Daniel Dunbar | 2df4ceb | 2010-03-19 10:43:15 +0000 | [diff] [blame^] | 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/MachObjectWriter.h" |
| 11 | #include "llvm/ADT/StringMap.h" |
| 12 | #include "llvm/ADT/Twine.h" |
| 13 | #include "llvm/MC/MCAssembler.h" |
| 14 | #include "llvm/MC/MCExpr.h" |
| 15 | #include "llvm/MC/MCObjectWriter.h" |
| 16 | #include "llvm/MC/MCSectionMachO.h" |
| 17 | #include "llvm/MC/MCSymbol.h" |
| 18 | #include "llvm/MC/MCValue.h" |
| 19 | #include "llvm/Support/ErrorHandling.h" |
| 20 | #include "llvm/Support/MachO.h" |
| 21 | #include "llvm/Target/TargetAsmBackend.h" |
| 22 | |
| 23 | // FIXME: Gross. |
| 24 | #include "../Target/X86/X86FixupKinds.h" |
| 25 | |
| 26 | #include <vector> |
| 27 | using namespace llvm; |
| 28 | |
| 29 | static unsigned getFixupKindLog2Size(unsigned Kind) { |
| 30 | switch (Kind) { |
| 31 | default: llvm_unreachable("invalid fixup kind!"); |
| 32 | case X86::reloc_pcrel_1byte: |
| 33 | case FK_Data_1: return 0; |
| 34 | case FK_Data_2: return 1; |
| 35 | case X86::reloc_pcrel_4byte: |
| 36 | case X86::reloc_riprel_4byte: |
| 37 | case FK_Data_4: return 2; |
| 38 | case FK_Data_8: return 3; |
| 39 | } |
| 40 | } |
| 41 | |
| 42 | static bool isFixupKindPCRel(unsigned Kind) { |
| 43 | switch (Kind) { |
| 44 | default: |
| 45 | return false; |
| 46 | case X86::reloc_pcrel_1byte: |
| 47 | case X86::reloc_pcrel_4byte: |
| 48 | case X86::reloc_riprel_4byte: |
| 49 | return true; |
| 50 | } |
| 51 | } |
| 52 | |
| 53 | namespace { |
| 54 | |
| 55 | class MachObjectWriterImpl { |
| 56 | // See <mach-o/loader.h>. |
| 57 | enum { |
| 58 | Header_Magic32 = 0xFEEDFACE, |
| 59 | Header_Magic64 = 0xFEEDFACF |
| 60 | }; |
| 61 | |
| 62 | enum { |
| 63 | Header32Size = 28, |
| 64 | Header64Size = 32, |
| 65 | SegmentLoadCommand32Size = 56, |
| 66 | SegmentLoadCommand64Size = 72, |
| 67 | Section32Size = 68, |
| 68 | Section64Size = 80, |
| 69 | SymtabLoadCommandSize = 24, |
| 70 | DysymtabLoadCommandSize = 80, |
| 71 | Nlist32Size = 12, |
| 72 | Nlist64Size = 16, |
| 73 | RelocationInfoSize = 8 |
| 74 | }; |
| 75 | |
| 76 | enum HeaderFileType { |
| 77 | HFT_Object = 0x1 |
| 78 | }; |
| 79 | |
| 80 | enum HeaderFlags { |
| 81 | HF_SubsectionsViaSymbols = 0x2000 |
| 82 | }; |
| 83 | |
| 84 | enum LoadCommandType { |
| 85 | LCT_Segment = 0x1, |
| 86 | LCT_Symtab = 0x2, |
| 87 | LCT_Dysymtab = 0xb, |
| 88 | LCT_Segment64 = 0x19 |
| 89 | }; |
| 90 | |
| 91 | // See <mach-o/nlist.h>. |
| 92 | enum SymbolTypeType { |
| 93 | STT_Undefined = 0x00, |
| 94 | STT_Absolute = 0x02, |
| 95 | STT_Section = 0x0e |
| 96 | }; |
| 97 | |
| 98 | enum SymbolTypeFlags { |
| 99 | // If any of these bits are set, then the entry is a stab entry number (see |
| 100 | // <mach-o/stab.h>. Otherwise the other masks apply. |
| 101 | STF_StabsEntryMask = 0xe0, |
| 102 | |
| 103 | STF_TypeMask = 0x0e, |
| 104 | STF_External = 0x01, |
| 105 | STF_PrivateExtern = 0x10 |
| 106 | }; |
| 107 | |
| 108 | /// IndirectSymbolFlags - Flags for encoding special values in the indirect |
| 109 | /// symbol entry. |
| 110 | enum IndirectSymbolFlags { |
| 111 | ISF_Local = 0x80000000, |
| 112 | ISF_Absolute = 0x40000000 |
| 113 | }; |
| 114 | |
| 115 | /// RelocationFlags - Special flags for addresses. |
| 116 | enum RelocationFlags { |
| 117 | RF_Scattered = 0x80000000 |
| 118 | }; |
| 119 | |
| 120 | enum RelocationInfoType { |
| 121 | RIT_Vanilla = 0, |
| 122 | RIT_Pair = 1, |
| 123 | RIT_Difference = 2, |
| 124 | RIT_PreboundLazyPointer = 3, |
| 125 | RIT_LocalDifference = 4 |
| 126 | }; |
| 127 | |
| 128 | /// MachSymbolData - Helper struct for containing some precomputed information |
| 129 | /// on symbols. |
| 130 | struct MachSymbolData { |
| 131 | MCSymbolData *SymbolData; |
| 132 | uint64_t StringIndex; |
| 133 | uint8_t SectionIndex; |
| 134 | |
| 135 | // Support lexicographic sorting. |
| 136 | bool operator<(const MachSymbolData &RHS) const { |
| 137 | const std::string &Name = SymbolData->getSymbol().getName(); |
| 138 | return Name < RHS.SymbolData->getSymbol().getName(); |
| 139 | } |
| 140 | }; |
| 141 | |
| 142 | /// @name Relocation Data |
| 143 | /// @{ |
| 144 | |
| 145 | struct MachRelocationEntry { |
