MachObjectWriter.cpp revision 263508
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/MCMachOSymbolFlags.h" 19#include "llvm/MC/MCObjectWriter.h" 20#include "llvm/MC/MCSectionMachO.h" 21#include "llvm/MC/MCSymbol.h" 22#include "llvm/MC/MCValue.h" 23#include "llvm/Support/Debug.h" 24#include "llvm/Support/ErrorHandling.h" 25#include "llvm/Support/MachO.h" 26#include <vector> 27using namespace llvm; 28 29void MachObjectWriter::reset() { 30 Relocations.clear(); 31 IndirectSymBase.clear(); 32 StringTable.clear(); 33 LocalSymbolData.clear(); 34 ExternalSymbolData.clear(); 35 UndefinedSymbolData.clear(); 36 MCObjectWriter::reset(); 37} 38 39bool MachObjectWriter:: 40doesSymbolRequireExternRelocation(const MCSymbolData *SD) { 41 // Undefined symbols are always extern. 42 if (SD->Symbol->isUndefined()) 43 return true; 44 45 // References to weak definitions require external relocation entries; the 46 // definition may not always be the one in the same object file. 47 if (SD->getFlags() & SF_WeakDefinition) 48 return true; 49 50 // Otherwise, we can use an internal relocation. 51 return false; 52} 53 54bool MachObjectWriter:: 55MachSymbolData::operator<(const MachSymbolData &RHS) const { 56 return SymbolData->getSymbol().getName() < 57 RHS.SymbolData->getSymbol().getName(); 58} 59 60bool 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 67uint64_t MachObjectWriter::getFragmentAddress(const MCFragment *Fragment, 68 const MCAsmLayout &Layout) const { 69 return getSectionAddress(Fragment->getParent()) + 70 Layout.getFragmentOffset(Fragment); 71} 72 73uint64_t MachObjectWriter::getSymbolAddress(const MCSymbolData* SD, 74 const MCAsmLayout &Layout) const { 75 const MCSymbol &S = SD->getSymbol(); 76 77 // If this is a variable, then recursively evaluate now. 78 if (S.isVariable()) { 79 if (const MCConstantExpr *C = 80 dyn_cast<const MCConstantExpr>(S.getVariableValue())) 81 return C->getValue(); 82 83 84 MCValue Target; 85 if (!S.getVariableValue()->EvaluateAsRelocatable(Target, Layout)) 86 report_fatal_error("unable to evaluate offset for variable '" + 87 S.getName() + "'"); 88 89 // Verify that any used symbols are defined. 90 if (Target.getSymA() && Target.getSymA()->getSymbol().isUndefined()) 91 report_fatal_error("unable to evaluate offset to undefined symbol '" + 92 Target.getSymA()->getSymbol().getName() + "'"); 93 if (Target.getSymB() && Target.getSymB()->getSymbol().isUndefined()) 94 report_fatal_error("unable to evaluate offset to undefined symbol '" + 95 Target.getSymB()->getSymbol().getName() + "'"); 96 97 uint64_t Address = Target.getConstant(); 98 if (Target.getSymA()) 99 Address += getSymbolAddress(&Layout.getAssembler().getSymbolData( 100 Target.getSymA()->getSymbol()), Layout); 101 if (Target.getSymB()) 102 Address += getSymbolAddress(&Layout.getAssembler().getSymbolData( 103 Target.getSymB()->getSymbol()), Layout); 104 return Address; 105 } 106 107 return getSectionAddress(SD->getFragment()->getParent()) + 108 Layout.getSymbolOffset(SD); 109} 110 111uint64_t MachObjectWriter::getPaddingSize(const MCSectionData *SD, 112 const MCAsmLayout &Layout) const { 113 uint64_t EndAddr = getSectionAddress(SD) + Layout.getSectionAddressSize(SD); 114 unsigned Next = SD->getLayoutOrder() + 1; 115 if (Next >= Layout.getSectionOrder().size()) 116 return 0; 117 118 const MCSectionData &NextSD = *Layout.getSectionOrder()[Next]; 119 if (NextSD.getSection().isVirtualSection()) 120 return 0; 121 return OffsetToAlignment(EndAddr, NextSD.getAlignment()); 122} 123 124void MachObjectWriter::WriteHeader(unsigned NumLoadCommands, 125 unsigned LoadCommandsSize, 126 bool SubsectionsViaSymbols) { 127 uint32_t Flags = 0; 128 129 if (SubsectionsViaSymbols) 130 Flags |= MachO::MH_SUBSECTIONS_VIA_SYMBOLS; 131 132 // struct mach_header (28 bytes) or 133 // struct mach_header_64 (32 bytes) 134 135 uint64_t Start = OS.tell(); 136 (void) Start; 137 138 Write32(is64Bit() ? MachO::MH_MAGIC_64 : MachO::MH_MAGIC); 139 140 Write32(TargetObjectWriter->getCPUType()); 141 Write32(TargetObjectWriter->getCPUSubtype()); 142 143 Write32(MachO::MH_OBJECT); 144 Write32(NumLoadCommands); 145 Write32(LoadCommandsSize); 146 Write32(Flags); 147 if (is64Bit()) 148 Write32(0); // reserved 149 150 assert(OS.tell() - Start == 151 (is64Bit()?sizeof(MachO::mach_header_64): sizeof(MachO::mach_header))); 152} 153 154/// WriteSegmentLoadCommand - Write a segment load command. 