1 //===-- llvm-objdump.cpp - Object file dumping utility for llvm -----------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This program is a utility that works like binutils "objdump", that is, it 10 // dumps out a plethora of information about an object file depending on the 11 // flags. 12 // 13 // The flags and output of this program should be near identical to those of 14 // binutils objdump. 15 // 16 //===----------------------------------------------------------------------===// 17 18 #include "llvm-objdump.h" 19 #include "COFFDump.h" 20 #include "ELFDump.h" 21 #include "MachODump.h" 22 #include "ObjdumpOptID.h" 23 #include "OffloadDump.h" 24 #include "SourcePrinter.h" 25 #include "WasmDump.h" 26 #include "XCOFFDump.h" 27 #include "llvm/ADT/IndexedMap.h" 28 #include "llvm/ADT/Optional.h" 29 #include "llvm/ADT/STLExtras.h" 30 #include "llvm/ADT/SetOperations.h" 31 #include "llvm/ADT/SmallSet.h" 32 #include "llvm/ADT/StringExtras.h" 33 #include "llvm/ADT/StringSet.h" 34 #include "llvm/ADT/Triple.h" 35 #include "llvm/ADT/Twine.h" 36 #include "llvm/DebugInfo/DWARF/DWARFContext.h" 37 #include "llvm/DebugInfo/Symbolize/SymbolizableModule.h" 38 #include "llvm/DebugInfo/Symbolize/Symbolize.h" 39 #include "llvm/Demangle/Demangle.h" 40 #include "llvm/MC/MCAsmInfo.h" 41 #include "llvm/MC/MCContext.h" 42 #include "llvm/MC/MCDisassembler/MCDisassembler.h" 43 #include "llvm/MC/MCDisassembler/MCRelocationInfo.h" 44 #include "llvm/MC/MCInst.h" 45 #include "llvm/MC/MCInstPrinter.h" 46 #include "llvm/MC/MCInstrAnalysis.h" 47 #include "llvm/MC/MCInstrInfo.h" 48 #include "llvm/MC/MCObjectFileInfo.h" 49 #include "llvm/MC/MCRegisterInfo.h" 50 #include "llvm/MC/MCSubtargetInfo.h" 51 #include "llvm/MC/MCTargetOptions.h" 52 #include "llvm/MC/TargetRegistry.h" 53 #include "llvm/Object/Archive.h" 54 #include "llvm/Object/COFF.h" 55 #include "llvm/Object/COFFImportFile.h" 56 #include "llvm/Object/ELFObjectFile.h" 57 #include "llvm/Object/ELFTypes.h" 58 #include "llvm/Object/FaultMapParser.h" 59 #include "llvm/Object/MachO.h" 60 #include "llvm/Object/MachOUniversal.h" 61 #include "llvm/Object/ObjectFile.h" 62 #include "llvm/Object/OffloadBinary.h" 63 #include "llvm/Object/Wasm.h" 64 #include "llvm/Option/Arg.h" 65 #include "llvm/Option/ArgList.h" 66 #include "llvm/Option/Option.h" 67 #include "llvm/Support/Casting.h" 68 #include "llvm/Support/Debug.h" 69 #include "llvm/Support/Errc.h" 70 #include "llvm/Support/FileSystem.h" 71 #include "llvm/Support/Format.h" 72 #include "llvm/Support/FormatVariadic.h" 73 #include "llvm/Support/GraphWriter.h" 74 #include "llvm/Support/Host.h" 75 #include "llvm/Support/InitLLVM.h" 76 #include "llvm/Support/MemoryBuffer.h" 77 #include "llvm/Support/SourceMgr.h" 78 #include "llvm/Support/StringSaver.h" 79 #include "llvm/Support/TargetSelect.h" 80 #include "llvm/Support/WithColor.h" 81 #include "llvm/Support/raw_ostream.h" 82 #include <algorithm> 83 #include <cctype> 84 #include <cstring> 85 #include <system_error> 86 #include <unordered_map> 87 #include <utility> 88 89 using namespace llvm; 90 using namespace llvm::object; 91 using namespace llvm::objdump; 92 using namespace llvm::opt; 93 94 namespace { 95 96 class CommonOptTable : public opt::OptTable { 97 public: 98 CommonOptTable(ArrayRef<Info> OptionInfos, const char *Usage, 99 const char *Description) 100 : OptTable(OptionInfos), Usage(Usage), Description(Description) { 101 setGroupedShortOptions(true); 102 } 103 104 void printHelp(StringRef Argv0, bool ShowHidden = false) const { 105 Argv0 = sys::path::filename(Argv0); 106 opt::OptTable::printHelp(outs(), (Argv0 + Usage).str().c_str(), Description, 107 ShowHidden, ShowHidden); 108 // TODO Replace this with OptTable API once it adds extrahelp support. 109 outs() << "\nPass @FILE as argument to read options from FILE.\n"; 110 } 111 112 private: 113 const char *Usage; 114 const char *Description; 115 }; 116 117 // ObjdumpOptID is in ObjdumpOptID.h 118 119 #define PREFIX(NAME, VALUE) const char *const OBJDUMP_##NAME[] = VALUE; 120 #include "ObjdumpOpts.inc" 121 #undef PREFIX 122 123 static constexpr opt::OptTable::Info ObjdumpInfoTable[] = { 124 #define OBJDUMP_nullptr nullptr 125 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 126 HELPTEXT, METAVAR, VALUES) \ 127 {OBJDUMP_##PREFIX, NAME, HELPTEXT, \ 128 METAVAR, OBJDUMP_##ID, opt::Option::KIND##Class, \ 129 PARAM, FLAGS, OBJDUMP_##GROUP, \ 130 OBJDUMP_##ALIAS, ALIASARGS, VALUES}, 131 #include "ObjdumpOpts.inc" 132 #undef OPTION 133 #undef OBJDUMP_nullptr 134 }; 135 136 class ObjdumpOptTable : public CommonOptTable { 137 public: 138 ObjdumpOptTable() 139 : CommonOptTable(ObjdumpInfoTable, " [options] <input object files>", 140 "llvm object file dumper") {} 141 }; 142 143 enum OtoolOptID { 144 OTOOL_INVALID = 0, // This is not an option ID. 145 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 146 HELPTEXT, METAVAR, VALUES) \ 147 OTOOL_##ID, 148 #include "OtoolOpts.inc" 149 #undef OPTION 150 }; 151 152 #define PREFIX(NAME, VALUE) const char *const OTOOL_##NAME[] = VALUE; 153 #include "OtoolOpts.inc" 154 #undef PREFIX 155 156 static constexpr opt::OptTable::Info OtoolInfoTable[] = { 157 #define OTOOL_nullptr nullptr 158 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 159 HELPTEXT, METAVAR, VALUES) \ 160 {OTOOL_##PREFIX, NAME, HELPTEXT, \ 161 METAVAR, OTOOL_##ID, opt::Option::KIND##Class, \ 162 PARAM, FLAGS, OTOOL_##GROUP, \ 163 OTOOL_##ALIAS, ALIASARGS, VALUES}, 164 #include "OtoolOpts.inc" 165 #undef OPTION 166 #undef OTOOL_nullptr 167 }; 168 169 class OtoolOptTable : public CommonOptTable { 170 public: 171 OtoolOptTable() 172 : CommonOptTable(OtoolInfoTable, " [option...] [file...]", 173 "Mach-O object file displaying tool") {} 174 }; 175 176 } // namespace 177 178 #define DEBUG_TYPE "objdump" 179 180 static uint64_t AdjustVMA; 181 static bool AllHeaders; 182 static std::string ArchName; 183 bool objdump::ArchiveHeaders; 184 bool objdump::Demangle; 185 bool objdump::Disassemble; 186 bool objdump::DisassembleAll; 187 bool objdump::SymbolDescription; 188 static std::vector<std::string> DisassembleSymbols; 189 static bool DisassembleZeroes; 190 static std::vector<std::string> DisassemblerOptions; 191 DIDumpType objdump::DwarfDumpType; 192 static bool DynamicRelocations; 193 static bool FaultMapSection; 194 static bool FileHeaders; 195 bool objdump::SectionContents; 196 static std::vector<std::string> InputFilenames; 197 bool objdump::PrintLines; 198 static bool MachOOpt; 199 std::string objdump::MCPU; 200 std::vector<std::string> objdump::MAttrs; 201 bool objdump::ShowRawInsn; 202 bool objdump::LeadingAddr; 203 static bool Offloading; 204 static bool RawClangAST; 205 bool objdump::Relocations; 206 bool objdump::PrintImmHex; 207 bool objdump::PrivateHeaders; 208 std::vector<std::string> objdump::FilterSections; 209 bool objdump::SectionHeaders; 210 static bool ShowLMA; 211 bool objdump::PrintSource; 212 213 static uint64_t StartAddress; 214 static bool HasStartAddressFlag; 215 static uint64_t StopAddress = UINT64_MAX; 216 static bool HasStopAddressFlag; 217 218 bool objdump::SymbolTable; 219 static bool SymbolizeOperands; 220 static bool DynamicSymbolTable; 221 std::string objdump::TripleName; 222 bool objdump::UnwindInfo; 223 static bool Wide; 224 std::string objdump::Prefix; 225 uint32_t objdump::PrefixStrip; 226 227 DebugVarsFormat objdump::DbgVariables = DVDisabled; 228 229 int objdump::DbgIndent = 52; 230 231 static StringSet<> DisasmSymbolSet; 232 StringSet<> objdump::FoundSectionSet; 233 static StringRef ToolName; 234 235 namespace { 236 struct FilterResult { 237 // True if the section should not be skipped. 238 bool Keep; 239 240 // True if the index counter should be incremented, even if the section should 241 // be skipped. For example, sections may be skipped if they are not included 242 // in the --section flag, but we still want those to count toward the section 243 // count. 244 bool IncrementIndex; 245 }; 246 } // namespace 247 248 static FilterResult checkSectionFilter(object::SectionRef S) { 249 if (FilterSections.empty()) 250 return {/*Keep=*/true, /*IncrementIndex=*/true}; 251 252 Expected<StringRef> SecNameOrErr = S.getName(); 253 if (!SecNameOrErr) { 254 consumeError(SecNameOrErr.takeError()); 255 return {/*Keep=*/false, /*IncrementIndex=*/false}; 256 } 257 StringRef SecName = *SecNameOrErr; 258 259 // StringSet does not allow empty key so avoid adding sections with 260 // no name (such as the section with index 0) here. 261 if (!SecName.empty()) 262 FoundSectionSet.insert(SecName); 263 264 // Only show the section if it's in the FilterSections list, but always 265 // increment so the indexing is stable. 266 return {/*Keep=*/is_contained(FilterSections, SecName), 267 /*IncrementIndex=*/true}; 268 } 269 270 SectionFilter objdump::ToolSectionFilter(object::ObjectFile const &O, 271 uint64_t *Idx) { 272 // Start at UINT64_MAX so that the first index returned after an increment is 273 // zero (after the unsigned wrap). 274 if (Idx) 275 *Idx = UINT64_MAX; 276 return SectionFilter( 277 [Idx](object::SectionRef S) { 278 FilterResult Result = checkSectionFilter(S); 279 if (Idx != nullptr && Result.IncrementIndex) 280 *Idx += 1; 281 return Result.Keep; 282 }, 283 O); 284 } 285 286 std::string objdump::getFileNameForError(const object::Archive::Child &C, 287 unsigned Index) { 288 Expected<StringRef> NameOrErr = C.getName(); 289 if (NameOrErr) 290 return std::string(NameOrErr.get()); 291 // If we have an error getting the name then we print the index of the archive 292 // member. Since we are already in an error state, we just ignore this error. 293 consumeError(NameOrErr.takeError()); 294 return "<file index: " + std::to_string(Index) + ">"; 295 } 296 297 void objdump::reportWarning(const Twine &Message, StringRef File) { 298 // Output order between errs() and outs() matters especially for archive 299 // files where the output is per member object. 300 outs().flush(); 301 WithColor::warning(errs(), ToolName) 302 << "'" << File << "': " << Message << "\n"; 303 } 304 305 [[noreturn]] void objdump::reportError(StringRef File, const Twine &Message) { 306 outs().flush(); 307 WithColor::error(errs(), ToolName) << "'" << File << "': " << Message << "\n"; 308 exit(1); 309 } 310 311 [[noreturn]] void objdump::reportError(Error E, StringRef FileName, 312 StringRef ArchiveName, 313 StringRef ArchitectureName) { 314 assert(E); 315 outs().flush(); 316 WithColor::error(errs(), ToolName); 317 if (ArchiveName != "") 318 errs() << ArchiveName << "(" << FileName << ")"; 319 else 320 errs() << "'" << FileName << "'"; 321 if (!ArchitectureName.empty()) 322 errs() << " (for architecture " << ArchitectureName << ")"; 323 errs() << ": "; 324 logAllUnhandledErrors(std::move(E), errs()); 325 exit(1); 326 } 327 328 static void reportCmdLineWarning(const Twine &Message) { 329 WithColor::warning(errs(), ToolName) << Message << "\n"; 330 } 331 332 [[noreturn]] static void reportCmdLineError(const Twine &Message) { 333 WithColor::error(errs(), ToolName) << Message << "\n"; 334 exit(1); 335 } 336 337 static void warnOnNoMatchForSections() { 338 SetVector<StringRef> MissingSections; 339 for (StringRef S : FilterSections) { 340 if (FoundSectionSet.count(S)) 341 return; 342 // User may specify a unnamed section. Don't warn for it. 343 if (!S.empty()) 344 MissingSections.insert(S); 345 } 346 347 // Warn only if no section in FilterSections is matched. 348 for (StringRef S : MissingSections) 349 reportCmdLineWarning("section '" + S + 350 "' mentioned in a -j/--section option, but not " 351 "found in any input file"); 352 } 353 354 static const Target *getTarget(const ObjectFile *Obj) { 355 // Figure out the target triple. 356 Triple TheTriple("unknown-unknown-unknown"); 357 if (TripleName.empty()) { 358 TheTriple = Obj->makeTriple(); 359 } else { 360 TheTriple.setTriple(Triple::normalize(TripleName)); 361 auto Arch = Obj->getArch(); 362 if (Arch == Triple::arm || Arch == Triple::armeb) 363 Obj->setARMSubArch(TheTriple); 364 } 365 366 // Get the target specific parser. 367 std::string Error; 368 const Target *TheTarget = TargetRegistry::lookupTarget(ArchName, TheTriple, 369 Error); 370 if (!TheTarget) 371 reportError(Obj->getFileName(), "can't find target: " + Error); 372 373 // Update the triple name and return the found target. 