| 146 | uint32_t Word0; |
| 147 | uint32_t Word1; |
| 148 | }; |
| 149 | |
| 150 | llvm::DenseMap<const MCSectionData*, |
| 151 | std::vector<MachRelocationEntry> > Relocations; |
| 152 | |
| 153 | /// @} |
| 154 | /// @name Symbol Table Data |
| 155 | /// @{ |
| 156 | |
| 157 | SmallString<256> StringTable; |
| 158 | std::vector<MachSymbolData> LocalSymbolData; |
| 159 | std::vector<MachSymbolData> ExternalSymbolData; |
| 160 | std::vector<MachSymbolData> UndefinedSymbolData; |
| 161 | |
| 162 | /// @} |
| 163 | |
| 164 | MachObjectWriter *Writer; |
| 165 | |
| 166 | raw_ostream &OS; |
| 167 | |
| 168 | unsigned Is64Bit : 1; |
| 169 | |
| 170 | public: |
| 171 | MachObjectWriterImpl(MachObjectWriter *_Writer, bool _Is64Bit) |
| 172 | : Writer(_Writer), OS(Writer->getStream()), Is64Bit(_Is64Bit) { |
| 173 | } |
| 174 | |
| 175 | void Write8(uint8_t Value) { Writer->Write8(Value); } |
| 176 | void Write16(uint16_t Value) { Writer->Write16(Value); } |
| 177 | void Write32(uint32_t Value) { Writer->Write32(Value); } |
| 178 | void Write64(uint64_t Value) { Writer->Write64(Value); } |
| 179 | void WriteZeros(unsigned N) { Writer->WriteZeros(N); } |
| 180 | void WriteBytes(StringRef Str, unsigned ZeroFillSize = 0) { |
| 181 | Writer->WriteBytes(Str, ZeroFillSize); |
| 182 | } |
| 183 | |
| 184 | void WriteHeader(unsigned NumLoadCommands, unsigned LoadCommandsSize, |
| 185 | bool SubsectionsViaSymbols) { |
| 186 | uint32_t Flags = 0; |
| 187 | |
| 188 | if (SubsectionsViaSymbols) |
| 189 | Flags |= HF_SubsectionsViaSymbols; |
| 190 | |
| 191 | // struct mach_header (28 bytes) or |
| 192 | // struct mach_header_64 (32 bytes) |
| 193 | |
| 194 | uint64_t Start = OS.tell(); |
| 195 | (void) Start; |
| 196 | |
| 197 | Write32(Is64Bit ? Header_Magic64 : Header_Magic32); |
| 198 | |
| 199 | // FIXME: Support cputype. |
| 200 | Write32(Is64Bit ? MachO::CPUTypeX86_64 : MachO::CPUTypeI386); |
| 201 | // FIXME: Support cpusubtype. |
| 202 | Write32(MachO::CPUSubType_I386_ALL); |
| 203 | Write32(HFT_Object); |
| 204 | Write32(NumLoadCommands); // Object files have a single load command, the |
| 205 | // segment. |
| 206 | Write32(LoadCommandsSize); |
| 207 | Write32(Flags); |
| 208 | if (Is64Bit) |
| 209 | Write32(0); // reserved |
| 210 | |
| 211 | assert(OS.tell() - Start == Is64Bit ? Header64Size : Header32Size); |
| 212 | } |
| 213 | |
| 214 | /// WriteSegmentLoadCommand - Write a segment load command. |
| 215 | /// |
| 216 | /// \arg NumSections - The number of sections in this segment. |
| 217 | /// \arg SectionDataSize - The total size of the sections. |
| 218 | void WriteSegmentLoadCommand(unsigned NumSections, |
| 219 | uint64_t VMSize, |
| 220 | uint64_t SectionDataStartOffset, |
| 221 | uint64_t SectionDataSize) { |
| 222 | // struct segment_command (56 bytes) or |
| 223 | // struct segment_command_64 (72 bytes) |
| 224 | |
| 225 | uint64_t Start = OS.tell(); |
| 226 | (void) Start; |
| 227 | |
| 228 | unsigned SegmentLoadCommandSize = Is64Bit ? SegmentLoadCommand64Size : |
| 229 | SegmentLoadCommand32Size; |
| 230 | Write32(Is64Bit ? LCT_Segment64 : LCT_Segment); |
| 231 | Write32(SegmentLoadCommandSize + |
| 232 | NumSections * (Is64Bit ? Section64Size : Section32Size)); |
| 233 | |
| 234 | WriteBytes("", 16); |
| 235 | if (Is64Bit) { |
| 236 | Write64(0); // vmaddr |
| 237 | Write64(VMSize); // vmsize |
| 238 | Write64(SectionDataStartOffset); // file offset |
| 239 | Write64(SectionDataSize); // file size |
| 240 | } else { |
| 241 | Write32(0); // vmaddr |
| 242 | Write32(VMSize); // vmsize |
| 243 | Write32(SectionDataStartOffset); // file offset |
| 244 | Write32(SectionDataSize); // file size |
| 245 | } |
| 246 | Write32(0x7); // maxprot |
| 247 | Write32(0x7); // initprot |
| 248 | Write32(NumSections); |
| 249 | Write32(0); // flags |
| 250 | |
| 251 | assert(OS.tell() - Start == SegmentLoadCommandSize); |
| 252 | } |
| 253 | |
| 254 | void WriteSection(const MCAssembler &Asm, const MCSectionData &SD, |
| 255 | uint64_t FileOffset, uint64_t RelocationsStart, |
| 256 | unsigned NumRelocations) { |
| 257 | // The offset is unused for virtual sections. |
| 258 | if (Asm.getBackend().isVirtualSection(SD.getSection())) { |