155/// 156/// \param NumSections The number of sections in this segment. 157/// \param SectionDataSize The total size of the sections. 158void MachObjectWriter::WriteSegmentLoadCommand(unsigned NumSections, 159 uint64_t VMSize, 160 uint64_t SectionDataStartOffset, 161 uint64_t SectionDataSize) { 162 // struct segment_command (56 bytes) or 163 // struct segment_command_64 (72 bytes) 164 165 uint64_t Start = OS.tell(); 166 (void) Start; 167 168 unsigned SegmentLoadCommandSize = 169 is64Bit() ? sizeof(MachO::segment_command_64): 170 sizeof(MachO::segment_command); 171 Write32(is64Bit() ? MachO::LC_SEGMENT_64 : MachO::LC_SEGMENT); 172 Write32(SegmentLoadCommandSize + 173 NumSections * (is64Bit() ? sizeof(MachO::section_64) : 174 sizeof(MachO::section))); 175 176 WriteBytes("", 16); 177 if (is64Bit()) { 178 Write64(0); // vmaddr 179 Write64(VMSize); // vmsize 180 Write64(SectionDataStartOffset); // file offset 181 Write64(SectionDataSize); // file size 182 } else { 183 Write32(0); // vmaddr 184 Write32(VMSize); // vmsize 185 Write32(SectionDataStartOffset); // file offset 186 Write32(SectionDataSize); // file size 187 } 188 // maxprot 189 Write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE); 190 // initprot 191 Write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE); 192 Write32(NumSections); 193 Write32(0); // flags 194 195 assert(OS.tell() - Start == SegmentLoadCommandSize); 196} 197 198void MachObjectWriter::WriteSection(const MCAssembler &Asm, 199 const MCAsmLayout &Layout, 200 const MCSectionData &SD, 201 uint64_t FileOffset, 202 uint64_t RelocationsStart, 203 unsigned NumRelocations) { 204 uint64_t SectionSize = Layout.getSectionAddressSize(&SD); 205 206 // The offset is unused for virtual sections. 207 if (SD.getSection().isVirtualSection()) { 208 assert(Layout.getSectionFileSize(&SD) == 0 && "Invalid file size!"); 209 FileOffset = 0; 210 } 211 212 // struct section (68 bytes) or 213 // struct section_64 (80 bytes) 214 215 uint64_t Start = OS.tell(); 216 (void) Start; 217 218 const MCSectionMachO &Section = cast<MCSectionMachO>(SD.getSection()); 219 WriteBytes(Section.getSectionName(), 16); 220 WriteBytes(Section.getSegmentName(), 16); 221 if (is64Bit()) { 222 Write64(getSectionAddress(&SD)); // address 223 Write64(SectionSize); // size 224 } else { 225 Write32(getSectionAddress(&SD)); // address 226 Write32(SectionSize); // size 227 } 228 Write32(FileOffset); 229 230 unsigned Flags = Section.getTypeAndAttributes(); 231 if (SD.hasInstructions()) 232 Flags |= MCSectionMachO::S_ATTR_SOME_INSTRUCTIONS; 233 234 assert(isPowerOf2_32(SD.getAlignment()) && "Invalid alignment!"); 235 Write32(Log2_32(SD.getAlignment())); 236 Write32(NumRelocations ? RelocationsStart : 0); 237 Write32(NumRelocations); 238 Write32(Flags); 239 Write32(IndirectSymBase.lookup(&SD)); // reserved1 240 Write32(Section.getStubSize()); // reserved2 241 if (is64Bit()) 242 Write32(0); // reserved3 243 244 assert(OS.tell() - Start == (is64Bit() ? sizeof(MachO::section_64) : 245 sizeof(MachO::section))); 246} 247 248void MachObjectWriter::WriteSymtabLoadCommand(uint32_t SymbolOffset, 249 uint32_t NumSymbols, 250 uint32_t StringTableOffset, 251 uint32_t StringTableSize) { 252 // struct symtab_command (24 bytes) 253 254 uint64_t Start = OS.tell(); 255 (void) Start; 256 257 Write32(MachO::LC_SYMTAB); 258 Write32(sizeof(MachO::symtab_command)); 259 Write32(SymbolOffset); 260 Write32(NumSymbols); 261 Write32(StringTableOffset); 262 Write32(StringTableSize); 263 264 assert(OS.tell() - Start == sizeof(MachO::symtab_command)); 265} 266 267void MachObjectWriter::WriteDysymtabLoadCommand(uint32_t