1 //===-- llvm-objdump.cpp - Object file dumping utility for llvm -----------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This program is a utility that works like binutils "objdump", that is, it 10 // dumps out a plethora of information about an object file depending on the 11 // flags. 12 // 13 // The flags and output of this program should be near identical to those of 14 // binutils objdump. 15 // 16 //===----------------------------------------------------------------------===// 17 18 #include "llvm-objdump.h" 19 #include "COFFDump.h" 20 #include "ELFDump.h" 21 #include "MachODump.h" 22 #include "ObjdumpOptID.h" 23 #include "OffloadDump.h" 24 #include "SourcePrinter.h" 25 #include "WasmDump.h" 26 #include "XCOFFDump.h" 27 #include "llvm/ADT/IndexedMap.h" 28 #include "llvm/ADT/Optional.h" 29 #include "llvm/ADT/STLExtras.h" 30 #include "llvm/ADT/SetOperations.h" 31 #include "llvm/ADT/SmallSet.h" 32 #include "llvm/ADT/StringExtras.h" 33 #include "llvm/ADT/StringSet.h" 34 #include "llvm/ADT/Triple.h" 35 #include "llvm/ADT/Twine.h" 36 #include "llvm/DebugInfo/DWARF/DWARFContext.h" 37 #include "llvm/DebugInfo/Symbolize/SymbolizableModule.h" 38 #include "llvm/DebugInfo/Symbolize/Symbolize.h" 39 #include "llvm/Debuginfod/BuildIDFetcher.h" 40 #include "llvm/Debuginfod/Debuginfod.h" 41 #include "llvm/Debuginfod/HTTPClient.h" 42 #include "llvm/Demangle/Demangle.h" 43 #include "llvm/MC/MCAsmInfo.h" 44 #include "llvm/MC/MCContext.h" 45 #include "llvm/MC/MCDisassembler/MCDisassembler.h" 46 #include "llvm/MC/MCDisassembler/MCRelocationInfo.h" 47 #include "llvm/MC/MCInst.h" 48 #include "llvm/MC/MCInstPrinter.h" 49 #include "llvm/MC/MCInstrAnalysis.h" 50 #include "llvm/MC/MCInstrInfo.h" 51 #include "llvm/MC/MCObjectFileInfo.h" 52 #include "llvm/MC/MCRegisterInfo.h" 53 #include "llvm/MC/MCSubtargetInfo.h" 54 #include "llvm/MC/MCTargetOptions.h" 55 #include "llvm/MC/TargetRegistry.h" 56 #include "llvm/Object/Archive.h" 57 #include "llvm/Object/BuildID.h" 58 #include "llvm/Object/COFF.h" 59 #include "llvm/Object/COFFImportFile.h" 60 #include "llvm/Object/ELFObjectFile.h" 61 #include "llvm/Object/ELFTypes.h" 62 #include "llvm/Object/FaultMapParser.h" 63 #include "llvm/Object/MachO.h" 64 #include "llvm/Object/MachOUniversal.h" 65 #include "llvm/Object/ObjectFile.h" 66 #include "llvm/Object/OffloadBinary.h" 67 #include "llvm/Object/Wasm.h" 68 #include "llvm/Option/Arg.h" 69 #include "llvm/Option/ArgList.h" 70 #include "llvm/Option/Option.h" 71 #include "llvm/Support/Casting.h" 72 #include "llvm/Support/Debug.h" 73 #include "llvm/Support/Errc.h" 74 #include "llvm/Support/FileSystem.h" 75 #include "llvm/Support/Format.h" 76 #include "llvm/Support/FormatVariadic.h" 77 #include "llvm/Support/GraphWriter.h" 78 #include "llvm/Support/Host.h" 79 #include "llvm/Support/InitLLVM.h" 80 #include "llvm/Support/MemoryBuffer.h" 81 #include "llvm/Support/SourceMgr.h" 82 #include "llvm/Support/StringSaver.h" 83 #include "llvm/Support/TargetSelect.h" 84 #include "llvm/Support/WithColor.h" 85 #include "llvm/Support/raw_ostream.h" 86 #include <algorithm> 87 #include <cctype> 88 #include <cstring> 89 #include <optional> 90 #include <system_error> 91 #include <unordered_map> 92 #include <utility> 93 94 using namespace llvm; 95 using namespace llvm::object; 96 using namespace llvm::objdump; 97 using namespace llvm::opt; 98 99 namespace { 100 101 class CommonOptTable : public opt::OptTable { 102 public: 103 CommonOptTable(ArrayRef<Info> OptionInfos, const char *Usage, 104 const char *Description) 105 : OptTable(OptionInfos), Usage(Usage), Description(Description) { 106 setGroupedShortOptions(true); 107 } 108 109 void printHelp(StringRef Argv0, bool ShowHidden = false) const { 110 Argv0 = sys::path::filename(Argv0); 111 opt::OptTable::printHelp(outs(), (Argv0 + Usage).str().c_str(), Description, 112 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 123 124 #define PREFIX(NAME, VALUE) const char *const OBJDUMP_##NAME[] = VALUE; 125 #include "ObjdumpOpts.inc" 126 #undef PREFIX 127 128 static constexpr opt::OptTable::Info ObjdumpInfoTable[] = { 129 #define OBJDUMP_nullptr nullptr 130 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 131 HELPTEXT, METAVAR, VALUES) \ 132 {OBJDUMP_##PREFIX, NAME, HELPTEXT, \ 133 METAVAR, OBJDUMP_##ID, opt::Option::KIND##Class, \ 134 PARAM, FLAGS, OBJDUMP_##GROUP, \ 135 OBJDUMP_##ALIAS, ALIASARGS, VALUES}, 136 #include "ObjdumpOpts.inc" 137 #undef OPTION 138 #undef OBJDUMP_nullptr 139 }; 140 141 class ObjdumpOptTable : public CommonOptTable { 142 public: 143 ObjdumpOptTable() 144 : CommonOptTable(ObjdumpInfoTable, " [options] <input object files>", 145 "llvm object file dumper") {} 146 }; 147 148 enum OtoolOptID { 149 OTOOL_INVALID = 0, // This is not an option ID. 