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/STLExtras.h" 29 #include "llvm/ADT/SetOperations.h" 30 #include "llvm/ADT/SmallSet.h" 31 #include "llvm/ADT/StringExtras.h" 32 #include "llvm/ADT/StringSet.h" 33 #include "llvm/ADT/Twine.h" 34 #include "llvm/DebugInfo/DWARF/DWARFContext.h" 35 #include "llvm/DebugInfo/Symbolize/SymbolizableModule.h" 36 #include "llvm/DebugInfo/Symbolize/Symbolize.h" 37 #include "llvm/Debuginfod/BuildIDFetcher.h" 38 #include "llvm/Debuginfod/Debuginfod.h" 39 #include "llvm/Debuginfod/HTTPClient.h" 40 #include "llvm/Demangle/Demangle.h" 41 #include "llvm/MC/MCAsmInfo.h" 42 #include "llvm/MC/MCContext.h" 43 #include "llvm/MC/MCDisassembler/MCDisassembler.h" 44 #include "llvm/MC/MCDisassembler/MCRelocationInfo.h" 45 #include "llvm/MC/MCInst.h" 46 #include "llvm/MC/MCInstPrinter.h" 47 #include "llvm/MC/MCInstrAnalysis.h" 48 #include "llvm/MC/MCInstrInfo.h" 49 #include "llvm/MC/MCObjectFileInfo.h" 50 #include "llvm/MC/MCRegisterInfo.h" 51 #include "llvm/MC/MCTargetOptions.h" 52 #include "llvm/MC/TargetRegistry.h" 53 #include "llvm/Object/Archive.h" 54 #include "llvm/Object/BuildID.h" 55 #include "llvm/Object/COFF.h" 56 #include "llvm/Object/COFFImportFile.h" 57 #include "llvm/Object/ELFObjectFile.h" 58 #include "llvm/Object/ELFTypes.h" 59 #include "llvm/Object/FaultMapParser.h" 60 #include "llvm/Object/MachO.h" 61 #include "llvm/Object/MachOUniversal.h" 62 #include "llvm/Object/ObjectFile.h" 63 #include "llvm/Object/OffloadBinary.h" 64 #include "llvm/Object/Wasm.h" 65 #include "llvm/Option/Arg.h" 66 #include "llvm/Option/ArgList.h" 67 #include "llvm/Option/Option.h" 68 #include "llvm/Support/Casting.h" 69 #include "llvm/Support/Debug.h" 70 #include "llvm/Support/Errc.h" 71 #include "llvm/Support/FileSystem.h" 72 #include "llvm/Support/Format.h" 73 #include "llvm/Support/FormatVariadic.h" 74 #include "llvm/Support/GraphWriter.h" 75 #include "llvm/Support/InitLLVM.h" 76 #include "llvm/Support/LLVMDriver.h" 77 #include "llvm/Support/MemoryBuffer.h" 78 #include "llvm/Support/SourceMgr.h" 79 #include "llvm/Support/StringSaver.h" 80 #include "llvm/Support/TargetSelect.h" 81 #include "llvm/Support/WithColor.h" 82 #include "llvm/Support/raw_ostream.h" 83 #include "llvm/TargetParser/Host.h" 84 #include "llvm/TargetParser/Triple.h" 85 #include <algorithm> 86 #include <cctype> 87 #include <cstring> 88 #include <optional> 89 #include <system_error> 90 #include <unordered_map> 91 #include <utility> 92 93 using namespace llvm; 94 using namespace llvm::object; 95 using namespace llvm::objdump; 96 using namespace llvm::opt; 97 98 namespace { 99 100 class CommonOptTable : public opt::GenericOptTable { 101 public: 102 CommonOptTable(ArrayRef<Info> OptionInfos, const char *Usage, 103 const char *Description) 104 : opt::GenericOptTable(OptionInfos), Usage(Usage), 105 Description(Description) { 106 setGroupedShortOptions(true); 107 } 108 109 void printHelp(StringRef Argv0, bool ShowHidden = false) const { 110 Argv0 = sys::path::filename(Argv0); 111 opt::GenericOptTable::printHelp(outs(), (Argv0 + Usage).str().c_str(), 112 Description, ShowHidden, ShowHidden); 113 // TODO Replace this with OptTable API once it adds extrahelp support. 114 outs() << "\nPass @FILE as argument to read options from FILE.\n"; 115 } 116 117 private: 118 const char *Usage; 119 const char *Description; 120 }; 121 122 // ObjdumpOptID is in ObjdumpOptID.h 123 namespace objdump_opt { 124 #define PREFIX(NAME, VALUE) \ 125 static constexpr StringLiteral NAME##_init[] = VALUE; \ 126 static constexpr ArrayRef<StringLiteral> NAME(NAME##_init, \ 127 std::size(NAME##_init) - 1); 128 #include "ObjdumpOpts.inc" 129 #undef PREFIX 130 131 static constexpr opt::OptTable::Info ObjdumpInfoTable[] = { 132 #define OPTION(...) \ 133 LLVM_CONSTRUCT_OPT_INFO_WITH_ID_PREFIX(OBJDUMP_, __VA_ARGS__), 134 #include "ObjdumpOpts.inc" 135 #undef OPTION 136 }; 137 } // namespace objdump_opt 138 139 class ObjdumpOptTable : public CommonOptTable { 140 public: 141 ObjdumpOptTable() 142 : CommonOptTable(objdump_opt::ObjdumpInfoTable, 143 " [options] <input object files>", 144 "llvm object file dumper") {} 145 }; 146 147 enum OtoolOptID { 148 OTOOL_INVALID = 0, // This is not an option ID. 149 #define OPTION(...) LLVM_MAKE_OPT_ID_WITH_ID_PREFIX(OTOOL_, __VA_ARGS__), 150 #include "OtoolOpts.inc" 151 #undef OPTION 152 }; 153 154 namespace otool { 155 #define PREFIX(NAME, VALUE) \ 156 static constexpr StringLiteral NAME##_init[] = VALUE; \ 157 static constexpr ArrayRef<StringLiteral> NAME(NAME##_init, \ 158 std::size(NAME##_init) - 1); 159 #include "OtoolOpts.inc" 160 #undef PREFIX 161 162 static constexpr opt::OptTable::Info OtoolInfoTable[] = { 163 #define OPTION(...) LLVM_CONSTRUCT_OPT_INFO_WITH_ID_PREFIX(OTOOL_, __VA_ARGS__), 164 #include "OtoolOpts.inc" 165 #undef OPTION 166 }; 167 } // namespace otool 168 169 class OtoolOptTable : public CommonOptTable { 170 public: 171 OtoolOptTable() 172 : CommonOptTable(otool::OtoolInfoTable, " [option...] [file...]", 173 "Mach-O object file displaying tool") {} 174 }; 175 176 } // namespace 177 178 #define DEBUG_TYPE "objdump" 179 180 enum class ColorOutput { 181 Auto, 182 Enable, 183 Disable, 184 Invalid, 185 }; 186 187 static uint64_t AdjustVMA; 188 static bool AllHeaders; 189 static std::string ArchName; 190 bool objdump::ArchiveHeaders; 191 bool objdump::Demangle; 192 bool objdump::Disassemble; 193 bool objdump::DisassembleAll; 194 bool objdump::SymbolDescription; 195 bool objdump::TracebackTable; 196 static std::vector<std::string> DisassembleSymbols; 197 static bool DisassembleZeroes; 198 static std::vector<std::string> DisassemblerOptions; 199 static ColorOutput DisassemblyColor; 200 DIDumpType objdump::DwarfDumpType; 201 static bool DynamicRelocations; 202 static bool FaultMapSection; 203 static bool FileHeaders; 204 bool objdump::SectionContents; 205 static std::vector<std::string> InputFilenames; 206 bool objdump::PrintLines; 207 static bool MachOOpt; 208 std::string objdump::MCPU; 209 std::vector<std::string> objdump::MAttrs; 210 bool objdump::ShowRawInsn; 211 bool objdump::LeadingAddr; 212 static bool Offloading; 213 static bool RawClangAST; 214 bool objdump::Relocations; 215 bool objdump::PrintImmHex; 216 bool objdump::PrivateHeaders; 217 std::vector<std::string> objdump::FilterSections; 218 bool objdump::SectionHeaders; 219 static bool ShowAllSymbols; 220 static bool ShowLMA; 221 bool objdump::PrintSource; 222 223 static uint64_t StartAddress; 224 static bool HasStartAddressFlag; 225 static uint64_t StopAddress = UINT64_MAX; 226 static bool HasStopAddressFlag; 227 228 bool objdump::SymbolTable; 229 static bool SymbolizeOperands; 230 static bool DynamicSymbolTable; 231 std::string objdump::TripleName; 232 bool objdump::UnwindInfo; 233 static bool Wide; 234 std::string objdump::Prefix; 235 uint32_t objdump::PrefixStrip; 236 237 DebugVarsFormat objdump::DbgVariables = DVDisabled; 238 239 int objdump::DbgIndent = 52; 240 241 static StringSet<> DisasmSymbolSet; 242 StringSet<> objdump::FoundSectionSet; 243 static StringRef ToolName; 244 245 std::unique_ptr<BuildIDFetcher> BIDFetcher; 246 247 Dumper::Dumper(const object::ObjectFile &O) : O(O) { 248 WarningHandler = [this](const Twine &Msg) { 249 if (Warnings.insert(Msg.str()).second) 250 reportWarning(Msg, this->O.getFileName()); 251 return Error::success(); 252 }; 253 } 254 255 void Dumper::reportUniqueWarning(Error Err) { 256 reportUniqueWarning(toString(std::move(Err))); 257 } 258 259 void Dumper::reportUniqueWarning(const Twine &Msg) { 260 cantFail(WarningHandler(Msg)); 261 } 262 263 static Expected<std::unique_ptr<Dumper>> createDumper(const ObjectFile &Obj) { 264 if (const auto *O = dyn_cast<COFFObjectFile>(&Obj)) 265 return createCOFFDumper(*O); 266 if (const auto *O = dyn_cast<ELFObjectFileBase>(&Obj)) 267 return createELFDumper(*O); 268 if (const auto *O = dyn_cast<MachOObjectFile>(&Obj)) 269 return createMachODumper(*O); 270 if (const auto *O = dyn_cast<WasmObjectFile>(&Obj)) 271 return createWasmDumper(*O); 272 if (const auto *O = dyn_cast<XCOFFObjectFile>(&Obj)) 273 return createXCOFFDumper(*O); 274 275 return createStringError(errc::invalid_argument, 276 "unsupported object file format"); 277 } 278 279 namespace { 280 struct FilterResult { 281 // True if the section should not be skipped. 282 bool Keep; 283 284 // True if the index counter should be incremented, even if the section should 285 // be skipped. For example, sections may be skipped if they are not included 286 // in the --section flag, but we still want those to count toward the section 287 // count. 288 bool IncrementIndex; 289 }; 290 } // namespace 291 292 static FilterResult checkSectionFilter(object::SectionRef S) { 293 if (FilterSections.empty()) 294 return {/*Keep=*/true, /*IncrementIndex=*/true}; 295 296 Expected<StringRef> SecNameOrErr = S.getName(); 297 if (!SecNameOrErr) { 298 consumeError(SecNameOrErr.takeError()); 299 return {/*Keep=*/false, /*IncrementIndex=*/false}; 300 } 301 StringRef SecName = *SecNameOrErr; 302 303 // StringSet does not allow empty key so avoid adding sections with 304 // no name (such as the section with index 0) here. 305 if (!SecName.empty()) 306 FoundSectionSet.insert(SecName); 307 308 // Only show the section if it's in the FilterSections list, but always 309 // increment so the indexing is stable. 310 return {/*Keep=*/is_contained(FilterSections, SecName), 311 /*IncrementIndex=*/true}; 312 } 313 314 SectionFilter objdump::ToolSectionFilter(object::ObjectFile const &O, 315 uint64_t *Idx) { 316 // Start at UINT64_MAX so that the first index returned after an increment is 317 // zero (after the unsigned wrap). 318 if (Idx) 319 *Idx = UINT64_MAX; 320 return SectionFilter( 321 [Idx](object::SectionRef S) { 322 FilterResult Result = checkSectionFilter(S); 323 if (Idx != nullptr && Result.IncrementIndex) 324 *Idx += 1; 325 return Result.Keep; 326 }, 327 O); 328 } 329 330 std::string objdump::getFileNameForError(const object::Archive::Child &C, 331 unsigned Index) { 332 Expected<StringRef> NameOrErr = C.getName(); 333 if (NameOrErr) 334 return std::string(NameOrErr.get()); 335 // If we have an error getting the name then we print the index of the archive 336 // member. Since we are already in an error state, we just ignore this error. 337 consumeError(NameOrErr.takeError()); 338 return "<file index: " + std::to_string(Index) + ">"; 339 } 340 341 void objdump::reportWarning(const Twine &Message, StringRef File) { 342 // Output order between errs() and outs() matters especially for archive 343 // files where the output is per member object. 344 outs().flush(); 345 WithColor::warning(errs(), ToolName) 346 << "'" << File << "': " << Message << "\n"; 347 } 348 349 [[noreturn]] void objdump::reportError(StringRef File, const Twine &Message) { 350 outs().flush(); 351 WithColor::error(errs(), ToolName) << "'" << File << "': " << Message << "\n"; 352 exit(1); 353 } 354 355 [[noreturn]] void objdump::reportError(Error E, StringRef FileName, 356 StringRef ArchiveName, 357 StringRef ArchitectureName) { 358 assert(E); 359 outs().flush(); 360 WithColor::error(errs(), ToolName); 361 if (ArchiveName != "") 362 errs() << ArchiveName << "(" << FileName << ")"; 363 else 364 errs() << "'" << FileName << "'"; 365 if (!ArchitectureName.empty()) 366 errs() << " (for architecture " << ArchitectureName << ")"; 367 errs() << ": "; 368 logAllUnhandledErrors(std::move(E), errs()); 369 exit(1); 370 } 371 372 static void reportCmdLineWarning(const Twine &Message) { 373 WithColor::warning(errs(), ToolName) << Message << "\n"; 374 } 375 376 [[noreturn]] static void reportCmdLineError(const Twine &Message) { 377 WithColor::error(errs(), ToolName) << Message << "\n"; 378 exit(1); 379 } 380 381 static void warnOnNoMatchForSections() { 382 SetVector<StringRef> MissingSections; 383 for (StringRef S : FilterSections) { 384 if (FoundSectionSet.count(S)) 385 return; 386 // User may specify a unnamed section. Don't warn for it. 387 if (!S.empty()) 388 MissingSections.insert(S); 389 } 390 391 // Warn only if no section in FilterSections is matched. 392 for (StringRef S : MissingSections) 393 reportCmdLineWarning("section '" + S + 394 "' mentioned in a -j/--section option, but not " 395 "found in any input file"); 396 } 397 398 static const Target *getTarget(const ObjectFile *Obj) { 399 // Figure out the target triple. 400 Triple TheTriple("unknown-unknown-unknown"); 401 if (TripleName.empty()) { 402 TheTriple = Obj->makeTriple(); 403 } else { 404 TheTriple.setTriple(Triple::normalize(TripleName)); 405 auto Arch = Obj->getArch(); 406 if (Arch == Triple::arm || Arch == Triple::armeb) 407 Obj->setARMSubArch(TheTriple); 408 } 409 410 // Get the target specific parser. 411 std::string Error; 412 const Target *TheTarget = TargetRegistry::lookupTarget(ArchName, TheTriple, 413 Error); 414 if (!TheTarget) 415 reportError(Obj->getFileName(), "can't find target: " + Error); 416 417 // Update the triple name and return the found target. 418 TripleName = TheTriple.getTriple(); 419 return TheTarget; 420 } 421 422 bool objdump::isRelocAddressLess(RelocationRef A, RelocationRef B) { 423 return A.getOffset() < B.getOffset(); 424 } 425 426 static Error getRelocationValueString(const RelocationRef &Rel, 427 SmallVectorImpl<char> &Result) { 428 const ObjectFile *Obj = Rel.getObject(); 429 if (auto *ELF = dyn_cast<ELFObjectFileBase>(Obj)) 430 return getELFRelocationValueString(ELF, Rel, Result); 431 if (auto *COFF = dyn_cast<COFFObjectFile>(Obj)) 432 return getCOFFRelocationValueString(COFF, Rel, Result); 433 if (auto *Wasm = dyn_cast<WasmObjectFile>(Obj)) 434 return getWasmRelocationValueString(Wasm, Rel, Result); 435 if (auto *MachO = dyn_cast<MachOObjectFile>(Obj)) 436 return getMachORelocationValueString(MachO, Rel, Result); 437 if (auto *XCOFF = dyn_cast<XCOFFObjectFile>(Obj)) 438 return getXCOFFRelocationValueString(*XCOFF, Rel, Result); 439 llvm_unreachable("unknown object file format"); 440 } 441 442 /// Indicates whether this relocation should hidden when listing 443 /// relocations, usually because it is the trailing part of a multipart 444 /// relocation that will be printed as part of the leading relocation. 445 static bool getHidden(RelocationRef RelRef) { 446 auto *MachO = dyn_cast<MachOObjectFile>(RelRef.getObject()); 447 if (!MachO) 448 return false; 449 450 unsigned Arch = MachO->getArch(); 451 DataRefImpl Rel = RelRef.getRawDataRefImpl(); 452 uint64_t Type = MachO->getRelocationType(Rel); 453 454 // On arches that use the generic relocations, GENERIC_RELOC_PAIR 455 // is always hidden. 456 if (Arch == Triple::x86 || Arch == Triple::arm || Arch == Triple::ppc) 457 return Type == MachO::GENERIC_RELOC_PAIR; 458 459 if (Arch == Triple::x86_64) { 460 // On x86_64, X86_64_RELOC_UNSIGNED is hidden only when it follows 461 // an X86_64_RELOC_SUBTRACTOR. 462 if (Type == MachO::X86_64_RELOC_UNSIGNED && Rel.d.a > 0) { 463 DataRefImpl RelPrev = Rel; 464 RelPrev.d.a--; 465 uint64_t PrevType = MachO->getRelocationType(RelPrev); 466 if (PrevType == MachO::X86_64_RELOC_SUBTRACTOR) 467 return true; 468 } 469 } 470 471 return false; 472 } 473 474 /// Get the column at which we want to start printing the instruction 475 /// disassembly, taking into account anything which appears to the left of it. 476 unsigned objdump::getInstStartColumn(const MCSubtargetInfo &STI) { 477 return !ShowRawInsn ? 16 : STI.getTargetTriple().isX86() ? 