FirstLocalSymbol, 268 uint32_t NumLocalSymbols, 269 uint32_t FirstExternalSymbol, 270 uint32_t NumExternalSymbols, 271 uint32_t FirstUndefinedSymbol, 272 uint32_t NumUndefinedSymbols, 273 uint32_t IndirectSymbolOffset, 274 uint32_t NumIndirectSymbols) { 275 // struct dysymtab_command (80 bytes) 276 277 uint64_t Start = OS.tell(); 278 (void) Start; 279 280 Write32(MachO::LC_DYSYMTAB); 281 Write32(sizeof(MachO::dysymtab_command)); 282 Write32(FirstLocalSymbol); 283 Write32(NumLocalSymbols); 284 Write32(FirstExternalSymbol); 285 Write32(NumExternalSymbols); 286 Write32(FirstUndefinedSymbol); 287 Write32(NumUndefinedSymbols); 288 Write32(0); // tocoff 289 Write32(0); // ntoc 290 Write32(0); // modtaboff 291 Write32(0); // nmodtab 292 Write32(0); // extrefsymoff 293 Write32(0); // nextrefsyms 294 Write32(IndirectSymbolOffset); 295 Write32(NumIndirectSymbols); 296 Write32(0); // extreloff 297 Write32(0); // nextrel 298 Write32(0); // locreloff 299 Write32(0); // nlocrel 300 301 assert(OS.tell() - Start == sizeof(MachO::dysymtab_command)); 302} 303 304void MachObjectWriter::WriteNlist(MachSymbolData &MSD, 305 const MCAsmLayout &Layout) { 306 MCSymbolData &Data = *MSD.SymbolData; 307 const MCSymbol &Symbol = Data.getSymbol(); 308 uint8_t Type = 0; 309 uint16_t Flags = Data.getFlags(); 310 uint64_t Address = 0; 311 312 // Set the N_TYPE bits. See <mach-o/nlist.h>. 313 // 314 // FIXME: Are the prebound or indirect fields possible here? 315 if (Symbol.isUndefined()) 316 Type = MachO::N_UNDF; 317 else if (Symbol.isAbsolute()) 318 Type = MachO::N_ABS; 319 else 320 Type = MachO::N_SECT; 321 322 // FIXME: Set STAB bits. 323 324 if (Data.isPrivateExtern()) 325 Type |= MachO::N_PEXT; 326 327 // Set external bit. 328 if (Data.isExternal() || Symbol.isUndefined()) 329 Type |= MachO::N_EXT; 330 331 // Compute the symbol address. 332 if (Symbol.isDefined()) { 333 Address = getSymbolAddress(&Data, Layout); 334 } else if (Data.isCommon()) { 335 // Common symbols are encoded with the size in the address 336 // field, and their alignment in the flags. 337 Address = Data.getCommonSize(); 338 339 // Common alignment is packed into the 'desc' bits. 340 if (unsigned Align = Data.getCommonAlignment()) { 341 unsigned Log2Size = Log2_32(Align); 342 assert((1U << Log2Size) == Align && "Invalid 'common' alignment!"); 343 if (Log2Size > 15) 344 report_fatal_error("invalid 'common' alignment '" + 345 Twine(Align) + "' for '" + Symbol.getName() + "'", 346 false); 347 // FIXME: Keep this mask with the SymbolFlags enumeration. 348 Flags = (Flags & 0xF0FF) | (Log2Size << 8); 349 } 350 } 351 352 // struct nlist (12 bytes) 353 354 Write32(MSD.StringIndex); 355 Write8(Type); 356 Write8(MSD.SectionIndex); 357 358 // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc' 359 // value. 360 Write16(Flags); 361 if (is64Bit()) 362 Write64(Address); 363 else 364 Write32(Address); 365} 366 367void MachObjectWriter::WriteLinkeditLoadCommand(uint32_t Type, 368 uint32_t DataOffset, 369 uint32_t DataSize) { 370 uint64_t Start = OS.tell(); 371 (void) Start; 372 373 Write32(Type); 374 Write32(sizeof(MachO::linkedit_data_command)); 375 Write32(DataOffset); 376 Write32(DataSize); 377 378 assert(OS.tell() - Start == sizeof(MachO::linkedit_data_command)); 379} 380 381static unsigned ComputeLinkerOptionsLoadCommandSize( 382 const std::vector<std::string> &Options, bool is64Bit) 383{ 384 unsigned Size = sizeof(MachO::linker_options_command); 385 for (unsigned i = 0, e = Options.size(); i != e; ++i) 386 Size += Options[i].size() + 1; 387 return RoundUpToAlignment(Size, is64Bit ? 8 : 4); 388} 389 390void MachObjectWriter::WriteLinkerOptionsLoadCommand( 391 const std::vector<std::string> &Options) 392{ 393 unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit()); 394 uint64_t Start = OS.tell(); 395 (void) Start; 396 397 Write32(MachO::LC_LINKER_OPTIONS); 398 Write32(Size); 399 Write32(Options.size()); 400 uint64_t BytesWritten = sizeof(MachO::linker_options_command); 401 for (unsigned i = 0, e = Options.size(); i != e; ++i) { 402 // Write each string, including the null byte. 403 const std::string &Option = Options[i]; 404 WriteBytes(Option.c_str(), Option.size() + 1); 405 BytesWritten += Option.size() + 1; 406 } 407 408 // Pad to a multiple of the pointer size. 409 WriteBytes("", OffsetToAlignment(BytesWritten, is64Bit() ? 