| 259 | assert(SD.getFileSize() == 0 && "Invalid file size!"); |
| 260 | FileOffset = 0; |
| 261 | } |
| 262 | |
| 263 | // struct section (68 bytes) or |
| 264 | // struct section_64 (80 bytes) |
| 265 | |
| 266 | uint64_t Start = OS.tell(); |
| 267 | (void) Start; |
| 268 | |
| 269 | // FIXME: cast<> support! |
| 270 | const MCSectionMachO &Section = |
| 271 | static_cast<const MCSectionMachO&>(SD.getSection()); |
| 272 | WriteBytes(Section.getSectionName(), 16); |
| 273 | WriteBytes(Section.getSegmentName(), 16); |
| 274 | if (Is64Bit) { |
| 275 | Write64(SD.getAddress()); // address |
| 276 | Write64(SD.getSize()); // size |
| 277 | } else { |
| 278 | Write32(SD.getAddress()); // address |
| 279 | Write32(SD.getSize()); // size |
| 280 | } |
| 281 | Write32(FileOffset); |
| 282 | |
| 283 | unsigned Flags = Section.getTypeAndAttributes(); |
| 284 | if (SD.hasInstructions()) |
| 285 | Flags |= MCSectionMachO::S_ATTR_SOME_INSTRUCTIONS; |
| 286 | |
| 287 | assert(isPowerOf2_32(SD.getAlignment()) && "Invalid alignment!"); |
| 288 | Write32(Log2_32(SD.getAlignment())); |
| 289 | Write32(NumRelocations ? RelocationsStart : 0); |
| 290 | Write32(NumRelocations); |
| 291 | Write32(Flags); |
| 292 | Write32(0); // reserved1 |
| 293 | Write32(Section.getStubSize()); // reserved2 |
| 294 | if (Is64Bit) |
| 295 | Write32(0); // reserved3 |
| 296 | |
| 297 | assert(OS.tell() - Start == Is64Bit ? Section64Size : Section32Size); |
| 298 | } |
| 299 | |
| 300 | void WriteSymtabLoadCommand(uint32_t SymbolOffset, uint32_t NumSymbols, |
| 301 | uint32_t StringTableOffset, |
| 302 | uint32_t StringTableSize) { |
| 303 | // struct symtab_command (24 bytes) |
| 304 | |
| 305 | uint64_t Start = OS.tell(); |
| 306 | (void) Start; |
| 307 | |
| 308 | Write32(LCT_Symtab); |
| 309 | Write32(SymtabLoadCommandSize); |
| 310 | Write32(SymbolOffset); |
| 311 | Write32(NumSymbols); |
| 312 | Write32(StringTableOffset); |
| 313 | Write32(StringTableSize); |
| 314 | |
| 315 | assert(OS.tell() - Start == SymtabLoadCommandSize); |
| 316 | } |
| 317 | |
| 318 | void WriteDysymtabLoadCommand(uint32_t FirstLocalSymbol, |
| 319 | uint32_t NumLocalSymbols, |
| 320 | uint32_t FirstExternalSymbol, |
| 321 | uint32_t NumExternalSymbols, |
| 322 | uint32_t FirstUndefinedSymbol, |
| 323 | uint32_t NumUndefinedSymbols, |
| 324 | uint32_t IndirectSymbolOffset, |
| 325 | uint32_t NumIndirectSymbols) { |
| 326 | // struct dysymtab_command (80 bytes) |
| 327 | |
| 328 | uint64_t Start = OS.tell(); |
| 329 | (void) Start; |
| 330 | |
| 331 | Write32(LCT_Dysymtab); |
| 332 | Write32(DysymtabLoadCommandSize); |
| 333 | Write32(FirstLocalSymbol); |
| 334 | Write32(NumLocalSymbols); |
| 335 | Write32(FirstExternalSymbol); |
| 336 | Write32(NumExternalSymbols); |
| 337 | Write32(FirstUndefinedSymbol); |
| 338 | Write32(NumUndefinedSymbols); |
| 339 | Write32(0); // tocoff |
| 340 | Write32(0); // ntoc |
| 341 | Write32(0); // modtaboff |
| 342 | Write32(0); // nmodtab |
| 343 | Write32(0); // extrefsymoff |
| 344 | Write32(0); // nextrefsyms |
| 345 | Write32(IndirectSymbolOffset); |
| 346 | Write32(NumIndirectSymbols); |
| 347 | Write32(0); // extreloff |
| 348 | Write32(0); // nextrel |
| 349 | Write32(0); // locreloff |
| 350 | Write32(0); // nlocrel |
| 351 | |
| 352 | assert(OS.tell() - Start == DysymtabLoadCommandSize); |
| 353 | } |
| 354 | |
| 355 | void WriteNlist(MachSymbolData &MSD) { |
| 356 | MCSymbolData &Data = *MSD.SymbolData; |
| 357 | const MCSymbol &Symbol = Data.getSymbol(); |
| 358 | uint8_t Type = 0; |
| 359 | uint16_t Flags = Data.getFlags(); |
| 360 | uint32_t Address = 0; |
| 361 | |
| 362 | // Set the N_TYPE bits. See <mach-o/nlist.h>. |
| 363 | // |
| 364 | // FIXME: Are the prebound or indirect fields possible here? |
| 365 | if (Symbol.isUndefined()) |
| 366 | Type = STT_Undefined; |
| 367 | else if (Symbol.isAbsolute()) |
| 368 | Type = STT_Absolute; |
| 369 | else |
| 370 | Type = STT_Section; |
| 371 | |
| 372 | // FIXME: Set STAB bits. |
| 373 | |
| 374 | if (Data.isPrivateExtern()) |
| 375 | Type |= STF_PrivateExtern; |
| 376 | |
| 377 | // Set external bit. |
| 378 | if (Data.isExternal() || Symbol.isUndefined()) |