40 : 24; 478 } 479 480 static void AlignToInstStartColumn(size_t Start, const MCSubtargetInfo &STI, 481 raw_ostream &OS) { 482 // The output of printInst starts with a tab. Print some spaces so that 483 // the tab has 1 column and advances to the target tab stop. 484 unsigned TabStop = getInstStartColumn(STI); 485 unsigned Column = OS.tell() - Start; 486 OS.indent(Column < TabStop - 1 ? TabStop - 1 - Column : 7 - Column % 8); 487 } 488 489 void objdump::printRawData(ArrayRef<uint8_t> Bytes, uint64_t Address, 490 formatted_raw_ostream &OS, 491 MCSubtargetInfo const &STI) { 492 size_t Start = OS.tell(); 493 if (LeadingAddr) 494 OS << format("%8" PRIx64 ":", Address); 495 if (ShowRawInsn) { 496 OS << ' '; 497 dumpBytes(Bytes, OS); 498 } 499 AlignToInstStartColumn(Start, STI, OS); 500 } 501 502 namespace { 503 504 static bool isAArch64Elf(const ObjectFile &Obj) { 505 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 506 return Elf && Elf->getEMachine() == ELF::EM_AARCH64; 507 } 508 509 static bool isArmElf(const ObjectFile &Obj) { 510 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 511 return Elf && Elf->getEMachine() == ELF::EM_ARM; 512 } 513 514 static bool isCSKYElf(const ObjectFile &Obj) { 515 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 516 return Elf && Elf->getEMachine() == ELF::EM_CSKY; 517 } 518 519 static bool hasMappingSymbols(const ObjectFile &Obj) { 520 return isArmElf(Obj) || isAArch64Elf(Obj) || isCSKYElf(Obj) ; 521 } 522 523 static void printRelocation(formatted_raw_ostream &OS, StringRef FileName, 524 const RelocationRef &Rel, uint64_t Address, 525 bool Is64Bits) { 526 StringRef Fmt = Is64Bits ? "%016" PRIx64 ": " : "%08" PRIx64 ": "; 527 SmallString<16> Name; 528 SmallString<32> Val; 529 Rel.getTypeName(Name); 530 if (Error E = getRelocationValueString(Rel, Val)) 531 reportError(std::move(E), FileName); 532 OS << (Is64Bits || !LeadingAddr ? "\t\t" : "\t\t\t"); 533 if (LeadingAddr) 534 OS << format(Fmt.data(), Address); 535 OS << Name << "\t" << Val; 536 } 537 538 class PrettyPrinter { 539 public: 540 virtual ~PrettyPrinter() = default; 541 virtual void 542 printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 543 object::SectionedAddress Address, formatted_raw_ostream &OS, 544 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 545 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 546 LiveVariablePrinter &LVP) { 547 if (SP && (PrintSource || PrintLines)) 548 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 549 LVP.printBetweenInsts(OS, false); 550 551 printRawData(Bytes, Address.Address, OS, STI); 552 553 if (MI) { 554 // See MCInstPrinter::printInst. On targets where a PC relative immediate 555 // is relative to the next instruction and the length of a MCInst is 556 // difficult to measure (x86), this is the address of the next 557 // instruction. 558 uint64_t Addr = 559 Address.Address + (STI.getTargetTriple().isX86() ? Bytes.size() : 0); 560 IP.printInst(MI, Addr, "", STI, OS); 561 } else 562 OS << "\t<unknown>"; 563 } 564 }; 565 PrettyPrinter PrettyPrinterInst; 566 567 class HexagonPrettyPrinter : public PrettyPrinter { 568 public: 569 void printLead(ArrayRef<uint8_t> Bytes, uint64_t Address, 570 formatted_raw_ostream &OS) { 571 uint32_t opcode = 572 (Bytes[3] << 24) | (Bytes[2] << 16) | (Bytes[1] << 8) | Bytes[0]; 573 if (LeadingAddr) 574 OS << format("%8" PRIx64 ":", Address); 575 if (ShowRawInsn) { 576 OS << "\t"; 577 dumpBytes(Bytes.slice(0, 4), OS); 578 OS << format("\t%08" PRIx32, opcode); 579 } 580 } 581 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 582 object::SectionedAddress Address, formatted_raw_ostream &OS, 583 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 584 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 585 LiveVariablePrinter &LVP) override { 586 if (SP && (PrintSource || PrintLines)) 587 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 588 if (!MI) { 589 printLead(Bytes, Address.Address, OS); 590 OS << " <unknown>"; 591 return; 592 } 593 std::string Buffer; 594 { 595 raw_string_ostream TempStream(Buffer); 596 IP.printInst(MI, Address.Address, "", STI, TempStream); 597 } 598 StringRef Contents(Buffer); 599 // Split off bundle attributes 600 auto PacketBundle = Contents.rsplit('\n'); 601 // Split off first instruction from the rest 602 auto HeadTail = PacketBundle.first.split('\n'); 603 auto Preamble = " { "; 604 auto Separator = ""; 605 606 // Hexagon's packets require relocations to be inline rather than 607 // clustered at the end of the packet. 608 std::vector<RelocationRef>::const_iterator RelCur = Rels->begin(); 609 std::vector<RelocationRef>::const_iterator RelEnd = Rels->end(); 610 auto PrintReloc = [&]() -> void { 611 while ((RelCur != RelEnd) && (RelCur->getOffset() <= Address.Address)) { 612 if (RelCur->getOffset() == Address.Address) { 613 printRelocation(OS, ObjectFilename, *RelCur, Address.Address, false); 614 return; 615 } 616 ++RelCur; 617 } 618 }; 619 620 while (!HeadTail.first.empty()) { 621 OS << Separator; 622 Separator = "\n"; 623 if (SP && (PrintSource || PrintLines)) 624 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 625 printLead(Bytes, Address.Address, OS); 626 OS << Preamble; 627 Preamble = " "; 628 StringRef Inst; 629 auto Duplex = HeadTail.first.split('\v'); 630 if (!Duplex.second.empty()) { 631 OS << Duplex.first; 632 OS << "; "; 633 Inst = Duplex.second; 634 } 635 else 636 Inst = HeadTail.first; 637 OS << Inst; 638 HeadTail = HeadTail.second.split('\n'); 639 if (HeadTail.first.empty()) 640 OS << " } " << PacketBundle.second; 641 PrintReloc(); 642 Bytes = Bytes.slice(4); 643 Address.Address += 4; 644 } 645 } 646 }; 647 HexagonPrettyPrinter HexagonPrettyPrinterInst; 648 649 class AMDGCNPrettyPrinter : public PrettyPrinter { 650 public: 651 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 652 object::SectionedAddress Address, formatted_raw_ostream &OS, 653 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 654 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 655 LiveVariablePrinter &LVP) override { 656 if (SP && (PrintSource || PrintLines)) 657 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 658 659 if (MI) { 660 SmallString<40> InstStr; 661 raw_svector_ostream IS(InstStr); 662 663 IP.printInst(MI, Address.Address, "", STI, IS); 664 665 OS << left_justify(IS.str(), 60); 666 } else { 667 // an unrecognized encoding - this is probably data so represent it 668 // using the .long directive, or .byte directive if fewer than 4 bytes 669 // remaining 670 if (Bytes.size() >= 4) { 671 OS << format("\t.long 0x%08" PRIx32 " ", 672 support::endian::read32<support::little>(Bytes.data())); 673 OS.indent(42); 674 } else { 675 OS << format("\t.byte 0x%02" PRIx8, Bytes[0]); 676 for (unsigned int i = 1; i < Bytes.size(); i++) 677 OS << format(", 0x%02" PRIx8, Bytes[i]); 678 OS.indent(55 - (6 * Bytes.size())); 679 } 680 } 681 682 OS << format("// %012" PRIX64 ":", Address.Address); 683 if (Bytes.size() >= 4) { 684 // D should be casted to uint32_t here as it is passed by format to 685 // snprintf as vararg. 686 for (uint32_t D : 687 ArrayRef(reinterpret_cast<const support::little32_t *>(Bytes.data()), 688 Bytes.size() / 4)) 689 OS << format(" %08" PRIX32, D); 690 } else { 691 for (unsigned char B : Bytes) 692 OS << format(" %02" PRIX8, B); 693 } 694 695 if (!Annot.empty()) 696 OS << " // " << Annot; 697 } 698 }; 699 AMDGCNPrettyPrinter AMDGCNPrettyPrinterInst; 700 701 class BPFPrettyPrinter : public PrettyPrinter { 702 public: 703 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 704 object::SectionedAddress Address, formatted_raw_ostream &OS, 705 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 706 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 707 LiveVariablePrinter &LVP) override { 708 if (SP && (PrintSource || PrintLines)) 709 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 710 if (LeadingAddr) 711 OS << format("%8" PRId64 ":", Address.Address / 8); 712 if (ShowRawInsn) { 713 OS << "\t"; 714 dumpBytes(Bytes, OS); 715 } 716 if (MI) 717 IP.printInst(MI, Address.Address, "", STI, OS); 718 else 719 OS << "\t<unknown>"; 720 } 721 }; 722 BPFPrettyPrinter BPFPrettyPrinterInst; 723 724 class ARMPrettyPrinter : public PrettyPrinter { 725 public: 726 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 727 object::SectionedAddress Address, formatted_raw_ostream &OS, 728 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 729 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 730 LiveVariablePrinter &LVP) override { 731 if (SP && (PrintSource || PrintLines)) 732 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 733 LVP.printBetweenInsts(OS, false); 734 735 size_t Start = OS.tell(); 736 if (LeadingAddr) 737 OS << format("%8" PRIx64 ":", Address.Address); 738 if (ShowRawInsn) { 739 size_t Pos = 0, End = Bytes.size(); 740 if (STI.checkFeatures("+thumb-mode")) { 741 for (; Pos + 2 <= End; Pos += 2) 742 OS << ' ' 743 << format_hex_no_prefix( 744 llvm::support::endian::read<uint16_t>( 745 Bytes.data() + Pos, InstructionEndianness), 746 4); 747 } else { 748 for (; Pos + 4 <= End; Pos += 4) 749 OS << ' ' 750 << format_hex_no_prefix( 751 llvm::support::endian::read<uint32_t>( 752 Bytes.data() + Pos, InstructionEndianness), 753 8); 754 } 755 if (Pos < End) { 756 OS << ' '; 757 dumpBytes(Bytes.slice(Pos), OS); 758 } 759 } 760 761 AlignToInstStartColumn(Start, STI, OS); 762 763 if (MI) { 764 IP.printInst(MI, Address.Address, "", STI, OS); 765 } else 766 OS << "\t<unknown>"; 767 } 768 769 void setInstructionEndianness(llvm::support::endianness Endianness) { 770 InstructionEndianness = Endianness; 771 } 772 773 private: 774 llvm::support::endianness InstructionEndianness = llvm::support::little; 775 }; 776 ARMPrettyPrinter ARMPrettyPrinterInst; 777 778 class AArch64PrettyPrinter : public PrettyPrinter { 779 public: 780 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 781 object::SectionedAddress Address, formatted_raw_ostream &OS, 782 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 783 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 784 LiveVariablePrinter &LVP) override { 785 if (SP && (PrintSource || PrintLines)) 786 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 787 LVP.printBetweenInsts(OS, false); 788 789 size_t Start = OS.tell(); 790 if (LeadingAddr) 791 OS << format("%8" PRIx64 ":", Address.Address); 792 if (ShowRawInsn) { 793 size_t Pos = 0, End = Bytes.size(); 794 for (; Pos + 4 <= End; Pos += 4) 795 OS << ' ' 796 << format_hex_no_prefix( 797 llvm::support::endian::read<uint32_t>(Bytes.data() + Pos, 798 llvm::support::little), 799 8); 800 if (Pos < End) { 801 OS << ' '; 802 dumpBytes(Bytes.slice(Pos), OS); 803 } 804 } 805 806 AlignToInstStartColumn(Start, STI, OS); 807 808 if (MI) { 809 IP.printInst(MI, Address.Address, "", STI, OS); 810 } else 811 OS << "\t<unknown>"; 812 } 813 }; 814 AArch64PrettyPrinter AArch64PrettyPrinterInst; 815 816 PrettyPrinter &selectPrettyPrinter(Triple const &Triple) { 817 switch(Triple.getArch()) { 818 default: 819 return PrettyPrinterInst; 820 case Triple::hexagon: 821 return HexagonPrettyPrinterInst; 822 case Triple::amdgcn: 823 return AMDGCNPrettyPrinterInst; 824 case Triple::bpfel: 825 case Triple::bpfeb: 826 return BPFPrettyPrinterInst; 827 case Triple::arm: 828 case Triple::armeb: 829 case Triple::thumb: 830 case Triple::thumbeb: 831 return ARMPrettyPrinterInst; 832 case Triple::aarch64: 833 case Triple::aarch64_be: 834 case Triple::aarch64_32: 835 return AArch64PrettyPrinterInst; 836 } 837 } 838 839 class DisassemblerTarget { 840 public: 841 const Target *TheTarget; 842 std::unique_ptr<const MCSubtargetInfo> SubtargetInfo; 843 std::shared_ptr<MCContext> Context; 844 std::unique_ptr<MCDisassembler> DisAsm; 845 std::shared_ptr<const MCInstrAnalysis> InstrAnalysis; 846 std::shared_ptr<MCInstPrinter> InstPrinter; 847 PrettyPrinter *Printer; 848 849 DisassemblerTarget(const Target *TheTarget, ObjectFile &Obj, 850 StringRef TripleName, StringRef MCPU, 851 SubtargetFeatures &Features); 852 DisassemblerTarget(DisassemblerTarget &Other, SubtargetFeatures &Features); 853 854 private: 855 MCTargetOptions Options; 856 std::shared_ptr<const MCRegisterInfo> RegisterInfo; 857 std::shared_ptr<const MCAsmInfo> AsmInfo; 858 std::shared_ptr<const MCInstrInfo> InstrInfo; 859 std::shared_ptr<MCObjectFileInfo> ObjectFileInfo; 860 }; 861 862 DisassemblerTarget::DisassemblerTarget(const Target *TheTarget, ObjectFile &Obj, 863 StringRef TripleName, StringRef MCPU, 864 SubtargetFeatures &Features) 865 : TheTarget(TheTarget), 866 Printer(&selectPrettyPrinter(Triple(TripleName))), 867 RegisterInfo(TheTarget->createMCRegInfo(TripleName)) { 868 if (!RegisterInfo) 869 reportError(Obj.getFileName(), "no register info for target " + TripleName); 870 871 // Set up disassembler. 872 AsmInfo.reset(TheTarget->createMCAsmInfo(*RegisterInfo, TripleName, Options)); 873 if (!AsmInfo) 874 reportError(Obj.getFileName(), "no assembly info for target " + TripleName); 875 876 SubtargetInfo.reset( 877 TheTarget->createMCSubtargetInfo(TripleName, MCPU, Features.getString())); 878 if (!SubtargetInfo) 879 reportError(Obj.getFileName(), 880 "no subtarget info for target " + TripleName); 881 InstrInfo.reset(TheTarget->createMCInstrInfo()); 882 if (!InstrInfo) 883 reportError(Obj.getFileName(), 884 "no instruction info for target " + TripleName); 885 Context = 886 std::make_shared<MCContext>(Triple(TripleName), AsmInfo.get(), 887 RegisterInfo.get(), SubtargetInfo.get()); 888 889 // FIXME: for now initialize MCObjectFileInfo with default values 890 ObjectFileInfo.reset( 891 TheTarget->createMCObjectFileInfo(*Context, /*PIC=*/false)); 892 Context->setObjectFileInfo(ObjectFileInfo.get()); 893 894 DisAsm.reset(TheTarget->createMCDisassembler(*SubtargetInfo, *Context)); 895 if (!DisAsm) 896 reportError(Obj.getFileName(), "no disassembler for target " + TripleName); 897 898 InstrAnalysis.reset(TheTarget->createMCInstrAnalysis(InstrInfo.get())); 899 900 int AsmPrinterVariant = AsmInfo->getAssemblerDialect(); 901 InstPrinter.reset(TheTarget->createMCInstPrinter(Triple(TripleName), 902 AsmPrinterVariant, *AsmInfo, 903 *InstrInfo, *RegisterInfo)); 904 if (!InstPrinter) 905 reportError(Obj.getFileName(), 906 "no instruction printer for target " + TripleName); 907 InstPrinter->setPrintImmHex(PrintImmHex); 908 InstPrinter->setPrintBranchImmAsAddress(true); 909 InstPrinter->setSymbolizeOperands(SymbolizeOperands); 910 InstPrinter->setMCInstrAnalysis(InstrAnalysis.get()); 911 912 switch (DisassemblyColor) { 913 case ColorOutput::Enable: 914 InstPrinter->setUseColor(true); 915 break; 916 case ColorOutput::Auto: 917 InstPrinter->setUseColor(outs().has_colors()); 918 break; 919 case ColorOutput::Disable: 920 case ColorOutput::Invalid: 921 InstPrinter->setUseColor(false); 922 break; 923 }; 924 } 925 926 DisassemblerTarget::DisassemblerTarget(DisassemblerTarget &Other, 927 SubtargetFeatures &Features) 928 : TheTarget(Other.TheTarget), 929 SubtargetInfo(TheTarget->createMCSubtargetInfo(TripleName, MCPU, 930 Features.getString())), 931 Context(Other.Context), 932 DisAsm(TheTarget->createMCDisassembler(*SubtargetInfo, *Context)), 933 InstrAnalysis(Other.InstrAnalysis), InstPrinter(Other.InstPrinter), 934 Printer(Other.Printer), RegisterInfo(Other.RegisterInfo), 935 AsmInfo(Other.AsmInfo), InstrInfo(Other.InstrInfo), 936 ObjectFileInfo(Other.ObjectFileInfo) {} 937 } // namespace 938 939 static uint8_t getElfSymbolType(const ObjectFile &Obj, const SymbolRef &Sym) { 940 assert(Obj.isELF()); 941 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 942 return unwrapOrError(Elf32LEObj->getSymbol(Sym.getRawDataRefImpl()), 943 Obj.getFileName()) 944 ->getType(); 945 if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 946 return unwrapOrError(Elf64LEObj->getSymbol(Sym.getRawDataRefImpl()), 947 Obj.getFileName()) 948 ->getType(); 949 if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 950 return unwrapOrError(Elf32BEObj->getSymbol(Sym.getRawDataRefImpl()), 951 Obj.getFileName()) 952 ->getType(); 953 if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 954 return unwrapOrError(Elf64BEObj->getSymbol(Sym.getRawDataRefImpl()), 955 Obj.getFileName()) 956 ->getType(); 957 llvm_unreachable("Unsupported binary format"); 958 } 959 960 template <class ELFT> 961 static void 962 addDynamicElfSymbols(const ELFObjectFile<ELFT> &Obj, 963 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 964 for (auto Symbol : Obj.getDynamicSymbolIterators()) { 965 uint8_t SymbolType = Symbol.getELFType(); 966 if (SymbolType == ELF::STT_SECTION) 967 continue; 968 969 uint64_t Address = unwrapOrError(Symbol.getAddress(), Obj.getFileName()); 970 // ELFSymbolRef::getAddress() returns size instead of value for common 971 // symbols which is not desirable for disassembly output. Overriding. 