374 TripleName = TheTriple.getTriple(); 375 return TheTarget; 376 } 377 378 bool objdump::isRelocAddressLess(RelocationRef A, RelocationRef B) { 379 return A.getOffset() < B.getOffset(); 380 } 381 382 static Error getRelocationValueString(const RelocationRef &Rel, 383 SmallVectorImpl<char> &Result) { 384 const ObjectFile *Obj = Rel.getObject(); 385 if (auto *ELF = dyn_cast<ELFObjectFileBase>(Obj)) 386 return getELFRelocationValueString(ELF, Rel, Result); 387 if (auto *COFF = dyn_cast<COFFObjectFile>(Obj)) 388 return getCOFFRelocationValueString(COFF, Rel, Result); 389 if (auto *Wasm = dyn_cast<WasmObjectFile>(Obj)) 390 return getWasmRelocationValueString(Wasm, Rel, Result); 391 if (auto *MachO = dyn_cast<MachOObjectFile>(Obj)) 392 return getMachORelocationValueString(MachO, Rel, Result); 393 if (auto *XCOFF = dyn_cast<XCOFFObjectFile>(Obj)) 394 return getXCOFFRelocationValueString(*XCOFF, Rel, Result); 395 llvm_unreachable("unknown object file format"); 396 } 397 398 /// Indicates whether this relocation should hidden when listing 399 /// relocations, usually because it is the trailing part of a multipart 400 /// relocation that will be printed as part of the leading relocation. 401 static bool getHidden(RelocationRef RelRef) { 402 auto *MachO = dyn_cast<MachOObjectFile>(RelRef.getObject()); 403 if (!MachO) 404 return false; 405 406 unsigned Arch = MachO->getArch(); 407 DataRefImpl Rel = RelRef.getRawDataRefImpl(); 408 uint64_t Type = MachO->getRelocationType(Rel); 409 410 // On arches that use the generic relocations, GENERIC_RELOC_PAIR 411 // is always hidden. 412 if (Arch == Triple::x86 || Arch == Triple::arm || Arch == Triple::ppc) 413 return Type == MachO::GENERIC_RELOC_PAIR; 414 415 if (Arch == Triple::x86_64) { 416 // On x86_64, X86_64_RELOC_UNSIGNED is hidden only when it follows 417 // an X86_64_RELOC_SUBTRACTOR. 418 if (Type == MachO::X86_64_RELOC_UNSIGNED && Rel.d.a > 0) { 419 DataRefImpl RelPrev = Rel; 420 RelPrev.d.a--; 421 uint64_t PrevType = MachO->getRelocationType(RelPrev); 422 if (PrevType == MachO::X86_64_RELOC_SUBTRACTOR) 423 return true; 424 } 425 } 426 427 return false; 428 } 429 430 namespace { 431 432 /// Get the column at which we want to start printing the instruction 433 /// disassembly, taking into account anything which appears to the left of it. 434 unsigned getInstStartColumn(const MCSubtargetInfo &STI) { 435 return !ShowRawInsn ? 16 : STI.getTargetTriple().isX86() ? 40 : 24; 436 } 437 438 static bool isAArch64Elf(const ObjectFile &Obj) { 439 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 440 return Elf && Elf->getEMachine() == ELF::EM_AARCH64; 441 } 442 443 static bool isArmElf(const ObjectFile &Obj) { 444 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 445 return Elf && Elf->getEMachine() == ELF::EM_ARM; 446 } 447 448 static bool isCSKYElf(const ObjectFile &Obj) { 449 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 450 return Elf && Elf->getEMachine() == ELF::EM_CSKY; 451 } 452 453 static bool hasMappingSymbols(const ObjectFile &Obj) { 454 return isArmElf(Obj) || isAArch64Elf(Obj) || isCSKYElf(Obj) ; 455 } 456 457 static void printRelocation(formatted_raw_ostream &OS, StringRef FileName, 458 const RelocationRef &Rel, uint64_t Address, 459 bool Is64Bits) { 460 StringRef Fmt = Is64Bits ? "\t\t%016" PRIx64 ": " : "\t\t\t%08" PRIx64 ": "; 461 SmallString<16> Name; 462 SmallString<32> Val; 463 Rel.getTypeName(Name); 464 if (Error E = getRelocationValueString(Rel, Val)) 465 reportError(std::move(E), FileName); 466 OS << format(Fmt.data(), Address) << Name << "\t" << Val; 467 } 468 469 class PrettyPrinter { 470 public: 471 virtual ~PrettyPrinter() = default; 472 virtual void 473 printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 474 object::SectionedAddress Address, formatted_raw_ostream &OS, 475 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 476 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 477 LiveVariablePrinter &LVP) { 478 if (SP && (PrintSource || PrintLines)) 479 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 480 LVP.printBetweenInsts(OS, false); 481 482 size_t Start = OS.tell(); 483 if (LeadingAddr) 484 OS << format("%8" PRIx64 ":", Address.Address); 485 if (ShowRawInsn) { 486 OS << ' '; 487 dumpBytes(Bytes, OS); 488 } 489 490 // The output of printInst starts with a tab. Print some spaces so that 491 // the tab has 1 column and advances to the target tab stop. 492 unsigned TabStop = getInstStartColumn(STI); 493 unsigned Column = OS.tell() - Start; 494 OS.indent(Column < TabStop - 1 ? TabStop - 1 - Column : 7 - Column % 8); 495 496 if (MI) { 497 // See MCInstPrinter::printInst. On targets where a PC relative immediate 498 // is relative to the next instruction and the length of a MCInst is 499 // difficult to measure (x86), this is the address of the next 500 // instruction. 501 uint64_t Addr = 502 Address.Address + (STI.getTargetTriple().isX86() ? Bytes.size() : 0); 503 IP.printInst(MI, Addr, "", STI, OS); 504 } else 505 OS << "\t<unknown>"; 506 } 507 }; 508 PrettyPrinter PrettyPrinterInst; 509 510 class HexagonPrettyPrinter : public PrettyPrinter { 511 public: 512 void printLead(ArrayRef<uint8_t> Bytes, uint64_t Address, 513 formatted_raw_ostream &OS) { 514 uint32_t opcode = 515 (Bytes[3] << 24) | (Bytes[2] << 16) | (Bytes[1] << 8) | Bytes[0]; 516 if (LeadingAddr) 517 OS << format("%8" PRIx64 ":", Address); 518 if (ShowRawInsn) { 519 OS << "\t"; 520 dumpBytes(Bytes.slice(0, 4), OS); 521 OS << format("\t%08" PRIx32, opcode); 522 } 523 } 524 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 525 object::SectionedAddress Address, formatted_raw_ostream &OS, 526 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 527 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 528 LiveVariablePrinter &LVP) override { 529 if (SP && (PrintSource || PrintLines)) 530 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 531 if (!MI) { 532 printLead(Bytes, Address.Address, OS); 533 OS << " <unknown>"; 534 return; 535 } 536 std::string Buffer; 537 { 538 raw_string_ostream TempStream(Buffer); 539 IP.printInst(MI, Address.Address, "", STI, TempStream); 540 } 541 StringRef Contents(Buffer); 542 // Split off bundle attributes 543 auto PacketBundle = Contents.rsplit('\n'); 544 // Split off first instruction from the rest 545 auto HeadTail = PacketBundle.first.split('\n'); 546 auto Preamble = " { "; 547 auto Separator = ""; 548 549 // Hexagon's packets require relocations to be inline rather than 550 // clustered at the end of the packet. 551 std::vector<RelocationRef>::const_iterator RelCur = Rels->begin(); 552 std::vector<RelocationRef>::const_iterator RelEnd = Rels->end(); 553 auto PrintReloc = [&]() -> void { 554 while ((RelCur != RelEnd) && (RelCur->getOffset() <= Address.Address)) { 555 if (RelCur->getOffset() == Address.Address) { 556 printRelocation(OS, ObjectFilename, *RelCur, Address.Address, false); 557 return; 558 } 559 ++RelCur; 560 } 561 }; 562 563 while (!HeadTail.first.empty()) { 564 OS << Separator; 565 Separator = "\n"; 566 if (SP && (PrintSource || PrintLines)) 567 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 568 printLead(Bytes, Address.Address, OS); 569 OS << Preamble; 570 Preamble = " "; 571 StringRef Inst; 572 auto Duplex = HeadTail.first.split('\v'); 573 if (!Duplex.second.empty()) { 574 OS << Duplex.first; 575 OS << "; "; 576 Inst = Duplex.second; 577 } 578 else 579 Inst = HeadTail.first; 580 OS << Inst; 581 HeadTail = HeadTail.second.split('\n'); 582 if (HeadTail.first.empty()) 583 OS << " } " << PacketBundle.second; 584 PrintReloc(); 585 Bytes = Bytes.slice(4); 586 Address.Address += 4; 587 } 588 } 589 }; 590 HexagonPrettyPrinter HexagonPrettyPrinterInst; 591 592 class AMDGCNPrettyPrinter : public PrettyPrinter { 593 public: 594 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 595 object::SectionedAddress Address, formatted_raw_ostream &OS, 596 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 597 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 598 LiveVariablePrinter &LVP) override { 599 if (SP && (PrintSource || PrintLines)) 600 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 601 602 if (MI) { 603 SmallString<40> InstStr; 604 raw_svector_ostream IS(InstStr); 605 606 IP.printInst(MI, Address.Address, "", STI, IS); 607 608 OS << left_justify(IS.str(), 60); 609 } else { 610 // an unrecognized encoding - this is probably data so represent it 611 // using the .long directive, or .byte directive if fewer than 4 bytes 612 // remaining 613 if (Bytes.size() >= 4) { 614 OS << format("\t.long 0x%08" PRIx32 " ", 615 support::endian::read32<support::little>(Bytes.data())); 616 OS.indent(42); 617 } else { 618 OS << format("\t.byte 0x%02" PRIx8, Bytes[0]); 619 for (unsigned int i = 1; i < Bytes.size(); i++) 620 OS << format(", 0x%02" PRIx8, Bytes[i]); 621 OS.indent(55 - (6 * Bytes.size())); 622 } 623 } 624 625 OS << format("// %012" PRIX64 ":", Address.Address); 626 if (Bytes.size() >= 4) { 627 // D should be casted to uint32_t here as it is passed by format to 628 // snprintf as vararg. 629 for (uint32_t D : makeArrayRef( 630 reinterpret_cast<const support::little32_t *>(Bytes.data()), 631 Bytes.size() / 4)) 632 OS << format(" %08" PRIX32, D); 633 } else { 634 for (unsigned char B : Bytes) 635 OS << format(" %02" PRIX8, B); 636 } 637 638 if (!Annot.empty()) 639 OS << " // " << Annot; 640 } 641 }; 642 AMDGCNPrettyPrinter AMDGCNPrettyPrinterInst; 643 644 class BPFPrettyPrinter : public PrettyPrinter { 645 public: 646 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 647 object::SectionedAddress Address, formatted_raw_ostream &OS, 648 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 649 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 650 LiveVariablePrinter &LVP) override { 651 if (SP && (PrintSource || PrintLines)) 652 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 653 if (LeadingAddr) 654 OS << format("%8" PRId64 ":", Address.Address / 8); 655 if (ShowRawInsn) { 656 OS << "\t"; 657 dumpBytes(Bytes, OS); 658 } 659 if (MI) 660 IP.printInst(MI, Address.Address, "", STI, OS); 661 else 662 OS << "\t<unknown>"; 663 } 664 }; 665 BPFPrettyPrinter BPFPrettyPrinterInst; 666 667 PrettyPrinter &selectPrettyPrinter(Triple const &Triple) { 668 switch(Triple.getArch()) { 669 default: 670 return PrettyPrinterInst; 671 case Triple::hexagon: 672 return HexagonPrettyPrinterInst; 673 case Triple::amdgcn: 674 return AMDGCNPrettyPrinterInst; 675 case Triple::bpfel: 676 case Triple::bpfeb: 677 return BPFPrettyPrinterInst; 678 } 679 } 680 } 681 682 static uint8_t getElfSymbolType(const ObjectFile &Obj, const SymbolRef &Sym) { 683 assert(Obj.isELF()); 684 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 685 return unwrapOrError(Elf32LEObj->getSymbol(Sym.getRawDataRefImpl()), 686 Obj.getFileName()) 687 ->getType(); 688 if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 689 return unwrapOrError(Elf64LEObj->getSymbol(Sym.getRawDataRefImpl()), 690 Obj.getFileName()) 691 ->getType(); 692 if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 693 return unwrapOrError(Elf32BEObj->getSymbol(Sym.getRawDataRefImpl()), 694 Obj.getFileName()) 695 ->getType(); 696 if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 697 return unwrapOrError(Elf64BEObj->getSymbol(Sym.getRawDataRefImpl()), 698 Obj.getFileName()) 699 ->getType(); 700 llvm_unreachable("Unsupported binary format"); 701 } 702 703 template <class ELFT> 704 static void 705 addDynamicElfSymbols(const ELFObjectFile<ELFT> &Obj, 706 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 707 for (auto Symbol : Obj.getDynamicSymbolIterators()) { 708 uint8_t SymbolType = Symbol.getELFType(); 709 if (SymbolType == ELF::STT_SECTION) 710 continue; 711 712 uint64_t Address = unwrapOrError(Symbol.getAddress(), Obj.getFileName()); 713 // ELFSymbolRef::getAddress() returns size instead of value for common 714 // symbols which is not desirable for disassembly output. Overriding. 715 if (SymbolType == ELF::STT_COMMON) 716 Address = unwrapOrError(Obj.getSymbol(Symbol.getRawDataRefImpl()), 717 Obj.getFileName()) 718 ->st_value; 719 720 StringRef Name = unwrapOrError(Symbol.getName(), Obj.getFileName()); 721 if (Name.empty()) 722 continue; 723 724 section_iterator SecI = 725 unwrapOrError(Symbol.getSection(), Obj.getFileName()); 726 if (SecI == Obj.section_end()) 727 continue; 728 729 AllSymbols[*SecI].emplace_back(Address, Name, SymbolType); 730 } 731 } 732 733 static void 734 addDynamicElfSymbols(const ELFObjectFileBase &Obj, 735 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 736 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 737 addDynamicElfSymbols(*Elf32LEObj, AllSymbols); 738 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 739 addDynamicElfSymbols(*Elf64LEObj, AllSymbols); 740 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 741 addDynamicElfSymbols(*Elf32BEObj, AllSymbols); 742 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 743 addDynamicElfSymbols(*Elf64BEObj, AllSymbols); 744 else 745 llvm_unreachable("Unsupported binary format"); 746 } 747 748 static Optional<SectionRef> getWasmCodeSection(const WasmObjectFile &Obj) { 749 for (auto SecI : Obj.sections()) { 750 const WasmSection &Section = Obj.getWasmSection(SecI); 751 if (Section.Type == wasm::WASM_SEC_CODE) 752 return SecI; 753 } 754 return None; 755 } 756 757 static void 758 addMissingWasmCodeSymbols(const WasmObjectFile &Obj, 759 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 760 Optional<SectionRef> Section = getWasmCodeSection(Obj); 761 if (!Section) 762 return; 763 SectionSymbolsTy &Symbols = AllSymbols[*Section]; 764 765 std::set<uint64_t> SymbolAddresses; 766 for (const auto &Sym : Symbols) 767 SymbolAddresses.insert(Sym.Addr); 768 769 for (const wasm::WasmFunction &Function : Obj.functions()) { 770 uint64_t Address = Function.CodeSectionOffset; 771 // Only add fallback symbols for functions not already present in the symbol 772 // table. 