| 379 | Type |= STF_External; |
| 380 | |
| 381 | // Compute the symbol address. |
| 382 | if (Symbol.isDefined()) { |
| 383 | if (Symbol.isAbsolute()) { |
| 384 | llvm_unreachable("FIXME: Not yet implemented!"); |
| 385 | } else { |
| 386 | Address = Data.getAddress(); |
| 387 | } |
| 388 | } else if (Data.isCommon()) { |
| 389 | // Common symbols are encoded with the size in the address |
| 390 | // field, and their alignment in the flags. |
| 391 | Address = Data.getCommonSize(); |
| 392 | |
| 393 | // Common alignment is packed into the 'desc' bits. |
| 394 | if (unsigned Align = Data.getCommonAlignment()) { |
| 395 | unsigned Log2Size = Log2_32(Align); |
| 396 | assert((1U << Log2Size) == Align && "Invalid 'common' alignment!"); |
| 397 | if (Log2Size > 15) |
| 398 | llvm_report_error("invalid 'common' alignment '" + |
| 399 | Twine(Align) + "'"); |
| 400 | // FIXME: Keep this mask with the SymbolFlags enumeration. |
| 401 | Flags = (Flags & 0xF0FF) | (Log2Size << 8); |
| 402 | } |
| 403 | } |
| 404 | |
| 405 | // struct nlist (12 bytes) |
| 406 | |
| 407 | Write32(MSD.StringIndex); |
| 408 | Write8(Type); |
| 409 | Write8(MSD.SectionIndex); |
| 410 | |
| 411 | // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc' |
| 412 | // value. |
| 413 | Write16(Flags); |
| 414 | if (Is64Bit) |
| 415 | Write64(Address); |
| 416 | else |
| 417 | Write32(Address); |
| 418 | } |
| 419 | |
| 420 | void RecordScatteredRelocation(const MCAssembler &Asm, |
| 421 | const MCFragment &Fragment, |
| 422 | const MCAsmFixup &Fixup, MCValue Target, |
| 423 | uint64_t &FixedValue) { |
| 424 | uint32_t Address = Fragment.getOffset() + Fixup.Offset; |
| 425 | unsigned IsPCRel = isFixupKindPCRel(Fixup.Kind); |
| 426 | unsigned Log2Size = getFixupKindLog2Size(Fixup.Kind); |
| 427 | unsigned Type = RIT_Vanilla; |
| 428 | |
| 429 | // See <reloc.h>. |
| 430 | const MCSymbol *A = &Target.getSymA()->getSymbol(); |
| 431 | MCSymbolData *A_SD = &Asm.getSymbolData(*A); |
| 432 | |
| 433 | if (!A_SD->getFragment()) |
| 434 | llvm_report_error("symbol '" + A->getName() + |
| 435 | "' can not be undefined in a subtraction expression"); |
| 436 | |
| 437 | uint32_t Value = A_SD->getAddress(); |
| 438 | uint32_t Value2 = 0; |
| 439 | |
| 440 | if (const MCSymbolRefExpr *B = Target.getSymB()) { |
| 441 | MCSymbolData *B_SD = &Asm.getSymbolData(B->getSymbol()); |
| 442 | |
| 443 | if (!B_SD->getFragment()) |
| 444 | llvm_report_error("symbol '" + B->getSymbol().getName() + |
| 445 | "' can not be undefined in a subtraction expression"); |
| 446 | |
| 447 | // Select the appropriate difference relocation type. |
| 448 | // |
| 449 | // Note that there is no longer any semantic difference between these two |
| 450 | // relocation types from the linkers point of view, this is done solely |
| 451 | // for pedantic compatibility with 'as'. |
| 452 | Type = A_SD->isExternal() ? RIT_Difference : RIT_LocalDifference; |
| 453 | Value2 = B_SD->getAddress(); |
| 454 | } |
| 455 | |
| 456 | // Relocations are written out in reverse order, so the PAIR comes first. |
| 457 | if (Type == RIT_Difference || Type == RIT_LocalDifference) { |
| 458 | MachRelocationEntry MRE; |
| 459 | MRE.Word0 = ((0 << 0) | |
| 460 | (RIT_Pair << 24) | |
| 461 | (Log2Size << 28) | |
| 462 | (IsPCRel << 30) | |
| 463 | RF_Scattered); |
| 464 | MRE.Word1 = Value2; |
| 465 | Relocations[Fragment.getParent()].push_back(MRE); |
| 466 | } |
| 467 | |
| 468 | MachRelocationEntry MRE; |
| 469 | MRE.Word0 = ((Address << 0) | |
| 470 | (Type << 24) | |
| 471 | (Log2Size << 28) | |
| 472 | (IsPCRel << 30) | |
| 473 | RF_Scattered); |
| 474 | MRE.Word1 = Value; |
| 475 | Relocations[Fragment.getParent()].push_back(MRE); |
| 476 | } |
| 477 | |
| 478 | virtual void RecordRelocation(const MCAssembler &Asm, |
| 479 | const MCDataFragment &Fragment, |
| 480 | const MCAsmFixup &Fixup, MCValue Target, |
| 481 | uint64_t &FixedValue) { |
| 482 | unsigned IsPCRel = isFixupKindPCRel(Fixup.Kind); |
| 483 | unsigned Log2Size = getFixupKindLog2Size(Fixup.Kind); |
| 484 | |
| 485 | // If this is a difference or a defined symbol plus an offset, then we need |
| 486 | // a scattered relocation entry. |
| 487 | uint32_t Offset = Target.getConstant(); |