150 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 151 HELPTEXT, METAVAR, VALUES) \ 152 OTOOL_##ID, 153 #include "OtoolOpts.inc" 154 #undef OPTION 155 }; 156 157 #define PREFIX(NAME, VALUE) const char *const OTOOL_##NAME[] = VALUE; 158 #include "OtoolOpts.inc" 159 #undef PREFIX 160 161 static constexpr opt::OptTable::Info OtoolInfoTable[] = { 162 #define OTOOL_nullptr nullptr 163 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 164 HELPTEXT, METAVAR, VALUES) \ 165 {OTOOL_##PREFIX, NAME, HELPTEXT, \ 166 METAVAR, OTOOL_##ID, opt::Option::KIND##Class, \ 167 PARAM, FLAGS, OTOOL_##GROUP, \ 168 OTOOL_##ALIAS, ALIASARGS, VALUES}, 169 #include "OtoolOpts.inc" 170 #undef OPTION 171 #undef OTOOL_nullptr 172 }; 173 174 class OtoolOptTable : public CommonOptTable { 175 public: 176 OtoolOptTable() 177 : CommonOptTable(OtoolInfoTable, " [option...] [file...]", 178 "Mach-O object file displaying tool") {} 179 }; 180 181 } // namespace 182 183 #define DEBUG_TYPE "objdump" 184 185 static uint64_t AdjustVMA; 186 static bool AllHeaders; 187 static std::string ArchName; 188 bool objdump::ArchiveHeaders; 189 bool objdump::Demangle; 190 bool objdump::Disassemble; 191 bool objdump::DisassembleAll; 192 bool objdump::SymbolDescription; 193 static std::vector<std::string> DisassembleSymbols; 194 static bool DisassembleZeroes; 195 static std::vector<std::string> DisassemblerOptions; 196 DIDumpType objdump::DwarfDumpType; 197 static bool DynamicRelocations; 198 static bool FaultMapSection; 199 static bool FileHeaders; 200 bool objdump::SectionContents; 201 static std::vector<std::string> InputFilenames; 202 bool objdump::PrintLines; 203 static bool MachOOpt; 204 std::string objdump::MCPU; 205 std::vector<std::string> objdump::MAttrs; 206 bool objdump::ShowRawInsn; 207 bool objdump::LeadingAddr; 208 static bool Offloading; 209 static bool RawClangAST; 210 bool objdump::Relocations; 211 bool objdump::PrintImmHex; 212 bool objdump::PrivateHeaders; 213 std::vector<std::string> objdump::FilterSections; 214 bool objdump::SectionHeaders; 215 static bool ShowAllSymbols; 216 static bool ShowLMA; 217 bool objdump::PrintSource; 218 219 static uint64_t StartAddress; 220 static bool HasStartAddressFlag; 221 static uint64_t StopAddress = UINT64_MAX; 222 static bool HasStopAddressFlag; 223 224 bool objdump::SymbolTable; 225 static bool SymbolizeOperands; 226 static bool DynamicSymbolTable; 227 std::string objdump::TripleName; 228 bool objdump::UnwindInfo; 229 static bool Wide; 230 std::string objdump::Prefix; 231 uint32_t objdump::PrefixStrip; 232 233 DebugVarsFormat objdump::DbgVariables = DVDisabled; 234 235 int objdump::DbgIndent = 52; 236 237 static StringSet<> DisasmSymbolSet; 238 StringSet<> objdump::FoundSectionSet; 239 static StringRef ToolName; 240 241 std::unique_ptr<BuildIDFetcher> BIDFetcher; 242 ExitOnError ExitOnErr; 243 244 namespace { 245 struct FilterResult { 246 // True if the section should not be skipped. 247 bool Keep; 248 249 // True if the index counter should be incremented, even if the section should 250 // be skipped. For example, sections may be skipped if they are not included 251 // in the --section flag, but we still want those to count toward the section 252 // count. 253 bool IncrementIndex; 254 }; 255 } // namespace 256 257 static FilterResult checkSectionFilter(object::SectionRef S) { 258 if (FilterSections.empty()) 259 return {/*Keep=*/true, /*IncrementIndex=*/true}; 260 261 Expected<StringRef> SecNameOrErr = S.getName(); 262 if (!SecNameOrErr) { 263 consumeError(SecNameOrErr.takeError()); 264 return {/*Keep=*/false, /*IncrementIndex=*/false}; 265 } 266 StringRef SecName = *SecNameOrErr; 267 268 // StringSet does not allow empty key so avoid adding sections with 269 // no name (such as the section with index 0) here. 270 if (!SecName.empty()) 271 FoundSectionSet.insert(SecName); 272 273 // Only show the section if it's in the FilterSections list, but always 274 // increment so the indexing is stable. 275 return {/*Keep=*/is_contained(FilterSections, SecName), 276 /*IncrementIndex=*/true}; 277 } 278 279 SectionFilter objdump::ToolSectionFilter(object::ObjectFile const &O, 280 uint64_t *Idx) { 281 // Start at UINT64_MAX so that the first index returned after an increment is 282 // zero (after the unsigned wrap). 283 if (Idx) 284 *Idx = UINT64_MAX; 285 return SectionFilter( 286 [Idx](object::SectionRef S) { 287 FilterResult Result = checkSectionFilter(S); 288 if (Idx != nullptr && Result.IncrementIndex) 289 *Idx += 1; 290 return Result.Keep; 291 }, 292 O); 293 } 294 295 std::string objdump::getFileNameForError(const object::Archive::Child &C, 296 unsigned Index) { 297 Expected<StringRef> NameOrErr = C.getName(); 298 if (NameOrErr) 299 return std::string(NameOrErr.get()); 300 // If we have an error getting the name then we print the index of the archive 301 // member. Since we are already in an error state, we just ignore this error. 