773 if (SymbolAddresses.count(Address)) 774 continue; 775 // This function has no symbol, so it should have no SymbolName. 776 assert(Function.SymbolName.empty()); 777 // We use DebugName for the name, though it may be empty if there is no 778 // "name" custom section, or that section is missing a name for this 779 // function. 780 StringRef Name = Function.DebugName; 781 Symbols.emplace_back(Address, Name, ELF::STT_NOTYPE); 782 } 783 } 784 785 static void addPltEntries(const ObjectFile &Obj, 786 std::map<SectionRef, SectionSymbolsTy> &AllSymbols, 787 StringSaver &Saver) { 788 Optional<SectionRef> Plt = None; 789 for (const SectionRef &Section : Obj.sections()) { 790 Expected<StringRef> SecNameOrErr = Section.getName(); 791 if (!SecNameOrErr) { 792 consumeError(SecNameOrErr.takeError()); 793 continue; 794 } 795 if (*SecNameOrErr == ".plt") 796 Plt = Section; 797 } 798 if (!Plt) 799 return; 800 if (auto *ElfObj = dyn_cast<ELFObjectFileBase>(&Obj)) { 801 for (auto PltEntry : ElfObj->getPltAddresses()) { 802 if (PltEntry.first) { 803 SymbolRef Symbol(*PltEntry.first, ElfObj); 804 uint8_t SymbolType = getElfSymbolType(Obj, Symbol); 805 if (Expected<StringRef> NameOrErr = Symbol.getName()) { 806 if (!NameOrErr->empty()) 807 AllSymbols[*Plt].emplace_back( 808 PltEntry.second, Saver.save((*NameOrErr + "@plt").str()), 809 SymbolType); 810 continue; 811 } else { 812 // The warning has been reported in disassembleObject(). 813 consumeError(NameOrErr.takeError()); 814 } 815 } 816 reportWarning("PLT entry at 0x" + Twine::utohexstr(PltEntry.second) + 817 " references an invalid symbol", 818 Obj.getFileName()); 819 } 820 } 821 } 822 823 // Normally the disassembly output will skip blocks of zeroes. This function 824 // returns the number of zero bytes that can be skipped when dumping the 825 // disassembly of the instructions in Buf. 826 static size_t countSkippableZeroBytes(ArrayRef<uint8_t> Buf) { 827 // Find the number of leading zeroes. 828 size_t N = 0; 829 while (N < Buf.size() && !Buf[N]) 830 ++N; 831 832 // We may want to skip blocks of zero bytes, but unless we see 833 // at least 8 of them in a row. 834 if (N < 8) 835 return 0; 836 837 // We skip zeroes in multiples of 4 because do not want to truncate an 838 // instruction if it starts with a zero byte. 839 return N & ~0x3; 840 } 841 842 // Returns a map from sections to their relocations. 843 static std::map<SectionRef, std::vector<RelocationRef>> 844 getRelocsMap(object::ObjectFile const &Obj) { 845 std::map<SectionRef, std::vector<RelocationRef>> Ret; 846 uint64_t I = (uint64_t)-1; 847 for (SectionRef Sec : Obj.sections()) { 848 ++I; 849 Expected<section_iterator> RelocatedOrErr = Sec.getRelocatedSection(); 850 if (!RelocatedOrErr) 851 reportError(Obj.getFileName(), 852 "section (" + Twine(I) + 853 "): failed to get a relocated section: " + 854 toString(RelocatedOrErr.takeError())); 855 856 section_iterator Relocated = *RelocatedOrErr; 857 if (Relocated == Obj.section_end() || !checkSectionFilter(*Relocated).Keep) 858 continue; 859 std::vector<RelocationRef> &V = Ret[*Relocated]; 860 append_range(V, Sec.relocations()); 861 // Sort relocations by address. 862 llvm::stable_sort(V, isRelocAddressLess); 863 } 864 return Ret; 865 } 866 867 // Used for --adjust-vma to check if address should be adjusted by the 868 // specified value for a given section. 869 // For ELF we do not adjust non-allocatable sections like debug ones, 870 // because they are not loadable. 871 // TODO: implement for other file formats. 872 static bool shouldAdjustVA(const SectionRef &Section) { 873 const ObjectFile *Obj = Section.getObject(); 874 if (Obj->isELF()) 875 return ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC; 876 return false; 877 } 878 879 880 typedef std::pair<uint64_t, char> MappingSymbolPair; 881 static char getMappingSymbolKind(ArrayRef<MappingSymbolPair> MappingSymbols, 882 uint64_t Address) { 883 auto It = 884 partition_point(MappingSymbols, [Address](const MappingSymbolPair &Val) { 885 return Val.first <= Address; 886 }); 887 // Return zero for any address before the first mapping symbol; this means 888 // we should use the default disassembly mode, depending on the target. 889 if (It == MappingSymbols.begin()) 890 return '\x00'; 891 return (It - 1)->second; 892 } 893 894 static uint64_t dumpARMELFData(uint64_t SectionAddr, uint64_t Index, 895 uint64_t End, const ObjectFile &Obj, 896 ArrayRef<uint8_t> Bytes, 897 ArrayRef<MappingSymbolPair> MappingSymbols, 898 raw_ostream &OS) { 899 support::endianness Endian = 900 Obj.isLittleEndian() ? support::little : support::big; 901 OS << format("%8" PRIx64 ":\t", SectionAddr + Index); 902 if (Index + 4 <= End) { 903 dumpBytes(Bytes.slice(Index, 4), OS); 904 OS << "\t.word\t" 905 << format_hex(support::endian::read32(Bytes.data() + Index, Endian), 906 10); 907 return 4; 908 } 909 if (Index + 2 <= End) { 910 dumpBytes(Bytes.slice(Index, 2), OS); 911 OS << "\t\t.short\t" 912 << format_hex(support::endian::read16(Bytes.data() + Index, Endian), 913 6); 914 return 2; 915 } 916 dumpBytes(Bytes.slice(Index, 1), OS); 917 OS << "\t\t.byte\t" << format_hex(Bytes[0], 4); 918 return 1; 919 } 920 921 static void dumpELFData(uint64_t SectionAddr, uint64_t Index, uint64_t End, 922 ArrayRef<uint8_t> Bytes) { 923 // print out data up to 8 bytes at a time in hex and ascii 924 uint8_t AsciiData[9] = {'\0'}; 925 uint8_t Byte; 926 int NumBytes = 0; 927 928 for (; Index < End; ++Index) { 929 if (NumBytes == 0) 930 outs() << format("%8" PRIx64 ":", SectionAddr + Index); 931 Byte = Bytes.slice(Index)[0]; 932 outs() << format(" %02x", Byte); 933 AsciiData[NumBytes] = isPrint(Byte) ? Byte : '.'; 934 935 uint8_t IndentOffset = 0; 936 NumBytes++; 937 if (Index == End - 1 || NumBytes > 8) { 938 // Indent the space for less than 8 bytes data. 939 // 2 spaces for byte and one for space between bytes 940 IndentOffset = 3 * (8 - NumBytes); 941 for (int Excess = NumBytes; Excess < 8; Excess++) 942 AsciiData[Excess] = '\0'; 943 NumBytes = 8; 944 } 945 if (NumBytes == 8) { 946 AsciiData[8] = '\0'; 947 outs() << std::string(IndentOffset, ' ') << " "; 948 outs() << reinterpret_cast<char *>(AsciiData); 949 outs() << '\n'; 950 NumBytes = 0; 951 } 952 } 953 } 954 955 SymbolInfoTy objdump::createSymbolInfo(const ObjectFile &Obj, 956 const SymbolRef &Symbol) { 957 const StringRef FileName = Obj.getFileName(); 958 const uint64_t Addr = unwrapOrError(Symbol.getAddress(), FileName); 959 const StringRef Name = unwrapOrError(Symbol.getName(), FileName); 960 961 if (Obj.isXCOFF() && SymbolDescription) { 962 const auto &XCOFFObj = cast<XCOFFObjectFile>(Obj); 963 DataRefImpl SymbolDRI = Symbol.getRawDataRefImpl(); 964 965 const uint32_t SymbolIndex = XCOFFObj.getSymbolIndex(SymbolDRI.p); 966 Optional<XCOFF::StorageMappingClass> Smc = 967 getXCOFFSymbolCsectSMC(XCOFFObj, Symbol); 968 return SymbolInfoTy(Addr, Name, Smc, SymbolIndex, 969 isLabel(XCOFFObj, Symbol)); 970 } else if (Obj.isXCOFF()) { 971 const SymbolRef::Type SymType = unwrapOrError(Symbol.getType(), FileName); 972 return SymbolInfoTy(Addr, Name, SymType, true); 973 } else 974 return SymbolInfoTy(Addr, Name, 975 Obj.isELF() ? getElfSymbolType(Obj, Symbol) 976 : (uint8_t)ELF::STT_NOTYPE); 977 } 978 979 static SymbolInfoTy createDummySymbolInfo(const ObjectFile &Obj, 980 const uint64_t Addr, StringRef &Name, 981 uint8_t Type) { 982 if (Obj.isXCOFF() && SymbolDescription) 983 return SymbolInfoTy(Addr, Name, None, None, false); 984 else 985 return SymbolInfoTy(Addr, Name, Type); 986 } 987 988 static void 989 collectBBAddrMapLabels(const std::unordered_map<uint64_t, BBAddrMap> &AddrToBBAddrMap, 990 uint64_t SectionAddr, uint64_t Start, uint64_t End, 991 std::unordered_map<uint64_t, std::vector<std::string>> &Labels) { 992 if (AddrToBBAddrMap.empty()) 993 return; 994 Labels.clear(); 995 uint64_t StartAddress = SectionAddr + Start; 996 uint64_t EndAddress = SectionAddr + End; 997 auto Iter = AddrToBBAddrMap.find(StartAddress); 998 if (Iter == AddrToBBAddrMap.end()) 999 return; 1000 for (unsigned I = 0, Size = Iter->second.BBEntries.size(); I < Size; ++I) { 1001 uint64_t BBAddress = Iter->second.BBEntries[I].Offset + Iter->second.Addr; 1002 if (BBAddress >= EndAddress) 1003 continue; 1004 Labels[BBAddress].push_back(("BB" + Twine(I)).str()); 1005 } 1006 } 1007 1008 static void collectLocalBranchTargets( 1009 ArrayRef<uint8_t> Bytes, const MCInstrAnalysis *MIA, MCDisassembler *DisAsm, 1010 MCInstPrinter *IP, const MCSubtargetInfo *STI, uint64_t SectionAddr, 1011 uint64_t Start, uint64_t End, std::unordered_map<uint64_t, std::string> &Labels) { 1012 // So far only supports PowerPC and X86. 1013 if (!STI->getTargetTriple().isPPC() && !STI->getTargetTriple().isX86()) 1014 return; 1015 1016 Labels.clear(); 1017 unsigned LabelCount = 0; 1018 Start += SectionAddr; 1019 End += SectionAddr; 1020 uint64_t Index = Start; 1021 while (Index < End) { 1022 // Disassemble a real instruction and record function-local branch labels. 1023 MCInst Inst; 1024 uint64_t Size; 1025 bool Disassembled = DisAsm->getInstruction( 1026 Inst, Size, Bytes.slice(Index - SectionAddr), Index, nulls()); 1027 if (Size == 0) 1028 Size = 1; 1029 1030 if (Disassembled && MIA) { 1031 uint64_t Target; 1032 bool TargetKnown = MIA->evaluateBranch(Inst, Index, Size, Target); 1033 // On PowerPC, if the address of a branch is the same as the target, it 1034 // means that it's a function call. Do not mark the label for this case. 1035 if (TargetKnown && (Target >= Start && Target < End) && 1036 !Labels.count(Target) && 1037 !(STI->getTargetTriple().isPPC() && Target == Index)) 1038 Labels[Target] = ("L" + Twine(LabelCount++)).str(); 1039 } 1040 Index += Size; 1041 } 1042 } 1043 1044 // Create an MCSymbolizer for the target and add it to the MCDisassembler. 1045 // This is currently only used on AMDGPU, and assumes the format of the 1046 // void * argument passed to AMDGPU's createMCSymbolizer. 1047 static void addSymbolizer( 1048 MCContext &Ctx, const Target *Target, StringRef TripleName, 1049 MCDisassembler *DisAsm, uint64_t SectionAddr, ArrayRef<uint8_t> Bytes, 1050 SectionSymbolsTy &Symbols, 1051 std::vector<std::unique_ptr<std::string>> &SynthesizedLabelNames) { 1052 1053 std::unique_ptr<MCRelocationInfo> RelInfo( 1054 Target->createMCRelocationInfo(TripleName, Ctx)); 1055 if (!RelInfo) 1056 return; 1057 std::unique_ptr<MCSymbolizer> Symbolizer(Target->createMCSymbolizer( 1058 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1059 MCSymbolizer *SymbolizerPtr = &*Symbolizer; 1060 DisAsm->setSymbolizer(std::move(Symbolizer)); 1061 1062 if (!SymbolizeOperands) 1063 return; 1064 1065 // Synthesize labels referenced by branch instructions by 1066 // disassembling, discarding the output, and collecting the referenced 1067 // addresses from the symbolizer. 1068 for (size_t Index = 0; Index != Bytes.size();) { 1069 MCInst Inst; 1070 uint64_t Size; 1071 DisAsm->getInstruction(Inst, Size, Bytes.slice(Index), SectionAddr + Index, 1072 nulls()); 1073 if (Size == 0) 1074 Size = 1; 1075 Index += Size; 1076 } 1077 ArrayRef<uint64_t> LabelAddrsRef = SymbolizerPtr->getReferencedAddresses(); 1078 // Copy and sort to remove duplicates. 1079 std::vector<uint64_t> LabelAddrs; 1080 LabelAddrs.insert(LabelAddrs.end(), LabelAddrsRef.begin(), 1081 LabelAddrsRef.end()); 1082 llvm::sort(LabelAddrs); 1083 LabelAddrs.resize(std::unique(LabelAddrs.begin(), LabelAddrs.end()) - 1084 LabelAddrs.begin()); 1085 // Add the labels. 1086 for (unsigned LabelNum = 0; LabelNum != LabelAddrs.size(); ++LabelNum) { 1087 auto Name = std::make_unique<std::string>(); 1088 *Name = (Twine("L") + Twine(LabelNum)).str(); 1089 SynthesizedLabelNames.push_back(std::move(Name)); 1090 Symbols.push_back(SymbolInfoTy( 1091 LabelAddrs[LabelNum], *SynthesizedLabelNames.back(), ELF::STT_NOTYPE)); 1092 } 1093 llvm::stable_sort(Symbols); 1094 // Recreate the symbolizer with the new symbols list. 1095 RelInfo.reset(Target->createMCRelocationInfo(TripleName, Ctx)); 1096 Symbolizer.reset(Target->createMCSymbolizer( 1097 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1098 DisAsm->setSymbolizer(std::move(Symbolizer)); 1099 } 1100 1101 static StringRef getSegmentName(const MachOObjectFile *MachO, 1102 const SectionRef &Section) { 1103 if (MachO) { 1104 DataRefImpl DR = Section.getRawDataRefImpl(); 1105 StringRef SegmentName = MachO->getSectionFinalSegmentName(DR); 1106 return SegmentName; 1107 } 1108 return ""; 1109 } 1110 1111 static void emitPostInstructionInfo(formatted_raw_ostream &FOS, 1112 const MCAsmInfo &MAI, 1113 const MCSubtargetInfo &STI, 1114 StringRef Comments, 1115 LiveVariablePrinter &LVP) { 1116 do { 1117 if (!Comments.empty()) { 1118 // Emit a line of comments. 