972 if (SymbolType == ELF::STT_COMMON) 973 Address = unwrapOrError(Obj.getSymbol(Symbol.getRawDataRefImpl()), 974 Obj.getFileName()) 975 ->st_value; 976 977 StringRef Name = unwrapOrError(Symbol.getName(), Obj.getFileName()); 978 if (Name.empty()) 979 continue; 980 981 section_iterator SecI = 982 unwrapOrError(Symbol.getSection(), Obj.getFileName()); 983 if (SecI == Obj.section_end()) 984 continue; 985 986 AllSymbols[*SecI].emplace_back(Address, Name, SymbolType); 987 } 988 } 989 990 static void 991 addDynamicElfSymbols(const ELFObjectFileBase &Obj, 992 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 993 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 994 addDynamicElfSymbols(*Elf32LEObj, AllSymbols); 995 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 996 addDynamicElfSymbols(*Elf64LEObj, AllSymbols); 997 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 998 addDynamicElfSymbols(*Elf32BEObj, AllSymbols); 999 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 1000 addDynamicElfSymbols(*Elf64BEObj, AllSymbols); 1001 else 1002 llvm_unreachable("Unsupported binary format"); 1003 } 1004 1005 static std::optional<SectionRef> getWasmCodeSection(const WasmObjectFile &Obj) { 1006 for (auto SecI : Obj.sections()) { 1007 const WasmSection &Section = Obj.getWasmSection(SecI); 1008 if (Section.Type == wasm::WASM_SEC_CODE) 1009 return SecI; 1010 } 1011 return std::nullopt; 1012 } 1013 1014 static void 1015 addMissingWasmCodeSymbols(const WasmObjectFile &Obj, 1016 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 1017 std::optional<SectionRef> Section = getWasmCodeSection(Obj); 1018 if (!Section) 1019 return; 1020 SectionSymbolsTy &Symbols = AllSymbols[*Section]; 1021 1022 std::set<uint64_t> SymbolAddresses; 1023 for (const auto &Sym : Symbols) 1024 SymbolAddresses.insert(Sym.Addr); 1025 1026 for (const wasm::WasmFunction &Function : Obj.functions()) { 1027 uint64_t Address = Function.CodeSectionOffset; 1028 // Only add fallback symbols for functions not already present in the symbol 1029 // table. 1030 if (SymbolAddresses.count(Address)) 1031 continue; 1032 // This function has no symbol, so it should have no SymbolName. 1033 assert(Function.SymbolName.empty()); 1034 // We use DebugName for the name, though it may be empty if there is no 1035 // "name" custom section, or that section is missing a name for this 1036 // function. 1037 StringRef Name = Function.DebugName; 1038 Symbols.emplace_back(Address, Name, ELF::STT_NOTYPE); 1039 } 1040 } 1041 1042 static void addPltEntries(const ObjectFile &Obj, 1043 std::map<SectionRef, SectionSymbolsTy> &AllSymbols, 1044 StringSaver &Saver) { 1045 auto *ElfObj = dyn_cast<ELFObjectFileBase>(&Obj); 1046 if (!ElfObj) 1047 return; 1048 DenseMap<StringRef, SectionRef> Sections; 1049 for (SectionRef Section : Obj.sections()) { 1050 Expected<StringRef> SecNameOrErr = Section.getName(); 1051 if (!SecNameOrErr) { 1052 consumeError(SecNameOrErr.takeError()); 1053 continue; 1054 } 1055 Sections[*SecNameOrErr] = Section; 1056 } 1057 for (auto Plt : ElfObj->getPltEntries()) { 1058 if (Plt.Symbol) { 1059 SymbolRef Symbol(*Plt.Symbol, ElfObj); 1060 uint8_t SymbolType = getElfSymbolType(Obj, Symbol); 1061 if (Expected<StringRef> NameOrErr = Symbol.getName()) { 1062 if (!NameOrErr->empty()) 1063 AllSymbols[Sections[Plt.Section]].emplace_back( 1064 Plt.Address, Saver.save((*NameOrErr + "@plt").str()), SymbolType); 1065 continue; 1066 } else { 1067 // The warning has been reported in disassembleObject(). 1068 consumeError(NameOrErr.takeError()); 1069 } 1070 } 1071 reportWarning("PLT entry at 0x" + Twine::utohexstr(Plt.Address) + 1072 " references an invalid symbol", 1073 Obj.getFileName()); 1074 } 1075 } 1076 1077 // Normally the disassembly output will skip blocks of zeroes. This function 1078 // returns the number of zero bytes that can be skipped when dumping the 1079 // disassembly of the instructions in Buf. 1080 static size_t countSkippableZeroBytes(ArrayRef<uint8_t> Buf) { 1081 // Find the number of leading zeroes. 1082 size_t N = 0; 1083 while (N < Buf.size() && !Buf[N]) 1084 ++N; 1085 1086 // We may want to skip blocks of zero bytes, but unless we see 1087 // at least 8 of them in a row. 1088 if (N < 8) 1089 return 0; 1090 1091 // We skip zeroes in multiples of 4 because do not want to truncate an 1092 // instruction if it starts with a zero byte. 1093 return N & ~0x3; 1094 } 1095 1096 // Returns a map from sections to their relocations. 1097 static std::map<SectionRef, std::vector<RelocationRef>> 1098 getRelocsMap(object::ObjectFile const &Obj) { 1099 std::map<SectionRef, std::vector<RelocationRef>> Ret; 1100 uint64_t I = (uint64_t)-1; 1101 for (SectionRef Sec : Obj.sections()) { 1102 ++I; 1103 Expected<section_iterator> RelocatedOrErr = Sec.getRelocatedSection(); 1104 if (!RelocatedOrErr) 1105 reportError(Obj.getFileName(), 1106 "section (" + Twine(I) + 1107 "): failed to get a relocated section: " + 1108 toString(RelocatedOrErr.takeError())); 1109 1110 section_iterator Relocated = *RelocatedOrErr; 1111 if (Relocated == Obj.section_end() || !checkSectionFilter(*Relocated).Keep) 1112 continue; 1113 std::vector<RelocationRef> &V = Ret[*Relocated]; 1114 append_range(V, Sec.relocations()); 1115 // Sort relocations by address. 1116 llvm::stable_sort(V, isRelocAddressLess); 1117 } 1118 return Ret; 1119 } 1120 1121 // Used for --adjust-vma to check if address should be adjusted by the 1122 // specified value for a given section. 1123 // For ELF we do not adjust non-allocatable sections like debug ones, 1124 // because they are not loadable. 1125 // TODO: implement for other file formats. 1126 static bool shouldAdjustVA(const SectionRef &Section) { 1127 const ObjectFile *Obj = Section.getObject(); 1128 if (Obj->isELF()) 1129 return ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC; 1130 return false; 1131 } 1132 1133 1134 typedef std::pair<uint64_t, char> MappingSymbolPair; 1135 static char getMappingSymbolKind(ArrayRef<MappingSymbolPair> MappingSymbols, 1136 uint64_t Address) { 1137 auto It = 1138 partition_point(MappingSymbols, [Address](const MappingSymbolPair &Val) { 1139 return Val.first <= Address; 1140 }); 1141 // Return zero for any address before the first mapping symbol; this means 1142 // we should use the default disassembly mode, depending on the target. 1143 if (It == MappingSymbols.begin()) 1144 return '\x00'; 1145 return (It - 1)->second; 1146 } 1147 1148 static uint64_t dumpARMELFData(uint64_t SectionAddr, uint64_t Index, 1149 uint64_t End, const ObjectFile &Obj, 1150 ArrayRef<uint8_t> Bytes, 1151 ArrayRef<MappingSymbolPair> MappingSymbols, 1152 const MCSubtargetInfo &STI, raw_ostream &OS) { 1153 support::endianness Endian = 1154 Obj.isLittleEndian() ? support::little : support::big; 1155 size_t Start = OS.tell(); 1156 OS << format("%8" PRIx64 ": ", SectionAddr + Index); 1157 if (Index + 4 <= End) { 1158 dumpBytes(Bytes.slice(Index, 4), OS); 1159 AlignToInstStartColumn(Start, STI, OS); 1160 OS << "\t.word\t" 1161 << format_hex(support::endian::read32(Bytes.data() + Index, Endian), 1162 10); 1163 return 4; 1164 } 1165 if (Index + 2 <= End) { 1166 dumpBytes(Bytes.slice(Index, 2), OS); 1167 AlignToInstStartColumn(Start, STI, OS); 1168 OS << "\t.short\t" 1169 << format_hex(support::endian::read16(Bytes.data() + Index, Endian), 6); 1170 return 2; 1171 } 1172 dumpBytes(Bytes.slice(Index, 1), OS); 1173 AlignToInstStartColumn(Start, STI, OS); 1174 OS << "\t.byte\t" << format_hex(Bytes[Index], 4); 1175 return 1; 1176 } 1177 1178 static void dumpELFData(uint64_t SectionAddr, uint64_t Index, uint64_t End, 1179 ArrayRef<uint8_t> Bytes) { 1180 // print out data up to 8 bytes at a time in hex and ascii 1181 uint8_t AsciiData[9] = {'\0'}; 1182 uint8_t Byte; 1183 int NumBytes = 0; 1184 1185 for (; Index < End; ++Index) { 1186 if (NumBytes == 0) 1187 outs() << format("%8" PRIx64 ":", SectionAddr + Index); 1188 Byte = Bytes.slice(Index)[0]; 1189 outs() << format(" %02x", Byte); 1190 AsciiData[NumBytes] = isPrint(Byte) ? Byte : '.'; 1191 1192 uint8_t IndentOffset = 0; 1193 NumBytes++; 1194 if (Index == End - 1 || NumBytes > 8) { 1195 // Indent the space for less than 8 bytes data. 1196 // 2 spaces for byte and one for space between bytes 1197 IndentOffset = 3 * (8 - NumBytes); 1198 for (int Excess = NumBytes; Excess < 8; Excess++) 1199 AsciiData[Excess] = '\0'; 1200 NumBytes = 8; 1201 } 1202 if (NumBytes == 8) { 1203 AsciiData[8] = '\0'; 1204 outs() << std::string(IndentOffset, ' ') << " "; 1205 outs() << reinterpret_cast<char *>(AsciiData); 1206 outs() << '\n'; 1207 NumBytes = 0; 1208 } 1209 } 1210 } 1211 1212 SymbolInfoTy objdump::createSymbolInfo(const ObjectFile &Obj, 1213 const SymbolRef &Symbol, 1214 bool IsMappingSymbol) { 1215 const StringRef FileName = Obj.getFileName(); 1216 const uint64_t Addr = unwrapOrError(Symbol.getAddress(), FileName); 1217 const StringRef Name = unwrapOrError(Symbol.getName(), FileName); 1218 1219 if (Obj.isXCOFF() && (SymbolDescription || TracebackTable)) { 1220 const auto &XCOFFObj = cast<XCOFFObjectFile>(Obj); 1221 DataRefImpl SymbolDRI = Symbol.getRawDataRefImpl(); 1222 1223 const uint32_t SymbolIndex = XCOFFObj.getSymbolIndex(SymbolDRI.p); 1224 std::optional<XCOFF::StorageMappingClass> Smc = 1225 getXCOFFSymbolCsectSMC(XCOFFObj, Symbol); 1226 return SymbolInfoTy(Smc, Addr, Name, SymbolIndex, 1227 isLabel(XCOFFObj, Symbol)); 1228 } else if (Obj.isXCOFF()) { 1229 const SymbolRef::Type SymType = unwrapOrError(Symbol.getType(), FileName); 1230 return SymbolInfoTy(Addr, Name, SymType, /*IsMappingSymbol=*/false, 1231 /*IsXCOFF=*/true); 1232 } else { 1233 uint8_t Type = 1234 Obj.isELF() ? getElfSymbolType(Obj, Symbol) : (uint8_t)ELF::STT_NOTYPE; 1235 return SymbolInfoTy(Addr, Name, Type, IsMappingSymbol); 1236 } 1237 } 1238 1239 static SymbolInfoTy createDummySymbolInfo(const ObjectFile &Obj, 1240 const uint64_t Addr, StringRef &Name, 1241 uint8_t Type) { 1242 if (Obj.isXCOFF() && (SymbolDescription || TracebackTable)) 1243 return SymbolInfoTy(std::nullopt, Addr, Name, std::nullopt, false); 1244 else 1245 return SymbolInfoTy(Addr, Name, Type); 1246 } 1247 1248 static void 1249 collectBBAddrMapLabels(const std::unordered_map<uint64_t, BBAddrMap> &AddrToBBAddrMap, 1250 uint64_t SectionAddr, uint64_t Start, uint64_t End, 1251 std::unordered_map<uint64_t, std::vector<std::string>> &Labels) { 1252 if (AddrToBBAddrMap.empty()) 1253 return; 1254 Labels.clear(); 1255 uint64_t StartAddress = SectionAddr + Start; 1256 uint64_t EndAddress = SectionAddr + End; 1257 auto Iter = AddrToBBAddrMap.find(StartAddress); 1258 if (Iter == AddrToBBAddrMap.end()) 1259 return; 1260 for (const BBAddrMap::BBEntry &BBEntry : Iter->second.BBEntries) { 1261 uint64_t BBAddress = BBEntry.Offset + Iter->second.Addr; 1262 if (BBAddress >= EndAddress) 1263 continue; 1264 Labels[BBAddress].push_back(("BB" + Twine(BBEntry.ID)).str()); 1265 } 1266 } 1267 1268 static void collectLocalBranchTargets( 1269 ArrayRef<uint8_t> Bytes, const MCInstrAnalysis *MIA, MCDisassembler *DisAsm, 1270 MCInstPrinter *IP, const MCSubtargetInfo *STI, uint64_t SectionAddr, 1271 uint64_t Start, uint64_t End, std::unordered_map<uint64_t, std::string> &Labels) { 1272 // So far only supports PowerPC and X86. 1273 if (!STI->getTargetTriple().isPPC() && !STI->getTargetTriple().isX86()) 1274 return; 1275 1276 Labels.clear(); 1277 unsigned LabelCount = 0; 1278 Start += SectionAddr; 1279 End += SectionAddr; 1280 uint64_t Index = Start; 1281 while (Index < End) { 1282 // Disassemble a real instruction and record function-local branch labels. 1283 MCInst Inst; 1284 uint64_t Size; 1285 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index - SectionAddr); 1286 bool Disassembled = 1287 DisAsm->getInstruction(Inst, Size, ThisBytes, Index, nulls()); 1288 if (Size == 0) 1289 Size = std::min<uint64_t>(ThisBytes.size(), 1290 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1291 1292 if (Disassembled && MIA) { 1293 uint64_t Target; 1294 bool TargetKnown = MIA->evaluateBranch(Inst, Index, Size, Target); 1295 // On PowerPC, if the address of a branch is the same as the target, it 1296 // means that it's a function call. Do not mark the label for this case. 1297 if (TargetKnown && (Target >= Start && Target < End) && 1298 !Labels.count(Target) && 1299 !(STI->getTargetTriple().isPPC() && Target == Index)) 1300 Labels[Target] = ("L" + Twine(LabelCount++)).str(); 1301 } 1302 Index += Size; 1303 } 1304 } 1305 1306 // Create an MCSymbolizer for the target and add it to the MCDisassembler. 1307 // This is currently only used on AMDGPU, and assumes the format of the 1308 // void * argument passed to AMDGPU's createMCSymbolizer. 1309 static void addSymbolizer( 1310 MCContext &Ctx, const Target *Target, StringRef TripleName, 1311 MCDisassembler *DisAsm, uint64_t SectionAddr, ArrayRef<uint8_t> Bytes, 1312 SectionSymbolsTy &Symbols, 1313 std::vector<std::unique_ptr<std::string>> &SynthesizedLabelNames) { 1314 1315 std::unique_ptr<MCRelocationInfo> RelInfo( 1316 Target->createMCRelocationInfo(TripleName, Ctx)); 1317 if (!RelInfo) 1318 return; 1319 std::unique_ptr<MCSymbolizer> Symbolizer(Target->createMCSymbolizer( 1320 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1321 MCSymbolizer *SymbolizerPtr = &*Symbolizer; 1322 DisAsm->setSymbolizer(std::move(Symbolizer)); 1323 1324 if (!SymbolizeOperands) 1325 return; 1326 1327 // Synthesize labels referenced by branch instructions by 1328 // disassembling, discarding the output, and collecting the referenced 1329 // addresses from the symbolizer. 1330 for (size_t Index = 0; Index != Bytes.size();) { 1331 MCInst Inst; 1332 uint64_t Size; 1333 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index); 1334 const uint64_t ThisAddr = SectionAddr + Index; 1335 DisAsm->getInstruction(Inst, Size, ThisBytes, ThisAddr, nulls()); 1336 if (Size == 0) 1337 Size = std::min<uint64_t>(ThisBytes.size(), 1338 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1339 Index += Size; 1340 } 1341 ArrayRef<uint64_t> LabelAddrsRef = SymbolizerPtr->getReferencedAddresses(); 1342 // Copy and sort to remove duplicates. 