302 consumeError(NameOrErr.takeError()); 303 return "<file index: " + std::to_string(Index) + ">"; 304 } 305 306 void objdump::reportWarning(const Twine &Message, StringRef File) { 307 // Output order between errs() and outs() matters especially for archive 308 // files where the output is per member object. 309 outs().flush(); 310 WithColor::warning(errs(), ToolName) 311 << "'" << File << "': " << Message << "\n"; 312 } 313 314 [[noreturn]] void objdump::reportError(StringRef File, const Twine &Message) { 315 outs().flush(); 316 WithColor::error(errs(), ToolName) << "'" << File << "': " << Message << "\n"; 317 exit(1); 318 } 319 320 [[noreturn]] void objdump::reportError(Error E, StringRef FileName, 321 StringRef ArchiveName, 322 StringRef ArchitectureName) { 323 assert(E); 324 outs().flush(); 325 WithColor::error(errs(), ToolName); 326 if (ArchiveName != "") 327 errs() << ArchiveName << "(" << FileName << ")"; 328 else 329 errs() << "'" << FileName << "'"; 330 if (!ArchitectureName.empty()) 331 errs() << " (for architecture " << ArchitectureName << ")"; 332 errs() << ": "; 333 logAllUnhandledErrors(std::move(E), errs()); 334 exit(1); 335 } 336 337 static void reportCmdLineWarning(const Twine &Message) { 338 WithColor::warning(errs(), ToolName) << Message << "\n"; 339 } 340 341 [[noreturn]] static void reportCmdLineError(const Twine &Message) { 342 WithColor::error(errs(), ToolName) << Message << "\n"; 343 exit(1); 344 } 345 346 static void warnOnNoMatchForSections() { 347 SetVector<StringRef> MissingSections; 348 for (StringRef S : FilterSections) { 349 if (FoundSectionSet.count(S)) 350 return; 351 // User may specify a unnamed section. Don't warn for it. 352 if (!S.empty()) 353 MissingSections.insert(S); 354 } 355 356 // Warn only if no section in FilterSections is matched. 357 for (StringRef S : MissingSections) 358 reportCmdLineWarning("section '" + S + 359 "' mentioned in a -j/--section option, but not " 360 "found in any input file"); 361 } 362 363 static const Target *getTarget(const ObjectFile *Obj) { 364 // Figure out the target triple. 365 Triple TheTriple("unknown-unknown-unknown"); 366 if (TripleName.empty()) { 367 TheTriple = Obj->makeTriple(); 368 } else { 369 TheTriple.setTriple(Triple::normalize(TripleName)); 370 auto Arch = Obj->getArch(); 371 if (Arch == Triple::arm || Arch == Triple::armeb) 372 Obj->setARMSubArch(TheTriple); 373 } 374 375 // Get the target specific parser. 376 std::string Error; 377 const Target *TheTarget = TargetRegistry::lookupTarget(ArchName, TheTriple, 378 Error); 379 if (!TheTarget) 380 reportError(Obj->getFileName(), "can't find target: " + Error); 381 382 // Update the triple name and return the found target. 383 TripleName = TheTriple.getTriple(); 384 return TheTarget; 385 } 386 387 bool objdump::isRelocAddressLess(RelocationRef A, RelocationRef B) { 388 return A.getOffset() < B.getOffset(); 389 } 390 391 static Error getRelocationValueString(const RelocationRef &Rel, 392 SmallVectorImpl<char> &Result) { 393 const ObjectFile *Obj = Rel.getObject(); 394 if (auto *ELF = dyn_cast<ELFObjectFileBase>(Obj)) 395 return getELFRelocationValueString(ELF, Rel, Result); 396 if (auto *COFF = dyn_cast<COFFObjectFile>(Obj)) 397 return getCOFFRelocationValueString(COFF, Rel, Result); 398 if (auto *Wasm = dyn_cast<WasmObjectFile>(Obj)) 399 return getWasmRelocationValueString(Wasm, Rel, Result); 400 if (auto *MachO = dyn_cast<MachOObjectFile>(Obj)) 401 return getMachORelocationValueString(MachO, Rel, Result); 402 if (auto *XCOFF = dyn_cast<XCOFFObjectFile>(Obj)) 403 return getXCOFFRelocationValueString(*XCOFF, Rel, Result); 404 llvm_unreachable("unknown object file format"); 405 } 406 407 /// Indicates whether this relocation should hidden when listing 408 /// relocations, usually because it is the trailing part of a multipart 409 /// relocation that will be printed as part of the leading relocation. 410 static bool getHidden(RelocationRef RelRef) { 411 auto *MachO = dyn_cast<MachOObjectFile>(RelRef.getObject()); 412 if (!MachO) 413 return false; 414 415 unsigned Arch = MachO->getArch(); 416 DataRefImpl Rel = RelRef.getRawDataRefImpl(); 417 uint64_t Type = MachO->getRelocationType(Rel); 418 419 // On arches that use the generic relocations, GENERIC_RELOC_PAIR 420 // is always hidden. 421 if (Arch == Triple::x86 || Arch == Triple::arm || Arch == Triple::ppc) 422 return Type == MachO::GENERIC_RELOC_PAIR; 423 424 if (Arch == Triple::x86_64) { 425 // On x86_64, X86_64_RELOC_UNSIGNED is hidden only when it follows 426 // an X86_64_RELOC_SUBTRACTOR. 