1119 StringRef Comment; 1120 std::tie(Comment, Comments) = Comments.split('\n'); 1121 // MAI.getCommentColumn() assumes that instructions are printed at the 1122 // position of 8, while getInstStartColumn() returns the actual position. 1123 unsigned CommentColumn = 1124 MAI.getCommentColumn() - 8 + getInstStartColumn(STI); 1125 FOS.PadToColumn(CommentColumn); 1126 FOS << MAI.getCommentString() << ' ' << Comment; 1127 } 1128 LVP.printAfterInst(FOS); 1129 FOS << '\n'; 1130 } while (!Comments.empty()); 1131 FOS.flush(); 1132 } 1133 1134 static void createFakeELFSections(ObjectFile &Obj) { 1135 assert(Obj.isELF()); 1136 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 1137 Elf32LEObj->createFakeSections(); 1138 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 1139 Elf64LEObj->createFakeSections(); 1140 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 1141 Elf32BEObj->createFakeSections(); 1142 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 1143 Elf64BEObj->createFakeSections(); 1144 else 1145 llvm_unreachable("Unsupported binary format"); 1146 } 1147 1148 static void disassembleObject(const Target *TheTarget, ObjectFile &Obj, 1149 MCContext &Ctx, MCDisassembler *PrimaryDisAsm, 1150 MCDisassembler *SecondaryDisAsm, 1151 const MCInstrAnalysis *MIA, MCInstPrinter *IP, 1152 const MCSubtargetInfo *PrimarySTI, 1153 const MCSubtargetInfo *SecondarySTI, 1154 PrettyPrinter &PIP, SourcePrinter &SP, 1155 bool InlineRelocs) { 1156 const MCSubtargetInfo *STI = PrimarySTI; 1157 MCDisassembler *DisAsm = PrimaryDisAsm; 1158 bool PrimaryIsThumb = false; 1159 if (isArmElf(Obj)) 1160 PrimaryIsThumb = STI->checkFeatures("+thumb-mode"); 1161 1162 std::map<SectionRef, std::vector<RelocationRef>> RelocMap; 1163 if (InlineRelocs) 1164 RelocMap = getRelocsMap(Obj); 1165 bool Is64Bits = Obj.getBytesInAddress() > 4; 1166 1167 // Create a mapping from virtual address to symbol name. This is used to 1168 // pretty print the symbols while disassembling. 1169 std::map<SectionRef, SectionSymbolsTy> AllSymbols; 1170 SectionSymbolsTy AbsoluteSymbols; 1171 const StringRef FileName = Obj.getFileName(); 1172 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&Obj); 1173 for (const SymbolRef &Symbol : Obj.symbols()) { 1174 Expected<StringRef> NameOrErr = Symbol.getName(); 1175 if (!NameOrErr) { 1176 reportWarning(toString(NameOrErr.takeError()), FileName); 1177 continue; 1178 } 1179 if (NameOrErr->empty() && !(Obj.isXCOFF() && SymbolDescription)) 1180 continue; 1181 1182 if (Obj.isELF() && getElfSymbolType(Obj, Symbol) == ELF::STT_SECTION) 1183 continue; 1184 1185 if (MachO) { 1186 // __mh_(execute|dylib|dylinker|bundle|preload|object)_header are special 1187 // symbols that support MachO header introspection. They do not bind to 1188 // code locations and are irrelevant for disassembly. 1189 if (NameOrErr->startswith("__mh_") && NameOrErr->endswith("_header")) 1190 continue; 1191 // Don't ask a Mach-O STAB symbol for its section unless you know that 1192 // STAB symbol's section field refers to a valid section index. Otherwise 1193 // the symbol may error trying to load a section that does not exist. 1194 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 1195 uint8_t NType = (MachO->is64Bit() ? 1196 MachO->getSymbol64TableEntry(SymDRI).n_type: 1197 MachO->getSymbolTableEntry(SymDRI).n_type); 1198 if (NType & MachO::N_STAB) 1199 continue; 1200 } 1201 1202 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1203 if (SecI != Obj.section_end()) 1204 AllSymbols[*SecI].push_back(createSymbolInfo(Obj, Symbol)); 1205 else 1206 AbsoluteSymbols.push_back(createSymbolInfo(Obj, Symbol)); 1207 } 1208 1209 if (AllSymbols.empty() && Obj.isELF()) 1210 addDynamicElfSymbols(cast<ELFObjectFileBase>(Obj), AllSymbols); 1211 1212 if (Obj.isWasm()) 1213 addMissingWasmCodeSymbols(cast<WasmObjectFile>(Obj), AllSymbols); 1214 1215 if (Obj.isELF() && Obj.sections().empty()) 1216 createFakeELFSections(Obj); 1217 1218 BumpPtrAllocator A; 1219 StringSaver Saver(A); 1220 addPltEntries(Obj, AllSymbols, Saver); 1221 1222 // Create a mapping from virtual address to section. An empty section can 1223 // cause more than one section at the same address. Sort such sections to be 1224 // before same-addressed non-empty sections so that symbol lookups prefer the 1225 // non-empty section. 1226 std::vector<std::pair<uint64_t, SectionRef>> SectionAddresses; 1227 for (SectionRef Sec : Obj.sections()) 1228 SectionAddresses.emplace_back(Sec.getAddress(), Sec); 1229 llvm::stable_sort(SectionAddresses, [](const auto &LHS, const auto &RHS) { 1230 if (LHS.first != RHS.first) 1231 return LHS.first < RHS.first; 1232 return LHS.second.getSize() < RHS.second.getSize(); 1233 }); 1234 1235 // Linked executables (.exe and .dll files) typically don't include a real 1236 // symbol table but they might contain an export table. 1237 if (const auto *COFFObj = dyn_cast<COFFObjectFile>(&Obj)) { 1238 for (const auto &ExportEntry : COFFObj->export_directories()) { 1239 StringRef Name; 1240 if (Error E = ExportEntry.getSymbolName(Name)) 1241 reportError(std::move(E), Obj.getFileName()); 1242 if (Name.empty()) 1243 continue; 1244 1245 uint32_t RVA; 1246 if (Error E = ExportEntry.getExportRVA(RVA)) 1247 reportError(std::move(E), Obj.getFileName()); 1248 1249 uint64_t VA = COFFObj->getImageBase() + RVA; 1250 auto Sec = partition_point( 1251 SectionAddresses, [VA](const std::pair<uint64_t, SectionRef> &O) { 1252 return O.first <= VA; 1253 }); 1254 if (Sec != SectionAddresses.begin()) { 1255 --Sec; 1256 AllSymbols[Sec->second].emplace_back(VA, Name, ELF::STT_NOTYPE); 1257 } else 1258 AbsoluteSymbols.emplace_back(VA, Name, ELF::STT_NOTYPE); 1259 } 1260 } 1261 1262 // Sort all the symbols, this allows us to use a simple binary search to find 1263 // Multiple symbols can have the same address. Use a stable sort to stabilize 1264 // the output. 1265 StringSet<> FoundDisasmSymbolSet; 1266 for (std::pair<const SectionRef, SectionSymbolsTy> &SecSyms : AllSymbols) 1267 llvm::stable_sort(SecSyms.second); 1268 llvm::stable_sort(AbsoluteSymbols); 1269 1270 std::unique_ptr<DWARFContext> DICtx; 1271 LiveVariablePrinter LVP(*Ctx.getRegisterInfo(), *STI); 1272 1273 if (DbgVariables != DVDisabled) { 1274 DICtx = DWARFContext::create(Obj); 1275 for (const std::unique_ptr<DWARFUnit> &CU : DICtx->compile_units()) 1276 LVP.addCompileUnit(CU->getUnitDIE(false)); 1277 } 1278 1279 LLVM_DEBUG(LVP.dump()); 1280 1281 std::unordered_map<uint64_t, BBAddrMap> AddrToBBAddrMap; 1282 auto ReadBBAddrMap = [&](Optional<unsigned> SectionIndex = None) { 1283 AddrToBBAddrMap.clear(); 1284 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj)) { 1285 auto BBAddrMapsOrErr = Elf->readBBAddrMap(SectionIndex); 1286 if (!BBAddrMapsOrErr) 1287 reportWarning(toString(BBAddrMapsOrErr.takeError()), 1288 Obj.getFileName()); 1289 for (auto &FunctionBBAddrMap : *BBAddrMapsOrErr) 1290 AddrToBBAddrMap.emplace(FunctionBBAddrMap.Addr, 1291 std::move(FunctionBBAddrMap)); 1292 } 1293 }; 1294 1295 // For non-relocatable objects, Read all LLVM_BB_ADDR_MAP sections into a 1296 // single mapping, since they don't have any conflicts. 1297 if (SymbolizeOperands && !Obj.isRelocatableObject()) 1298 ReadBBAddrMap(); 1299 1300 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 1301 if (FilterSections.empty() && !DisassembleAll && 1302 (!Section.isText() || Section.isVirtual())) 1303 continue; 1304 1305 uint64_t SectionAddr = Section.getAddress(); 1306 uint64_t SectSize = Section.getSize(); 1307 if (!SectSize) 1308 continue; 1309 1310 // For relocatable object files, read the LLVM_BB_ADDR_MAP section 1311 // corresponding to this section, if present. 1312 if (SymbolizeOperands && Obj.isRelocatableObject()) 1313 ReadBBAddrMap(Section.getIndex()); 1314 1315 // Get the list of all the symbols in this section. 1316 SectionSymbolsTy &Symbols = AllSymbols[Section]; 1317 std::vector<MappingSymbolPair> MappingSymbols; 1318 if (hasMappingSymbols(Obj)) { 1319 for (const auto &Symb : Symbols) { 1320 uint64_t Address = Symb.Addr; 1321 StringRef Name = Symb.Name; 1322 if (Name.startswith("$d")) 1323 MappingSymbols.emplace_back(Address - SectionAddr, 'd'); 1324 if (Name.startswith("$x")) 1325 MappingSymbols.emplace_back(Address - SectionAddr, 'x'); 1326 if (Name.startswith("$a")) 1327 MappingSymbols.emplace_back(Address - SectionAddr, 'a'); 1328 if (Name.startswith("$t")) 1329 MappingSymbols.emplace_back(Address - SectionAddr, 't'); 1330 } 1331 } 1332 1333 llvm::sort(MappingSymbols); 1334 1335 ArrayRef<uint8_t> Bytes = arrayRefFromStringRef( 1336 unwrapOrError(Section.getContents(), Obj.getFileName())); 1337 1338 std::vector<std::unique_ptr<std::string>> SynthesizedLabelNames; 1339 if (Obj.isELF() && Obj.getArch() == Triple::amdgcn) { 1340 // AMDGPU disassembler uses symbolizer for printing labels 1341 addSymbolizer(Ctx, TheTarget, TripleName, DisAsm, SectionAddr, Bytes, 1342 Symbols, SynthesizedLabelNames); 1343 } 1344 1345 StringRef SegmentName = getSegmentName(MachO, Section); 1346 StringRef SectionName = unwrapOrError(Section.getName(), Obj.getFileName()); 1347 // If the section has no symbol at the start, just insert a dummy one. 1348 if (Symbols.empty() || Symbols[0].Addr != 0) { 1349 Symbols.insert(Symbols.begin(), 1350 createDummySymbolInfo(Obj, SectionAddr, SectionName, 1351 Section.isText() ? ELF::STT_FUNC 1352 : ELF::STT_OBJECT)); 1353 } 1354 1355 SmallString<40> Comments; 1356 raw_svector_ostream CommentStream(Comments); 1357 1358 uint64_t VMAAdjustment = 0; 1359 if (shouldAdjustVA(Section)) 1360 VMAAdjustment = AdjustVMA; 1361 1362 // In executable and shared objects, r_offset holds a virtual address. 1363 // Subtract SectionAddr from the r_offset field of a relocation to get 1364 // the section offset. 1365 uint64_t RelAdjustment = Obj.isRelocatableObject() ? 0 : SectionAddr; 1366 uint64_t Size; 1367 uint64_t Index; 1368 bool PrintedSection = false; 1369 std::vector<RelocationRef> Rels = RelocMap[Section]; 1370 std::vector<RelocationRef>::const_iterator RelCur = Rels.begin(); 1371 std::vector<RelocationRef>::const_iterator RelEnd = Rels.end(); 1372 // Disassemble symbol by symbol. 1373 for (unsigned SI = 0, SE = Symbols.size(); SI != SE; ++SI) { 1374 std::string SymbolName = Symbols[SI].Name.str(); 1375 if (Demangle) 1376 SymbolName = demangle(SymbolName); 1377 1378 // Skip if --disassemble-symbols is not empty and the symbol is not in 1379 // the list. 1380 if (!DisasmSymbolSet.empty() && !DisasmSymbolSet.count(SymbolName)) 1381 continue; 1382 1383 uint64_t Start = Symbols[SI].Addr; 1384 if (Start < SectionAddr || StopAddress <= Start) 1385 continue; 1386 else 1387 FoundDisasmSymbolSet.insert(SymbolName); 1388 1389 // The end is the section end, the beginning of the next symbol, or 1390 // --stop-address. 1391 uint64_t End = std::min<uint64_t>(SectionAddr + SectSize, StopAddress); 1392 if (SI + 1 < SE) 1393 End = std::min(End, Symbols[SI + 1].Addr); 1394 if (Start >= End || End <= StartAddress) 1395 continue; 1396 Start -= SectionAddr; 1397 End -= SectionAddr; 1398 1399 if (!PrintedSection) { 1400 PrintedSection = true; 1401 outs() << "\nDisassembly of section "; 1402 if (!SegmentName.empty()) 1403 outs() << SegmentName << ","; 1404 outs() << SectionName << ":\n"; 1405 } 1406 1407 outs() << '\n'; 1408 if (LeadingAddr) 1409 outs() << format(Is64Bits ? "%016" PRIx64 " " : "%08" PRIx64 " ", 1410 SectionAddr + Start + VMAAdjustment); 1411 if (Obj.isXCOFF() && SymbolDescription) { 1412 outs() << getXCOFFSymbolDescription(Symbols[SI], SymbolName) << ":\n"; 1413 } else 1414 outs() << '<' << SymbolName << ">:\n"; 1415 1416 // Don't print raw contents of a virtual section. A virtual section 1417 // doesn't have any contents in the file. 1418 if (Section.isVirtual()) { 1419 outs() << "...\n"; 1420 continue; 1421 } 1422 1423 auto Status = DisAsm->onSymbolStart(Symbols[SI], Size, 1424 Bytes.slice(Start, End - Start), 1425 SectionAddr + Start, CommentStream); 1426 // To have round trippable disassembly, we fall back to decoding the 1427 // remaining bytes as instructions. 1428 // 1429 // If there is a failure, we disassemble the failed region as bytes before 1430 // falling back. The target is expected to print nothing in this case. 1431 // 1432 // If there is Success or SoftFail i.e no 'real' failure, we go ahead by 1433 // Size bytes before falling back. 1434 // So if the entire symbol is 'eaten' by the target: 1435 // Start += Size // Now Start = End and we will never decode as 1436 // // instructions 1437 // 1438 // Right now, most targets return None i.e ignore to treat a symbol 1439 // separately. But WebAssembly decodes preludes for some symbols. 