1343 std::vector<uint64_t> LabelAddrs; 1344 LabelAddrs.insert(LabelAddrs.end(), LabelAddrsRef.begin(), 1345 LabelAddrsRef.end()); 1346 llvm::sort(LabelAddrs); 1347 LabelAddrs.resize(std::unique(LabelAddrs.begin(), LabelAddrs.end()) - 1348 LabelAddrs.begin()); 1349 // Add the labels. 1350 for (unsigned LabelNum = 0; LabelNum != LabelAddrs.size(); ++LabelNum) { 1351 auto Name = std::make_unique<std::string>(); 1352 *Name = (Twine("L") + Twine(LabelNum)).str(); 1353 SynthesizedLabelNames.push_back(std::move(Name)); 1354 Symbols.push_back(SymbolInfoTy( 1355 LabelAddrs[LabelNum], *SynthesizedLabelNames.back(), ELF::STT_NOTYPE)); 1356 } 1357 llvm::stable_sort(Symbols); 1358 // Recreate the symbolizer with the new symbols list. 1359 RelInfo.reset(Target->createMCRelocationInfo(TripleName, Ctx)); 1360 Symbolizer.reset(Target->createMCSymbolizer( 1361 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1362 DisAsm->setSymbolizer(std::move(Symbolizer)); 1363 } 1364 1365 static StringRef getSegmentName(const MachOObjectFile *MachO, 1366 const SectionRef &Section) { 1367 if (MachO) { 1368 DataRefImpl DR = Section.getRawDataRefImpl(); 1369 StringRef SegmentName = MachO->getSectionFinalSegmentName(DR); 1370 return SegmentName; 1371 } 1372 return ""; 1373 } 1374 1375 static void emitPostInstructionInfo(formatted_raw_ostream &FOS, 1376 const MCAsmInfo &MAI, 1377 const MCSubtargetInfo &STI, 1378 StringRef Comments, 1379 LiveVariablePrinter &LVP) { 1380 do { 1381 if (!Comments.empty()) { 1382 // Emit a line of comments. 1383 StringRef Comment; 1384 std::tie(Comment, Comments) = Comments.split('\n'); 1385 // MAI.getCommentColumn() assumes that instructions are printed at the 1386 // position of 8, while getInstStartColumn() returns the actual position. 1387 unsigned CommentColumn = 1388 MAI.getCommentColumn() - 8 + getInstStartColumn(STI); 1389 FOS.PadToColumn(CommentColumn); 1390 FOS << MAI.getCommentString() << ' ' << Comment; 1391 } 1392 LVP.printAfterInst(FOS); 1393 FOS << '\n'; 1394 } while (!Comments.empty()); 1395 FOS.flush(); 1396 } 1397 1398 static void createFakeELFSections(ObjectFile &Obj) { 1399 assert(Obj.isELF()); 1400 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 1401 Elf32LEObj->createFakeSections(); 1402 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 1403 Elf64LEObj->createFakeSections(); 1404 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 1405 Elf32BEObj->createFakeSections(); 1406 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 1407 Elf64BEObj->createFakeSections(); 1408 else 1409 llvm_unreachable("Unsupported binary format"); 1410 } 1411 1412 // Tries to fetch a more complete version of the given object file using its 1413 // Build ID. Returns std::nullopt if nothing was found. 1414 static std::optional<OwningBinary<Binary>> 1415 fetchBinaryByBuildID(const ObjectFile &Obj) { 1416 object::BuildIDRef BuildID = getBuildID(&Obj); 1417 if (BuildID.empty()) 1418 return std::nullopt; 1419 std::optional<std::string> Path = BIDFetcher->fetch(BuildID); 1420 if (!Path) 1421 return std::nullopt; 1422 Expected<OwningBinary<Binary>> DebugBinary = createBinary(*Path); 1423 if (!DebugBinary) { 1424 reportWarning(toString(DebugBinary.takeError()), *Path); 1425 return std::nullopt; 1426 } 1427 return std::move(*DebugBinary); 1428 } 1429 1430 static void 1431 disassembleObject(ObjectFile &Obj, const ObjectFile &DbgObj, 1432 DisassemblerTarget &PrimaryTarget, 1433 std::optional<DisassemblerTarget> &SecondaryTarget, 1434 SourcePrinter &SP, bool InlineRelocs) { 1435 DisassemblerTarget *DT = &PrimaryTarget; 1436 bool PrimaryIsThumb = false; 1437 SmallVector<std::pair<uint64_t, uint64_t>, 0> CHPECodeMap; 1438 1439 if (SecondaryTarget) { 1440 if (isArmElf(Obj)) { 1441 PrimaryIsThumb = 1442 PrimaryTarget.SubtargetInfo->checkFeatures("+thumb-mode"); 1443 } else if (const auto *COFFObj = dyn_cast<COFFObjectFile>(&Obj)) { 1444 const chpe_metadata *CHPEMetadata = COFFObj->getCHPEMetadata(); 1445 if (CHPEMetadata && CHPEMetadata->CodeMapCount) { 1446 uintptr_t CodeMapInt; 1447 cantFail(COFFObj->getRvaPtr(CHPEMetadata->CodeMap, CodeMapInt)); 1448 auto CodeMap = reinterpret_cast<const chpe_range_entry *>(CodeMapInt); 1449 1450 for (uint32_t i = 0; i < CHPEMetadata->CodeMapCount; ++i) { 1451 if (CodeMap[i].getType() == chpe_range_type::Amd64 && 1452 CodeMap[i].Length) { 1453 // Store x86_64 CHPE code ranges. 1454 uint64_t Start = CodeMap[i].getStart() + COFFObj->getImageBase(); 1455 CHPECodeMap.emplace_back(Start, Start + CodeMap[i].Length); 1456 } 1457 } 1458 llvm::sort(CHPECodeMap); 1459 } 1460 } 1461 } 1462 1463 std::map<SectionRef, std::vector<RelocationRef>> RelocMap; 1464 if (InlineRelocs) 1465 RelocMap = getRelocsMap(Obj); 1466 bool Is64Bits = Obj.getBytesInAddress() > 4; 1467 1468 // Create a mapping from virtual address to symbol name. This is used to 1469 // pretty print the symbols while disassembling. 1470 std::map<SectionRef, SectionSymbolsTy> AllSymbols; 1471 std::map<SectionRef, SmallVector<MappingSymbolPair, 0>> AllMappingSymbols; 1472 SectionSymbolsTy AbsoluteSymbols; 1473 const StringRef FileName = Obj.getFileName(); 1474 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&Obj); 1475 for (const SymbolRef &Symbol : Obj.symbols()) { 1476 Expected<StringRef> NameOrErr = Symbol.getName(); 1477 if (!NameOrErr) { 1478 reportWarning(toString(NameOrErr.takeError()), FileName); 1479 continue; 1480 } 1481 if (NameOrErr->empty() && !(Obj.isXCOFF() && SymbolDescription)) 1482 continue; 1483 1484 if (Obj.isELF() && 1485 (cantFail(Symbol.getFlags()) & SymbolRef::SF_FormatSpecific)) { 1486 // Symbol is intended not to be displayed by default (STT_FILE, 1487 // STT_SECTION, or a mapping symbol). Ignore STT_SECTION symbols. We will 1488 // synthesize a section symbol if no symbol is defined at offset 0. 1489 // 1490 // For a mapping symbol, store it within both AllSymbols and 1491 // AllMappingSymbols. If --show-all-symbols is unspecified, its label will 1492 // not be printed in disassembly listing. 1493 if (getElfSymbolType(Obj, Symbol) != ELF::STT_SECTION && 1494 hasMappingSymbols(Obj)) { 1495 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1496 if (SecI != Obj.section_end()) { 1497 uint64_t SectionAddr = SecI->getAddress(); 1498 uint64_t Address = cantFail(Symbol.getAddress()); 1499 StringRef Name = *NameOrErr; 1500 if (Name.consume_front("$") && Name.size() && 1501 strchr("adtx", Name[0])) { 1502 AllMappingSymbols[*SecI].emplace_back(Address - SectionAddr, 1503 Name[0]); 1504 AllSymbols[*SecI].push_back( 1505 createSymbolInfo(Obj, Symbol, /*MappingSymbol=*/true)); 1506 } 1507 } 1508 } 1509 continue; 1510 } 1511 1512 if (MachO) { 1513 // __mh_(execute|dylib|dylinker|bundle|preload|object)_header are special 1514 // symbols that support MachO header introspection. They do not bind to 1515 // code locations and are irrelevant for disassembly. 1516 if (NameOrErr->startswith("__mh_") && NameOrErr->endswith("_header")) 1517 continue; 1518 // Don't ask a Mach-O STAB symbol for its section unless you know that 1519 // STAB symbol's section field refers to a valid section index. Otherwise 1520 // the symbol may error trying to load a section that does not exist. 1521 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 1522 uint8_t NType = (MachO->is64Bit() ? 1523 MachO->getSymbol64TableEntry(SymDRI).n_type: 1524 MachO->getSymbolTableEntry(SymDRI).n_type); 1525 if (NType & MachO::N_STAB) 1526 continue; 1527 } 1528 1529 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1530 if (SecI != Obj.section_end()) 1531 AllSymbols[*SecI].push_back(createSymbolInfo(Obj, Symbol)); 1532 else 1533 AbsoluteSymbols.push_back(createSymbolInfo(Obj, Symbol)); 1534 } 1535 1536 if (AllSymbols.empty() && Obj.isELF()) 1537 addDynamicElfSymbols(cast<ELFObjectFileBase>(Obj), AllSymbols); 1538 1539 if (Obj.isWasm()) 1540 addMissingWasmCodeSymbols(cast<WasmObjectFile>(Obj), AllSymbols); 1541 1542 if (Obj.isELF() && Obj.sections().empty()) 1543 createFakeELFSections(Obj); 1544 1545 BumpPtrAllocator A; 1546 StringSaver Saver(A); 1547 addPltEntries(Obj, AllSymbols, Saver); 1548 1549 // Create a mapping from virtual address to section. An empty section can 1550 // cause more than one section at the same address. Sort such sections to be 1551 // before same-addressed non-empty sections so that symbol lookups prefer the 1552 // non-empty section. 1553 std::vector<std::pair<uint64_t, SectionRef>> SectionAddresses; 1554 for (SectionRef Sec : Obj.sections()) 1555 SectionAddresses.emplace_back(Sec.getAddress(), Sec); 1556 llvm::stable_sort(SectionAddresses, [](const auto &LHS, const auto &RHS) { 1557 if (LHS.first != RHS.first) 1558 return LHS.first < RHS.first; 1559 return LHS.second.getSize() < RHS.second.getSize(); 1560 }); 1561 1562 // Linked executables (.exe and .dll files) typically don't include a real 1563 // symbol table but they might contain an export table. 1564 if (const auto *COFFObj = dyn_cast<COFFObjectFile>(&Obj)) { 1565 for (const auto &ExportEntry : COFFObj->export_directories()) { 1566 StringRef Name; 1567 if (Error E = ExportEntry.getSymbolName(Name)) 1568 reportError(std::move(E), Obj.getFileName()); 1569 if (Name.empty()) 1570 continue; 1571 1572 uint32_t RVA; 1573 if (Error E = ExportEntry.getExportRVA(RVA)) 1574 reportError(std::move(E), Obj.getFileName()); 1575 1576 uint64_t VA = COFFObj->getImageBase() + RVA; 1577 auto Sec = partition_point( 1578 SectionAddresses, [VA](const std::pair<uint64_t, SectionRef> &O) { 1579 return O.first <= VA; 1580 }); 1581 if (Sec != SectionAddresses.begin()) { 1582 --Sec; 1583 AllSymbols[Sec->second].emplace_back(VA, Name, ELF::STT_NOTYPE); 1584 } else 1585 AbsoluteSymbols.emplace_back(VA, Name, ELF::STT_NOTYPE); 1586 } 1587 } 1588 1589 // Sort all the symbols, this allows us to use a simple binary search to find 1590 // Multiple symbols can have the same address. Use a stable sort to stabilize 1591 // the output. 1592 StringSet<> FoundDisasmSymbolSet; 1593 for (std::pair<const SectionRef, SectionSymbolsTy> &SecSyms : AllSymbols) 1594 llvm::stable_sort(SecSyms.second); 1595 llvm::stable_sort(AbsoluteSymbols); 1596 1597 std::unique_ptr<DWARFContext> DICtx; 1598 LiveVariablePrinter LVP(*DT->Context->getRegisterInfo(), *DT->SubtargetInfo); 1599 1600 if (DbgVariables != DVDisabled) { 1601 DICtx = DWARFContext::create(DbgObj); 1602 for (const std::unique_ptr<DWARFUnit> &CU : DICtx->compile_units()) 1603 LVP.addCompileUnit(CU->getUnitDIE(false)); 1604 } 1605 1606 LLVM_DEBUG(LVP.dump()); 1607 1608 std::unordered_map<uint64_t, BBAddrMap> AddrToBBAddrMap; 1609 auto ReadBBAddrMap = [&](std::optional<unsigned> SectionIndex = 1610 std::nullopt) { 1611 AddrToBBAddrMap.clear(); 1612 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj)) { 1613 auto BBAddrMapsOrErr = Elf->readBBAddrMap(SectionIndex); 1614 if (!BBAddrMapsOrErr) { 1615 reportWarning(toString(BBAddrMapsOrErr.takeError()), Obj.getFileName()); 1616 return; 1617 } 1618 for (auto &FunctionBBAddrMap : *BBAddrMapsOrErr) 1619 AddrToBBAddrMap.emplace(FunctionBBAddrMap.Addr, 1620 std::move(FunctionBBAddrMap)); 1621 } 1622 }; 1623 1624 // For non-relocatable objects, Read all LLVM_BB_ADDR_MAP sections into a 1625 // single mapping, since they don't have any conflicts. 1626 if (SymbolizeOperands && !Obj.isRelocatableObject()) 1627 ReadBBAddrMap(); 1628 1629 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 1630 if (FilterSections.empty() && !DisassembleAll && 1631 (!Section.isText() || Section.isVirtual())) 1632 continue; 1633 1634 uint64_t SectionAddr = Section.getAddress(); 1635 uint64_t SectSize = Section.getSize(); 1636 if (!SectSize) 1637 continue; 1638 1639 // For relocatable object files, read the LLVM_BB_ADDR_MAP section 1640 // corresponding to this section, if present. 1641 if (SymbolizeOperands && Obj.isRelocatableObject()) 1642 ReadBBAddrMap(Section.getIndex()); 1643 1644 // Get the list of all the symbols in this section. 1645 SectionSymbolsTy &Symbols = AllSymbols[Section]; 1646 auto &MappingSymbols = AllMappingSymbols[Section]; 1647 llvm::sort(MappingSymbols); 1648 1649 ArrayRef<uint8_t> Bytes = arrayRefFromStringRef( 1650 unwrapOrError(Section.getContents(), Obj.getFileName())); 1651 1652 std::vector<std::unique_ptr<std::string>> SynthesizedLabelNames; 1653 if (Obj.isELF() && Obj.getArch() == Triple::amdgcn) { 1654 // AMDGPU disassembler uses symbolizer for printing labels 1655 addSymbolizer(*DT->Context, DT->TheTarget, TripleName, DT->DisAsm.get(), 1656 SectionAddr, Bytes, Symbols, SynthesizedLabelNames); 1657 } 1658 1659 StringRef SegmentName = getSegmentName(MachO, Section); 1660 StringRef SectionName = unwrapOrError(Section.getName(), Obj.getFileName()); 1661 // If the section has no symbol at the start, just insert a dummy one. 1662 // Without --show-all-symbols, also insert one if all symbols at the start 1663 // are mapping symbols. 1664 bool CreateDummy = Symbols.empty(); 1665 if (!CreateDummy) { 1666 CreateDummy = true; 1667 for (auto &Sym : Symbols) { 1668 if (Sym.Addr != SectionAddr) 1669 break; 1670 if (!Sym.IsMappingSymbol || ShowAllSymbols) 1671 CreateDummy = false; 1672 } 1673 } 1674 if (CreateDummy) { 1675 SymbolInfoTy Sym = createDummySymbolInfo( 1676 Obj, SectionAddr, SectionName, 1677 Section.isText() ? ELF::STT_FUNC : ELF::STT_OBJECT); 1678 if (Obj.isXCOFF()) 1679 Symbols.insert(Symbols.begin(), Sym); 1680 else 1681 Symbols.insert(llvm::lower_bound(Symbols, Sym), Sym); 1682 } 1683 1684 SmallString<40> Comments; 1685 raw_svector_ostream CommentStream(Comments); 1686 1687 uint64_t VMAAdjustment = 0; 1688 if (shouldAdjustVA(Section)) 1689 VMAAdjustment = AdjustVMA; 1690 1691 // In executable and shared objects, r_offset holds a virtual address. 1692 // Subtract SectionAddr from the r_offset field of a relocation to get 1693 // the section offset. 1694 uint64_t RelAdjustment = Obj.isRelocatableObject() ? 0 : SectionAddr; 1695 uint64_t Size; 1696 uint64_t Index; 1697 bool PrintedSection = false; 1698 std::vector<RelocationRef> Rels = RelocMap[Section]; 1699 std::vector<RelocationRef>::const_iterator RelCur = Rels.begin(); 1700 std::vector<RelocationRef>::const_iterator RelEnd = Rels.end(); 1701 1702 // Loop over each chunk of code between two points where at least 1703 // one symbol is defined. 1704 for (size_t SI = 0, SE = Symbols.size(); SI != SE;) { 1705 // Advance SI past all the symbols starting at the same address, 1706 // and make an ArrayRef of them. 1707 unsigned FirstSI = SI; 1708 uint64_t Start = Symbols[SI].Addr; 1709 ArrayRef<SymbolInfoTy> SymbolsHere; 1710 while (SI != SE && Symbols[SI].Addr == Start) 1711 ++SI; 1712 SymbolsHere = ArrayRef<SymbolInfoTy>(&Symbols[FirstSI], SI - FirstSI); 1713 1714 // Get the demangled names of all those symbols. We end up with a vector 1715 // of StringRef that holds the names we're going to use, and a vector of 1716 // std::string that stores the new strings returned by demangle(), if 1717 // any. If we don't call demangle() then that vector can stay empty. 1718 std::vector<StringRef> SymNamesHere; 1719 std::vector<std::string> DemangledSymNamesHere; 1720 if (Demangle) { 1721 // Fetch the demangled names and store them locally. 