427 if (Type == MachO::X86_64_RELOC_UNSIGNED && Rel.d.a > 0) { 428 DataRefImpl RelPrev = Rel; 429 RelPrev.d.a--; 430 uint64_t PrevType = MachO->getRelocationType(RelPrev); 431 if (PrevType == MachO::X86_64_RELOC_SUBTRACTOR) 432 return true; 433 } 434 } 435 436 return false; 437 } 438 439 namespace { 440 441 /// Get the column at which we want to start printing the instruction 442 /// disassembly, taking into account anything which appears to the left of it. 443 unsigned getInstStartColumn(const MCSubtargetInfo &STI) { 444 return !ShowRawInsn ? 16 : STI.getTargetTriple().isX86() ? 40 : 24; 445 } 446 447 static bool isAArch64Elf(const ObjectFile &Obj) { 448 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 449 return Elf && Elf->getEMachine() == ELF::EM_AARCH64; 450 } 451 452 static bool isArmElf(const ObjectFile &Obj) { 453 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 454 return Elf && Elf->getEMachine() == ELF::EM_ARM; 455 } 456 457 static bool isCSKYElf(const ObjectFile &Obj) { 458 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 459 return Elf && Elf->getEMachine() == ELF::EM_CSKY; 460 } 461 462 static bool hasMappingSymbols(const ObjectFile &Obj) { 463 return isArmElf(Obj) || isAArch64Elf(Obj) || isCSKYElf(Obj) ; 464 } 465 466 static void printRelocation(formatted_raw_ostream &OS, StringRef FileName, 467 const RelocationRef &Rel, uint64_t Address, 468 bool Is64Bits) { 469 StringRef Fmt = Is64Bits ? "%016" PRIx64 ": " : "%08" PRIx64 ": "; 470 SmallString<16> Name; 471 SmallString<32> Val; 472 Rel.getTypeName(Name); 473 if (Error E = getRelocationValueString(Rel, Val)) 474 reportError(std::move(E), FileName); 475 OS << (Is64Bits || !LeadingAddr ? "\t\t" : "\t\t\t"); 476 if (LeadingAddr) 477 OS << format(Fmt.data(), Address); 478 OS << Name << "\t" << Val; 479 } 480 481 static void AlignToInstStartColumn(size_t Start, const MCSubtargetInfo &STI, 482 raw_ostream &OS) { 483 // The output of printInst starts with a tab. Print some spaces so that 484 // the tab has 1 column and advances to the target tab stop. 485 unsigned TabStop = getInstStartColumn(STI); 486 unsigned Column = OS.tell() - Start; 487 OS.indent(Column < TabStop - 1 ? TabStop - 1 - Column : 7 - Column % 8); 488 } 489 490 class PrettyPrinter { 491 public: 492 virtual ~PrettyPrinter() = default; 493 virtual void 494 printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 495 object::SectionedAddress Address, formatted_raw_ostream &OS, 496 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 497 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 498 LiveVariablePrinter &LVP) { 499 if (SP && (PrintSource || PrintLines)) 500 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 501 LVP.printBetweenInsts(OS, false); 502 503 size_t Start = OS.tell(); 504 if (LeadingAddr) 505 OS << format("%8" PRIx64 ":", Address.Address); 506 if (ShowRawInsn) { 507 OS << ' '; 508 dumpBytes(Bytes, OS); 509 } 510 511 AlignToInstStartColumn(Start, STI, OS); 512 513 if (MI) { 514 // See MCInstPrinter::printInst. On targets where a PC relative immediate 515 // is relative to the next instruction and the length of a MCInst is 516 // difficult to measure (x86), this is the address of the next 517 // instruction. 518 uint64_t Addr = 519 Address.Address + (STI.getTargetTriple().isX86() ? Bytes.size() : 0); 520 IP.printInst(MI, Addr, "", STI, OS); 521 } else 522 OS << "\t<unknown>"; 523 } 524 }; 525 PrettyPrinter PrettyPrinterInst; 526 527 class HexagonPrettyPrinter : public PrettyPrinter { 528 public: 529 void printLead(ArrayRef<uint8_t> Bytes, uint64_t Address, 530 formatted_raw_ostream &OS) { 531 uint32_t opcode = 532 (Bytes[3] << 24) | (Bytes[2] << 16) | (Bytes[1] << 8) | Bytes[0]; 533 if (LeadingAddr) 534 OS << format("%8" PRIx64 ":", Address); 535 if (ShowRawInsn) { 536 OS << "\t"; 537 dumpBytes(Bytes.slice(0, 4), OS); 538 OS << format("\t%08" PRIx32, opcode); 539 } 540 } 541 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 542 object::SectionedAddress Address, formatted_raw_ostream &OS, 543 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 544 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 545 LiveVariablePrinter &LVP) override { 546 if (SP && (PrintSource || PrintLines)) 547 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 548 if (!MI) { 549 printLead(Bytes, Address.Address, OS); 550 OS << " <unknown>"; 551 return; 552 } 553 std::string Buffer; 554 { 555 raw_string_ostream TempStream(Buffer); 556 IP.printInst(MI, Address.Address, "", STI, TempStream); 557 } 558 StringRef Contents(Buffer); 559 // Split off bundle attributes 560 auto PacketBundle = Contents.rsplit('\n'); 561 // Split off first instruction from the rest 562 auto HeadTail = PacketBundle.first.split('\n'); 563 auto Preamble = " { "; 564 auto Separator = ""; 565 566 // Hexagon's packets require relocations to be inline rather than 567 // clustered at the end of the packet. 