1440 // 1441 if (Status) { 1442 if (Status.value() == MCDisassembler::Fail) { 1443 outs() << "// Error in decoding " << SymbolName 1444 << " : Decoding failed region as bytes.\n"; 1445 for (uint64_t I = 0; I < Size; ++I) { 1446 outs() << "\t.byte\t " << format_hex(Bytes[I], 1, /*Upper=*/true) 1447 << "\n"; 1448 } 1449 } 1450 } else { 1451 Size = 0; 1452 } 1453 1454 Start += Size; 1455 1456 Index = Start; 1457 if (SectionAddr < StartAddress) 1458 Index = std::max<uint64_t>(Index, StartAddress - SectionAddr); 1459 1460 // If there is a data/common symbol inside an ELF text section and we are 1461 // only disassembling text (applicable all architectures), we are in a 1462 // situation where we must print the data and not disassemble it. 1463 if (Obj.isELF() && !DisassembleAll && Section.isText()) { 1464 uint8_t SymTy = Symbols[SI].Type; 1465 if (SymTy == ELF::STT_OBJECT || SymTy == ELF::STT_COMMON) { 1466 dumpELFData(SectionAddr, Index, End, Bytes); 1467 Index = End; 1468 } 1469 } 1470 1471 bool CheckARMELFData = hasMappingSymbols(Obj) && 1472 Symbols[SI].Type != ELF::STT_OBJECT && 1473 !DisassembleAll; 1474 bool DumpARMELFData = false; 1475 formatted_raw_ostream FOS(outs()); 1476 1477 std::unordered_map<uint64_t, std::string> AllLabels; 1478 std::unordered_map<uint64_t, std::vector<std::string>> BBAddrMapLabels; 1479 if (SymbolizeOperands) { 1480 collectLocalBranchTargets(Bytes, MIA, DisAsm, IP, PrimarySTI, 1481 SectionAddr, Index, End, AllLabels); 1482 collectBBAddrMapLabels(AddrToBBAddrMap, SectionAddr, Index, End, 1483 BBAddrMapLabels); 1484 } 1485 1486 while (Index < End) { 1487 // ARM and AArch64 ELF binaries can interleave data and text in the 1488 // same section. We rely on the markers introduced to understand what 1489 // we need to dump. If the data marker is within a function, it is 1490 // denoted as a word/short etc. 1491 if (CheckARMELFData) { 1492 char Kind = getMappingSymbolKind(MappingSymbols, Index); 1493 DumpARMELFData = Kind == 'd'; 1494 if (SecondarySTI) { 1495 if (Kind == 'a') { 1496 STI = PrimaryIsThumb ? SecondarySTI : PrimarySTI; 1497 DisAsm = PrimaryIsThumb ? SecondaryDisAsm : PrimaryDisAsm; 1498 } else if (Kind == 't') { 1499 STI = PrimaryIsThumb ? PrimarySTI : SecondarySTI; 1500 DisAsm = PrimaryIsThumb ? PrimaryDisAsm : SecondaryDisAsm; 1501 } 1502 } 1503 } 1504 1505 if (DumpARMELFData) { 1506 Size = dumpARMELFData(SectionAddr, Index, End, Obj, Bytes, 1507 MappingSymbols, FOS); 1508 } else { 1509 // When -z or --disassemble-zeroes are given we always dissasemble 1510 // them. Otherwise we might want to skip zero bytes we see. 1511 if (!DisassembleZeroes) { 1512 uint64_t MaxOffset = End - Index; 1513 // For --reloc: print zero blocks patched by relocations, so that 1514 // relocations can be shown in the dump. 1515 if (RelCur != RelEnd) 1516 MaxOffset = std::min(RelCur->getOffset() - RelAdjustment - Index, 1517 MaxOffset); 1518 1519 if (size_t N = 1520 countSkippableZeroBytes(Bytes.slice(Index, MaxOffset))) { 1521 FOS << "\t\t..." << '\n'; 1522 Index += N; 1523 continue; 1524 } 1525 } 1526 1527 // Print local label if there's any. 1528 auto Iter1 = BBAddrMapLabels.find(SectionAddr + Index); 1529 if (Iter1 != BBAddrMapLabels.end()) { 1530 for (StringRef Label : Iter1->second) 1531 FOS << "<" << Label << ">:\n"; 1532 } else { 1533 auto Iter2 = AllLabels.find(SectionAddr + Index); 1534 if (Iter2 != AllLabels.end()) 1535 FOS << "<" << Iter2->second << ">:\n"; 1536 } 1537 1538 // Disassemble a real instruction or a data when disassemble all is 1539 // provided 1540 MCInst Inst; 1541 bool Disassembled = 1542 DisAsm->getInstruction(Inst, Size, Bytes.slice(Index), 1543 SectionAddr + Index, CommentStream); 1544 if (Size == 0) 1545 Size = 1; 1546 1547 LVP.update({Index, Section.getIndex()}, 1548 {Index + Size, Section.getIndex()}, Index + Size != End); 1549 1550 IP->setCommentStream(CommentStream); 1551 1552 PIP.printInst( 1553 *IP, Disassembled ? &Inst : nullptr, Bytes.slice(Index, Size), 1554 {SectionAddr + Index + VMAAdjustment, Section.getIndex()}, FOS, 1555 "", *STI, &SP, Obj.getFileName(), &Rels, LVP); 1556 1557 IP->setCommentStream(llvm::nulls()); 1558 1559 // If disassembly has failed, avoid analysing invalid/incomplete 1560 // instruction information. Otherwise, try to resolve the target 1561 // address (jump target or memory operand address) and print it on the 1562 // right of the instruction. 1563 if (Disassembled && MIA) { 1564 // Branch targets are printed just after the instructions. 1565 llvm::raw_ostream *TargetOS = &FOS; 1566 uint64_t Target; 1567 bool PrintTarget = 1568 MIA->evaluateBranch(Inst, SectionAddr + Index, Size, Target); 1569 if (!PrintTarget) 1570 if (Optional<uint64_t> MaybeTarget = 1571 MIA->evaluateMemoryOperandAddress( 1572 Inst, STI, SectionAddr + Index, Size)) { 1573 Target = *MaybeTarget; 1574 PrintTarget = true; 1575 // Do not print real address when symbolizing. 1576 if (!SymbolizeOperands) { 1577 // Memory operand addresses are printed as comments. 1578 TargetOS = &CommentStream; 1579 *TargetOS << "0x" << Twine::utohexstr(Target); 1580 } 1581 } 1582 if (PrintTarget) { 1583 // In a relocatable object, the target's section must reside in 1584 // the same section as the call instruction or it is accessed 1585 // through a relocation. 1586 // 1587 // In a non-relocatable object, the target may be in any section. 1588 // In that case, locate the section(s) containing the target 1589 // address and find the symbol in one of those, if possible. 1590 // 1591 // N.B. We don't walk the relocations in the relocatable case yet. 1592 std::vector<const SectionSymbolsTy *> TargetSectionSymbols; 1593 if (!Obj.isRelocatableObject()) { 1594 auto It = llvm::partition_point( 1595 SectionAddresses, 1596 [=](const std::pair<uint64_t, SectionRef> &O) { 1597 return O.first <= Target; 1598 }); 1599 uint64_t TargetSecAddr = 0; 1600 while (It != SectionAddresses.begin()) { 1601 --It; 1602 if (TargetSecAddr == 0) 1603 TargetSecAddr = It->first; 1604 if (It->first != TargetSecAddr) 1605 break; 1606 TargetSectionSymbols.push_back(&AllSymbols[It->second]); 1607 } 1608 } else { 1609 TargetSectionSymbols.push_back(&Symbols); 1610 } 1611 TargetSectionSymbols.push_back(&AbsoluteSymbols); 1612 1613 // Find the last symbol in the first candidate section whose 1614 // offset is less than or equal to the target. If there are no 1615 // such symbols, try in the next section and so on, before finally 1616 // using the nearest preceding absolute symbol (if any), if there 1617 // are no other valid symbols. 1618 const SymbolInfoTy *TargetSym = nullptr; 1619 for (const SectionSymbolsTy *TargetSymbols : 1620 TargetSectionSymbols) { 1621 auto It = llvm::partition_point( 1622 *TargetSymbols, 1623 [=](const SymbolInfoTy &O) { return O.Addr <= Target; }); 1624 if (It != TargetSymbols->begin()) { 1625 TargetSym = &*(It - 1); 1626 break; 1627 } 1628 } 1629 1630 // Print the labels corresponding to the target if there's any. 1631 bool BBAddrMapLabelAvailable = BBAddrMapLabels.count(Target); 1632 bool LabelAvailable = AllLabels.count(Target); 1633 if (TargetSym != nullptr) { 1634 uint64_t TargetAddress = TargetSym->Addr; 1635 uint64_t Disp = Target - TargetAddress; 1636 std::string TargetName = TargetSym->Name.str(); 1637 if (Demangle) 1638 TargetName = demangle(TargetName); 1639 1640 *TargetOS << " <"; 1641 if (!Disp) { 1642 // Always Print the binary symbol precisely corresponding to 1643 // the target address. 1644 *TargetOS << TargetName; 1645 } else if (BBAddrMapLabelAvailable) { 1646 *TargetOS << BBAddrMapLabels[Target].front(); 1647 } else if (LabelAvailable) { 1648 *TargetOS << AllLabels[Target]; 1649 } else { 1650 // Always Print the binary symbol plus an offset if there's no 1651 // local label corresponding to the target address. 1652 *TargetOS << TargetName << "+0x" << Twine::utohexstr(Disp); 1653 } 1654 *TargetOS << ">"; 1655 } else if (BBAddrMapLabelAvailable) { 1656 *TargetOS << " <" << BBAddrMapLabels[Target].front() << ">"; 1657 } else if (LabelAvailable) { 1658 *TargetOS << " <" << AllLabels[Target] << ">"; 1659 } 1660 // By convention, each record in the comment stream should be 1661 // terminated. 1662 if (TargetOS == &CommentStream) 1663 *TargetOS << "\n"; 1664 } 1665 } 1666 } 1667 1668 assert(Ctx.getAsmInfo()); 1669 emitPostInstructionInfo(FOS, *Ctx.getAsmInfo(), *STI, 1670 CommentStream.str(), LVP); 1671 Comments.clear(); 1672 1673 // Hexagon does this in pretty printer 1674 if (Obj.getArch() != Triple::hexagon) { 1675 // Print relocation for instruction and data. 1676 while (RelCur != RelEnd) { 1677 uint64_t Offset = RelCur->getOffset() - RelAdjustment; 1678 // If this relocation is hidden, skip it. 1679 if (getHidden(*RelCur) || SectionAddr + Offset < StartAddress) { 1680 ++RelCur; 1681 continue; 1682 } 1683 1684 // Stop when RelCur's offset is past the disassembled 1685 // instruction/data. Note that it's possible the disassembled data 1686 // is not the complete data: we might see the relocation printed in 1687 // the middle of the data, but this matches the binutils objdump 1688 // output. 1689 if (Offset >= Index + Size) 1690 break; 1691 1692 // When --adjust-vma is used, update the address printed. 1693 if (RelCur->getSymbol() != Obj.symbol_end()) { 1694 Expected<section_iterator> SymSI = 1695 RelCur->getSymbol()->getSection(); 1696 if (SymSI && *SymSI != Obj.section_end() && 1697 shouldAdjustVA(**SymSI)) 1698 Offset += AdjustVMA; 1699 } 1700 1701 printRelocation(FOS, Obj.getFileName(), *RelCur, 1702 SectionAddr + Offset, Is64Bits); 1703 LVP.printAfterOtherLine(FOS, true); 1704 ++RelCur; 1705 } 1706 } 1707 1708 Index += Size; 1709 } 1710 } 1711 } 1712 StringSet<> MissingDisasmSymbolSet = 1713 set_difference(DisasmSymbolSet, FoundDisasmSymbolSet); 1714 for (StringRef Sym : MissingDisasmSymbolSet.keys()) 1715 reportWarning("failed to disassemble missing symbol " + Sym, FileName); 1716 } 1717 1718 static void disassembleObject(ObjectFile *Obj, bool InlineRelocs) { 1719 const Target *TheTarget = getTarget(Obj); 1720 1721 // Package up features to be passed to target/subtarget 1722 SubtargetFeatures Features = Obj->getFeatures(); 1723 if (!MAttrs.empty()) { 1724 for (unsigned I = 0; I != MAttrs.size(); ++I) 1725 Features.AddFeature(MAttrs[I]); 1726 } else if (MCPU.empty() && Obj->getArch() == llvm::Triple::aarch64) { 1727 Features.AddFeature("+all"); 1728 } 1729 1730 std::unique_ptr<const MCRegisterInfo> MRI( 1731 TheTarget->createMCRegInfo(TripleName)); 1732 if (!MRI) 1733 reportError(Obj->getFileName(), 1734 "no register info for target " + TripleName); 1735 1736 // Set up disassembler. 1737 MCTargetOptions MCOptions; 1738 std::unique_ptr<const MCAsmInfo> AsmInfo( 1739 TheTarget->createMCAsmInfo(*MRI, TripleName, MCOptions)); 1740 if (!AsmInfo) 1741 reportError(Obj->getFileName(), 1742 "no assembly info for target " + TripleName); 1743 1744 if (MCPU.empty()) 1745 MCPU = Obj->tryGetCPUName().value_or("").str(); 1746 1747 std::unique_ptr<const MCSubtargetInfo> STI( 1748 TheTarget->createMCSubtargetInfo(TripleName, MCPU, Features.getString())); 1749 if (!STI) 1750 reportError(Obj->getFileName(), 1751 "no subtarget info for target " + TripleName); 1752 std::unique_ptr<const MCInstrInfo> MII(TheTarget->createMCInstrInfo()); 1753 if (!MII) 1754 reportError(Obj->getFileName(), 1755 "no instruction info for target " + TripleName); 1756 MCContext Ctx(Triple(TripleName), AsmInfo.get(), MRI.get(), STI.get()); 1757 // FIXME: for now initialize MCObjectFileInfo with default values 1758 std::unique_ptr<MCObjectFileInfo> MOFI( 1759 TheTarget->createMCObjectFileInfo(Ctx, /*PIC=*/false)); 1760 Ctx.setObjectFileInfo(MOFI.get()); 1761 1762 std::unique_ptr<MCDisassembler> DisAsm( 1763 TheTarget->createMCDisassembler(*STI, Ctx)); 1764 if (!DisAsm) 1765 reportError(Obj->getFileName(), "no disassembler for target " + TripleName); 1766 1767 // If we have an ARM object file, we need a second disassembler, because 1768 // ARM CPUs have two different instruction sets: ARM mode, and Thumb mode. 1769 // We use mapping symbols to switch between the two assemblers, where 1770 // appropriate. 