8 : 4)); 410 411 assert(OS.tell() - Start == Size); 412} 413 414 415void MachObjectWriter::RecordRelocation(const MCAssembler &Asm, 416 const MCAsmLayout &Layout, 417 const MCFragment *Fragment, 418 const MCFixup &Fixup, 419 MCValue Target, 420 uint64_t &FixedValue) { 421 TargetObjectWriter->RecordRelocation(this, Asm, Layout, Fragment, Fixup, 422 Target, FixedValue); 423} 424 425void MachObjectWriter::BindIndirectSymbols(MCAssembler &Asm) { 426 // This is the point where 'as' creates actual symbols for indirect symbols 427 // (in the following two passes). It would be easier for us to do this sooner 428 // when we see the attribute, but that makes getting the order in the symbol 429 // table much more complicated than it is worth. 430 // 431 // FIXME: Revisit this when the dust settles. 432 433 // Report errors for use of .indirect_symbol not in a symbol pointer section 434 // or stub section. 435 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(), 436 ie = Asm.indirect_symbol_end(); it != ie; ++it) { 437 const MCSectionMachO &Section = 438 cast<MCSectionMachO>(it->SectionData->getSection()); 439 440 if (Section.getType() != MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS && 441 Section.getType() != MCSectionMachO::S_LAZY_SYMBOL_POINTERS && 442 Section.getType() != MCSectionMachO::S_SYMBOL_STUBS) { 443 MCSymbol &Symbol = *it->Symbol; 444 report_fatal_error("indirect symbol '" + Symbol.getName() + 445 "' not in a symbol pointer or stub section"); 446 } 447 } 448 449 // Bind non lazy symbol pointers first. 450 unsigned IndirectIndex = 0; 451 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(), 452 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) { 453 const MCSectionMachO &Section = 454 cast<MCSectionMachO>(it->SectionData->getSection()); 455 456 if (Section.getType() != MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS) 457 continue; 458 459 // Initialize the section indirect symbol base, if necessary. 460 IndirectSymBase.insert(std::make_pair(it->SectionData, IndirectIndex)); 461 462 Asm.getOrCreateSymbolData(*it->Symbol); 463 } 464 465 // Then lazy symbol pointers and symbol stubs. 466 IndirectIndex = 0; 467 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(), 468 ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) { 469 const MCSectionMachO &Section = 470 cast<MCSectionMachO>(it->SectionData->getSection()); 471 472 if (Section.getType() != MCSectionMachO::S_LAZY_SYMBOL_POINTERS && 473 Section.getType() != MCSectionMachO::S_SYMBOL_STUBS) 474 continue; 475 476 // Initialize the section indirect symbol base, if necessary. 477 IndirectSymBase.insert(std::make_pair(it->SectionData, IndirectIndex)); 478 479 // Set the symbol type to undefined lazy, but only on construction. 480 // 481 // FIXME: Do not hardcode. 482 bool Created; 483 MCSymbolData &Entry = Asm.getOrCreateSymbolData(*it->Symbol, &Created); 484 if (Created) 485 Entry.setFlags(Entry.getFlags() | 0x0001); 486 } 487} 488 489/// ComputeSymbolTable - Compute the symbol table data 490/// 491/// \param StringTable [out] - The string table data. 492/// \param StringIndexMap [out] - Map from symbol names to offsets in the 493/// string table. 494void MachObjectWriter:: 495ComputeSymbolTable(MCAssembler &Asm, SmallString<256> &StringTable, 496 std::vector<MachSymbolData> &LocalSymbolData, 497 std::vector<MachSymbolData> &ExternalSymbolData, 498 std::vector<MachSymbolData> &UndefinedSymbolData) { 499 // Build section lookup table. 500 DenseMap<const MCSection*, uint8_t> SectionIndexMap; 501 unsigned Index = 1; 502 for (MCAssembler::iterator it = Asm.begin(), 503 ie = Asm.end(); it != ie; ++it, ++Index) 504 SectionIndexMap[&it->getSection()] = Index; 505 assert(Index <= 256 && "Too many sections!"); 506 507 // Index 0 is always the empty string. 