1722 for (const SymbolInfoTy &Symbol : SymbolsHere) 1723 DemangledSymNamesHere.push_back(demangle(Symbol.Name)); 1724 // Now we've finished modifying that vector, it's safe to make 1725 // a vector of StringRefs pointing into it. 1726 SymNamesHere.insert(SymNamesHere.begin(), DemangledSymNamesHere.begin(), 1727 DemangledSymNamesHere.end()); 1728 } else { 1729 for (const SymbolInfoTy &Symbol : SymbolsHere) 1730 SymNamesHere.push_back(Symbol.Name); 1731 } 1732 1733 // Distinguish ELF data from code symbols, which will be used later on to 1734 // decide whether to 'disassemble' this chunk as a data declaration via 1735 // dumpELFData(), or whether to treat it as code. 1736 // 1737 // If data _and_ code symbols are defined at the same address, the code 1738 // takes priority, on the grounds that disassembling code is our main 1739 // purpose here, and it would be a worse failure to _not_ interpret 1740 // something that _was_ meaningful as code than vice versa. 1741 // 1742 // Any ELF symbol type that is not clearly data will be regarded as code. 1743 // In particular, one of the uses of STT_NOTYPE is for branch targets 1744 // inside functions, for which STT_FUNC would be inaccurate. 1745 // 1746 // So here, we spot whether there's any non-data symbol present at all, 1747 // and only set the DisassembleAsData flag if there isn't. Also, we use 1748 // this distinction to inform the decision of which symbol to print at 1749 // the head of the section, so that if we're printing code, we print a 1750 // code-related symbol name to go with it. 1751 bool DisassembleAsData = false; 1752 size_t DisplaySymIndex = SymbolsHere.size() - 1; 1753 if (Obj.isELF() && !DisassembleAll && Section.isText()) { 1754 DisassembleAsData = true; // unless we find a code symbol below 1755 1756 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 1757 uint8_t SymTy = SymbolsHere[i].Type; 1758 if (SymTy != ELF::STT_OBJECT && SymTy != ELF::STT_COMMON) { 1759 DisassembleAsData = false; 1760 DisplaySymIndex = i; 1761 } 1762 } 1763 } 1764 1765 // Decide which symbol(s) from this collection we're going to print. 1766 std::vector<bool> SymsToPrint(SymbolsHere.size(), false); 1767 // If the user has given the --disassemble-symbols option, then we must 1768 // display every symbol in that set, and no others. 1769 if (!DisasmSymbolSet.empty()) { 1770 bool FoundAny = false; 1771 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 1772 if (DisasmSymbolSet.count(SymNamesHere[i])) { 1773 SymsToPrint[i] = true; 1774 FoundAny = true; 1775 } 1776 } 1777 1778 // And if none of the symbols here is one that the user asked for, skip 1779 // disassembling this entire chunk of code. 1780 if (!FoundAny) 1781 continue; 1782 } else if (!SymbolsHere[DisplaySymIndex].IsMappingSymbol) { 1783 // Otherwise, print whichever symbol at this location is last in the 1784 // Symbols array, because that array is pre-sorted in a way intended to 1785 // correlate with priority of which symbol to display. 1786 SymsToPrint[DisplaySymIndex] = true; 1787 } 1788 1789 // Now that we know we're disassembling this section, override the choice 1790 // of which symbols to display by printing _all_ of them at this address 1791 // if the user asked for all symbols. 1792 // 1793 // That way, '--show-all-symbols --disassemble-symbol=foo' will print 1794 // only the chunk of code headed by 'foo', but also show any other 1795 // symbols defined at that address, such as aliases for 'foo', or the ARM 1796 // mapping symbol preceding its code. 1797 if (ShowAllSymbols) { 1798 for (size_t i = 0; i < SymbolsHere.size(); ++i) 1799 SymsToPrint[i] = true; 1800 } 1801 1802 if (Start < SectionAddr || StopAddress <= Start) 1803 continue; 1804 1805 for (size_t i = 0; i < SymbolsHere.size(); ++i) 1806 FoundDisasmSymbolSet.insert(SymNamesHere[i]); 1807 1808 // The end is the section end, the beginning of the next symbol, or 1809 // --stop-address. 1810 uint64_t End = std::min<uint64_t>(SectionAddr + SectSize, StopAddress); 1811 if (SI < SE) 1812 End = std::min(End, Symbols[SI].Addr); 1813 if (Start >= End || End <= StartAddress) 1814 continue; 1815 Start -= SectionAddr; 1816 End -= SectionAddr; 1817 1818 if (!PrintedSection) { 1819 PrintedSection = true; 1820 outs() << "\nDisassembly of section "; 1821 if (!SegmentName.empty()) 1822 outs() << SegmentName << ","; 1823 outs() << SectionName << ":\n"; 1824 } 1825 1826 bool PrintedLabel = false; 1827 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 1828 if (!SymsToPrint[i]) 1829 continue; 1830 1831 const SymbolInfoTy &Symbol = SymbolsHere[i]; 1832 const StringRef SymbolName = SymNamesHere[i]; 1833 1834 if (!PrintedLabel) { 1835 outs() << '\n'; 1836 PrintedLabel = true; 1837 } 1838 if (LeadingAddr) 1839 outs() << format(Is64Bits ? "%016" PRIx64 " " : "%08" PRIx64 " ", 1840 SectionAddr + Start + VMAAdjustment); 1841 if (Obj.isXCOFF() && SymbolDescription) { 1842 outs() << getXCOFFSymbolDescription(Symbol, SymbolName) << ":\n"; 1843 } else 1844 outs() << '<' << SymbolName << ">:\n"; 1845 } 1846 1847 // Don't print raw contents of a virtual section. A virtual section 1848 // doesn't have any contents in the file. 1849 if (Section.isVirtual()) { 1850 outs() << "...\n"; 1851 continue; 1852 } 1853 1854 // See if any of the symbols defined at this location triggers target- 1855 // specific disassembly behavior, e.g. of special descriptors or function 1856 // prelude information. 1857 // 1858 // We stop this loop at the first symbol that triggers some kind of 1859 // interesting behavior (if any), on the assumption that if two symbols 1860 // defined at the same address trigger two conflicting symbol handlers, 1861 // the object file is probably confused anyway, and it would make even 1862 // less sense to present the output of _both_ handlers, because that 1863 // would describe the same data twice. 1864 for (size_t SHI = 0; SHI < SymbolsHere.size(); ++SHI) { 1865 SymbolInfoTy Symbol = SymbolsHere[SHI]; 1866 1867 auto Status = DT->DisAsm->onSymbolStart( 1868 Symbol, Size, Bytes.slice(Start, End - Start), SectionAddr + Start, 1869 CommentStream); 1870 1871 if (!Status) { 1872 // If onSymbolStart returns std::nullopt, that means it didn't trigger 1873 // any interesting handling for this symbol. Try the other symbols 1874 // defined at this address. 1875 continue; 1876 } 1877 1878 if (*Status == MCDisassembler::Fail) { 1879 // If onSymbolStart returns Fail, that means it identified some kind 1880 // of special data at this address, but wasn't able to disassemble it 1881 // meaningfully. So we fall back to disassembling the failed region 1882 // as bytes, assuming that the target detected the failure before 1883 // printing anything. 1884 // 1885 // Return values Success or SoftFail (i.e no 'real' failure) are 1886 // expected to mean that the target has emitted its own output. 1887 // 1888 // Either way, 'Size' will have been set to the amount of data 1889 // covered by whatever prologue the target identified. So we advance 1890 // our own position to beyond that. Sometimes that will be the entire 1891 // distance to the next symbol, and sometimes it will be just a 1892 // prologue and we should start disassembling instructions from where 1893 // it left off. 1894 outs() << DT->Context->getAsmInfo()->getCommentString() 1895 << " error in decoding " << SymNamesHere[SHI] 1896 << " : decoding failed region as bytes.\n"; 1897 for (uint64_t I = 0; I < Size; ++I) { 1898 outs() << "\t.byte\t " << format_hex(Bytes[I], 1, /*Upper=*/true) 1899 << "\n"; 1900 } 1901 } 1902 Start += Size; 1903 break; 1904 } 1905 1906 Index = Start; 1907 if (SectionAddr < StartAddress) 1908 Index = std::max<uint64_t>(Index, StartAddress - SectionAddr); 1909 1910 if (DisassembleAsData) { 1911 dumpELFData(SectionAddr, Index, End, Bytes); 1912 Index = End; 1913 continue; 1914 } 1915 1916 bool DumpARMELFData = false; 1917 bool DumpTracebackTableForXCOFFFunction = 1918 Obj.isXCOFF() && Section.isText() && TracebackTable && 1919 Symbols[SI - 1].XCOFFSymInfo.StorageMappingClass && 1920 (*Symbols[SI - 1].XCOFFSymInfo.StorageMappingClass == XCOFF::XMC_PR); 1921 1922 formatted_raw_ostream FOS(outs()); 1923 1924 // FIXME: Workaround for bug in formatted_raw_ostream. Color escape codes 1925 // are (incorrectly) written directly to the unbuffered raw_ostream 1926 // wrapped by the formatted_raw_ostream. 1927 if (DisassemblyColor == ColorOutput::Enable || 1928 DisassemblyColor == ColorOutput::Auto) 1929 FOS.SetUnbuffered(); 1930 1931 std::unordered_map<uint64_t, std::string> AllLabels; 1932 std::unordered_map<uint64_t, std::vector<std::string>> BBAddrMapLabels; 1933 if (SymbolizeOperands) { 1934 collectLocalBranchTargets(Bytes, DT->InstrAnalysis.get(), 1935 DT->DisAsm.get(), DT->InstPrinter.get(), 1936 PrimaryTarget.SubtargetInfo.get(), 1937 SectionAddr, Index, End, AllLabels); 1938 collectBBAddrMapLabels(AddrToBBAddrMap, SectionAddr, Index, End, 1939 BBAddrMapLabels); 1940 } 1941 1942 while (Index < End) { 1943 // ARM and AArch64 ELF binaries can interleave data and text in the 1944 // same section. We rely on the markers introduced to understand what 1945 // we need to dump. If the data marker is within a function, it is 1946 // denoted as a word/short etc. 1947 if (!MappingSymbols.empty()) { 1948 char Kind = getMappingSymbolKind(MappingSymbols, Index); 1949 DumpARMELFData = Kind == 'd'; 1950 if (SecondaryTarget) { 1951 if (Kind == 'a') { 1952 DT = PrimaryIsThumb ? &*SecondaryTarget : &PrimaryTarget; 1953 } else if (Kind == 't') { 1954 DT = PrimaryIsThumb ? &PrimaryTarget : &*SecondaryTarget; 1955 } 1956 } 1957 } else if (!CHPECodeMap.empty()) { 1958 uint64_t Address = SectionAddr + Index; 1959 auto It = partition_point( 1960 CHPECodeMap, 1961 [Address](const std::pair<uint64_t, uint64_t> &Entry) { 1962 return Entry.first <= Address; 1963 }); 1964 if (It != CHPECodeMap.begin() && Address < (It - 1)->second) { 1965 DT = &*SecondaryTarget; 1966 } else { 1967 DT = &PrimaryTarget; 1968 // X64 disassembler range may have left Index unaligned, so 1969 // make sure that it's aligned when we switch back to ARM64 1970 // code. 1971 Index = llvm::alignTo(Index, 4); 1972 if (Index >= End) 1973 break; 1974 } 1975 } 1976 1977 if (DumpARMELFData) { 1978 Size = dumpARMELFData(SectionAddr, Index, End, Obj, Bytes, 1979 MappingSymbols, *DT->SubtargetInfo, FOS); 1980 } else { 1981 // When -z or --disassemble-zeroes are given we always dissasemble 1982 // them. Otherwise we might want to skip zero bytes we see. 1983 if (!DisassembleZeroes) { 1984 uint64_t MaxOffset = End - Index; 1985 // For --reloc: print zero blocks patched by relocations, so that 1986 // relocations can be shown in the dump. 1987 if (RelCur != RelEnd) 1988 MaxOffset = std::min(RelCur->getOffset() - RelAdjustment - Index, 1989 MaxOffset); 1990 1991 if (size_t N = 1992 countSkippableZeroBytes(Bytes.slice(Index, MaxOffset))) { 1993 FOS << "\t\t..." << '\n'; 1994 Index += N; 1995 continue; 1996 } 1997 } 1998 1999 if (DumpTracebackTableForXCOFFFunction && 2000 doesXCOFFTracebackTableBegin(Bytes.slice(Index, 4))) { 2001 dumpTracebackTable(Bytes.slice(Index), 2002 SectionAddr + Index + VMAAdjustment, FOS, 2003 SectionAddr + End + VMAAdjustment, 2004 *DT->SubtargetInfo, cast<XCOFFObjectFile>(&Obj)); 2005 Index = End; 2006 continue; 2007 } 2008 2009 // Print local label if there's any. 2010 auto Iter1 = BBAddrMapLabels.find(SectionAddr + Index); 2011 if (Iter1 != BBAddrMapLabels.end()) { 2012 for (StringRef Label : Iter1->second) 2013 FOS << "<" << Label << ">:\n"; 2014 } else { 2015 auto Iter2 = AllLabels.find(SectionAddr + Index); 2016 if (Iter2 != AllLabels.end()) 2017 FOS << "<" << Iter2->second << ">:\n"; 2018 } 2019 2020 // Disassemble a real instruction or a data when disassemble all is 2021 // provided 2022 MCInst Inst; 2023 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index); 2024 uint64_t ThisAddr = SectionAddr + Index; 2025 bool Disassembled = DT->DisAsm->getInstruction( 2026 Inst, Size, ThisBytes, ThisAddr, CommentStream); 2027 if (Size == 0) 2028 Size = std::min<uint64_t>( 2029 ThisBytes.size(), 2030 DT->DisAsm->suggestBytesToSkip(ThisBytes, ThisAddr)); 2031 2032 LVP.update({Index, Section.getIndex()}, 2033 {Index + Size, Section.getIndex()}, Index + Size != End); 2034 2035 DT->InstPrinter->setCommentStream(CommentStream); 2036 2037 DT->Printer->printInst( 2038 *DT->InstPrinter, Disassembled ? &Inst : nullptr, 2039 Bytes.slice(Index, Size), 2040 {SectionAddr + Index + VMAAdjustment, Section.getIndex()}, FOS, 2041 "", *DT->SubtargetInfo, &SP, Obj.getFileName(), &Rels, LVP); 2042 2043 DT->InstPrinter->setCommentStream(llvm::nulls()); 2044 2045 // If disassembly has failed, avoid analysing invalid/incomplete 2046 // instruction information. Otherwise, try to resolve the target 2047 // address (jump target or memory operand address) and print it on the 2048 // right of the instruction. 2049 if (Disassembled && DT->InstrAnalysis) { 2050 // Branch targets are printed just after the instructions. 2051 llvm::raw_ostream *TargetOS = &FOS; 2052 uint64_t Target; 2053 bool PrintTarget = DT->InstrAnalysis->evaluateBranch( 2054 Inst, SectionAddr + Index, Size, Target); 2055 if (!PrintTarget) 2056 if (std::optional<uint64_t> MaybeTarget = 2057 DT->InstrAnalysis->evaluateMemoryOperandAddress( 2058 Inst, DT->SubtargetInfo.get(), SectionAddr + Index, 2059 Size)) { 2060 Target = *MaybeTarget; 2061 PrintTarget = true; 2062 // Do not print real address when symbolizing. 2063 if (!SymbolizeOperands) { 2064 // Memory operand addresses are printed as comments. 2065 TargetOS = &CommentStream; 2066 *TargetOS << "0x" << Twine::utohexstr(Target); 2067 } 2068 } 2069 if (PrintTarget) { 2070 // In a relocatable object, the target's section must reside in 2071 // the same section as the call instruction or it is accessed 2072 // through a relocation. 2073 // 2074 // In a non-relocatable object, the target may be in any section. 2075 // In that case, locate the section(s) containing the target 2076 // address and find the symbol in one of those, if possible. 2077 // 2078 // N.B. We don't walk the relocations in the relocatable case yet. 2079 std::vector<const SectionSymbolsTy *> TargetSectionSymbols; 2080 if (!Obj.isRelocatableObject()) { 2081 auto It = llvm::partition_point( 2082 SectionAddresses, 2083 [=](const std::pair<uint64_t, SectionRef> &O) { 2084 return O.first <= Target; 2085 }); 2086 uint64_t TargetSecAddr = 0; 2087 while (It != SectionAddresses.begin()) { 2088 --It; 2089 if (TargetSecAddr == 0) 2090 TargetSecAddr = It->first; 2091 if (It->first != TargetSecAddr) 2092 break; 2093 TargetSectionSymbols.push_back(&AllSymbols[It->second]); 2094 } 2095 } else { 2096 TargetSectionSymbols.push_back(&Symbols); 2097 } 2098 TargetSectionSymbols.push_back(&AbsoluteSymbols); 2099 2100 // Find the last symbol in the first candidate section whose 2101 // offset is less than or equal to the target. If there are no 2102 // such symbols, try in the next section and so on, before finally 2103 // using the nearest preceding absolute symbol (if any), if there 2104 // are no other valid symbols. 2105 const SymbolInfoTy *TargetSym = nullptr; 2106 for (const SectionSymbolsTy *TargetSymbols : 2107 TargetSectionSymbols) { 2108 auto It = llvm::partition_point( 2109 *TargetSymbols, 2110 [=](const SymbolInfoTy &O) { return O.Addr <= Target; }); 2111 while (It != TargetSymbols->begin()) { 2112 --It; 2113 // Skip mapping symbols to avoid possible ambiguity as they 2114 // do not allow uniquely identifying the target address. 