568 std::vector<RelocationRef>::const_iterator RelCur = Rels->begin(); 569 std::vector<RelocationRef>::const_iterator RelEnd = Rels->end(); 570 auto PrintReloc = [&]() -> void { 571 while ((RelCur != RelEnd) && (RelCur->getOffset() <= Address.Address)) { 572 if (RelCur->getOffset() == Address.Address) { 573 printRelocation(OS, ObjectFilename, *RelCur, Address.Address, false); 574 return; 575 } 576 ++RelCur; 577 } 578 }; 579 580 while (!HeadTail.first.empty()) { 581 OS << Separator; 582 Separator = "\n"; 583 if (SP && (PrintSource || PrintLines)) 584 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 585 printLead(Bytes, Address.Address, OS); 586 OS << Preamble; 587 Preamble = " "; 588 StringRef Inst; 589 auto Duplex = HeadTail.first.split('\v'); 590 if (!Duplex.second.empty()) { 591 OS << Duplex.first; 592 OS << "; "; 593 Inst = Duplex.second; 594 } 595 else 596 Inst = HeadTail.first; 597 OS << Inst; 598 HeadTail = HeadTail.second.split('\n'); 599 if (HeadTail.first.empty()) 600 OS << " } " << PacketBundle.second; 601 PrintReloc(); 602 Bytes = Bytes.slice(4); 603 Address.Address += 4; 604 } 605 } 606 }; 607 HexagonPrettyPrinter HexagonPrettyPrinterInst; 608 609 class AMDGCNPrettyPrinter : public PrettyPrinter { 610 public: 611 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 612 object::SectionedAddress Address, formatted_raw_ostream &OS, 613 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 614 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 615 LiveVariablePrinter &LVP) override { 616 if (SP && (PrintSource || PrintLines)) 617 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 618 619 if (MI) { 620 SmallString<40> InstStr; 621 raw_svector_ostream IS(InstStr); 622 623 IP.printInst(MI, Address.Address, "", STI, IS); 624 625 OS << left_justify(IS.str(), 60); 626 } else { 627 // an unrecognized encoding - this is probably data so represent it 628 // using the .long directive, or .byte directive if fewer than 4 bytes 629 // remaining 630 if (Bytes.size() >= 4) { 631 OS << format("\t.long 0x%08" PRIx32 " ", 632 support::endian::read32<support::little>(Bytes.data())); 633 OS.indent(42); 634 } else { 635 OS << format("\t.byte 0x%02" PRIx8, Bytes[0]); 636 for (unsigned int i = 1; i < Bytes.size(); i++) 637 OS << format(", 0x%02" PRIx8, Bytes[i]); 638 OS.indent(55 - (6 * Bytes.size())); 639 } 640 } 641 642 OS << format("// %012" PRIX64 ":", Address.Address); 643 if (Bytes.size() >= 4) { 644 // D should be casted to uint32_t here as it is passed by format to 645 // snprintf as vararg. 646 for (uint32_t D : makeArrayRef( 647 reinterpret_cast<const support::little32_t *>(Bytes.data()), 648 Bytes.size() / 4)) 649 OS << format(" %08" PRIX32, D); 650 } else { 651 for (unsigned char B : Bytes) 652 OS << format(" %02" PRIX8, B); 653 } 654 655 if (!Annot.empty()) 656 OS << " // " << Annot; 657 } 658 }; 659 AMDGCNPrettyPrinter AMDGCNPrettyPrinterInst; 660 661 class BPFPrettyPrinter : public PrettyPrinter { 662 public: 663 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 664 object::SectionedAddress Address, formatted_raw_ostream &OS, 665 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 666 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 667 LiveVariablePrinter &LVP) override { 668 if (SP && (PrintSource || PrintLines)) 669 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 670 if (LeadingAddr) 671 OS << format("%8" PRId64 ":", Address.Address / 8); 672 if (ShowRawInsn) { 673 OS << "\t"; 674 dumpBytes(Bytes, OS); 675 } 676 if (MI) 677 IP.printInst(MI, Address.Address, "", STI, OS); 678 else 679 OS << "\t<unknown>"; 680 } 681 }; 682 BPFPrettyPrinter BPFPrettyPrinterInst; 683 684 class ARMPrettyPrinter : public PrettyPrinter { 685 public: 686 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 687 object::SectionedAddress Address, formatted_raw_ostream &OS, 688 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 689 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 690 LiveVariablePrinter &LVP) override { 691 if (SP && (PrintSource || PrintLines)) 692 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 693 LVP.printBetweenInsts(OS, false); 694 695 size_t Start = OS.tell(); 696 if (LeadingAddr) 697 OS << format("%8" PRIx64 ":", Address.Address); 698 if (ShowRawInsn) { 699 size_t Pos = 0, End = Bytes.size(); 700 if (STI.checkFeatures("+thumb-mode")) { 701 for (; Pos + 2 <= End; Pos += 2) 702 OS << ' ' 703 << format_hex_no_prefix( 704 llvm::support::endian::read<uint16_t>( 705 Bytes.data() + Pos, InstructionEndianness), 706 4); 707 } else { 708 for (; Pos + 4 <= End; Pos += 4) 709 OS << ' ' 710 << format_hex_no_prefix( 711 