508 StringMap<uint64_t> StringIndexMap; 509 StringTable += '\x00'; 510 511 // Build the symbol arrays and the string table, but only for non-local 512 // symbols. 513 // 514 // The particular order that we collect the symbols and create the string 515 // table, then sort the symbols is chosen to match 'as'. Even though it 516 // doesn't matter for correctness, this is important for letting us diff .o 517 // files. 518 for (MCAssembler::symbol_iterator it = Asm.symbol_begin(), 519 ie = Asm.symbol_end(); it != ie; ++it) { 520 const MCSymbol &Symbol = it->getSymbol(); 521 522 // Ignore non-linker visible symbols. 523 if (!Asm.isSymbolLinkerVisible(it->getSymbol())) 524 continue; 525 526 if (!it->isExternal() && !Symbol.isUndefined()) 527 continue; 528 529 uint64_t &Entry = StringIndexMap[Symbol.getName()]; 530 if (!Entry) { 531 Entry = StringTable.size(); 532 StringTable += Symbol.getName(); 533 StringTable += '\x00'; 534 } 535 536 MachSymbolData MSD; 537 MSD.SymbolData = it; 538 MSD.StringIndex = Entry; 539 540 if (Symbol.isUndefined()) { 541 MSD.SectionIndex = 0; 542 UndefinedSymbolData.push_back(MSD); 543 } else if (Symbol.isAbsolute()) { 544 MSD.SectionIndex = 0; 545 ExternalSymbolData.push_back(MSD); 546 } else { 547 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection()); 548 assert(MSD.SectionIndex && "Invalid section index!"); 549 ExternalSymbolData.push_back(MSD); 550 } 551 } 552 553 // Now add the data for local symbols. 554 for (MCAssembler::symbol_iterator it = Asm.symbol_begin(), 555 ie = Asm.symbol_end(); it != ie; ++it) { 556 const MCSymbol &Symbol = it->getSymbol(); 557 558 // Ignore non-linker visible symbols. 559 if (!Asm.isSymbolLinkerVisible(it->getSymbol())) 560 continue; 561 562 if (it->isExternal() || Symbol.isUndefined()) 563 continue; 564 565 uint64_t &Entry = StringIndexMap[Symbol.getName()]; 566 if (!Entry) { 567 Entry = StringTable.size(); 568 StringTable += Symbol.getName(); 569 StringTable += '\x00'; 570 } 571 572 MachSymbolData MSD; 573 MSD.SymbolData = it; 574 MSD.StringIndex = Entry; 575 576 if (Symbol.isAbsolute()) { 577 MSD.SectionIndex = 0; 578 LocalSymbolData.push_back(MSD); 579 } else { 580 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection()); 581 assert(MSD.SectionIndex && "Invalid section index!"); 582 LocalSymbolData.push_back(MSD); 583 } 584 } 585 586 // External and undefined symbols are required to be in lexicographic order. 587 std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end()); 588 std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end()); 589 590 // Set the symbol indices. 591 Index = 0; 592 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i) 593 LocalSymbolData[i].SymbolData->setIndex(Index++); 594 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i) 595 ExternalSymbolData[i].SymbolData->setIndex(Index++); 596 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i) 597 UndefinedSymbolData[i].SymbolData->setIndex(Index++); 598 599 // The string table is padded to a multiple of 4. 600 while (StringTable.size() % 4) 601 StringTable += '\x00'; 602} 603 604void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm, 605 const MCAsmLayout &Layout) { 606 uint64_t StartAddress = 0; 607 const SmallVectorImpl<MCSectionData*> &Order = Layout.getSectionOrder(); 608 for (int i = 0, n = Order.size(); i != n ; ++i) { 609 const MCSectionData *SD = Order[i]; 610 StartAddress = RoundUpToAlignment(StartAddress, SD->getAlignment()); 611 SectionAddress[SD] = StartAddress; 612 StartAddress += Layout.getSectionAddressSize(SD); 613 614 // Explicitly pad the section to match the alignment requirements of the 615 // following one. This is for 'gas' compatibility, it shouldn't 616 /// strictly be necessary. 