1771 std::unique_ptr<MCDisassembler> SecondaryDisAsm; 1772 std::unique_ptr<const MCSubtargetInfo> SecondarySTI; 1773 if (isArmElf(*Obj) && !STI->checkFeatures("+mclass")) { 1774 if (STI->checkFeatures("+thumb-mode")) 1775 Features.AddFeature("-thumb-mode"); 1776 else 1777 Features.AddFeature("+thumb-mode"); 1778 SecondarySTI.reset(TheTarget->createMCSubtargetInfo(TripleName, MCPU, 1779 Features.getString())); 1780 SecondaryDisAsm.reset(TheTarget->createMCDisassembler(*SecondarySTI, Ctx)); 1781 } 1782 1783 std::unique_ptr<const MCInstrAnalysis> MIA( 1784 TheTarget->createMCInstrAnalysis(MII.get())); 1785 1786 int AsmPrinterVariant = AsmInfo->getAssemblerDialect(); 1787 std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter( 1788 Triple(TripleName), AsmPrinterVariant, *AsmInfo, *MII, *MRI)); 1789 if (!IP) 1790 reportError(Obj->getFileName(), 1791 "no instruction printer for target " + TripleName); 1792 IP->setPrintImmHex(PrintImmHex); 1793 IP->setPrintBranchImmAsAddress(true); 1794 IP->setSymbolizeOperands(SymbolizeOperands); 1795 IP->setMCInstrAnalysis(MIA.get()); 1796 1797 PrettyPrinter &PIP = selectPrettyPrinter(Triple(TripleName)); 1798 SourcePrinter SP(Obj, TheTarget->getName()); 1799 1800 for (StringRef Opt : DisassemblerOptions) 1801 if (!IP->applyTargetSpecificCLOption(Opt)) 1802 reportError(Obj->getFileName(), 1803 "Unrecognized disassembler option: " + Opt); 1804 1805 disassembleObject(TheTarget, *Obj, Ctx, DisAsm.get(), SecondaryDisAsm.get(), 1806 MIA.get(), IP.get(), STI.get(), SecondarySTI.get(), PIP, SP, 1807 InlineRelocs); 1808 } 1809 1810 void objdump::printRelocations(const ObjectFile *Obj) { 1811 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : 1812 "%08" PRIx64; 1813 1814 // Build a mapping from relocation target to a vector of relocation 1815 // sections. Usually, there is an only one relocation section for 1816 // each relocated section. 1817 MapVector<SectionRef, std::vector<SectionRef>> SecToRelSec; 1818 uint64_t Ndx; 1819 for (const SectionRef &Section : ToolSectionFilter(*Obj, &Ndx)) { 1820 if (Obj->isELF() && (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC)) 1821 continue; 1822 if (Section.relocation_begin() == Section.relocation_end()) 1823 continue; 1824 Expected<section_iterator> SecOrErr = Section.getRelocatedSection(); 1825 if (!SecOrErr) 1826 reportError(Obj->getFileName(), 1827 "section (" + Twine(Ndx) + 1828 "): unable to get a relocation target: " + 1829 toString(SecOrErr.takeError())); 1830 SecToRelSec[**SecOrErr].push_back(Section); 1831 } 1832 1833 for (std::pair<SectionRef, std::vector<SectionRef>> &P : SecToRelSec) { 1834 StringRef SecName = unwrapOrError(P.first.getName(), Obj->getFileName()); 1835 outs() << "\nRELOCATION RECORDS FOR [" << SecName << "]:\n"; 1836 uint32_t OffsetPadding = (Obj->getBytesInAddress() > 4 ? 16 : 8); 1837 uint32_t TypePadding = 24; 1838 outs() << left_justify("OFFSET", OffsetPadding) << " " 1839 << left_justify("TYPE", TypePadding) << " " 1840 << "VALUE\n"; 1841 1842 for (SectionRef Section : P.second) { 1843 for (const RelocationRef &Reloc : Section.relocations()) { 1844 uint64_t Address = Reloc.getOffset(); 1845 SmallString<32> RelocName; 1846 SmallString<32> ValueStr; 1847 if (Address < StartAddress || Address > StopAddress || getHidden(Reloc)) 1848 continue; 1849 Reloc.getTypeName(RelocName); 1850 if (Error E = getRelocationValueString(Reloc, ValueStr)) 1851 reportError(std::move(E), Obj->getFileName()); 1852 1853 outs() << format(Fmt.data(), Address) << " " 1854 << left_justify(RelocName, TypePadding) << " " << ValueStr 1855 << "\n"; 1856 } 1857 } 1858 } 1859 } 1860 1861 void objdump::printDynamicRelocations(const ObjectFile *Obj) { 1862 // For the moment, this option is for ELF only 1863 if (!Obj->isELF()) 1864 return; 1865 1866 const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj); 1867 if (!Elf || !any_of(Elf->sections(), [](const ELFSectionRef Sec) { 1868 return Sec.getType() == ELF::SHT_DYNAMIC; 1869 })) { 1870 reportError(Obj->getFileName(), "not a dynamic object"); 1871 return; 1872 } 1873 1874 std::vector<SectionRef> DynRelSec = Obj->dynamic_relocation_sections(); 1875 if (DynRelSec.empty()) 1876 return; 1877 1878 outs() << "\nDYNAMIC RELOCATION RECORDS\n"; 1879 const uint32_t OffsetPadding = (Obj->getBytesInAddress() > 4 ? 16 : 8); 1880 const uint32_t TypePadding = 24; 1881 outs() << left_justify("OFFSET", OffsetPadding) << ' ' 1882 << left_justify("TYPE", TypePadding) << " VALUE\n"; 1883 1884 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 1885 for (const SectionRef &Section : DynRelSec) 1886 for (const RelocationRef &Reloc : Section.relocations()) { 1887 uint64_t Address = Reloc.getOffset(); 1888 SmallString<32> RelocName; 1889 SmallString<32> ValueStr; 1890 Reloc.getTypeName(RelocName); 1891 if (Error E = getRelocationValueString(Reloc, ValueStr)) 1892 reportError(std::move(E), Obj->getFileName()); 1893 outs() << format(Fmt.data(), Address) << ' ' 1894 << left_justify(RelocName, TypePadding) << ' ' << ValueStr << '\n'; 1895 } 1896 } 1897 1898 // Returns true if we need to show LMA column when dumping section headers. We 1899 // show it only when the platform is ELF and either we have at least one section 1900 // whose VMA and LMA are different and/or when --show-lma flag is used. 1901 static bool shouldDisplayLMA(const ObjectFile &Obj) { 1902 if (!Obj.isELF()) 1903 return false; 1904 for (const SectionRef &S : ToolSectionFilter(Obj)) 1905 if (S.getAddress() != getELFSectionLMA(S)) 1906 return true; 1907 return ShowLMA; 1908 } 1909 1910 static size_t getMaxSectionNameWidth(const ObjectFile &Obj) { 1911 // Default column width for names is 13 even if no names are that long. 1912 size_t MaxWidth = 13; 1913 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 1914 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 1915 MaxWidth = std::max(MaxWidth, Name.size()); 1916 } 1917 return MaxWidth; 1918 } 1919 1920 void objdump::printSectionHeaders(ObjectFile &Obj) { 1921 size_t NameWidth = getMaxSectionNameWidth(Obj); 1922 size_t AddressWidth = 2 * Obj.getBytesInAddress(); 1923 bool HasLMAColumn = shouldDisplayLMA(Obj); 1924 outs() << "\nSections:\n"; 1925 if (HasLMAColumn) 1926 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 1927 << left_justify("VMA", AddressWidth) << " " 1928 << left_justify("LMA", AddressWidth) << " Type\n"; 1929 else 1930 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 1931 << left_justify("VMA", AddressWidth) << " Type\n"; 1932 1933 if (Obj.isELF() && Obj.sections().empty()) 1934 createFakeELFSections(Obj); 1935 1936 uint64_t Idx; 1937 for (const SectionRef &Section : ToolSectionFilter(Obj, &Idx)) { 1938 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 1939 uint64_t VMA = Section.getAddress(); 1940 if (shouldAdjustVA(Section)) 1941 VMA += AdjustVMA; 1942 1943 uint64_t Size = Section.getSize(); 1944 1945 std::string Type = Section.isText() ? "TEXT" : ""; 1946 if (Section.isData()) 1947 Type += Type.empty() ? "DATA" : ", DATA"; 1948 if (Section.isBSS()) 1949 Type += Type.empty() ? "BSS" : ", BSS"; 1950 if (Section.isDebugSection()) 1951 Type += Type.empty() ? "DEBUG" : ", DEBUG"; 1952 1953 if (HasLMAColumn) 1954 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 1955 Name.str().c_str(), Size) 1956 << format_hex_no_prefix(VMA, AddressWidth) << " " 1957 << format_hex_no_prefix(getELFSectionLMA(Section), AddressWidth) 1958 << " " << Type << "\n"; 1959 else 1960 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 1961 Name.str().c_str(), Size) 1962 << format_hex_no_prefix(VMA, AddressWidth) << " " << Type << "\n"; 1963 } 1964 } 1965 1966 void objdump::printSectionContents(const ObjectFile *Obj) { 1967 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj); 1968 1969 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1970 StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName()); 1971 uint64_t BaseAddr = Section.getAddress(); 1972 uint64_t Size = Section.getSize(); 1973 if (!Size) 1974 continue; 1975 1976 outs() << "Contents of section "; 1977 StringRef SegmentName = getSegmentName(MachO, Section); 1978 if (!SegmentName.empty()) 1979 outs() << SegmentName << ","; 1980 outs() << Name << ":\n"; 1981 if (Section.isBSS()) { 1982 outs() << format("<skipping contents of bss section at [%04" PRIx64 1983 ", %04" PRIx64 ")>\n", 1984 BaseAddr, BaseAddr + Size); 1985 continue; 1986 } 1987 1988 StringRef Contents = unwrapOrError(Section.getContents(), Obj->getFileName()); 1989 1990 // Dump out the content as hex and printable ascii characters. 1991 for (std::size_t Addr = 0, End = Contents.size(); Addr < End; Addr += 16) { 1992 outs() << format(" %04" PRIx64 " ", BaseAddr + Addr); 1993 // Dump line of hex. 1994 for (std::size_t I = 0; I < 16; ++I) { 1995 if (I != 0 && I % 4 == 0) 1996 outs() << ' '; 1997 if (Addr + I < End) 1998 outs() << hexdigit((Contents[Addr + I] >> 4) & 0xF, true) 1999 << hexdigit(Contents[Addr + I] & 0xF, true); 2000 else 2001 outs() << " "; 2002 } 2003 // Print ascii. 2004 outs() << " "; 2005 for (std::size_t I = 0; I < 16 && Addr + I < End; ++I) { 2006 if (isPrint(static_cast<unsigned char>(Contents[Addr + I]) & 0xFF)) 2007 outs() << Contents[Addr + I]; 2008 else 2009 outs() << "."; 2010 } 2011 outs() << "\n"; 2012 } 2013 } 2014 } 2015 2016 void objdump::printSymbolTable(const ObjectFile &O, StringRef ArchiveName, 2017 StringRef ArchitectureName, bool DumpDynamic) { 2018 if (O.isCOFF() && !DumpDynamic) { 2019 outs() << "\nSYMBOL TABLE:\n"; 2020 printCOFFSymbolTable(cast<const COFFObjectFile>(O)); 2021 return; 2022 } 2023 2024 const StringRef FileName = O.getFileName(); 2025 2026 if (!DumpDynamic) { 2027 outs() << "\nSYMBOL TABLE:\n"; 2028 for (auto I = O.symbol_begin(); I != O.symbol_end(); ++I) 2029 printSymbol(O, *I, {}, FileName, ArchiveName, ArchitectureName, 2030 DumpDynamic); 2031 return; 2032 } 2033 2034 outs() << "\nDYNAMIC SYMBOL TABLE:\n"; 2035 if (!O.isELF()) { 2036 reportWarning( 2037 "this operation is not currently supported for this file format", 2038 FileName); 2039 return; 2040 } 2041 2042 const ELFObjectFileBase *ELF = cast<const ELFObjectFileBase>(&O); 2043 auto Symbols = ELF->getDynamicSymbolIterators(); 2044 Expected<std::vector<VersionEntry>> SymbolVersionsOrErr = 2045 ELF->readDynsymVersions(); 2046 if (!SymbolVersionsOrErr) { 2047 reportWarning(toString(SymbolVersionsOrErr.takeError()), FileName); 2048 SymbolVersionsOrErr = std::vector<VersionEntry>(); 2049 (void)!SymbolVersionsOrErr; 2050 } 2051 for (auto &Sym : Symbols) 2052 printSymbol(O, Sym, *SymbolVersionsOrErr, FileName, ArchiveName, 2053 ArchitectureName, DumpDynamic); 2054 } 2055 2056 void objdump::printSymbol(const ObjectFile &O, const SymbolRef &Symbol, 2057 ArrayRef<VersionEntry> SymbolVersions, 2058 StringRef FileName, StringRef ArchiveName, 2059 StringRef ArchitectureName, bool DumpDynamic) { 2060 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&O); 2061 uint64_t Address = unwrapOrError(Symbol.getAddress(), FileName, ArchiveName, 2062 ArchitectureName); 2063 if ((Address < StartAddress) || (Address > StopAddress)) 2064 return; 2065 SymbolRef::Type Type = 2066 unwrapOrError(Symbol.getType(), FileName, ArchiveName, ArchitectureName); 2067 uint32_t Flags = 2068 unwrapOrError(Symbol.getFlags(), FileName, ArchiveName, ArchitectureName); 2069 2070 // Don't ask a Mach-O STAB symbol for its section unless you know that 2071 // STAB symbol's section field refers to a valid section index. Otherwise 2072 // the symbol may error trying to load a section that does not exist. 2073 bool IsSTAB = false; 2074 if (MachO) { 2075 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 2076 uint8_t NType = 2077 (MachO->is64Bit() ? MachO->getSymbol64TableEntry(SymDRI).n_type 2078 : MachO->getSymbolTableEntry(SymDRI).n_type); 2079 if (NType & MachO::N_STAB) 2080 IsSTAB = true; 2081 } 2082 section_iterator Section = IsSTAB 2083 ? O.section_end() 2084 : unwrapOrError(Symbol.getSection(), FileName, 2085 ArchiveName, ArchitectureName); 2086 2087 StringRef Name; 2088 if (Type == SymbolRef::ST_Debug && Section != O.section_end()) { 2089 if (Expected<StringRef> NameOrErr = Section->getName()) 2090 Name = *NameOrErr; 2091 else 2092 consumeError(NameOrErr.takeError()); 2093 2094 } else { 2095 Name = unwrapOrError(Symbol.getName(), FileName, ArchiveName, 2096 ArchitectureName); 2097 } 2098 2099 bool Global = Flags & SymbolRef::SF_Global; 2100 bool Weak = Flags & SymbolRef::SF_Weak; 2101 bool Absolute = Flags & SymbolRef::SF_Absolute; 2102 bool Common = Flags & SymbolRef::SF_Common; 2103 bool Hidden = Flags & SymbolRef::SF_Hidden; 2104 2105 char GlobLoc = ' '; 2106 if ((Section != O.section_end() || Absolute) && !Weak) 2107 GlobLoc = Global ? 'g' : 'l'; 2108 char IFunc = ' '; 2109 if (O.isELF()) { 2110 if (ELFSymbolRef(Symbol).getELFType() == ELF::STT_GNU_IFUNC) 2111 IFunc = 'i'; 2112 if (ELFSymbolRef(Symbol).getBinding() == ELF::STB_GNU_UNIQUE) 2113 GlobLoc = 'u'; 2114 } 2115 2116 char Debug = ' '; 2117 if (DumpDynamic) 2118 Debug = 'D'; 2119 else if (Type == SymbolRef::ST_Debug || Type == SymbolRef::ST_File) 2120 Debug = 'd'; 2121 2122 char FileFunc = ' '; 2123 if (Type == SymbolRef::ST_File) 2124 FileFunc = 'f'; 2125 else if (Type == SymbolRef::ST_Function) 2126 FileFunc = 'F'; 2127 else if (Type == SymbolRef::ST_Data) 2128 FileFunc = 'O'; 2129 2130 const char *Fmt = O.getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2131 2132 outs() << format(Fmt, Address) << " " 2133 << GlobLoc // Local -> 'l', Global -> 'g', Neither -> ' ' 2134 << (Weak ? 