llvm::support::endian::read<uint32_t>( 712 Bytes.data() + Pos, InstructionEndianness), 713 8); 714 } 715 if (Pos < End) { 716 OS << ' '; 717 dumpBytes(Bytes.slice(Pos), OS); 718 } 719 } 720 721 AlignToInstStartColumn(Start, STI, OS); 722 723 if (MI) { 724 IP.printInst(MI, Address.Address, "", STI, OS); 725 } else 726 OS << "\t<unknown>"; 727 } 728 729 void setInstructionEndianness(llvm::support::endianness Endianness) { 730 InstructionEndianness = Endianness; 731 } 732 733 private: 734 llvm::support::endianness InstructionEndianness = llvm::support::little; 735 }; 736 ARMPrettyPrinter ARMPrettyPrinterInst; 737 738 class AArch64PrettyPrinter : public PrettyPrinter { 739 public: 740 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 741 object::SectionedAddress Address, formatted_raw_ostream &OS, 742 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 743 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 744 LiveVariablePrinter &LVP) override { 745 if (SP && (PrintSource || PrintLines)) 746 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 747 LVP.printBetweenInsts(OS, false); 748 749 size_t Start = OS.tell(); 750 if (LeadingAddr) 751 OS << format("%8" PRIx64 ":", Address.Address); 752 if (ShowRawInsn) { 753 size_t Pos = 0, End = Bytes.size(); 754 for (; Pos + 4 <= End; Pos += 4) 755 OS << ' ' 756 << format_hex_no_prefix( 757 llvm::support::endian::read<uint32_t>(Bytes.data() + Pos, 758 llvm::support::little), 759 8); 760 if (Pos < End) { 761 OS << ' '; 762 dumpBytes(Bytes.slice(Pos), OS); 763 } 764 } 765 766 AlignToInstStartColumn(Start, STI, OS); 767 768 if (MI) { 769 IP.printInst(MI, Address.Address, "", STI, OS); 770 } else 771 OS << "\t<unknown>"; 772 } 773 }; 774 AArch64PrettyPrinter AArch64PrettyPrinterInst; 775 776 PrettyPrinter &selectPrettyPrinter(Triple const &Triple) { 777 switch(Triple.getArch()) { 778 default: 779 return PrettyPrinterInst; 780 case Triple::hexagon: 781 return HexagonPrettyPrinterInst; 782 case Triple::amdgcn: 783 return AMDGCNPrettyPrinterInst; 784 case Triple::bpfel: 785 case Triple::bpfeb: 786 return BPFPrettyPrinterInst; 787 case Triple::arm: 788 case Triple::armeb: 789 case Triple::thumb: 790 case Triple::thumbeb: 791 return ARMPrettyPrinterInst; 792 case Triple::aarch64: 793 case Triple::aarch64_be: 794 case Triple::aarch64_32: 795 return AArch64PrettyPrinterInst; 796 } 797 } 798 } 799 800 static uint8_t getElfSymbolType(const ObjectFile &Obj, const SymbolRef &Sym) { 801 assert(Obj.isELF()); 802 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 803 return unwrapOrError(Elf32LEObj->getSymbol(Sym.getRawDataRefImpl()), 804 Obj.getFileName()) 805 ->getType(); 806 if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 807 return unwrapOrError(Elf64LEObj->getSymbol(Sym.getRawDataRefImpl()), 808 Obj.getFileName()) 809 ->getType(); 810 if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 811 return unwrapOrError(Elf32BEObj->getSymbol(Sym.getRawDataRefImpl()), 812 Obj.getFileName()) 813 ->getType(); 814 if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 815 return unwrapOrError(Elf64BEObj->getSymbol(Sym.getRawDataRefImpl()), 816 Obj.getFileName()) 817 ->getType(); 818 llvm_unreachable("Unsupported binary format"); 819 } 820 821 template <class ELFT> 822 static void 823 addDynamicElfSymbols(const ELFObjectFile<ELFT> &Obj, 824 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 825 for (auto Symbol : Obj.getDynamicSymbolIterators()) { 826 uint8_t SymbolType = Symbol.getELFType(); 827 if (SymbolType == ELF::STT_SECTION) 828 continue; 829 830 uint64_t Address = unwrapOrError(Symbol.getAddress(), Obj.getFileName()); 831 // ELFSymbolRef::getAddress() returns size instead of value for common 832 // symbols which is not desirable for disassembly output. Overriding. 833 if (SymbolType == ELF::STT_COMMON) 834 Address = unwrapOrError(Obj.getSymbol(Symbol.getRawDataRefImpl()), 835 Obj.getFileName()) 836 ->st_value; 837 838 StringRef Name = unwrapOrError(Symbol.getName(), Obj.getFileName()); 839 if (Name.empty()) 840 continue; 841 842 section_iterator SecI = 843 unwrapOrError(Symbol.getSection(), Obj.getFileName()); 844 if (SecI == Obj.section_end()) 845 continue; 846 847 AllSymbols[*SecI].emplace_back(Address, Name, SymbolType); 848 } 849 } 850 851 static void 852 addDynamicElfSymbols(const ELFObjectFileBase &Obj, 853 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 854 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 855 addDynamicElfSymbols(*Elf32LEObj, AllSymbols); 856 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 857 addDynamicElfSymbols(*Elf64LEObj, AllSymbols); 858 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 859 addDynamicElfSymbols(*Elf32BEObj, AllSymbols); 860 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 