| 488 | if (IsPCRel) |
| 489 | Offset += 1 << Log2Size; |
| 490 | if (Target.getSymB() || |
| 491 | (Target.getSymA() && !Target.getSymA()->getSymbol().isUndefined() && |
| 492 | Offset)) { |
| 493 | RecordScatteredRelocation(Asm, Fragment, Fixup, Target, FixedValue); |
| 494 | return; |
| 495 | } |
| 496 | |
| 497 | // See <reloc.h>. |
| 498 | uint32_t Address = Fragment.getOffset() + Fixup.Offset; |
| 499 | uint32_t Value = 0; |
| 500 | unsigned Index = 0; |
| 501 | unsigned IsExtern = 0; |
| 502 | unsigned Type = 0; |
| 503 | |
| 504 | if (Target.isAbsolute()) { // constant |
| 505 | // SymbolNum of 0 indicates the absolute section. |
| 506 | // |
| 507 | // FIXME: Currently, these are never generated (see code below). I cannot |
| 508 | // find a case where they are actually emitted. |
| 509 | Type = RIT_Vanilla; |
| 510 | Value = 0; |
| 511 | } else { |
| 512 | const MCSymbol *Symbol = &Target.getSymA()->getSymbol(); |
| 513 | MCSymbolData *SD = &Asm.getSymbolData(*Symbol); |
| 514 | |
| 515 | if (Symbol->isUndefined()) { |
| 516 | IsExtern = 1; |
| 517 | Index = SD->getIndex(); |
| 518 | Value = 0; |
| 519 | } else { |
| 520 | // The index is the section ordinal. |
| 521 | // |
| 522 | // FIXME: O(N) |
| 523 | Index = 1; |
| 524 | MCAssembler::const_iterator it = Asm.begin(), ie = Asm.end(); |
| 525 | for (; it != ie; ++it, ++Index) |
| 526 | if (&*it == SD->getFragment()->getParent()) |
| 527 | break; |
| 528 | assert(it != ie && "Unable to find section index!"); |
| 529 | Value = SD->getAddress(); |
| 530 | } |
| 531 | |
| 532 | Type = RIT_Vanilla; |
| 533 | } |
| 534 | |
| 535 | // struct relocation_info (8 bytes) |
| 536 | MachRelocationEntry MRE; |
| 537 | MRE.Word0 = Address; |
| 538 | MRE.Word1 = ((Index << 0) | |
| 539 | (IsPCRel << 24) | |
| 540 | (Log2Size << 25) | |
| 541 | (IsExtern << 27) | |
| 542 | (Type << 28)); |
| 543 | Relocations[Fragment.getParent()].push_back(MRE); |
| 544 | } |
| 545 | |
| 546 | void BindIndirectSymbols(MCAssembler &Asm) { |
| 547 | // This is the point where 'as' creates actual symbols for indirect symbols |
| 548 | // (in the following two passes). It would be easier for us to do this |
| 549 | // sooner when we see the attribute, but that makes getting the order in the |
| 550 | // symbol table much more complicated than it is worth. |
| 551 | // |
| 552 | // FIXME: Revisit this when the dust settles. |
| 553 | |
| 554 | // Bind non lazy symbol pointers first. |
| 555 | for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(), |
| 556 | ie = Asm.indirect_symbol_end(); it != ie; ++it) { |
| 557 | // FIXME: cast<> support! |
| 558 | const MCSectionMachO &Section = |
| 559 | static_cast<const MCSectionMachO&>(it->SectionData->getSection()); |
| 560 | |
| 561 | if (Section.getType() != MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS) |
| 562 | continue; |
| 563 | |
| 564 | Asm.getOrCreateSymbolData(*it->Symbol); |
| 565 | } |
| 566 | |
| 567 | // Then lazy symbol pointers and symbol stubs. |
| 568 | for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(), |
| 569 | ie = Asm.indirect_symbol_end(); it != ie; ++it) { |
| 570 | // FIXME: cast<> support! |
| 571 | const MCSectionMachO &Section = |
| 572 | static_cast<const MCSectionMachO&>(it->SectionData->getSection()); |
| 573 | |
| 574 | if (Section.getType() != MCSectionMachO::S_LAZY_SYMBOL_POINTERS && |
| 575 | Section.getType() != MCSectionMachO::S_SYMBOL_STUBS) |
| 576 | continue; |
| 577 | |
| 578 | // Set the symbol type to undefined lazy, but only on construction. |
| 579 | // |
| 580 | // FIXME: Do not hardcode. |
| 581 | bool Created; |
| 582 | MCSymbolData &Entry = Asm.getOrCreateSymbolData(*it->Symbol, &Created); |
| 583 | if (Created) |
| 584 | Entry.setFlags(Entry.getFlags() | 0x0001); |
| 585 | } |
| 586 | } |
| 587 | |
| 588 | /// ComputeSymbolTable - Compute the symbol table data |
| 589 | /// |
| 590 | /// \param StringTable [out] - The string table data. |
| 591 | /// \param StringIndexMap [out] - Map from symbol names to offsets in the |
| 592 | /// string table. |
| 593 | void ComputeSymbolTable(MCAssembler &Asm, SmallString<256> &StringTable, |
| 594 | std::vector<MachSymbolData> &LocalSymbolData, |