2115 if (!It->IsMappingSymbol) { 2116 TargetSym = &*It; 2117 break; 2118 } 2119 } 2120 if (TargetSym) 2121 break; 2122 } 2123 2124 // Print the labels corresponding to the target if there's any. 2125 bool BBAddrMapLabelAvailable = BBAddrMapLabels.count(Target); 2126 bool LabelAvailable = AllLabels.count(Target); 2127 if (TargetSym != nullptr) { 2128 uint64_t TargetAddress = TargetSym->Addr; 2129 uint64_t Disp = Target - TargetAddress; 2130 std::string TargetName = Demangle ? demangle(TargetSym->Name) 2131 : TargetSym->Name.str(); 2132 2133 *TargetOS << " <"; 2134 if (!Disp) { 2135 // Always Print the binary symbol precisely corresponding to 2136 // the target address. 2137 *TargetOS << TargetName; 2138 } else if (BBAddrMapLabelAvailable) { 2139 *TargetOS << BBAddrMapLabels[Target].front(); 2140 } else if (LabelAvailable) { 2141 *TargetOS << AllLabels[Target]; 2142 } else { 2143 // Always Print the binary symbol plus an offset if there's no 2144 // local label corresponding to the target address. 2145 *TargetOS << TargetName << "+0x" << Twine::utohexstr(Disp); 2146 } 2147 *TargetOS << ">"; 2148 } else if (BBAddrMapLabelAvailable) { 2149 *TargetOS << " <" << BBAddrMapLabels[Target].front() << ">"; 2150 } else if (LabelAvailable) { 2151 *TargetOS << " <" << AllLabels[Target] << ">"; 2152 } 2153 // By convention, each record in the comment stream should be 2154 // terminated. 2155 if (TargetOS == &CommentStream) 2156 *TargetOS << "\n"; 2157 } 2158 } 2159 } 2160 2161 assert(DT->Context->getAsmInfo()); 2162 emitPostInstructionInfo(FOS, *DT->Context->getAsmInfo(), 2163 *DT->SubtargetInfo, CommentStream.str(), LVP); 2164 Comments.clear(); 2165 2166 // Hexagon does this in pretty printer 2167 if (Obj.getArch() != Triple::hexagon) { 2168 // Print relocation for instruction and data. 2169 while (RelCur != RelEnd) { 2170 uint64_t Offset = RelCur->getOffset() - RelAdjustment; 2171 // If this relocation is hidden, skip it. 2172 if (getHidden(*RelCur) || SectionAddr + Offset < StartAddress) { 2173 ++RelCur; 2174 continue; 2175 } 2176 2177 // Stop when RelCur's offset is past the disassembled 2178 // instruction/data. Note that it's possible the disassembled data 2179 // is not the complete data: we might see the relocation printed in 2180 // the middle of the data, but this matches the binutils objdump 2181 // output. 2182 if (Offset >= Index + Size) 2183 break; 2184 2185 // When --adjust-vma is used, update the address printed. 2186 if (RelCur->getSymbol() != Obj.symbol_end()) { 2187 Expected<section_iterator> SymSI = 2188 RelCur->getSymbol()->getSection(); 2189 if (SymSI && *SymSI != Obj.section_end() && 2190 shouldAdjustVA(**SymSI)) 2191 Offset += AdjustVMA; 2192 } 2193 2194 printRelocation(FOS, Obj.getFileName(), *RelCur, 2195 SectionAddr + Offset, Is64Bits); 2196 LVP.printAfterOtherLine(FOS, true); 2197 ++RelCur; 2198 } 2199 } 2200 2201 Index += Size; 2202 } 2203 } 2204 } 2205 StringSet<> MissingDisasmSymbolSet = 2206 set_difference(DisasmSymbolSet, FoundDisasmSymbolSet); 2207 for (StringRef Sym : MissingDisasmSymbolSet.keys()) 2208 reportWarning("failed to disassemble missing symbol " + Sym, FileName); 2209 } 2210 2211 static void disassembleObject(ObjectFile *Obj, bool InlineRelocs) { 2212 // If information useful for showing the disassembly is missing, try to find a 2213 // more complete binary and disassemble that instead. 2214 OwningBinary<Binary> FetchedBinary; 2215 if (Obj->symbols().empty()) { 2216 if (std::optional<OwningBinary<Binary>> FetchedBinaryOpt = 2217 fetchBinaryByBuildID(*Obj)) { 2218 if (auto *O = dyn_cast<ObjectFile>(FetchedBinaryOpt->getBinary())) { 2219 if (!O->symbols().empty() || 2220 (!O->sections().empty() && Obj->sections().empty())) { 2221 FetchedBinary = std::move(*FetchedBinaryOpt); 2222 Obj = O; 2223 } 2224 } 2225 } 2226 } 2227 2228 const Target *TheTarget = getTarget(Obj); 2229 2230 // Package up features to be passed to target/subtarget 2231 Expected<SubtargetFeatures> FeaturesValue = Obj->getFeatures(); 2232 if (!FeaturesValue) 2233 reportError(FeaturesValue.takeError(), Obj->getFileName()); 2234 SubtargetFeatures Features = *FeaturesValue; 2235 if (!MAttrs.empty()) { 2236 for (unsigned I = 0; I != MAttrs.size(); ++I) 2237 Features.AddFeature(MAttrs[I]); 2238 } else if (MCPU.empty() && Obj->getArch() == llvm::Triple::aarch64) { 2239 Features.AddFeature("+all"); 2240 } 2241 2242 if (MCPU.empty()) 2243 MCPU = Obj->tryGetCPUName().value_or("").str(); 2244 2245 if (isArmElf(*Obj)) { 2246 // When disassembling big-endian Arm ELF, the instruction endianness is 2247 // determined in a complex way. In relocatable objects, AAELF32 mandates 2248 // that instruction endianness matches the ELF file endianness; in 2249 // executable images, that's true unless the file header has the EF_ARM_BE8 2250 // flag, in which case instructions are little-endian regardless of data 2251 // endianness. 2252 // 2253 // We must set the big-endian-instructions SubtargetFeature to make the 2254 // disassembler read the instructions the right way round, and also tell 2255 // our own prettyprinter to retrieve the encodings the same way to print in 2256 // hex. 2257 const auto *Elf32BE = dyn_cast<ELF32BEObjectFile>(Obj); 2258 2259 if (Elf32BE && (Elf32BE->isRelocatableObject() || 2260 !(Elf32BE->getPlatformFlags() & ELF::EF_ARM_BE8))) { 2261 Features.AddFeature("+big-endian-instructions"); 2262 ARMPrettyPrinterInst.setInstructionEndianness(llvm::support::big); 2263 } else { 2264 ARMPrettyPrinterInst.setInstructionEndianness(llvm::support::little); 2265 } 2266 } 2267 2268 DisassemblerTarget PrimaryTarget(TheTarget, *Obj, TripleName, MCPU, Features); 2269 2270 // If we have an ARM object file, we need a second disassembler, because 2271 // ARM CPUs have two different instruction sets: ARM mode, and Thumb mode. 2272 // We use mapping symbols to switch between the two assemblers, where 2273 // appropriate. 2274 std::optional<DisassemblerTarget> SecondaryTarget; 2275 2276 if (isArmElf(*Obj)) { 2277 if (!PrimaryTarget.SubtargetInfo->checkFeatures("+mclass")) { 2278 if (PrimaryTarget.SubtargetInfo->checkFeatures("+thumb-mode")) 2279 Features.AddFeature("-thumb-mode"); 2280 else 2281 Features.AddFeature("+thumb-mode"); 2282 SecondaryTarget.emplace(PrimaryTarget, Features); 2283 } 2284 } else if (const auto *COFFObj = dyn_cast<COFFObjectFile>(Obj)) { 2285 const chpe_metadata *CHPEMetadata = COFFObj->getCHPEMetadata(); 2286 if (CHPEMetadata && CHPEMetadata->CodeMapCount) { 2287 // Set up x86_64 disassembler for ARM64EC binaries. 2288 Triple X64Triple(TripleName); 2289 X64Triple.setArch(Triple::ArchType::x86_64); 2290 2291 std::string Error; 2292 const Target *X64Target = 2293 TargetRegistry::lookupTarget("", X64Triple, Error); 2294 if (X64Target) { 2295 SubtargetFeatures X64Features; 2296 SecondaryTarget.emplace(X64Target, *Obj, X64Triple.getTriple(), "", 2297 X64Features); 2298 } else { 2299 reportWarning(Error, Obj->getFileName()); 2300 } 2301 } 2302 } 2303 2304 const ObjectFile *DbgObj = Obj; 2305 if (!FetchedBinary.getBinary() && !Obj->hasDebugInfo()) { 2306 if (std::optional<OwningBinary<Binary>> DebugBinaryOpt = 2307 fetchBinaryByBuildID(*Obj)) { 2308 if (auto *FetchedObj = 2309 dyn_cast<const ObjectFile>(DebugBinaryOpt->getBinary())) { 2310 if (FetchedObj->hasDebugInfo()) { 2311 FetchedBinary = std::move(*DebugBinaryOpt); 2312 DbgObj = FetchedObj; 2313 } 2314 } 2315 } 2316 } 2317 2318 std::unique_ptr<object::Binary> DSYMBinary; 2319 std::unique_ptr<MemoryBuffer> DSYMBuf; 2320 if (!DbgObj->hasDebugInfo()) { 2321 if (const MachOObjectFile *MachOOF = dyn_cast<MachOObjectFile>(&*Obj)) { 2322 DbgObj = objdump::getMachODSymObject(MachOOF, Obj->getFileName(), 2323 DSYMBinary, DSYMBuf); 2324 if (!DbgObj) 2325 return; 2326 } 2327 } 2328 2329 SourcePrinter SP(DbgObj, TheTarget->getName()); 2330 2331 for (StringRef Opt : DisassemblerOptions) 2332 if (!PrimaryTarget.InstPrinter->applyTargetSpecificCLOption(Opt)) 2333 reportError(Obj->getFileName(), 2334 "Unrecognized disassembler option: " + Opt); 2335 2336 disassembleObject(*Obj, *DbgObj, PrimaryTarget, SecondaryTarget, SP, 2337 InlineRelocs); 2338 } 2339 2340 void Dumper::printRelocations() { 2341 StringRef Fmt = O.getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2342 2343 // Build a mapping from relocation target to a vector of relocation 2344 // sections. Usually, there is an only one relocation section for 2345 // each relocated section. 2346 MapVector<SectionRef, std::vector<SectionRef>> SecToRelSec; 2347 uint64_t Ndx; 2348 for (const SectionRef &Section : ToolSectionFilter(O, &Ndx)) { 2349 if (O.isELF() && (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC)) 2350 continue; 2351 if (Section.relocation_begin() == Section.relocation_end()) 2352 continue; 2353 Expected<section_iterator> SecOrErr = Section.getRelocatedSection(); 2354 if (!SecOrErr) 2355 reportError(O.getFileName(), 2356 "section (" + Twine(Ndx) + 2357 "): unable to get a relocation target: " + 2358 toString(SecOrErr.takeError())); 2359 SecToRelSec[**SecOrErr].push_back(Section); 2360 } 2361 2362 for (std::pair<SectionRef, std::vector<SectionRef>> &P : SecToRelSec) { 2363 StringRef SecName = unwrapOrError(P.first.getName(), O.getFileName()); 2364 outs() << "\nRELOCATION RECORDS FOR [" << SecName << "]:\n"; 2365 uint32_t OffsetPadding = (O.getBytesInAddress() > 4 ? 16 : 8); 2366 uint32_t TypePadding = 24; 2367 outs() << left_justify("OFFSET", OffsetPadding) << " " 2368 << left_justify("TYPE", TypePadding) << " " 2369 << "VALUE\n"; 2370 2371 for (SectionRef Section : P.second) { 2372 for (const RelocationRef &Reloc : Section.relocations()) { 2373 uint64_t Address = Reloc.getOffset(); 2374 SmallString<32> RelocName; 2375 SmallString<32> ValueStr; 2376 if (Address < StartAddress || Address > StopAddress || getHidden(Reloc)) 2377 continue; 2378 Reloc.getTypeName(RelocName); 2379 if (Error E = getRelocationValueString(Reloc, ValueStr)) 2380 reportUniqueWarning(std::move(E)); 2381 2382 outs() << format(Fmt.data(), Address) << " " 2383 << left_justify(RelocName, TypePadding) << " " << ValueStr 2384 << "\n"; 2385 } 2386 } 2387 } 2388 } 2389 2390 // Returns true if we need to show LMA column when dumping section headers. We 2391 // show it only when the platform is ELF and either we have at least one section 2392 // whose VMA and LMA are different and/or when --show-lma flag is used. 2393 static bool shouldDisplayLMA(const ObjectFile &Obj) { 2394 if (!Obj.isELF()) 2395 return false; 2396 for (const SectionRef &S : ToolSectionFilter(Obj)) 2397 if (S.getAddress() != getELFSectionLMA(S)) 2398 return true; 2399 return ShowLMA; 2400 } 2401 2402 static size_t getMaxSectionNameWidth(const ObjectFile &Obj) { 2403 // Default column width for names is 13 even if no names are that long. 2404 size_t MaxWidth = 13; 2405 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 2406 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 2407 MaxWidth = std::max(MaxWidth, Name.size()); 2408 } 2409 return MaxWidth; 2410 } 2411 2412 void objdump::printSectionHeaders(ObjectFile &Obj) { 2413 if (Obj.isELF() && Obj.sections().empty()) 2414 createFakeELFSections(Obj); 2415 2416 size_t NameWidth = getMaxSectionNameWidth(Obj); 2417 size_t AddressWidth = 2 * Obj.getBytesInAddress(); 2418 bool HasLMAColumn = shouldDisplayLMA(Obj); 2419 outs() << "\nSections:\n"; 2420 if (HasLMAColumn) 2421 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 2422 << left_justify("VMA", AddressWidth) << " " 2423 << left_justify("LMA", AddressWidth) << " Type\n"; 2424 else 2425 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 2426 << left_justify("VMA", AddressWidth) << " Type\n"; 2427 2428 uint64_t Idx; 2429 for (const SectionRef &Section : ToolSectionFilter(Obj, &Idx)) { 2430 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 2431 uint64_t VMA = Section.getAddress(); 2432 if (shouldAdjustVA(Section)) 2433 VMA += AdjustVMA; 2434 2435 uint64_t Size = Section.getSize(); 2436 2437 std::string Type = Section.isText() ? "TEXT" : ""; 2438 if (Section.isData()) 2439 Type += Type.empty() ? "DATA" : ", DATA"; 2440 if (Section.isBSS()) 2441 Type += Type.empty() ? "BSS" : ", BSS"; 2442 if (Section.isDebugSection()) 2443 Type += Type.empty() ? "DEBUG" : ", DEBUG"; 2444 2445 if (HasLMAColumn) 2446 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 2447 Name.str().c_str(), Size) 2448 << format_hex_no_prefix(VMA, AddressWidth) << " " 2449 << format_hex_no_prefix(getELFSectionLMA(Section), AddressWidth) 2450 << " " << Type << "\n"; 2451 else 2452 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 2453 Name.str().c_str(), Size) 2454 << format_hex_no_prefix(VMA, AddressWidth) << " " << Type << "\n"; 2455 } 2456 } 2457 2458 void objdump::printSectionContents(const ObjectFile *Obj) { 2459 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj); 2460 2461 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 2462 StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName()); 2463 uint64_t BaseAddr = Section.getAddress(); 2464 uint64_t Size = Section.getSize(); 2465 if (!Size) 2466 continue; 2467 2468 outs() << "Contents of section "; 2469 StringRef SegmentName = getSegmentName(MachO, Section); 2470 if (!SegmentName.empty()) 2471 outs() << SegmentName << ","; 2472 outs() << Name << ":\n"; 2473 if (Section.isBSS()) { 2474 outs() << format("<skipping contents of bss section at [%04" PRIx64 2475 ", %04" PRIx64 ")>\n", 2476 BaseAddr, BaseAddr + Size); 2477 continue; 2478 } 2479 2480 StringRef Contents = unwrapOrError(Section.getContents(), Obj->getFileName()); 2481 2482 // Dump out the content as hex and printable ascii characters. 2483 for (std::size_t Addr = 0, End = Contents.size(); Addr < End; Addr += 16) { 2484 outs() << format(" %04" PRIx64 " ", BaseAddr + Addr); 2485 // Dump line of hex. 2486 for (std::size_t I = 0; I < 16; ++I) { 2487 if (I != 0 && I % 4 == 0) 2488 outs() << ' '; 2489 if (Addr + I < End) 2490 outs() << hexdigit((Contents[Addr + I] >> 4) & 0xF, true) 2491 << hexdigit(Contents[Addr + I] & 0xF, true); 2492 else 2493 outs() << " "; 2494 } 2495 // Print ascii. 2496 outs() << " "; 2497 for (std::size_t I = 0; I < 16 && Addr + I < End; ++I) { 2498 if (isPrint(static_cast<unsigned char>(Contents[Addr + I]) & 0xFF)) 2499 outs() << Contents[Addr + I]; 2500 else 2501 outs() << "."; 2502 } 2503 outs() << "\n"; 2504 } 2505 } 2506 } 2507 2508 void Dumper::printSymbolTable(StringRef ArchiveName, StringRef ArchitectureName, 2509 bool DumpDynamic) { 2510 if (O.isCOFF() && !DumpDynamic) { 2511 outs() << "\nSYMBOL TABLE:\n"; 2512 printCOFFSymbolTable(cast<const COFFObjectFile>(O)); 2513 return; 2514 } 2515 2516 const StringRef FileName = O.getFileName(); 2517 2518 if (!DumpDynamic) { 2519 outs() << "\nSYMBOL TABLE:\n"; 2520 for (auto I = O.symbol_begin(); I != O.symbol_end(); ++I) 2521 printSymbol(*I, {}, FileName, ArchiveName, ArchitectureName, DumpDynamic); 2522 return; 2523 } 2524 2525 outs() << "\nDYNAMIC SYMBOL TABLE:\n"; 2526 if (!O.isELF()) { 2527 reportWarning( 2528 "this operation is not currently supported for this file format", 2529 FileName); 2530 return; 2531 } 2532 2533 const ELFObjectFileBase *ELF = cast<const ELFObjectFileBase>(&O); 2534 auto Symbols = ELF->getDynamicSymbolIterators(); 2535 Expected<std::vector<VersionEntry>> SymbolVersionsOrErr = 2536 ELF->readDynsymVersions(); 2537 if (!SymbolVersionsOrErr) { 2538 reportWarning(toString(SymbolVersionsOrErr.takeError()), FileName); 2539 SymbolVersionsOrErr = std::vector<VersionEntry>(); 2540 (void)!SymbolVersionsOrErr; 2541 } 2542 for (auto &Sym : Symbols) 2543 printSymbol(Sym, *SymbolVersionsOrErr, FileName, ArchiveName, 2544 ArchitectureName, DumpDynamic); 2545 } 2546 2547 void Dumper::printSymbol(const SymbolRef &Symbol, 2548 ArrayRef<VersionEntry> SymbolVersions, 2549 StringRef FileName, StringRef ArchiveName, 2550 StringRef ArchitectureName, bool DumpDynamic) { 2551 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&O); 2552 Expected<uint64_t> AddrOrErr = Symbol.getAddress(); 2553 if (!AddrOrErr) { 2554 reportUniqueWarning(AddrOrErr.takeError()); 2555 return; 2556 } 2557 uint64_t Address = *AddrOrErr; 2558 if ((Address < StartAddress) || (Address > StopAddress)) 2559 return; 2560 SymbolRef::Type Type = 2561 unwrapOrError(Symbol.getType(), FileName, ArchiveName, ArchitectureName); 2562 uint32_t Flags = 2563 unwrapOrError(Symbol.getFlags(), FileName, ArchiveName, ArchitectureName); 2564 2565 // Don't ask a Mach-O STAB symbol for its section unless you know that 2566 // STAB symbol's section field refers to a valid section index. Otherwise 2567 // the symbol may error trying to load a section that does not exist. 