection &Sec : Asm) { 897 Asm.writeSectionData(&Sec, Layout); 898 899 uint64_t Pad = getPaddingSize(&Sec, Layout); 900 WriteZeros(Pad); 901 } 902 903 // Write the extra padding. 904 WriteZeros(SectionDataPadding); 905 906 // Write the relocation entries. 907 for (const MCSection &Sec : Asm) { 908 // Write the section relocation entries, in reverse order to match 'as' 909 // (approximately, the exact algorithm is more complicated than this). 910 std::vector<RelAndSymbol> &Relocs = Relocations[&Sec]; 911 for (const RelAndSymbol &Rel : make_range(Relocs.rbegin(), Relocs.rend())) { 912 write32(Rel.MRE.r_word0); 913 write32(Rel.MRE.r_word1); 914 } 915 } 916 917 // Write out the data-in-code region payload, if there is one. 918 for (MCAssembler::const_data_region_iterator 919 it = Asm.data_region_begin(), ie = Asm.data_region_end(); 920 it != ie; ++it) { 921 const DataRegionData *Data = &(*it); 922 uint64_t Start = getSymbolAddress(*Data->Start, Layout); 923 uint64_t End = getSymbolAddress(*Data->End, Layout); 924 DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind 925 << " start: " << Start << "(" << Data->Start->getName() << ")" 926 << " end: " << End << "(" << Data->End->getName() << ")" 927 << " size: " << End - Start 928 << "\n"); 929 write32(Start); 930 write16(End - Start); 931 write16(Data->Kind); 932 } 933 934 // Write out the loh commands, if there is one. 935 if (LOHSize) { 936 #ifndef NDEBUG 937 unsigned Start = getStream().tell(); 938 #endif 939 Asm.getLOHContainer().emit(*this, Layout); 940 // Pad to a multiple of the pointer size. 941 writeBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4)); 942 assert(getStream().tell() - Start == LOHSize); 943 } 944 945 // Write the symbol table data, if used. 946 if (NumSymbols) { 947 // Write the indirect symbol entries. 948 for (MCAssembler::const_indirect_symbol_iterator 949 it = Asm.indirect_symbol_begin(), 950 ie = Asm.indirect_symbol_end(); it != ie; ++it) { 951 // Indirect symbols in the non-lazy symbol pointer section have some 952 // special handling. 953 const MCSectionMachO &Section = 954 static_cast<const MCSectionMachO &>(*it->Section); 955 if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) { 956 // If this symbol is defined and internal, mark it as such. 957 if (it->Symbol->isDefined() && !it->Symbol->isExternal()) { 958 uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL; 959 if (it->Symbol->isAbsolute()) 960 Flags |= MachO::INDIRECT_SYMBOL_ABS; 961 write32(Flags); 962 continue; 963 } 964 } 965 966 write32(it->Symbol->getIndex()); 967 } 968 969 // FIXME: Check that offsets match computed ones. 970 971 // Write the symbol table entries. 972 for (auto *SymbolData : 973 {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData}) 974 for (MachSymbolData &Entry : *SymbolData) 975 writeNlist(Entry, Layout); 976 977 // Write the string table. 978 getStream() << StringTable.data(); 979 } 980 } 981 982 MCObjectWriter *llvm::createMachObjectWriter(MCMachObjectTargetWriter *MOTW, 983 raw_pwrite_stream &OS, 984 bool IsLittleEndian) { 985 return new MachObjectWriter(MOTW, OS, IsLittleEndian); 986 } 987