861 addDynamicElfSymbols(*Elf64BEObj, AllSymbols); 862 else 863 llvm_unreachable("Unsupported binary format"); 864 } 865 866 static Optional<SectionRef> getWasmCodeSection(const WasmObjectFile &Obj) { 867 for (auto SecI : Obj.sections()) { 868 const WasmSection &Section = Obj.getWasmSection(SecI); 869 if (Section.Type == wasm::WASM_SEC_CODE) 870 return SecI; 871 } 872 return std::nullopt; 873 } 874 875 static void 876 addMissingWasmCodeSymbols(const WasmObjectFile &Obj, 877 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 878 Optional<SectionRef> Section = getWasmCodeSection(Obj); 879 if (!Section) 880 return; 881 SectionSymbolsTy &Symbols = AllSymbols[*Section]; 882 883 std::set<uint64_t> SymbolAddresses; 884 for (const auto &Sym : Symbols) 885 SymbolAddresses.insert(Sym.Addr); 886 887 for (const wasm::WasmFunction &Function : Obj.functions()) { 888 uint64_t Address = Function.CodeSectionOffset; 889 // Only add fallback symbols for functions not already present in the symbol 890 // table. 891 if (SymbolAddresses.count(Address)) 892 continue; 893 // This function has no symbol, so it should have no SymbolName. 894 assert(Function.SymbolName.empty()); 895 // We use DebugName for the name, though it may be empty if there is no 896 // "name" custom section, or that section is missing a name for this 897 // function. 898 StringRef Name = Function.DebugName; 899 Symbols.emplace_back(Address, Name, ELF::STT_NOTYPE); 900 } 901 } 902 903 static void addPltEntries(const ObjectFile &Obj, 904 std::map<SectionRef, SectionSymbolsTy> &AllSymbols, 905 StringSaver &Saver) { 906 std::optional<SectionRef> Plt; 907 for (const SectionRef &Section : Obj.sections()) { 908 Expected<StringRef> SecNameOrErr = Section.getName(); 909 if (!SecNameOrErr) { 910 consumeError(SecNameOrErr.takeError()); 911 continue; 912 } 913 if (*SecNameOrErr == ".plt") 914 Plt = Section; 915 } 916 if (!Plt) 917 return; 918 if (auto *ElfObj = dyn_cast<ELFObjectFileBase>(&Obj)) { 919 for (auto PltEntry : ElfObj->getPltAddresses()) { 920 if (PltEntry.first) { 921 SymbolRef Symbol(*PltEntry.first, ElfObj); 922 uint8_t SymbolType = getElfSymbolType(Obj, Symbol); 923 if (Expected<StringRef> NameOrErr = Symbol.getName()) { 924 if (!NameOrErr->empty()) 925 AllSymbols[*Plt].emplace_back( 926 PltEntry.second, Saver.save((*NameOrErr + "@plt").str()), 927 SymbolType); 928 continue; 929 } else { 930 // The warning has been reported in disassembleObject(). 931 consumeError(NameOrErr.takeError()); 932 } 933 } 934 reportWarning("PLT entry at 0x" + Twine::utohexstr(PltEntry.second) + 935 " references an invalid symbol", 936 Obj.getFileName()); 937 } 938 } 939 } 940 941 // Normally the disassembly output will skip blocks of zeroes. This function 942 // returns the number of zero bytes that can be skipped when dumping the 943 // disassembly of the instructions in Buf. 944 static size_t countSkippableZeroBytes(ArrayRef<uint8_t> Buf) { 945 // Find the number of leading zeroes. 946 size_t N = 0; 947 while (N < Buf.size() && !Buf[N]) 948 ++N; 949 950 // We may want to skip blocks of zero bytes, but unless we see 951 // at least 8 of them in a row. 952 if (N < 8) 953 return 0; 954 955 // We skip zeroes in multiples of 4 because do not want to truncate an 956 // instruction if it starts with a zero byte. 957 return N & ~0x3; 958 } 959 960 // Returns a map from sections to their relocations. 961 static std::map<SectionRef, std::vector<RelocationRef>> 962 getRelocsMap(object::ObjectFile const &Obj) { 963 std::map<SectionRef, std::vector<RelocationRef>> Ret; 964 uint64_t I = (uint64_t)-1; 965 for (SectionRef Sec : Obj.sections()) { 966 ++I; 967 Expected<section_iterator> RelocatedOrErr = Sec.getRelocatedSection(); 968 if (!RelocatedOrErr) 969 reportError(Obj.getFileName(), 970 "section (" + Twine(I) + 971 "): failed to get a relocated section: " + 972 toString(RelocatedOrErr.takeError())); 973 974 section_iterator Relocated = *RelocatedOrErr; 975 if (Relocated == Obj.section_end() || !checkSectionFilter(*Relocated).Keep) 976 continue; 977 std::vector<RelocationRef> &V = Ret[*Relocated]; 978 append_range(V, Sec.relocations()); 979 // Sort relocations by address. 980 llvm::stable_sort(V, isRelocAddressLess); 981 } 982 return Ret; 983 } 984 985 // Used for --adjust-vma to check if address should be adjusted by the 986 // specified value for a given section. 987 // For ELF we do not adjust non-allocatable sections like debug ones, 988 // because they are not loadable. 989 // TODO: implement for other file formats. 