| 595 | std::vector<MachSymbolData> &ExternalSymbolData, |
| 596 | std::vector<MachSymbolData> &UndefinedSymbolData) { |
| 597 | // Build section lookup table. |
| 598 | DenseMap<const MCSection*, uint8_t> SectionIndexMap; |
| 599 | unsigned Index = 1; |
| 600 | for (MCAssembler::iterator it = Asm.begin(), |
| 601 | ie = Asm.end(); it != ie; ++it, ++Index) |
| 602 | SectionIndexMap[&it->getSection()] = Index; |
| 603 | assert(Index <= 256 && "Too many sections!"); |
| 604 | |
| 605 | // Index 0 is always the empty string. |
| 606 | StringMap<uint64_t> StringIndexMap; |
| 607 | StringTable += '\x00'; |
| 608 | |
| 609 | // Build the symbol arrays and the string table, but only for non-local |
| 610 | // symbols. |
| 611 | // |
| 612 | // The particular order that we collect the symbols and create the string |
| 613 | // table, then sort the symbols is chosen to match 'as'. Even though it |
| 614 | // doesn't matter for correctness, this is important for letting us diff .o |
| 615 | // files. |
| 616 | for (MCAssembler::symbol_iterator it = Asm.symbol_begin(), |
| 617 | ie = Asm.symbol_end(); it != ie; ++it) { |
| 618 | const MCSymbol &Symbol = it->getSymbol(); |
| 619 | |
| 620 | // Ignore non-linker visible symbols. |
| 621 | if (!Asm.isSymbolLinkerVisible(it)) |
| 622 | continue; |
| 623 | |
| 624 | if (!it->isExternal() && !Symbol.isUndefined()) |
| 625 | continue; |
| 626 | |
| 627 | uint64_t &Entry = StringIndexMap[Symbol.getName()]; |
| 628 | if (!Entry) { |
| 629 | Entry = StringTable.size(); |
| 630 | StringTable += Symbol.getName(); |
| 631 | StringTable += '\x00'; |
| 632 | } |
| 633 | |
| 634 | MachSymbolData MSD; |
| 635 | MSD.SymbolData = it; |
| 636 | MSD.StringIndex = Entry; |
| 637 | |
| 638 | if (Symbol.isUndefined()) { |
| 639 | MSD.SectionIndex = 0; |
| 640 | UndefinedSymbolData.push_back(MSD); |
| 641 | } else if (Symbol.isAbsolute()) { |
| 642 | MSD.SectionIndex = 0; |
| 643 | ExternalSymbolData.push_back(MSD); |
| 644 | } else { |
| 645 | MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection()); |
| 646 | assert(MSD.SectionIndex && "Invalid section index!"); |
| 647 | ExternalSymbolData.push_back(MSD); |
| 648 | } |
| 649 | } |
| 650 | |
| 651 | // Now add the data for local symbols. |
| 652 | for (MCAssembler::symbol_iterator it = Asm.symbol_begin(), |
| 653 | ie = Asm.symbol_end(); it != ie; ++it) { |
| 654 | const MCSymbol &Symbol = it->getSymbol(); |
| 655 | |
| 656 | // Ignore non-linker visible symbols. |
| 657 | if (!Asm.isSymbolLinkerVisible(it)) |
| 658 | continue; |
| 659 | |
| 660 | if (it->isExternal() || Symbol.isUndefined()) |
| 661 | continue; |
| 662 | |
| 663 | uint64_t &Entry = StringIndexMap[Symbol.getName()]; |
| 664 | if (!Entry) { |
| 665 | Entry = StringTable.size(); |
| 666 | StringTable += Symbol.getName(); |
| 667 | StringTable += '\x00'; |
| 668 | } |
| 669 | |
| 670 | MachSymbolData MSD; |
| 671 | MSD.SymbolData = it; |
| 672 | MSD.StringIndex = Entry; |
| 673 | |
| 674 | if (Symbol.isAbsolute()) { |
| 675 | MSD.SectionIndex = 0; |
| 676 | LocalSymbolData.push_back(MSD); |
| 677 | } else { |
| 678 | MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection()); |
| 679 | assert(MSD.SectionIndex && "Invalid section index!"); |
| 680 | LocalSymbolData.push_back(MSD); |
| 681 | } |
| 682 | } |
| 683 | |
| 684 | // External and undefined symbols are required to be in lexicographic order. |
| 685 | std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end()); |
| 686 | std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end()); |
| 687 | |
| 688 | // Set the symbol indices. |
| 689 | Index = 0; |
| 690 | for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i) |
| 691 | LocalSymbolData[i].SymbolData->setIndex(Index++); |
| 692 | for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i) |
| 693 | ExternalSymbolData[i].SymbolData->setIndex(Index++); |
| 694 | for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i) |
| 695 | UndefinedSymbolData[i].SymbolData->setIndex(Index++); |
| 696 | |
| 697 | // The string table is padded to a multiple of 4. |
| 698 | while (StringTable.size() % 4) |
| 699 | StringTable += '\x00'; |
| 700 | } |