617 StartAddress += getPaddingSize(SD, Layout); 618 } 619} 620 621void MachObjectWriter::markAbsoluteVariableSymbols(MCAssembler &Asm, 622 const MCAsmLayout &Layout) { 623 for (MCAssembler::symbol_iterator i = Asm.symbol_begin(), 624 e = Asm.symbol_end(); 625 i != e; ++i) { 626 MCSymbolData &SD = *i; 627 if (!SD.getSymbol().isVariable()) 628 continue; 629 630 // Is the variable is a symbol difference (SA - SB + C) expression, 631 // and neither symbol is external, mark the variable as absolute. 632 const MCExpr *Expr = SD.getSymbol().getVariableValue(); 633 MCValue Value; 634 if (Expr->EvaluateAsRelocatable(Value, Layout)) { 635 if (Value.getSymA() && Value.getSymB()) 636 const_cast<MCSymbol*>(&SD.getSymbol())->setAbsolute(); 637 } 638 } 639} 640 641void MachObjectWriter::ExecutePostLayoutBinding(MCAssembler &Asm, 642 const MCAsmLayout &Layout) { 643 computeSectionAddresses(Asm, Layout); 644 645 // Create symbol data for any indirect symbols. 646 BindIndirectSymbols(Asm); 647 648 // Mark symbol difference expressions in variables (from .set or = directives) 649 // as absolute. 650 markAbsoluteVariableSymbols(Asm, Layout); 651 652 // Compute symbol table information and bind symbol indices. 653 ComputeSymbolTable(Asm, StringTable, LocalSymbolData, ExternalSymbolData, 654 UndefinedSymbolData); 655} 656 657bool MachObjectWriter:: 658IsSymbolRefDifferenceFullyResolvedImpl(const MCAssembler &Asm, 659 const MCSymbolData &DataA, 660 const MCFragment &FB, 661 bool InSet, 662 bool IsPCRel) const { 663 if (InSet) 664 return true; 665 666 // The effective address is 667 // addr(atom(A)) + offset(A) 668 // - addr(atom(B)) - offset(B) 669 // and the offsets are not relocatable, so the fixup is fully resolved when 670 // addr(atom(A)) - addr(atom(B)) == 0. 671 const MCSymbolData *A_Base = 0, *B_Base = 0; 672 673 const MCSymbol &SA = DataA.getSymbol().AliasedSymbol(); 674 const MCSection &SecA = SA.getSection(); 675 const MCSection &SecB = FB.getParent()->getSection(); 676 677 if (IsPCRel) { 678 // The simple (Darwin, except on x86_64) way of dealing with this was to 679 // assume that any reference to a temporary symbol *must* be a temporary 680 // symbol in the same atom, unless the sections differ. Therefore, any PCrel 681 // relocation to a temporary symbol (in the same section) is fully 682 // resolved. This also works in conjunction with absolutized .set, which 683 // requires the compiler to use .set to absolutize the differences between 684 // symbols which the compiler knows to be assembly time constants, so we 685 // don't need to worry about considering symbol differences fully resolved. 686 // 687 // If the file isn't using sub-sections-via-symbols, we can make the 688 // same assumptions about any symbol that we normally make about 689 // assembler locals. 690 691 if (!Asm.getBackend().hasReliableSymbolDifference()) { 692 if (!SA.isInSection() || &SecA != &SecB || 693 (!SA.isTemporary() && 694 FB.getAtom() != Asm.getSymbolData(SA).getFragment()->getAtom() && 695 Asm.getSubsectionsViaSymbols())) 696 return false; 697 return true; 698 } 699 // For Darwin x86_64, there is one special case when the reference IsPCRel. 700 // If the fragment with the reference does not have a base symbol but meets 701 // the simple way of dealing with this, in that it is a temporary symbol in 702 // the same atom then it is assumed to be fully resolved. This is needed so 703 // a relocation entry is not created and so the static linker does not 704 // mess up the reference later. 705 else if(!FB.getAtom() && 706 SA.isTemporary() && SA.isInSection() && &SecA == &SecB){ 707 return true; 708 } 709 } else { 710 if (!TargetObjectWriter->useAggressiveSymbolFolding()) 711 return false; 712 } 713 714 const MCFragment *FA = Asm.getSymbolData(SA).getFragment(); 715 716 // Bail if the symbol has no fragment. 717 if (!FA) 718 return false; 719 720 A_Base = FA->getAtom(); 721 if (!A_Base) 722 return false; 723 724 B_Base = FB.getAtom(); 725 if (!B_Base) 726 return false; 727 728 // If the atoms are the same, they are guaranteed to have the same address. 729 if (A_Base == B_Base) 730 return true; 731 732 // Otherwise, we can't prove this is fully resolved. 