2568 bool IsSTAB = false; 2569 if (MachO) { 2570 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 2571 uint8_t NType = 2572 (MachO->is64Bit() ? MachO->getSymbol64TableEntry(SymDRI).n_type 2573 : MachO->getSymbolTableEntry(SymDRI).n_type); 2574 if (NType & MachO::N_STAB) 2575 IsSTAB = true; 2576 } 2577 section_iterator Section = IsSTAB 2578 ? O.section_end() 2579 : unwrapOrError(Symbol.getSection(), FileName, 2580 ArchiveName, ArchitectureName); 2581 2582 StringRef Name; 2583 if (Type == SymbolRef::ST_Debug && Section != O.section_end()) { 2584 if (Expected<StringRef> NameOrErr = Section->getName()) 2585 Name = *NameOrErr; 2586 else 2587 consumeError(NameOrErr.takeError()); 2588 2589 } else { 2590 Name = unwrapOrError(Symbol.getName(), FileName, ArchiveName, 2591 ArchitectureName); 2592 } 2593 2594 bool Global = Flags & SymbolRef::SF_Global; 2595 bool Weak = Flags & SymbolRef::SF_Weak; 2596 bool Absolute = Flags & SymbolRef::SF_Absolute; 2597 bool Common = Flags & SymbolRef::SF_Common; 2598 bool Hidden = Flags & SymbolRef::SF_Hidden; 2599 2600 char GlobLoc = ' '; 2601 if ((Section != O.section_end() || Absolute) && !Weak) 2602 GlobLoc = Global ? 'g' : 'l'; 2603 char IFunc = ' '; 2604 if (O.isELF()) { 2605 if (ELFSymbolRef(Symbol).getELFType() == ELF::STT_GNU_IFUNC) 2606 IFunc = 'i'; 2607 if (ELFSymbolRef(Symbol).getBinding() == ELF::STB_GNU_UNIQUE) 2608 GlobLoc = 'u'; 2609 } 2610 2611 char Debug = ' '; 2612 if (DumpDynamic) 2613 Debug = 'D'; 2614 else if (Type == SymbolRef::ST_Debug || Type == SymbolRef::ST_File) 2615 Debug = 'd'; 2616 2617 char FileFunc = ' '; 2618 if (Type == SymbolRef::ST_File) 2619 FileFunc = 'f'; 2620 else if (Type == SymbolRef::ST_Function) 2621 FileFunc = 'F'; 2622 else if (Type == SymbolRef::ST_Data) 2623 FileFunc = 'O'; 2624 2625 const char *Fmt = O.getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2626 2627 outs() << format(Fmt, Address) << " " 2628 << GlobLoc // Local -> 'l', Global -> 'g', Neither -> ' ' 2629 << (Weak ? 'w' : ' ') // Weak? 2630 << ' ' // Constructor. Not supported yet. 2631 << ' ' // Warning. Not supported yet. 2632 << IFunc // Indirect reference to another symbol. 2633 << Debug // Debugging (d) or dynamic (D) symbol. 2634 << FileFunc // Name of function (F), file (f) or object (O). 2635 << ' '; 2636 if (Absolute) { 2637 outs() << "*ABS*"; 2638 } else if (Common) { 2639 outs() << "*COM*"; 2640 } else if (Section == O.section_end()) { 2641 if (O.isXCOFF()) { 2642 XCOFFSymbolRef XCOFFSym = cast<const XCOFFObjectFile>(O).toSymbolRef( 2643 Symbol.getRawDataRefImpl()); 2644 if (XCOFF::N_DEBUG == XCOFFSym.getSectionNumber()) 2645 outs() << "*DEBUG*"; 2646 else 2647 outs() << "*UND*"; 2648 } else 2649 outs() << "*UND*"; 2650 } else { 2651 StringRef SegmentName = getSegmentName(MachO, *Section); 2652 if (!SegmentName.empty()) 2653 outs() << SegmentName << ","; 2654 StringRef SectionName = unwrapOrError(Section->getName(), FileName); 2655 outs() << SectionName; 2656 if (O.isXCOFF()) { 2657 std::optional<SymbolRef> SymRef = 2658 getXCOFFSymbolContainingSymbolRef(cast<XCOFFObjectFile>(O), Symbol); 2659 if (SymRef) { 2660 2661 Expected<StringRef> NameOrErr = SymRef->getName(); 2662 2663 if (NameOrErr) { 2664 outs() << " (csect:"; 2665 std::string SymName = 2666 Demangle ? demangle(*NameOrErr) : NameOrErr->str(); 2667 2668 if (SymbolDescription) 2669 SymName = getXCOFFSymbolDescription(createSymbolInfo(O, *SymRef), 2670 SymName); 2671 2672 outs() << ' ' << SymName; 2673 outs() << ") "; 2674 } else 2675 reportWarning(toString(NameOrErr.takeError()), FileName); 2676 } 2677 } 2678 } 2679 2680 if (Common) 2681 outs() << '\t' << format(Fmt, static_cast<uint64_t>(Symbol.getAlignment())); 2682 else if (O.isXCOFF()) 2683 outs() << '\t' 2684 << format(Fmt, cast<XCOFFObjectFile>(O).getSymbolSize( 2685 Symbol.getRawDataRefImpl())); 2686 else if (O.isELF()) 2687 outs() << '\t' << format(Fmt, ELFSymbolRef(Symbol).getSize()); 2688 2689 if (O.isELF()) { 2690 if (!SymbolVersions.empty()) { 2691 const VersionEntry &Ver = 2692 SymbolVersions[Symbol.getRawDataRefImpl().d.b - 1]; 2693 std::string Str; 2694 if (!Ver.Name.empty()) 2695 Str = Ver.IsVerDef ? ' ' + Ver.Name : '(' + Ver.Name + ')'; 2696 outs() << ' ' << left_justify(Str, 12); 2697 } 2698 2699 uint8_t Other = ELFSymbolRef(Symbol).getOther(); 2700 switch (Other) { 2701 case ELF::STV_DEFAULT: 2702 break; 2703 case ELF::STV_INTERNAL: 2704 outs() << " .internal"; 2705 break; 2706 case ELF::STV_HIDDEN: 2707 outs() << " .hidden"; 2708 break; 2709 case ELF::STV_PROTECTED: 2710 outs() << " .protected"; 2711 break; 2712 default: 2713 outs() << format(" 0x%02x", Other); 2714 break; 2715 } 2716 } else if (Hidden) { 2717 outs() << " .hidden"; 2718 } 2719 2720 std::string SymName = Demangle ? demangle(Name) : Name.str(); 2721 if (O.isXCOFF() && SymbolDescription) 2722 SymName = getXCOFFSymbolDescription(createSymbolInfo(O, Symbol), SymName); 2723 2724 outs() << ' ' << SymName << '\n'; 2725 } 2726 2727 static void printUnwindInfo(const ObjectFile *O) { 2728 outs() << "Unwind info:\n\n"; 2729 2730 if (const COFFObjectFile *Coff = dyn_cast<COFFObjectFile>(O)) 2731 printCOFFUnwindInfo(Coff); 2732 else if (const MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(O)) 2733 printMachOUnwindInfo(MachO); 2734 else 2735 // TODO: Extract DWARF dump tool to objdump. 2736 WithColor::error(errs(), ToolName) 2737 << "This operation is only currently supported " 2738 "for COFF and MachO object files.\n"; 2739 } 2740 2741 /// Dump the raw contents of the __clangast section so the output can be piped 2742 /// into llvm-bcanalyzer. 2743 static void printRawClangAST(const ObjectFile *Obj) { 2744 if (outs().is_displayed()) { 2745 WithColor::error(errs(), ToolName) 2746 << "The -raw-clang-ast option will dump the raw binary contents of " 2747 "the clang ast section.\n" 2748 "Please redirect the output to a file or another program such as " 2749 "llvm-bcanalyzer.\n"; 2750 return; 2751 } 2752 2753 StringRef ClangASTSectionName("__clangast"); 2754 if (Obj->isCOFF()) { 2755 ClangASTSectionName = "clangast"; 2756 } 2757 2758 std::optional<object::SectionRef> ClangASTSection; 2759 for (auto Sec : ToolSectionFilter(*Obj)) { 2760 StringRef Name; 2761 if (Expected<StringRef> NameOrErr = Sec.getName()) 2762 Name = *NameOrErr; 2763 else 2764 consumeError(NameOrErr.takeError()); 2765 2766 if (Name == ClangASTSectionName) { 2767 ClangASTSection = Sec; 2768 break; 2769 } 2770 } 2771 if (!ClangASTSection) 2772 return; 2773 2774 StringRef ClangASTContents = 2775 unwrapOrError(ClangASTSection->getContents(), Obj->getFileName()); 2776 outs().write(ClangASTContents.data(), ClangASTContents.size()); 2777 } 2778 2779 static void printFaultMaps(const ObjectFile *Obj) { 2780 StringRef FaultMapSectionName; 2781 2782 if (Obj->isELF()) { 2783 FaultMapSectionName = ".llvm_faultmaps"; 2784 } else if (Obj->isMachO()) { 2785 FaultMapSectionName = "__llvm_faultmaps"; 2786 } else { 2787 WithColor::error(errs(), ToolName) 2788 << "This operation is only currently supported " 2789 "for ELF and Mach-O executable files.\n"; 2790 return; 2791 } 2792 2793 std::optional<object::SectionRef> FaultMapSection; 2794 2795 for (auto Sec : ToolSectionFilter(*Obj)) { 2796 StringRef Name; 2797 if (Expected<StringRef> NameOrErr = Sec.getName()) 2798 Name = *NameOrErr; 2799 else 2800 consumeError(NameOrErr.takeError()); 2801 2802 if (Name == FaultMapSectionName) { 2803 FaultMapSection = Sec; 2804 break; 2805 } 2806 } 2807 2808 outs() << "FaultMap table:\n"; 2809 2810 if (!FaultMapSection) { 2811 outs() << "<not found>\n"; 2812 return; 2813 } 2814 2815 StringRef FaultMapContents = 2816 unwrapOrError(FaultMapSection->getContents(), Obj->getFileName()); 2817 FaultMapParser FMP(FaultMapContents.bytes_begin(), 2818 FaultMapContents.bytes_end()); 2819 2820 outs() << FMP; 2821 } 2822 2823 void Dumper::printPrivateHeaders() { 2824 reportError(O.getFileName(), "Invalid/Unsupported object file format"); 2825 } 2826 2827 static void printFileHeaders(const ObjectFile *O) { 2828 if (!O->isELF() && !O->isCOFF()) 2829 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2830 2831 Triple::ArchType AT = O->getArch(); 2832 outs() << "architecture: " << Triple::getArchTypeName(AT) << "\n"; 2833 uint64_t Address = unwrapOrError(O->getStartAddress(), O->getFileName()); 2834 2835 StringRef Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2836 outs() << "start address: " 2837 << "0x" << format(Fmt.data(), Address) << "\n"; 2838 } 2839 2840 static void printArchiveChild(StringRef Filename, const Archive::Child &C) { 2841 Expected<sys::fs::perms> ModeOrErr = C.getAccessMode(); 2842 if (!ModeOrErr) { 2843 WithColor::error(errs(), ToolName) << "ill-formed archive entry.\n"; 2844 consumeError(ModeOrErr.takeError()); 2845 return; 2846 } 2847 sys::fs::perms Mode = ModeOrErr.get(); 2848 outs() << ((Mode & sys::fs::owner_read) ? "r" : "-"); 2849 outs() << ((Mode & sys::fs::owner_write) ? "w" : "-"); 2850 outs() << ((Mode & sys::fs::owner_exe) ? "x" : "-"); 2851 outs() << ((Mode & sys::fs::group_read) ? "r" : "-"); 2852 outs() << ((Mode & sys::fs::group_write) ? "w" : "-"); 2853 outs() << ((Mode & sys::fs::group_exe) ? "x" : "-"); 2854 outs() << ((Mode & sys::fs::others_read) ? "r" : "-"); 2855 outs() << ((Mode & sys::fs::others_write) ? "w" : "-"); 2856 outs() << ((Mode & sys::fs::others_exe) ? "x" : "-"); 2857 2858 outs() << " "; 2859 2860 outs() << format("%d/%d %6" PRId64 " ", unwrapOrError(C.getUID(), Filename), 2861 unwrapOrError(C.getGID(), Filename), 2862 unwrapOrError(C.getRawSize(), Filename)); 2863 2864 StringRef RawLastModified = C.getRawLastModified(); 2865 unsigned Seconds; 2866 if (RawLastModified.getAsInteger(10, Seconds)) 2867 outs() << "(date: \"" << RawLastModified 2868 << "\" contains non-decimal chars) "; 2869 else { 2870 // Since ctime(3) returns a 26 character string of the form: 2871 // "Sun Sep 16 01:03:52 1973\n\0" 2872 // just print 24 characters. 2873 time_t t = Seconds; 2874 outs() << format("%.24s ", ctime(&t)); 2875 } 2876 2877 StringRef Name = ""; 2878 Expected<StringRef> NameOrErr = C.getName(); 2879 if (!NameOrErr) { 2880 consumeError(NameOrErr.takeError()); 2881 Name = unwrapOrError(C.getRawName(), Filename); 2882 } else { 2883 Name = NameOrErr.get(); 2884 } 2885 outs() << Name << "\n"; 2886 } 2887 2888 // For ELF only now. 2889 static bool shouldWarnForInvalidStartStopAddress(ObjectFile *Obj) { 2890 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj)) { 2891 if (Elf->getEType() != ELF::ET_REL) 2892 return true; 2893 } 2894 return false; 2895 } 2896 2897 static void checkForInvalidStartStopAddress(ObjectFile *Obj, 2898 uint64_t Start, uint64_t Stop) { 2899 if (!shouldWarnForInvalidStartStopAddress(Obj)) 2900 return; 2901 2902 for (const SectionRef &Section : Obj->sections()) 2903 if (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC) { 2904 uint64_t BaseAddr = Section.getAddress(); 2905 uint64_t Size = Section.getSize(); 2906 if ((Start < BaseAddr + Size) && Stop > BaseAddr) 2907 return; 2908 } 2909 2910 if (!HasStartAddressFlag) 2911 reportWarning("no section has address less than 0x" + 2912 Twine::utohexstr(Stop) + " specified by --stop-address", 2913 Obj->getFileName()); 2914 else if (!HasStopAddressFlag) 2915 reportWarning("no section has address greater than or equal to 0x" + 2916 Twine::utohexstr(Start) + " specified by --start-address", 2917 Obj->getFileName()); 2918 else 2919 reportWarning("no section overlaps the range [0x" + 2920 Twine::utohexstr(Start) + ",0x" + Twine::utohexstr(Stop) + 2921 ") specified by --start-address/--stop-address", 2922 Obj->getFileName()); 2923 } 2924 2925 static void dumpObject(ObjectFile *O, const Archive *A = nullptr, 2926 const Archive::Child *C = nullptr) { 2927 Expected<std::unique_ptr<Dumper>> DumperOrErr = createDumper(*O); 2928 if (!DumperOrErr) { 2929 reportError(DumperOrErr.takeError(), O->getFileName(), 2930 A ? A->getFileName() : ""); 2931 return; 2932 } 2933 Dumper &D = **DumperOrErr; 2934 2935 // Avoid other output when using a raw option. 2936 if (!RawClangAST) { 2937 outs() << '\n'; 2938 if (A) 2939 outs() << A->getFileName() << "(" << O->getFileName() << ")"; 2940 else 2941 outs() << O->getFileName(); 2942 outs() << ":\tfile format " << O->getFileFormatName().lower() << "\n"; 2943 } 2944 2945 if (HasStartAddressFlag || HasStopAddressFlag) 2946 checkForInvalidStartStopAddress(O, StartAddress, StopAddress); 2947 2948 // TODO: Change print* free functions to Dumper member functions to utilitize 2949 // stateful functions like reportUniqueWarning. 2950 2951 // Note: the order here matches GNU objdump for compatability. 2952 StringRef ArchiveName = A ? A->getFileName() : ""; 2953 if (ArchiveHeaders && !MachOOpt && C) 2954 printArchiveChild(ArchiveName, *C); 2955 if (FileHeaders) 2956 printFileHeaders(O); 2957 if (PrivateHeaders || FirstPrivateHeader) 2958 D.printPrivateHeaders(); 2959 if (SectionHeaders) 2960 printSectionHeaders(*O); 2961 if (SymbolTable) 2962 D.printSymbolTable(ArchiveName); 2963 if (DynamicSymbolTable) 2964 D.printSymbolTable(ArchiveName, /*ArchitectureName=*/"", 2965 /*DumpDynamic=*/true); 2966 if (DwarfDumpType != DIDT_Null) { 2967 std::unique_ptr<DIContext> DICtx = DWARFContext::create(*O); 2968 // Dump the complete DWARF structure. 2969 DIDumpOptions DumpOpts; 2970 DumpOpts.DumpType = DwarfDumpType; 2971 DICtx->dump(outs(), DumpOpts); 2972 } 2973 if (Relocations && !Disassemble) 2974 D.printRelocations(); 2975 if (DynamicRelocations) 2976 D.printDynamicRelocations(); 2977 if (SectionContents) 2978 printSectionContents(O); 2979 if (Disassemble) 2980 disassembleObject(O, Relocations); 2981 if (UnwindInfo) 2982 printUnwindInfo(O); 2983 2984 // Mach-O specific options: 2985 if (ExportsTrie) 2986 printExportsTrie(O); 2987 if (Rebase) 2988 printRebaseTable(O); 2989 if (Bind) 2990 printBindTable(O); 2991 if (LazyBind) 2992 printLazyBindTable(O); 2993 if (WeakBind) 2994 printWeakBindTable(O); 2995 2996 // Other special sections: 2997 if (RawClangAST) 2998 printRawClangAST(O); 2999 if (FaultMapSection) 3000 printFaultMaps(O); 3001 if (Offloading) 3002 dumpOffloadBinary(*O); 3003 } 3004 3005 static void dumpObject(const COFFImportFile *I, const Archive *A, 3006 const Archive::Child *C = nullptr) { 3007 StringRef ArchiveName = A ? A->getFileName() : ""; 3008 3009 // Avoid other output when using a raw option. 