| 701 | |
| 702 | virtual void ExecutePostLayoutBinding(MCAssembler &Asm) { |
| 703 | // Create symbol data for any indirect symbols. |
| 704 | BindIndirectSymbols(Asm); |
| 705 | |
| 706 | // Compute symbol table information and bind symbol indices. |
| 707 | ComputeSymbolTable(Asm, StringTable, LocalSymbolData, ExternalSymbolData, |
| 708 | UndefinedSymbolData); |
| 709 | } |
| 710 | |
| 711 | virtual void WriteObject(const MCAssembler &Asm) { |
| 712 | unsigned NumSections = Asm.size(); |
| 713 | |
| 714 | // The section data starts after the header, the segment load command (and |
| 715 | // section headers) and the symbol table. |
| 716 | unsigned NumLoadCommands = 1; |
| 717 | uint64_t LoadCommandsSize = Is64Bit ? |
| 718 | SegmentLoadCommand64Size + NumSections * Section64Size : |
| 719 | SegmentLoadCommand32Size + NumSections * Section32Size; |
| 720 | |
| 721 | // Add the symbol table load command sizes, if used. |
| 722 | unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() + |
| 723 | UndefinedSymbolData.size(); |
| 724 | if (NumSymbols) { |
| 725 | NumLoadCommands += 2; |
| 726 | LoadCommandsSize += SymtabLoadCommandSize + DysymtabLoadCommandSize; |
| 727 | } |
| 728 | |
| 729 | // Compute the total size of the section data, as well as its file size and |
| 730 | // vm size. |
| 731 | uint64_t SectionDataStart = (Is64Bit ? Header64Size : Header32Size) |
| 732 | + LoadCommandsSize; |
| 733 | uint64_t SectionDataSize = 0; |
| 734 | uint64_t SectionDataFileSize = 0; |
| 735 | uint64_t VMSize = 0; |
| 736 | for (MCAssembler::const_iterator it = Asm.begin(), |
| 737 | ie = Asm.end(); it != ie; ++it) { |
| 738 | const MCSectionData &SD = *it; |
| 739 | |
| 740 | VMSize = std::max(VMSize, SD.getAddress() + SD.getSize()); |
| 741 | |
| 742 | if (Asm.getBackend().isVirtualSection(SD.getSection())) |
| 743 | continue; |
| 744 | |
| 745 | SectionDataSize = std::max(SectionDataSize, |
| 746 | SD.getAddress() + SD.getSize()); |
| 747 | SectionDataFileSize = std::max(SectionDataFileSize, |
| 748 | SD.getAddress() + SD.getFileSize()); |
| 749 | } |
| 750 | |
| 751 | // The section data is padded to 4 bytes. |
| 752 | // |
| 753 | // FIXME: Is this machine dependent? |
| 754 | unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4); |
| 755 | SectionDataFileSize += SectionDataPadding; |
| 756 | |
| 757 | // Write the prolog, starting with the header and load command... |
| 758 | WriteHeader(NumLoadCommands, LoadCommandsSize, |
| 759 | Asm.getSubsectionsViaSymbols()); |
| 760 | WriteSegmentLoadCommand(NumSections, VMSize, |
| 761 | SectionDataStart, SectionDataSize); |
| 762 | |
| 763 | // ... and then the section headers. |
| 764 | uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize; |
| 765 | for (MCAssembler::const_iterator it = Asm.begin(), |
| 766 | ie = Asm.end(); it != ie; ++it) { |
| 767 | std::vector<MachRelocationEntry> &Relocs = Relocations[it]; |
| 768 | unsigned NumRelocs = Relocs.size(); |
| 769 | uint64_t SectionStart = SectionDataStart + it->getAddress(); |
| 770 | WriteSection(Asm, *it, SectionStart, RelocTableEnd, NumRelocs); |
| 771 | RelocTableEnd += NumRelocs * RelocationInfoSize; |
| 772 | } |
| 773 | |
| 774 | // Write the symbol table load command, if used. |
| 775 | if (NumSymbols) { |
| 776 | unsigned FirstLocalSymbol = 0; |
| 777 | unsigned NumLocalSymbols = LocalSymbolData.size(); |
| 778 | unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols; |
| 779 | unsigned NumExternalSymbols = ExternalSymbolData.size(); |
| 780 | unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols; |
| 781 | unsigned NumUndefinedSymbols = UndefinedSymbolData.size(); |
| 782 | unsigned NumIndirectSymbols = Asm.indirect_symbol_size(); |
| 783 | unsigned NumSymTabSymbols = |
| 784 | NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols; |
| 785 | uint64_t IndirectSymbolSize = NumIndirectSymbols * 4; |
| 786 | uint64_t IndirectSymbolOffset = 0; |
| 787 | |
| 788 | // If used, the indirect symbols are written after the section data. |
| 789 | if (NumIndirectSymbols) |
| 790 | IndirectSymbolOffset = RelocTableEnd; |
| 791 | |
| 792 | // The symbol table is written after the indirect symbol data. |