'w' : ' ') // Weak? 2135 << ' ' // Constructor. Not supported yet. 2136 << ' ' // Warning. Not supported yet. 2137 << IFunc // Indirect reference to another symbol. 2138 << Debug // Debugging (d) or dynamic (D) symbol. 2139 << FileFunc // Name of function (F), file (f) or object (O). 2140 << ' '; 2141 if (Absolute) { 2142 outs() << "*ABS*"; 2143 } else if (Common) { 2144 outs() << "*COM*"; 2145 } else if (Section == O.section_end()) { 2146 if (O.isXCOFF()) { 2147 XCOFFSymbolRef XCOFFSym = cast<const XCOFFObjectFile>(O).toSymbolRef( 2148 Symbol.getRawDataRefImpl()); 2149 if (XCOFF::N_DEBUG == XCOFFSym.getSectionNumber()) 2150 outs() << "*DEBUG*"; 2151 else 2152 outs() << "*UND*"; 2153 } else 2154 outs() << "*UND*"; 2155 } else { 2156 StringRef SegmentName = getSegmentName(MachO, *Section); 2157 if (!SegmentName.empty()) 2158 outs() << SegmentName << ","; 2159 StringRef SectionName = unwrapOrError(Section->getName(), FileName); 2160 outs() << SectionName; 2161 if (O.isXCOFF()) { 2162 Optional<SymbolRef> SymRef = 2163 getXCOFFSymbolContainingSymbolRef(cast<XCOFFObjectFile>(O), Symbol); 2164 if (SymRef) { 2165 2166 Expected<StringRef> NameOrErr = SymRef->getName(); 2167 2168 if (NameOrErr) { 2169 outs() << " (csect:"; 2170 std::string SymName(NameOrErr.get()); 2171 2172 if (Demangle) 2173 SymName = demangle(SymName); 2174 2175 if (SymbolDescription) 2176 SymName = getXCOFFSymbolDescription( 2177 createSymbolInfo(O, SymRef.value()), SymName); 2178 2179 outs() << ' ' << SymName; 2180 outs() << ") "; 2181 } else 2182 reportWarning(toString(NameOrErr.takeError()), FileName); 2183 } 2184 } 2185 } 2186 2187 if (Common) 2188 outs() << '\t' << format(Fmt, static_cast<uint64_t>(Symbol.getAlignment())); 2189 else if (O.isXCOFF()) 2190 outs() << '\t' 2191 << format(Fmt, cast<XCOFFObjectFile>(O).getSymbolSize( 2192 Symbol.getRawDataRefImpl())); 2193 else if (O.isELF()) 2194 outs() << '\t' << format(Fmt, ELFSymbolRef(Symbol).getSize()); 2195 2196 if (O.isELF()) { 2197 if (!SymbolVersions.empty()) { 2198 const VersionEntry &Ver = 2199 SymbolVersions[Symbol.getRawDataRefImpl().d.b - 1]; 2200 std::string Str; 2201 if (!Ver.Name.empty()) 2202 Str = Ver.IsVerDef ? ' ' + Ver.Name : '(' + Ver.Name + ')'; 2203 outs() << ' ' << left_justify(Str, 12); 2204 } 2205 2206 uint8_t Other = ELFSymbolRef(Symbol).getOther(); 2207 switch (Other) { 2208 case ELF::STV_DEFAULT: 2209 break; 2210 case ELF::STV_INTERNAL: 2211 outs() << " .internal"; 2212 break; 2213 case ELF::STV_HIDDEN: 2214 outs() << " .hidden"; 2215 break; 2216 case ELF::STV_PROTECTED: 2217 outs() << " .protected"; 2218 break; 2219 default: 2220 outs() << format(" 0x%02x", Other); 2221 break; 2222 } 2223 } else if (Hidden) { 2224 outs() << " .hidden"; 2225 } 2226 2227 std::string SymName(Name); 2228 if (Demangle) 2229 SymName = demangle(SymName); 2230 2231 if (O.isXCOFF() && SymbolDescription) 2232 SymName = getXCOFFSymbolDescription(createSymbolInfo(O, Symbol), SymName); 2233 2234 outs() << ' ' << SymName << '\n'; 2235 } 2236 2237 static void printUnwindInfo(const ObjectFile *O) { 2238 outs() << "Unwind info:\n\n"; 2239 2240 if (const COFFObjectFile *Coff = dyn_cast<COFFObjectFile>(O)) 2241 printCOFFUnwindInfo(Coff); 2242 else if (const MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(O)) 2243 printMachOUnwindInfo(MachO); 2244 else 2245 // TODO: Extract DWARF dump tool to objdump. 2246 WithColor::error(errs(), ToolName) 2247 << "This operation is only currently supported " 2248 "for COFF and MachO object files.\n"; 2249 } 2250 2251 /// Dump the raw contents of the __clangast section so the output can be piped 2252 /// into llvm-bcanalyzer. 2253 static void printRawClangAST(const ObjectFile *Obj) { 2254 if (outs().is_displayed()) { 2255 WithColor::error(errs(), ToolName) 2256 << "The -raw-clang-ast option will dump the raw binary contents of " 2257 "the clang ast section.\n" 2258 "Please redirect the output to a file or another program such as " 2259 "llvm-bcanalyzer.\n"; 2260 return; 2261 } 2262 2263 StringRef ClangASTSectionName("__clangast"); 2264 if (Obj->isCOFF()) { 2265 ClangASTSectionName = "clangast"; 2266 } 2267 2268 Optional<object::SectionRef> ClangASTSection; 2269 for (auto Sec : ToolSectionFilter(*Obj)) { 2270 StringRef Name; 2271 if (Expected<StringRef> NameOrErr = Sec.getName()) 2272 Name = *NameOrErr; 2273 else 2274 consumeError(NameOrErr.takeError()); 2275 2276 if (Name == ClangASTSectionName) { 2277 ClangASTSection = Sec; 2278 break; 2279 } 2280 } 2281 if (!ClangASTSection) 2282 return; 2283 2284 StringRef ClangASTContents = 2285 unwrapOrError(ClangASTSection.value().getContents(), Obj->getFileName()); 2286 outs().write(ClangASTContents.data(), ClangASTContents.size()); 2287 } 2288 2289 static void printFaultMaps(const ObjectFile *Obj) { 2290 StringRef FaultMapSectionName; 2291 2292 if (Obj->isELF()) { 2293 FaultMapSectionName = ".llvm_faultmaps"; 2294 } else if (Obj->isMachO()) { 2295 FaultMapSectionName = "__llvm_faultmaps"; 2296 } else { 2297 WithColor::error(errs(), ToolName) 2298 << "This operation is only currently supported " 2299 "for ELF and Mach-O executable files.\n"; 2300 return; 2301 } 2302 2303 Optional<object::SectionRef> FaultMapSection; 2304 2305 for (auto Sec : ToolSectionFilter(*Obj)) { 2306 StringRef Name; 2307 if (Expected<StringRef> NameOrErr = Sec.getName()) 2308 Name = *NameOrErr; 2309 else 2310 consumeError(NameOrErr.takeError()); 2311 2312 if (Name == FaultMapSectionName) { 2313 FaultMapSection = Sec; 2314 break; 2315 } 2316 } 2317 2318 outs() << "FaultMap table:\n"; 2319 2320 if (!FaultMapSection) { 2321 outs() << "<not found>\n"; 2322 return; 2323 } 2324 2325 StringRef FaultMapContents = 2326 unwrapOrError(FaultMapSection->getContents(), Obj->getFileName()); 2327 FaultMapParser FMP(FaultMapContents.bytes_begin(), 2328 FaultMapContents.bytes_end()); 2329 2330 outs() << FMP; 2331 } 2332 2333 static void printPrivateFileHeaders(const ObjectFile *O, bool OnlyFirst) { 2334 if (O->isELF()) { 2335 printELFFileHeader(O); 2336 printELFDynamicSection(O); 2337 printELFSymbolVersionInfo(O); 2338 return; 2339 } 2340 if (O->isCOFF()) 2341 return printCOFFFileHeader(cast<object::COFFObjectFile>(*O)); 2342 if (O->isWasm()) 2343 return printWasmFileHeader(O); 2344 if (O->isMachO()) { 2345 printMachOFileHeader(O); 2346 if (!OnlyFirst) 2347 printMachOLoadCommands(O); 2348 return; 2349 } 2350 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2351 } 2352 2353 static void printFileHeaders(const ObjectFile *O) { 2354 if (!O->isELF() && !O->isCOFF()) 2355 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2356 2357 Triple::ArchType AT = O->getArch(); 2358 outs() << "architecture: " << Triple::getArchTypeName(AT) << "\n"; 2359 uint64_t Address = unwrapOrError(O->getStartAddress(), O->getFileName()); 2360 2361 StringRef Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2362 outs() << "start address: " 2363 << "0x" << format(Fmt.data(), Address) << "\n"; 2364 } 2365 2366 static void printArchiveChild(StringRef Filename, const Archive::Child &C) { 2367 Expected<sys::fs::perms> ModeOrErr = C.getAccessMode(); 2368 if (!ModeOrErr) { 2369 WithColor::error(errs(), ToolName) << "ill-formed archive entry.\n"; 2370 consumeError(ModeOrErr.takeError()); 2371 return; 2372 } 2373 sys::fs::perms Mode = ModeOrErr.get(); 2374 outs() << ((Mode & sys::fs::owner_read) ? "r" : "-"); 2375 outs() << ((Mode & sys::fs::owner_write) ? "w" : "-"); 2376 outs() << ((Mode & sys::fs::owner_exe) ? "x" : "-"); 2377 outs() << ((Mode & sys::fs::group_read) ? "r" : "-"); 2378 outs() << ((Mode & sys::fs::group_write) ? "w" : "-"); 2379 outs() << ((Mode & sys::fs::group_exe) ? "x" : "-"); 2380 outs() << ((Mode & sys::fs::others_read) ? "r" : "-"); 2381 outs() << ((Mode & sys::fs::others_write) ? "w" : "-"); 2382 outs() << ((Mode & sys::fs::others_exe) ? "x" : "-"); 2383 2384 outs() << " "; 2385 2386 outs() << format("%d/%d %6" PRId64 " ", unwrapOrError(C.getUID(), Filename), 2387 unwrapOrError(C.getGID(), Filename), 2388 unwrapOrError(C.getRawSize(), Filename)); 2389 2390 StringRef RawLastModified = C.getRawLastModified(); 2391 unsigned Seconds; 2392 if (RawLastModified.getAsInteger(10, Seconds)) 2393 outs() << "(date: \"" << RawLastModified 2394 << "\" contains non-decimal chars) "; 2395 else { 2396 // Since ctime(3) returns a 26 character string of the form: 2397 // "Sun Sep 16 01:03:52 1973\n\0" 2398 // just print 24 characters. 2399 time_t t = Seconds; 2400 outs() << format("%.24s ", ctime(&t)); 2401 } 2402 2403 StringRef Name = ""; 2404 Expected<StringRef> NameOrErr = C.getName(); 2405 if (!NameOrErr) { 2406 consumeError(NameOrErr.takeError()); 2407 Name = unwrapOrError(C.getRawName(), Filename); 2408 } else { 2409 Name = NameOrErr.get(); 2410 } 2411 outs() << Name << "\n"; 2412 } 2413 2414 // For ELF only now. 2415 static bool shouldWarnForInvalidStartStopAddress(ObjectFile *Obj) { 2416 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj)) { 2417 if (Elf->getEType() != ELF::ET_REL) 2418 return true; 2419 } 2420 return false; 2421 } 2422 2423 static void checkForInvalidStartStopAddress(ObjectFile *Obj, 2424 uint64_t Start, uint64_t Stop) { 2425 if (!shouldWarnForInvalidStartStopAddress(Obj)) 2426 return; 2427 2428 for (const SectionRef &Section : Obj->sections()) 2429 if (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC) { 2430 uint64_t BaseAddr = Section.getAddress(); 2431 uint64_t Size = Section.getSize(); 2432 if ((Start < BaseAddr + Size) && Stop > BaseAddr) 2433 return; 2434 } 2435 2436 if (!HasStartAddressFlag) 2437 reportWarning("no section has address less than 0x" + 2438 Twine::utohexstr(Stop) + " specified by --stop-address", 2439 Obj->getFileName()); 2440 else if (!HasStopAddressFlag) 2441 reportWarning("no section has address greater than or equal to 0x" + 2442 Twine::utohexstr(Start) + " specified by --start-address", 2443 Obj->getFileName()); 2444 else 2445 reportWarning("no section overlaps the range [0x" + 2446 Twine::utohexstr(Start) + ",0x" + Twine::utohexstr(Stop) + 2447 ") specified by --start-address/--stop-address", 2448 Obj->getFileName()); 2449 } 2450 2451 static void dumpObject(ObjectFile *O, const Archive *A = nullptr, 2452 const Archive::Child *C = nullptr) { 2453 // Avoid other output when using a raw option. 2454 if (!RawClangAST) { 2455 outs() << '\n'; 2456 if (A) 2457 outs() << A->getFileName() << "(" << O->getFileName() << ")"; 2458 else 2459 outs() << O->getFileName(); 2460 outs() << ":\tfile format " << O->getFileFormatName().lower() << "\n"; 2461 } 2462 2463 if (HasStartAddressFlag || HasStopAddressFlag) 2464 checkForInvalidStartStopAddress(O, StartAddress, StopAddress); 2465 2466 // Note: the order here matches GNU objdump for compatability. 2467 StringRef ArchiveName = A ? A->getFileName() : ""; 2468 if (ArchiveHeaders && !MachOOpt && C) 2469 printArchiveChild(ArchiveName, *C); 2470 if (FileHeaders) 2471 printFileHeaders(O); 2472 if (PrivateHeaders || FirstPrivateHeader) 2473 printPrivateFileHeaders(O, FirstPrivateHeader); 2474 if (SectionHeaders) 2475 printSectionHeaders(*O); 2476 if (SymbolTable) 2477 printSymbolTable(*O, ArchiveName); 2478 if (DynamicSymbolTable) 2479 printSymbolTable(*O, ArchiveName, /*ArchitectureName=*/"", 2480 /*DumpDynamic=*/true); 2481 if (DwarfDumpType != DIDT_Null) { 2482 std::unique_ptr<DIContext> DICtx = DWARFContext::create(*O); 2483 // Dump the complete DWARF structure. 2484 DIDumpOptions DumpOpts; 2485 DumpOpts.DumpType = DwarfDumpType; 2486 DICtx->dump(outs(), DumpOpts); 2487 } 2488 if (Relocations && !Disassemble) 2489 printRelocations(O); 2490 if (DynamicRelocations) 2491 printDynamicRelocations(O); 2492 if (SectionContents) 2493 printSectionContents(O); 2494 if (Disassemble) 2495 disassembleObject(O, Relocations); 2496 if (UnwindInfo) 2497 printUnwindInfo(O); 2498 2499 // Mach-O specific options: 2500 if (ExportsTrie) 2501 printExportsTrie(O); 2502 if (Rebase) 2503 printRebaseTable(O); 2504 if (Bind) 2505 printBindTable(O); 2506 if (LazyBind) 2507 printLazyBindTable(O); 2508 if (WeakBind) 2509 printWeakBindTable(O); 2510 2511 // Other special sections: 2512 if (RawClangAST) 2513 printRawClangAST(O); 2514 if (FaultMapSection) 2515 printFaultMaps(O); 2516 if (Offloading) 2517 dumpOffloadBinary(*O); 2518 } 2519 2520 static void dumpObject(const COFFImportFile *I, const Archive *A, 2521 const Archive::Child *C = nullptr) { 2522 StringRef ArchiveName = A ? A->getFileName() : ""; 2523 2524 // Avoid other output when using a raw option. 