733 return false; 734} 735 736void MachObjectWriter::WriteObject(MCAssembler &Asm, 737 const MCAsmLayout &Layout) { 738 unsigned NumSections = Asm.size(); 739 740 // The section data starts after the header, the segment load command (and 741 // section headers) and the symbol table. 742 unsigned NumLoadCommands = 1; 743 uint64_t LoadCommandsSize = is64Bit() ? 744 sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64): 745 sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section); 746 747 // Add the data-in-code load command size, if used. 748 unsigned NumDataRegions = Asm.getDataRegions().size(); 749 if (NumDataRegions) { 750 ++NumLoadCommands; 751 LoadCommandsSize += sizeof(MachO::linkedit_data_command); 752 } 753 754 // Add the symbol table load command sizes, if used. 755 unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() + 756 UndefinedSymbolData.size(); 757 if (NumSymbols) { 758 NumLoadCommands += 2; 759 LoadCommandsSize += (sizeof(MachO::symtab_command) + 760 sizeof(MachO::dysymtab_command)); 761 } 762 763 // Add the linker option load commands sizes. 764 const std::vector<std::vector<std::string> > &LinkerOptions = 765 Asm.getLinkerOptions(); 766 for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) { 767 ++NumLoadCommands; 768 LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(LinkerOptions[i], 769 is64Bit()); 770 } 771 772 // Compute the total size of the section data, as well as its file size and vm 773 // size. 774 uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) : 775 sizeof(MachO::mach_header)) + LoadCommandsSize; 776 uint64_t SectionDataSize = 0; 777 uint64_t SectionDataFileSize = 0; 778 uint64_t VMSize = 0; 779 for (MCAssembler::const_iterator it = Asm.begin(), 780 ie = Asm.end(); it != ie; ++it) { 781 const MCSectionData &SD = *it; 782 uint64_t Address = getSectionAddress(&SD); 783 uint64_t Size = Layout.getSectionAddressSize(&SD); 784 uint64_t FileSize = Layout.getSectionFileSize(&SD); 785 FileSize += getPaddingSize(&SD, Layout); 786 787 VMSize = std::max(VMSize, Address + Size); 788 789 if (SD.getSection().isVirtualSection()) 790 continue; 791 792 SectionDataSize = std::max(SectionDataSize, Address + Size); 793 SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize); 794 } 795 796 // The section data is padded to 4 bytes. 797 // 798 // FIXME: Is this machine dependent? 799 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4); 800 SectionDataFileSize += SectionDataPadding; 801 802 // Write the prolog, starting with the header and load command... 803 WriteHeader(NumLoadCommands, LoadCommandsSize, 804 Asm.getSubsectionsViaSymbols()); 805 WriteSegmentLoadCommand(NumSections, VMSize, 806 SectionDataStart, SectionDataSize); 807 808 // ... and then the section headers. 809 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize; 810 for (MCAssembler::const_iterator it = Asm.begin(), 811 ie = Asm.end(); it != ie; ++it) { 812 std::vector<MachO::any_relocation_info> &Relocs = Relocations[it]; 813 unsigned NumRelocs = Relocs.size(); 814 uint64_t SectionStart = SectionDataStart + getSectionAddress(it); 815 WriteSection(Asm, Layout, *it, SectionStart, RelocTableEnd, NumRelocs); 816 RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info); 817 } 818 819 // Write the data-in-code load command, if used. 820 uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8; 821 if (NumDataRegions) { 822 uint64_t DataRegionsOffset = RelocTableEnd; 823 uint64_t DataRegionsSize = NumDataRegions * 8; 824 WriteLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset, 825 DataRegionsSize); 826 } 827 828 // Write the symbol table load command, if used. 829 if (NumSymbols) { 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; 841 842 // If used, the indirect symbols are written after the section data. 843 if (NumIndirectSymbols) 844 IndirectSymbolOffset = DataInCodeTableEnd; 845 846 // The symbol table is written after the indirect symbol data. 847 uint64_t SymbolTableOffset = DataInCodeTableEnd + IndirectSymbolSize; 848 849 // The string table is written after symbol table. 