3010 if (!RawClangAST) 3011 outs() << '\n' 3012 << ArchiveName << "(" << I->getFileName() << ")" 3013 << ":\tfile format COFF-import-file" 3014 << "\n\n"; 3015 3016 if (ArchiveHeaders && !MachOOpt && C) 3017 printArchiveChild(ArchiveName, *C); 3018 if (SymbolTable) 3019 printCOFFSymbolTable(*I); 3020 } 3021 3022 /// Dump each object file in \a a; 3023 static void dumpArchive(const Archive *A) { 3024 Error Err = Error::success(); 3025 unsigned I = -1; 3026 for (auto &C : A->children(Err)) { 3027 ++I; 3028 Expected<std::unique_ptr<Binary>> ChildOrErr = C.getAsBinary(); 3029 if (!ChildOrErr) { 3030 if (auto E = isNotObjectErrorInvalidFileType(ChildOrErr.takeError())) 3031 reportError(std::move(E), getFileNameForError(C, I), A->getFileName()); 3032 continue; 3033 } 3034 if (ObjectFile *O = dyn_cast<ObjectFile>(&*ChildOrErr.get())) 3035 dumpObject(O, A, &C); 3036 else if (COFFImportFile *I = dyn_cast<COFFImportFile>(&*ChildOrErr.get())) 3037 dumpObject(I, A, &C); 3038 else 3039 reportError(errorCodeToError(object_error::invalid_file_type), 3040 A->getFileName()); 3041 } 3042 if (Err) 3043 reportError(std::move(Err), A->getFileName()); 3044 } 3045 3046 /// Open file and figure out how to dump it. 3047 static void dumpInput(StringRef file) { 3048 // If we are using the Mach-O specific object file parser, then let it parse 3049 // the file and process the command line options. So the -arch flags can 3050 // be used to select specific slices, etc. 3051 if (MachOOpt) { 3052 parseInputMachO(file); 3053 return; 3054 } 3055 3056 // Attempt to open the binary. 3057 OwningBinary<Binary> OBinary = unwrapOrError(createBinary(file), file); 3058 Binary &Binary = *OBinary.getBinary(); 3059 3060 if (Archive *A = dyn_cast<Archive>(&Binary)) 3061 dumpArchive(A); 3062 else if (ObjectFile *O = dyn_cast<ObjectFile>(&Binary)) 3063 dumpObject(O); 3064 else if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(&Binary)) 3065 parseInputMachO(UB); 3066 else if (OffloadBinary *OB = dyn_cast<OffloadBinary>(&Binary)) 3067 dumpOffloadSections(*OB); 3068 else 3069 reportError(errorCodeToError(object_error::invalid_file_type), file); 3070 } 3071 3072 template <typename T> 3073 static void parseIntArg(const llvm::opt::InputArgList &InputArgs, int ID, 3074 T &Value) { 3075 if (const opt::Arg *A = InputArgs.getLastArg(ID)) { 3076 StringRef V(A->getValue()); 3077 if (!llvm::to_integer(V, Value, 0)) { 3078 reportCmdLineError(A->getSpelling() + 3079 ": expected a non-negative integer, but got '" + V + 3080 "'"); 3081 } 3082 } 3083 } 3084 3085 static object::BuildID parseBuildIDArg(const opt::Arg *A) { 3086 StringRef V(A->getValue()); 3087 object::BuildID BID = parseBuildID(V); 3088 if (BID.empty()) 3089 reportCmdLineError(A->getSpelling() + ": expected a build ID, but got '" + 3090 V + "'"); 3091 return BID; 3092 } 3093 3094 void objdump::invalidArgValue(const opt::Arg *A) { 3095 reportCmdLineError("'" + StringRef(A->getValue()) + 3096 "' is not a valid value for '" + A->getSpelling() + "'"); 3097 } 3098 3099 static std::vector<std::string> 3100 commaSeparatedValues(const llvm::opt::InputArgList &InputArgs, int ID) { 3101 std::vector<std::string> Values; 3102 for (StringRef Value : InputArgs.getAllArgValues(ID)) { 3103 llvm::SmallVector<StringRef, 2> SplitValues; 3104 llvm::SplitString(Value, SplitValues, ","); 3105 for (StringRef SplitValue : SplitValues) 3106 Values.push_back(SplitValue.str()); 3107 } 3108 return Values; 3109 } 3110 3111 static void parseOtoolOptions(const llvm::opt::InputArgList &InputArgs) { 3112 MachOOpt = true; 3113 FullLeadingAddr = true; 3114 PrintImmHex = true; 3115 3116 ArchName = InputArgs.getLastArgValue(OTOOL_arch).str(); 3117 LinkOptHints = InputArgs.hasArg(OTOOL_C); 3118 if (InputArgs.hasArg(OTOOL_d)) 3119 FilterSections.push_back("__DATA,__data"); 3120 DylibId = InputArgs.hasArg(OTOOL_D); 3121 UniversalHeaders = InputArgs.hasArg(OTOOL_f); 3122 DataInCode = InputArgs.hasArg(OTOOL_G); 3123 FirstPrivateHeader = InputArgs.hasArg(OTOOL_h); 3124 IndirectSymbols = InputArgs.hasArg(OTOOL_I); 3125 ShowRawInsn = InputArgs.hasArg(OTOOL_j); 3126 PrivateHeaders = InputArgs.hasArg(OTOOL_l); 3127 DylibsUsed = InputArgs.hasArg(OTOOL_L); 3128 MCPU = InputArgs.getLastArgValue(OTOOL_mcpu_EQ).str(); 3129 ObjcMetaData = InputArgs.hasArg(OTOOL_o); 3130 DisSymName = InputArgs.getLastArgValue(OTOOL_p).str(); 3131 InfoPlist = InputArgs.hasArg(OTOOL_P); 3132 Relocations = InputArgs.hasArg(OTOOL_r); 3133 if (const Arg *A = InputArgs.getLastArg(OTOOL_s)) { 3134 auto Filter = (A->getValue(0) + StringRef(",") + A->getValue(1)).str(); 3135 FilterSections.push_back(Filter); 3136 } 3137 if (InputArgs.hasArg(OTOOL_t)) 3138 FilterSections.push_back("__TEXT,__text"); 3139 Verbose = InputArgs.hasArg(OTOOL_v) || InputArgs.hasArg(OTOOL_V) || 3140 InputArgs.hasArg(OTOOL_o); 3141 SymbolicOperands = InputArgs.hasArg(OTOOL_V); 3142 if (InputArgs.hasArg(OTOOL_x)) 3143 FilterSections.push_back(",__text"); 3144 LeadingAddr = LeadingHeaders = !InputArgs.hasArg(OTOOL_X); 3145 3146 ChainedFixups = InputArgs.hasArg(OTOOL_chained_fixups); 3147 DyldInfo = InputArgs.hasArg(OTOOL_dyld_info); 3148 3149 InputFilenames = InputArgs.getAllArgValues(OTOOL_INPUT); 3150 if (InputFilenames.empty()) 3151 reportCmdLineError("no input file"); 3152 3153 for (const Arg *A : InputArgs) { 3154 const Option &O = A->getOption(); 3155 if (O.getGroup().isValid() && O.getGroup().getID() == OTOOL_grp_obsolete) { 3156 reportCmdLineWarning(O.getPrefixedName() + 3157 " is obsolete and not implemented"); 3158 } 3159 } 3160 } 3161 3162 static void parseObjdumpOptions(const llvm::opt::InputArgList &InputArgs) { 3163 parseIntArg(InputArgs, OBJDUMP_adjust_vma_EQ, AdjustVMA); 3164 AllHeaders = InputArgs.hasArg(OBJDUMP_all_headers); 3165 ArchName = InputArgs.getLastArgValue(OBJDUMP_arch_name_EQ).str(); 3166 ArchiveHeaders = InputArgs.hasArg(OBJDUMP_archive_headers); 3167 Demangle = InputArgs.hasArg(OBJDUMP_demangle); 3168 Disassemble = InputArgs.hasArg(OBJDUMP_disassemble); 3169 DisassembleAll = InputArgs.hasArg(OBJDUMP_disassemble_all); 3170 SymbolDescription = InputArgs.hasArg(OBJDUMP_symbol_description); 3171 TracebackTable = InputArgs.hasArg(OBJDUMP_traceback_table); 3172 DisassembleSymbols = 3173 commaSeparatedValues(InputArgs, OBJDUMP_disassemble_symbols_EQ); 3174 DisassembleZeroes = InputArgs.hasArg(OBJDUMP_disassemble_zeroes); 3175 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_dwarf_EQ)) { 3176 DwarfDumpType = StringSwitch<DIDumpType>(A->getValue()) 3177 .Case("frames", DIDT_DebugFrame) 3178 .Default(DIDT_Null); 3179 if (DwarfDumpType == DIDT_Null) 3180 invalidArgValue(A); 3181 } 3182 DynamicRelocations = InputArgs.hasArg(OBJDUMP_dynamic_reloc); 3183 FaultMapSection = InputArgs.hasArg(OBJDUMP_fault_map_section); 3184 Offloading = InputArgs.hasArg(OBJDUMP_offloading); 3185 FileHeaders = InputArgs.hasArg(OBJDUMP_file_headers); 3186 SectionContents = InputArgs.hasArg(OBJDUMP_full_contents); 3187 PrintLines = InputArgs.hasArg(OBJDUMP_line_numbers); 3188 InputFilenames = InputArgs.getAllArgValues(OBJDUMP_INPUT); 3189 MachOOpt = InputArgs.hasArg(OBJDUMP_macho); 3190 MCPU = InputArgs.getLastArgValue(OBJDUMP_mcpu_EQ).str(); 3191 MAttrs = commaSeparatedValues(InputArgs, OBJDUMP_mattr_EQ); 3192 ShowRawInsn = !InputArgs.hasArg(OBJDUMP_no_show_raw_insn); 3193 LeadingAddr = !InputArgs.hasArg(OBJDUMP_no_leading_addr); 3194 RawClangAST = InputArgs.hasArg(OBJDUMP_raw_clang_ast); 3195 Relocations = InputArgs.hasArg(OBJDUMP_reloc); 3196 PrintImmHex = 3197 InputArgs.hasFlag(OBJDUMP_print_imm_hex, OBJDUMP_no_print_imm_hex, true); 3198 PrivateHeaders = InputArgs.hasArg(OBJDUMP_private_headers); 3199 FilterSections = InputArgs.getAllArgValues(OBJDUMP_section_EQ); 3200 SectionHeaders = InputArgs.hasArg(OBJDUMP_section_headers); 3201 ShowAllSymbols = InputArgs.hasArg(OBJDUMP_show_all_symbols); 3202 ShowLMA = InputArgs.hasArg(OBJDUMP_show_lma); 3203 PrintSource = InputArgs.hasArg(OBJDUMP_source); 3204 parseIntArg(InputArgs, OBJDUMP_start_address_EQ, StartAddress); 3205 HasStartAddressFlag = InputArgs.hasArg(OBJDUMP_start_address_EQ); 3206 parseIntArg(InputArgs, OBJDUMP_stop_address_EQ, StopAddress); 3207 HasStopAddressFlag = InputArgs.hasArg(OBJDUMP_stop_address_EQ); 3208 SymbolTable = InputArgs.hasArg(OBJDUMP_syms); 3209 SymbolizeOperands = InputArgs.hasArg(OBJDUMP_symbolize_operands); 3210 DynamicSymbolTable = InputArgs.hasArg(OBJDUMP_dynamic_syms); 3211 TripleName = InputArgs.getLastArgValue(OBJDUMP_triple_EQ).str(); 3212 UnwindInfo = InputArgs.hasArg(OBJDUMP_unwind_info); 3213 Wide = InputArgs.hasArg(OBJDUMP_wide); 3214 Prefix = InputArgs.getLastArgValue(OBJDUMP_prefix).str(); 3215 parseIntArg(InputArgs, OBJDUMP_prefix_strip, PrefixStrip); 3216 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_debug_vars_EQ)) { 3217 DbgVariables = StringSwitch<DebugVarsFormat>(A->getValue()) 3218 .Case("ascii", DVASCII) 3219 .Case("unicode", DVUnicode) 3220 .Default(DVInvalid); 3221 if (DbgVariables == DVInvalid) 3222 invalidArgValue(A); 3223 } 3224 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_disassembler_color_EQ)) { 3225 DisassemblyColor = StringSwitch<ColorOutput>(A->getValue()) 3226 .Case("on", ColorOutput::Enable) 3227 .Case("off", ColorOutput::Disable) 3228 .Case("terminal", ColorOutput::Auto) 3229 .Default(ColorOutput::Invalid); 3230 if (DisassemblyColor == ColorOutput::Invalid) 3231 invalidArgValue(A); 3232 } 3233 3234 parseIntArg(InputArgs, OBJDUMP_debug_vars_indent_EQ, DbgIndent); 3235 3236 parseMachOOptions(InputArgs); 3237 3238 // Parse -M (--disassembler-options) and deprecated 3239 // --x86-asm-syntax={att,intel}. 3240 // 3241 // Note, for x86, the asm dialect (AssemblerDialect) is initialized when the 3242 // MCAsmInfo is constructed. MCInstPrinter::applyTargetSpecificCLOption is 3243 // called too late. For now we have to use the internal cl::opt option. 3244 const char *AsmSyntax = nullptr; 3245 for (const auto *A : InputArgs.filtered(OBJDUMP_disassembler_options_EQ, 3246 OBJDUMP_x86_asm_syntax_att, 3247 OBJDUMP_x86_asm_syntax_intel)) { 3248 switch (A->getOption().getID()) { 3249 case OBJDUMP_x86_asm_syntax_att: 3250 AsmSyntax = "--x86-asm-syntax=att"; 3251 continue; 3252 case OBJDUMP_x86_asm_syntax_intel: 3253 AsmSyntax = "--x86-asm-syntax=intel"; 3254 continue; 3255 } 3256 3257 SmallVector<StringRef, 2> Values; 3258 llvm::SplitString(A->getValue(), Values, ","); 3259 for (StringRef V : Values) { 3260 if (V == "att") 3261 AsmSyntax = "--x86-asm-syntax=att"; 3262 else if (V == "intel") 3263 AsmSyntax = "--x86-asm-syntax=intel"; 3264 else 3265 DisassemblerOptions.push_back(V.str()); 3266 } 3267 } 3268 if (AsmSyntax) { 3269 const char *Argv[] = {"llvm-objdump", AsmSyntax}; 3270 llvm::cl::ParseCommandLineOptions(2, Argv); 3271 } 3272 3273 // Look up any provided build IDs, then append them to the input filenames. 3274 for (const opt::Arg *A : InputArgs.filtered(OBJDUMP_build_id)) { 3275 object::BuildID BuildID = parseBuildIDArg(A); 3276 std::optional<std::string> Path = BIDFetcher->fetch(BuildID); 3277 if (!Path) { 3278 reportCmdLineError(A->getSpelling() + ": could not find build ID '" + 3279 A->getValue() + "'"); 3280 } 3281 InputFilenames.push_back(std::move(*Path)); 3282 } 3283 3284 // objdump defaults to a.out if no filenames specified. 3285 if (InputFilenames.empty()) 3286 InputFilenames.push_back("a.out"); 3287 } 3288 3289 int llvm_objdump_main(int argc, char **argv, const llvm::ToolContext &) { 3290 using namespace llvm; 3291 InitLLVM X(argc, argv); 3292 3293 ToolName = argv[0]; 3294 std::unique_ptr<CommonOptTable> T; 3295 OptSpecifier Unknown, HelpFlag, HelpHiddenFlag, VersionFlag; 3296 3297 StringRef Stem = sys::path::stem(ToolName); 3298 auto Is = [=](StringRef Tool) { 3299 // We need to recognize the following filenames: 3300 // 3301 // llvm-objdump -> objdump 3302 // llvm-otool-10.exe -> otool 3303 // powerpc64-unknown-freebsd13-objdump -> objdump 3304 auto I = Stem.rfind_insensitive(Tool); 3305 return I != StringRef::npos && 3306 (I + Tool.size() == Stem.size() || !isAlnum(Stem[I + Tool.size()])); 3307 }; 3308 if (Is("otool")) { 3309 T = std::make_unique<OtoolOptTable>(); 3310 Unknown = OTOOL_UNKNOWN; 3311 HelpFlag = OTOOL_help; 3312 HelpHiddenFlag = OTOOL_help_hidden; 3313 VersionFlag = OTOOL_version; 3314 } else { 3315 T = std::make_unique<ObjdumpOptTable>(); 3316 Unknown = OBJDUMP_UNKNOWN; 3317 HelpFlag = OBJDUMP_help; 3318 HelpHiddenFlag = OBJDUMP_help_hidden; 3319 VersionFlag = OBJDUMP_version; 3320 } 3321 3322 BumpPtrAllocator A; 3323 StringSaver Saver(A); 3324 opt::InputArgList InputArgs = 3325 T->parseArgs(argc, argv, Unknown, Saver, 3326 [&](StringRef Msg) { reportCmdLineError(Msg); }); 3327 3328 if (InputArgs.size() == 0 || InputArgs.hasArg(HelpFlag)) { 3329 T->printHelp(ToolName); 3330 return 0; 3331 } 3332 if (InputArgs.hasArg(HelpHiddenFlag)) { 3333 T->printHelp(ToolName, /*ShowHidden=*/true); 3334 return 0; 3335 } 3336 3337 // Initialize targets and assembly printers/parsers. 3338 InitializeAllTargetInfos(); 3339 InitializeAllTargetMCs(); 3340 InitializeAllDisassemblers(); 3341 3342 if (InputArgs.hasArg(VersionFlag)) { 3343 cl::PrintVersionMessage(); 3344 if (!Is("otool")) { 3345 outs() << '\n'; 3346 TargetRegistry::printRegisteredTargetsForVersion(outs()); 3347 } 3348 return 0; 3349 } 3350 3351 // Initialize debuginfod. 3352 const bool ShouldUseDebuginfodByDefault = 3353 InputArgs.hasArg(OBJDUMP_build_id) || canUseDebuginfod(); 3354 std::vector<std::string> DebugFileDirectories = 3355 InputArgs.getAllArgValues(OBJDUMP_debug_file_directory); 3356 if (InputArgs.hasFlag(OBJDUMP_debuginfod, OBJDUMP_no_debuginfod, 3357 ShouldUseDebuginfodByDefault)) { 3358 HTTPClient::initialize(); 3359 BIDFetcher = 3360 std::make_unique<DebuginfodFetcher>(std::move(DebugFileDirectories)); 3361 } else { 3362 BIDFetcher = 3363 std::make_unique<BuildIDFetcher>(std::move(DebugFileDirectories)); 3364 } 3365 3366 if (Is("otool")) 3367 parseOtoolOptions(InputArgs); 3368 else 3369 parseObjdumpOptions(InputArgs); 3370 3371 if (StartAddress >= StopAddress) 3372 reportCmdLineError("start address should be less than stop address"); 3373 3374 // Removes trailing separators from prefix. 3375 while (!Prefix.empty() && sys::path::is_separator(Prefix.back())) 3376 Prefix.pop_back(); 3377 3378 if (AllHeaders) 3379 ArchiveHeaders = FileHeaders = PrivateHeaders = Relocations = 3380 SectionHeaders = SymbolTable = true; 3381 3382 if (DisassembleAll || PrintSource || PrintLines || TracebackTable || 3383 !DisassembleSymbols.empty()) 3384 Disassemble = true; 3385 3386 if (!ArchiveHeaders && !Disassemble && DwarfDumpType == DIDT_Null && 3387 !DynamicRelocations && !FileHeaders && !PrivateHeaders && !RawClangAST && 3388 !Relocations && !SectionHeaders && !SectionContents && !SymbolTable && 3389 !DynamicSymbolTable && !UnwindInfo && !FaultMapSection && !Offloading && 3390 !(MachOOpt && 3391 (Bind || DataInCode || ChainedFixups || DyldInfo || DylibId || 3392 DylibsUsed || ExportsTrie || FirstPrivateHeader || 3393 FunctionStartsType != FunctionStartsMode::None || IndirectSymbols || 3394 InfoPlist || LazyBind || LinkOptHints || ObjcMetaData || Rebase || 3395 Rpaths || UniversalHeaders || WeakBind || !FilterSections.empty()))) { 3396 T->printHelp(ToolName); 3397 return 2; 3398 } 3399 3400 DisasmSymbolSet.insert(DisassembleSymbols.begin(), DisassembleSymbols.end()); 3401 3402 llvm::for_each(InputFilenames, dumpInput); 3403 3404 warnOnNoMatchForSections(); 3405 3406 return EXIT_SUCCESS; 3407 } 3408