2525 if (!RawClangAST) 2526 outs() << '\n' 2527 << ArchiveName << "(" << I->getFileName() << ")" 2528 << ":\tfile format COFF-import-file" 2529 << "\n\n"; 2530 2531 if (ArchiveHeaders && !MachOOpt && C) 2532 printArchiveChild(ArchiveName, *C); 2533 if (SymbolTable) 2534 printCOFFSymbolTable(*I); 2535 } 2536 2537 /// Dump each object file in \a a; 2538 static void dumpArchive(const Archive *A) { 2539 Error Err = Error::success(); 2540 unsigned I = -1; 2541 for (auto &C : A->children(Err)) { 2542 ++I; 2543 Expected<std::unique_ptr<Binary>> ChildOrErr = C.getAsBinary(); 2544 if (!ChildOrErr) { 2545 if (auto E = isNotObjectErrorInvalidFileType(ChildOrErr.takeError())) 2546 reportError(std::move(E), getFileNameForError(C, I), A->getFileName()); 2547 continue; 2548 } 2549 if (ObjectFile *O = dyn_cast<ObjectFile>(&*ChildOrErr.get())) 2550 dumpObject(O, A, &C); 2551 else if (COFFImportFile *I = dyn_cast<COFFImportFile>(&*ChildOrErr.get())) 2552 dumpObject(I, A, &C); 2553 else 2554 reportError(errorCodeToError(object_error::invalid_file_type), 2555 A->getFileName()); 2556 } 2557 if (Err) 2558 reportError(std::move(Err), A->getFileName()); 2559 } 2560 2561 /// Open file and figure out how to dump it. 2562 static void dumpInput(StringRef file) { 2563 // If we are using the Mach-O specific object file parser, then let it parse 2564 // the file and process the command line options. So the -arch flags can 2565 // be used to select specific slices, etc. 2566 if (MachOOpt) { 2567 parseInputMachO(file); 2568 return; 2569 } 2570 2571 // Attempt to open the binary. 2572 OwningBinary<Binary> OBinary = unwrapOrError(createBinary(file), file); 2573 Binary &Binary = *OBinary.getBinary(); 2574 2575 if (Archive *A = dyn_cast<Archive>(&Binary)) 2576 dumpArchive(A); 2577 else if (ObjectFile *O = dyn_cast<ObjectFile>(&Binary)) 2578 dumpObject(O); 2579 else if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(&Binary)) 2580 parseInputMachO(UB); 2581 else if (OffloadBinary *OB = dyn_cast<OffloadBinary>(&Binary)) 2582 dumpOffloadSections(*OB); 2583 else 2584 reportError(errorCodeToError(object_error::invalid_file_type), file); 2585 } 2586 2587 template <typename T> 2588 static void parseIntArg(const llvm::opt::InputArgList &InputArgs, int ID, 2589 T &Value) { 2590 if (const opt::Arg *A = InputArgs.getLastArg(ID)) { 2591 StringRef V(A->getValue()); 2592 if (!llvm::to_integer(V, Value, 0)) { 2593 reportCmdLineError(A->getSpelling() + 2594 ": expected a non-negative integer, but got '" + V + 2595 "'"); 2596 } 2597 } 2598 } 2599 2600 static void invalidArgValue(const opt::Arg *A) { 2601 reportCmdLineError("'" + StringRef(A->getValue()) + 2602 "' is not a valid value for '" + A->getSpelling() + "'"); 2603 } 2604 2605 static std::vector<std::string> 2606 commaSeparatedValues(const llvm::opt::InputArgList &InputArgs, int ID) { 2607 std::vector<std::string> Values; 2608 for (StringRef Value : InputArgs.getAllArgValues(ID)) { 2609 llvm::SmallVector<StringRef, 2> SplitValues; 2610 llvm::SplitString(Value, SplitValues, ","); 2611 for (StringRef SplitValue : SplitValues) 2612 Values.push_back(SplitValue.str()); 2613 } 2614 return Values; 2615 } 2616 2617 static void parseOtoolOptions(const llvm::opt::InputArgList &InputArgs) { 2618 MachOOpt = true; 2619 FullLeadingAddr = true; 2620 PrintImmHex = true; 2621 2622 ArchName = InputArgs.getLastArgValue(OTOOL_arch).str(); 2623 LinkOptHints = InputArgs.hasArg(OTOOL_C); 2624 if (InputArgs.hasArg(OTOOL_d)) 2625 FilterSections.push_back("__DATA,__data"); 2626 DylibId = InputArgs.hasArg(OTOOL_D); 2627 UniversalHeaders = InputArgs.hasArg(OTOOL_f); 2628 DataInCode = InputArgs.hasArg(OTOOL_G); 2629 FirstPrivateHeader = InputArgs.hasArg(OTOOL_h); 2630 IndirectSymbols = InputArgs.hasArg(OTOOL_I); 2631 ShowRawInsn = InputArgs.hasArg(OTOOL_j); 2632 PrivateHeaders = InputArgs.hasArg(OTOOL_l); 2633 DylibsUsed = InputArgs.hasArg(OTOOL_L); 2634 MCPU = InputArgs.getLastArgValue(OTOOL_mcpu_EQ).str(); 2635 ObjcMetaData = InputArgs.hasArg(OTOOL_o); 2636 DisSymName = InputArgs.getLastArgValue(OTOOL_p).str(); 2637 InfoPlist = InputArgs.hasArg(OTOOL_P); 2638 Relocations = InputArgs.hasArg(OTOOL_r); 2639 if (const Arg *A = InputArgs.getLastArg(OTOOL_s)) { 2640 auto Filter = (A->getValue(0) + StringRef(",") + A->getValue(1)).str(); 2641 FilterSections.push_back(Filter); 2642 } 2643 if (InputArgs.hasArg(OTOOL_t)) 2644 FilterSections.push_back("__TEXT,__text"); 2645 Verbose = InputArgs.hasArg(OTOOL_v) || InputArgs.hasArg(OTOOL_V) || 2646 InputArgs.hasArg(OTOOL_o); 2647 SymbolicOperands = InputArgs.hasArg(OTOOL_V); 2648 if (InputArgs.hasArg(OTOOL_x)) 2649 FilterSections.push_back(",__text"); 2650 LeadingAddr = LeadingHeaders = !InputArgs.hasArg(OTOOL_X); 2651 2652 InputFilenames = InputArgs.getAllArgValues(OTOOL_INPUT); 2653 if (InputFilenames.empty()) 2654 reportCmdLineError("no input file"); 2655 2656 for (const Arg *A : InputArgs) { 2657 const Option &O = A->getOption(); 2658 if (O.getGroup().isValid() && O.getGroup().getID() == OTOOL_grp_obsolete) { 2659 reportCmdLineWarning(O.getPrefixedName() + 2660 " is obsolete and not implemented"); 2661 } 2662 } 2663 } 2664 2665 static void parseObjdumpOptions(const llvm::opt::InputArgList &InputArgs) { 2666 parseIntArg(InputArgs, OBJDUMP_adjust_vma_EQ, AdjustVMA); 2667 AllHeaders = InputArgs.hasArg(OBJDUMP_all_headers); 2668 ArchName = InputArgs.getLastArgValue(OBJDUMP_arch_name_EQ).str(); 2669 ArchiveHeaders = InputArgs.hasArg(OBJDUMP_archive_headers); 2670 Demangle = InputArgs.hasArg(OBJDUMP_demangle); 2671 Disassemble = InputArgs.hasArg(OBJDUMP_disassemble); 2672 DisassembleAll = InputArgs.hasArg(OBJDUMP_disassemble_all); 2673 SymbolDescription = InputArgs.hasArg(OBJDUMP_symbol_description); 2674 DisassembleSymbols = 2675 commaSeparatedValues(InputArgs, OBJDUMP_disassemble_symbols_EQ); 2676 DisassembleZeroes = InputArgs.hasArg(OBJDUMP_disassemble_zeroes); 2677 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_dwarf_EQ)) { 2678 DwarfDumpType = StringSwitch<DIDumpType>(A->getValue()) 2679 .Case("frames", DIDT_DebugFrame) 2680 .Default(DIDT_Null); 2681 if (DwarfDumpType == DIDT_Null) 2682 invalidArgValue(A); 2683 } 2684 DynamicRelocations = InputArgs.hasArg(OBJDUMP_dynamic_reloc); 2685 FaultMapSection = InputArgs.hasArg(OBJDUMP_fault_map_section); 2686 Offloading = InputArgs.hasArg(OBJDUMP_offloading); 2687 FileHeaders = InputArgs.hasArg(OBJDUMP_file_headers); 2688 SectionContents = InputArgs.hasArg(OBJDUMP_full_contents); 2689 PrintLines = InputArgs.hasArg(OBJDUMP_line_numbers); 2690 InputFilenames = InputArgs.getAllArgValues(OBJDUMP_INPUT); 2691 MachOOpt = InputArgs.hasArg(OBJDUMP_macho); 2692 MCPU = InputArgs.getLastArgValue(OBJDUMP_mcpu_EQ).str(); 2693 MAttrs = commaSeparatedValues(InputArgs, OBJDUMP_mattr_EQ); 2694 ShowRawInsn = !InputArgs.hasArg(OBJDUMP_no_show_raw_insn); 2695 LeadingAddr = !InputArgs.hasArg(OBJDUMP_no_leading_addr); 2696 RawClangAST = InputArgs.hasArg(OBJDUMP_raw_clang_ast); 2697 Relocations = InputArgs.hasArg(OBJDUMP_reloc); 2698 PrintImmHex = 2699 InputArgs.hasFlag(OBJDUMP_print_imm_hex, OBJDUMP_no_print_imm_hex, false); 2700 PrivateHeaders = InputArgs.hasArg(OBJDUMP_private_headers); 2701 FilterSections = InputArgs.getAllArgValues(OBJDUMP_section_EQ); 2702 SectionHeaders = InputArgs.hasArg(OBJDUMP_section_headers); 2703 ShowLMA = InputArgs.hasArg(OBJDUMP_show_lma); 2704 PrintSource = InputArgs.hasArg(OBJDUMP_source); 2705 parseIntArg(InputArgs, OBJDUMP_start_address_EQ, StartAddress); 2706 HasStartAddressFlag = InputArgs.hasArg(OBJDUMP_start_address_EQ); 2707 parseIntArg(InputArgs, OBJDUMP_stop_address_EQ, StopAddress); 2708 HasStopAddressFlag = InputArgs.hasArg(OBJDUMP_stop_address_EQ); 2709 SymbolTable = InputArgs.hasArg(OBJDUMP_syms); 2710 SymbolizeOperands = InputArgs.hasArg(OBJDUMP_symbolize_operands); 2711 DynamicSymbolTable = InputArgs.hasArg(OBJDUMP_dynamic_syms); 2712 TripleName = InputArgs.getLastArgValue(OBJDUMP_triple_EQ).str(); 2713 UnwindInfo = InputArgs.hasArg(OBJDUMP_unwind_info); 2714 Wide = InputArgs.hasArg(OBJDUMP_wide); 2715 Prefix = InputArgs.getLastArgValue(OBJDUMP_prefix).str(); 2716 parseIntArg(InputArgs, OBJDUMP_prefix_strip, PrefixStrip); 2717 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_debug_vars_EQ)) { 2718 DbgVariables = StringSwitch<DebugVarsFormat>(A->getValue()) 2719 .Case("ascii", DVASCII) 2720 .Case("unicode", DVUnicode) 2721 .Default(DVInvalid); 2722 if (DbgVariables == DVInvalid) 2723 invalidArgValue(A); 2724 } 2725 parseIntArg(InputArgs, OBJDUMP_debug_vars_indent_EQ, DbgIndent); 2726 2727 parseMachOOptions(InputArgs); 2728 2729 // Parse -M (--disassembler-options) and deprecated 2730 // --x86-asm-syntax={att,intel}. 2731 // 2732 // Note, for x86, the asm dialect (AssemblerDialect) is initialized when the 2733 // MCAsmInfo is constructed. MCInstPrinter::applyTargetSpecificCLOption is 2734 // called too late. For now we have to use the internal cl::opt option. 2735 const char *AsmSyntax = nullptr; 2736 for (const auto *A : InputArgs.filtered(OBJDUMP_disassembler_options_EQ, 2737 OBJDUMP_x86_asm_syntax_att, 2738 OBJDUMP_x86_asm_syntax_intel)) { 2739 switch (A->getOption().getID()) { 2740 case OBJDUMP_x86_asm_syntax_att: 2741 AsmSyntax = "--x86-asm-syntax=att"; 2742 continue; 2743 case OBJDUMP_x86_asm_syntax_intel: 2744 AsmSyntax = "--x86-asm-syntax=intel"; 2745 continue; 2746 } 2747 2748 SmallVector<StringRef, 2> Values; 2749 llvm::SplitString(A->getValue(), Values, ","); 2750 for (StringRef V : Values) { 2751 if (V == "att") 2752 AsmSyntax = "--x86-asm-syntax=att"; 2753 else if (V == "intel") 2754 AsmSyntax = "--x86-asm-syntax=intel"; 2755 else 2756 DisassemblerOptions.push_back(V.str()); 2757 } 2758 } 2759 if (AsmSyntax) { 2760 const char *Argv[] = {"llvm-objdump", AsmSyntax}; 2761 llvm::cl::ParseCommandLineOptions(2, Argv); 2762 } 2763 2764 // objdump defaults to a.out if no filenames specified. 2765 if (InputFilenames.empty()) 2766 InputFilenames.push_back("a.out"); 2767 } 2768 2769 int main(int argc, char **argv) { 2770 using namespace llvm; 2771 InitLLVM X(argc, argv); 2772 2773 ToolName = argv[0]; 2774 std::unique_ptr<CommonOptTable> T; 2775 OptSpecifier Unknown, HelpFlag, HelpHiddenFlag, VersionFlag; 2776 2777 StringRef Stem = sys::path::stem(ToolName); 2778 auto Is = [=](StringRef Tool) { 2779 // We need to recognize the following filenames: 2780 // 2781 // llvm-objdump -> objdump 2782 // llvm-otool-10.exe -> otool 2783 // powerpc64-unknown-freebsd13-objdump -> objdump 2784 auto I = Stem.rfind_insensitive(Tool); 2785 return I != StringRef::npos && 2786 (I + Tool.size() == Stem.size() || !isAlnum(Stem[I + Tool.size()])); 2787 }; 2788 if (Is("otool")) { 2789 T = std::make_unique<OtoolOptTable>(); 2790 Unknown = OTOOL_UNKNOWN; 2791 HelpFlag = OTOOL_help; 2792 HelpHiddenFlag = OTOOL_help_hidden; 2793 VersionFlag = OTOOL_version; 2794 } else { 2795 T = std::make_unique<ObjdumpOptTable>(); 2796 Unknown = OBJDUMP_UNKNOWN; 2797 HelpFlag = OBJDUMP_help; 2798 HelpHiddenFlag = OBJDUMP_help_hidden; 2799 VersionFlag = OBJDUMP_version; 2800 } 2801 2802 BumpPtrAllocator A; 2803 StringSaver Saver(A); 2804 opt::InputArgList InputArgs = 2805 T->parseArgs(argc, argv, Unknown, Saver, 2806 [&](StringRef Msg) { reportCmdLineError(Msg); }); 2807 2808 if (InputArgs.size() == 0 || InputArgs.hasArg(HelpFlag)) { 2809 T->printHelp(ToolName); 2810 return 0; 2811 } 2812 if (InputArgs.hasArg(HelpHiddenFlag)) { 2813 T->printHelp(ToolName, /*ShowHidden=*/true); 2814 return 0; 2815 } 2816 2817 // Initialize targets and assembly printers/parsers. 2818 InitializeAllTargetInfos(); 2819 InitializeAllTargetMCs(); 2820 InitializeAllDisassemblers(); 2821 2822 if (InputArgs.hasArg(VersionFlag)) { 2823 cl::PrintVersionMessage(); 2824 if (!Is("otool")) { 2825 outs() << '\n'; 2826 TargetRegistry::printRegisteredTargetsForVersion(outs()); 2827 } 2828 return 0; 2829 } 2830 2831 if (Is("otool")) 2832 parseOtoolOptions(InputArgs); 2833 else 2834 parseObjdumpOptions(InputArgs); 2835 2836 if (StartAddress >= StopAddress) 2837 reportCmdLineError("start address should be less than stop address"); 2838 2839 // Removes trailing separators from prefix. 2840 while (!Prefix.empty() && sys::path::is_separator(Prefix.back())) 2841 Prefix.pop_back(); 2842 2843 if (AllHeaders) 2844 ArchiveHeaders = FileHeaders = PrivateHeaders = Relocations = 2845 SectionHeaders = SymbolTable = true; 2846 2847 if (DisassembleAll || PrintSource || PrintLines || 2848 !DisassembleSymbols.empty()) 2849 Disassemble = true; 2850 2851 if (!ArchiveHeaders && !Disassemble && DwarfDumpType == DIDT_Null && 2852 !DynamicRelocations && !FileHeaders && !PrivateHeaders && !RawClangAST && 2853 !Relocations && !SectionHeaders && !SectionContents && !SymbolTable && 2854 !DynamicSymbolTable && !UnwindInfo && !FaultMapSection && !Offloading && 2855 !(MachOOpt && (Bind || DataInCode || DyldInfo || DylibId || DylibsUsed || 2856 ExportsTrie || FirstPrivateHeader || FunctionStarts || 2857 IndirectSymbols || InfoPlist || LazyBind || LinkOptHints || 2858 ObjcMetaData || Rebase || Rpaths || UniversalHeaders || 2859 WeakBind || !FilterSections.empty()))) { 2860 T->printHelp(ToolName); 2861 return 2; 2862 } 2863 2864 DisasmSymbolSet.insert(DisassembleSymbols.begin(), DisassembleSymbols.end()); 2865 2866 llvm::for_each(InputFilenames, dumpInput); 2867 2868 warnOnNoMatchForSections(); 2869 2870 return EXIT_SUCCESS; 2871 } 2872