| 793 | uint64_t SymbolTableOffset = RelocTableEnd + IndirectSymbolSize; |
| 794 | |
| 795 | // The string table is written after symbol table. |
| 796 | uint64_t StringTableOffset = |
| 797 | SymbolTableOffset + NumSymTabSymbols * (Is64Bit ? Nlist64Size : |
| 798 | Nlist32Size); |
| 799 | WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols, |
| 800 | StringTableOffset, StringTable.size()); |
| 801 | |
| 802 | WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols, |
| 803 | FirstExternalSymbol, NumExternalSymbols, |
| 804 | FirstUndefinedSymbol, NumUndefinedSymbols, |
| 805 | IndirectSymbolOffset, NumIndirectSymbols); |
| 806 | } |
| 807 | |
| 808 | // Write the actual section data. |
| 809 | for (MCAssembler::const_iterator it = Asm.begin(), |
| 810 | ie = Asm.end(); it != ie; ++it) |
| 811 | Asm.WriteSectionData(it, Writer); |
| 812 | |
| 813 | // Write the extra padding. |
| 814 | WriteZeros(SectionDataPadding); |
| 815 | |
| 816 | // Write the relocation entries. |
| 817 | for (MCAssembler::const_iterator it = Asm.begin(), |
| 818 | ie = Asm.end(); it != ie; ++it) { |
| 819 | // Write the section relocation entries, in reverse order to match 'as' |
| 820 | // (approximately, the exact algorithm is more complicated than this). |
| 821 | std::vector<MachRelocationEntry> &Relocs = Relocations[it]; |
| 822 | for (unsigned i = 0, e = Relocs.size(); i != e; ++i) { |
| 823 | Write32(Relocs[e - i - 1].Word0); |
| 824 | Write32(Relocs[e - i - 1].Word1); |
| 825 | } |
| 826 | } |
| 827 | |
| 828 | // Write the symbol table data, if used. |
| 829 | if (NumSymbols) { |
| 830 | // Write the indirect symbol entries. |
| 831 | for (MCAssembler::const_indirect_symbol_iterator |
| 832 | it = Asm.indirect_symbol_begin(), |
| 833 | ie = Asm.indirect_symbol_end(); it != ie; ++it) { |
| 834 | // Indirect symbols in the non lazy symbol pointer section have some |
| 835 | // special handling. |
| 836 | const MCSectionMachO &Section = |
| 837 | static_cast<const MCSectionMachO&>(it->SectionData->getSection()); |
| 838 | if (Section.getType() == MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS) { |
| 839 | // If this symbol is defined and internal, mark it as such. |
| 840 | if (it->Symbol->isDefined() && |
| 841 | !Asm.getSymbolData(*it->Symbol).isExternal()) { |
| 842 | uint32_t Flags = ISF_Local; |
| 843 | if (it->Symbol->isAbsolute()) |
| 844 | Flags |= ISF_Absolute; |
| 845 | Write32(Flags); |
| 846 | continue; |
| 847 | } |
| 848 | } |
| 849 | |
| 850 | Write32(Asm.getSymbolData(*it->Symbol).getIndex()); |
| 851 | } |
| 852 | |
| 853 | // FIXME: Check that offsets match computed ones. |
| 854 | |
| 855 | // Write the symbol table entries. |
| 856 | for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i) |
| 857 | WriteNlist(LocalSymbolData[i]); |
| 858 | for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i) |
| 859 | WriteNlist(ExternalSymbolData[i]); |
| 860 | for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i) |
| 861 | WriteNlist(UndefinedSymbolData[i]); |
| 862 | |
| 863 | // Write the string table. |
| 864 | OS << StringTable.str(); |
| 865 | } |
| 866 | } |
| 867 | }; |
| 868 | |
| 869 | } |
| 870 | |
| 871 | MachObjectWriter::MachObjectWriter(raw_ostream &OS, |
| 872 | bool Is64Bit, |
| 873 | bool IsLittleEndian) |
| 874 | : MCObjectWriter(OS, IsLittleEndian) |
| 875 | { |
| 876 | Impl = new MachObjectWriterImpl(this, Is64Bit); |
| 877 | } |
| 878 | |
| 879 | MachObjectWriter::~MachObjectWriter() { |
| 880 | delete (MachObjectWriterImpl*) Impl; |
| 881 | } |
| 882 | |
| 883 | void MachObjectWriter::ExecutePostLayoutBinding(MCAssembler &Asm) { |
| 884 | ((MachObjectWriterImpl*) Impl)->ExecutePostLayoutBinding(Asm); |
| 885 | } |
| 886 | |
| 887 | void MachObjectWriter::RecordRelocation(const MCAssembler &Asm, |
| 888 | const MCDataFragment &Fragment, |
| 889 | const MCAsmFixup &Fixup, MCValue Target, |
| 890 | uint64_t &FixedValue) { |
| 891 | ((MachObjectWriterImpl*) Impl)->RecordRelocation(Asm, Fragment, Fixup, |
| 892 | Target, FixedValue); |
| 893 | } |
| 894 | |
| 895 | void MachObjectWriter::WriteObject(const MCAssembler &Asm) { |
| 896 | ((MachObjectWriterImpl*) Impl)->WriteObject(Asm); |
| 897 | } |