850 uint64_t StringTableOffset = 851 SymbolTableOffset + NumSymTabSymbols * (is64Bit() ? 852 sizeof(MachO::nlist_64) : 853 sizeof(MachO::nlist)); 854 WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols, 855 StringTableOffset, StringTable.size()); 856 857 WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols, 858 FirstExternalSymbol, NumExternalSymbols, 859 FirstUndefinedSymbol, NumUndefinedSymbols, 860 IndirectSymbolOffset, NumIndirectSymbols); 861 } 862 863 // Write the linker options load commands. 864 for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) { 865 WriteLinkerOptionsLoadCommand(LinkerOptions[i]); 866 } 867 868 // Write the actual section data. 869 for (MCAssembler::const_iterator it = Asm.begin(), 870 ie = Asm.end(); it != ie; ++it) { 871 Asm.writeSectionData(it, Layout); 872 873 uint64_t Pad = getPaddingSize(it, Layout); 874 for (unsigned int i = 0; i < Pad; ++i) 875 Write8(0); 876 } 877 878 // Write the extra padding. 879 WriteZeros(SectionDataPadding); 880 881 // Write the relocation entries. 882 for (MCAssembler::const_iterator it = Asm.begin(), 883 ie = Asm.end(); it != ie; ++it) { 884 // Write the section relocation entries, in reverse order to match 'as' 885 // (approximately, the exact algorithm is more complicated than this). 886 std::vector<MachO::any_relocation_info> &Relocs = Relocations[it]; 887 for (unsigned i = 0, e = Relocs.size(); i != e; ++i) { 888 Write32(Relocs[e - i - 1].r_word0); 889 Write32(Relocs[e - i - 1].r_word1); 890 } 891 } 892 893 // Write out the data-in-code region payload, if there is one. 894 for (MCAssembler::const_data_region_iterator 895 it = Asm.data_region_begin(), ie = Asm.data_region_end(); 896 it != ie; ++it) { 897 const DataRegionData *Data = &(*it); 898 uint64_t Start = 899 getSymbolAddress(&Layout.getAssembler().getSymbolData(*Data->Start), 900 Layout); 901 uint64_t End = 902 getSymbolAddress(&Layout.getAssembler().getSymbolData(*Data->End), 903 Layout); 904 DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind 905 << " start: " << Start << "(" << Data->Start->getName() << ")" 906 << " end: " << End << "(" << Data->End->getName() << ")" 907 << " size: " << End - Start 908 << "\n"); 909 Write32(Start); 910 Write16(End - Start); 911 Write16(Data->Kind); 912 } 913 914 // Write the symbol table data, if used. 915 if (NumSymbols) { 916 // Write the indirect symbol entries. 917 for (MCAssembler::const_indirect_symbol_iterator 918 it = Asm.indirect_symbol_begin(), 919 ie = Asm.indirect_symbol_end(); it != ie; ++it) { 920 // Indirect symbols in the non lazy symbol pointer section have some 921 // special handling. 922 const MCSectionMachO &Section = 923 static_cast<const MCSectionMachO&>(it->SectionData->getSection()); 924 if (Section.getType() == MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS) { 925 // If this symbol is defined and internal, mark it as such. 926 if (it->Symbol->isDefined() && 927 !Asm.getSymbolData(*it->Symbol).isExternal()) { 928 uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL; 929 if (it->Symbol->isAbsolute()) 930 Flags |= MachO::INDIRECT_SYMBOL_ABS; 931 Write32(Flags); 932 continue; 933 } 934 } 935 936 Write32(Asm.getSymbolData(*it->Symbol).getIndex()); 937 } 938 939 // FIXME: Check that offsets match computed ones. 940 941 // Write the symbol table entries. 942 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i) 943 WriteNlist(LocalSymbolData[i], Layout); 944 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i) 945 WriteNlist(ExternalSymbolData[i], Layout); 946 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i) 947 WriteNlist(UndefinedSymbolData[i], Layout); 948 949 // Write the string table. 950 OS << StringTable.str(); 951 } 952} 953 954MCObjectWriter *llvm::createMachObjectWriter(MCMachObjectTargetWriter *MOTW, 955 raw_ostream &OS, 956 bool IsLittleEndian) { 957 return new MachObjectWriter(MOTW, OS, IsLittleEndian); 958} 959