990 static bool shouldAdjustVA(const SectionRef &Section) { 991 const ObjectFile *Obj = Section.getObject(); 992 if (Obj->isELF()) 993 return ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC; 994 return false; 995 } 996 997 998 typedef std::pair<uint64_t, char> MappingSymbolPair; 999 static char getMappingSymbolKind(ArrayRef<MappingSymbolPair> MappingSymbols, 1000 uint64_t Address) { 1001 auto It = 1002 partition_point(MappingSymbols, [Address](const MappingSymbolPair &Val) { 1003 return Val.first <= Address; 1004 }); 1005 // Return zero for any address before the first mapping symbol; this means 1006 // we should use the default disassembly mode, depending on the target. 1007 if (It == MappingSymbols.begin()) 1008 return '\x00'; 1009 return (It - 1)->second; 1010 } 1011 1012 static uint64_t dumpARMELFData(uint64_t SectionAddr, uint64_t Index, 1013 uint64_t End, const ObjectFile &Obj, 1014 ArrayRef<uint8_t> Bytes, 1015 ArrayRef<MappingSymbolPair> MappingSymbols, 1016 const MCSubtargetInfo &STI, raw_ostream &OS) { 1017 support::endianness Endian = 1018 Obj.isLittleEndian() ? support::little : support::big; 1019 size_t Start = OS.tell(); 1020 OS << format("%8" PRIx64 ": ", SectionAddr + Index); 1021 if (Index + 4 <= End) { 1022 dumpBytes(Bytes.slice(Index, 4), OS); 1023 AlignToInstStartColumn(Start, STI, OS); 1024 OS << "\t.word\t" 1025 << format_hex(support::endian::read32(Bytes.data() + Index, Endian), 1026 10); 1027 return 4; 1028 } 1029 if (Index + 2 <= End) { 1030 dumpBytes(Bytes.slice(Index, 2), OS); 1031 AlignToInstStartColumn(Start, STI, OS); 1032 OS << "\t.short\t" 1033 << format_hex(support::endian::read16(Bytes.data() + Index, Endian), 6); 1034 return 2; 1035 } 1036 dumpBytes(Bytes.slice(Index, 1), OS); 1037 AlignToInstStartColumn(Start, STI, OS); 1038 OS << "\t.byte\t" << format_hex(Bytes[Index], 4); 1039 return 1; 1040 } 1041 1042 static void dumpELFData(uint64_t SectionAddr, uint64_t Index, uint64_t End, 1043 ArrayRef<uint8_t> Bytes) { 1044 // print out data up to 8 bytes at a time in hex and ascii 1045 uint8_t AsciiData[9] = {'\0'}; 1046 uint8_t Byte; 1047 int NumBytes = 0; 1048 1049 for (; Index < End; ++Index) { 1050 if (NumBytes == 0) 1051 outs() << format("%8" PRIx64 ":", SectionAddr + Index); 1052 Byte = Bytes.slice(Index)[0]; 1053 outs() << format(" %02x", Byte); 1054 AsciiData[NumBytes] = isPrint(Byte) ? Byte : '.'; 1055 1056 uint8_t IndentOffset = 0; 1057 NumBytes++; 1058 if (Index == End - 1 || NumBytes > 8) { 1059 // Indent the space for less than 8 bytes data. 1060 // 2 spaces for byte and one for space between bytes 1061 IndentOffset = 3 * (8 - NumBytes); 1062 for (int Excess = NumBytes; Excess < 8; Excess++) 1063 AsciiData[Excess] = '\0'; 1064 NumBytes = 8; 1065 } 1066 if (NumBytes == 8) { 1067 AsciiData[8] = '\0'; 1068 outs() << std::string(IndentOffset, ' ') << " "; 1069 outs() << reinterpret_cast<char *>(AsciiData); 1070 outs() << '\n'; 1071 NumBytes = 0; 1072 } 1073 } 1074 } 1075 1076 SymbolInfoTy objdump::createSymbolInfo(const ObjectFile &Obj, 1077 const SymbolRef &Symbol) { 1078 const StringRef FileName = Obj.getFileName(); 1079 const uint64_t Addr = unwrapOrError(Symbol.getAddress(), FileName); 1080 const StringRef Name = unwrapOrError(Symbol.getName(), FileName); 1081 1082 if (Obj.isXCOFF() && SymbolDescription) { 1083 const auto &XCOFFObj = cast<XCOFFObjectFile>(Obj); 1084 DataRefImpl SymbolDRI = Symbol.getRawDataRefImpl(); 1085 1086 const uint32_t SymbolIndex = XCOFFObj.getSymbolIndex(SymbolDRI.p); 1087 Optional<XCOFF::StorageMappingClass> Smc = 1088 getXCOFFSymbolCsectSMC(XCOFFObj, Symbol); 1089 return SymbolInfoTy(Addr, Name, Smc, SymbolIndex, 1090 isLabel(XCOFFObj, Symbol)); 1091 } else if (Obj.isXCOFF()) { 1092 const SymbolRef::Type SymType = unwrapOrError(Symbol.getType(), FileName); 1093 return SymbolInfoTy(Addr, Name, SymType, true); 1094 } else 1095 return SymbolInfoTy(Addr, Name, 1096 Obj.isELF() ? getElfSymbolType(Obj, Symbol) 1097 : (uint8_t)ELF::STT_NOTYPE); 1098 } 1099 1100 static SymbolInfoTy createDummySymbolInfo(const ObjectFile &Obj, 1101 const uint64_t Addr, StringRef &Name, 1102 uint8_t Type) { 1103 if (Obj.isXCOFF() && SymbolDescription) 1104 return SymbolInfoTy(Addr, Name, std::nullopt, std::nullopt, false); 1105 else 1106 return SymbolInfoTy(Addr, Name, Type); 1107 } 1108 1109 static void 1110 collectBBAddrMapLabels(const std::unordered_map<uint64_t, BBAddrMap> &AddrToBBAddrMap, 1111 uint64_t SectionAddr, uint64_t Start, uint64_t End, 1112 std::unordered_map<uint64_t, std::vector<std::string>> &Labels) { 1113 if (AddrToBBAddrMap.empty()) 1114 return; 1115 Labels.clear(); 1116 uint64_t StartAddress = SectionAddr + Start; 1117 uint64_t EndAddress = SectionAddr + End; 1118 auto Iter = AddrToBBAddrMap.find(StartAddress); 1119 if (Iter == AddrToBBAddrMap.end()) 1120 return; 1121 for (unsigned I = 0, Size = Iter->second.BBEntries.size(); I < Size; ++I) { 1122 uint64_t BBAddress = Iter->second.BBEntries[I].Offset + Iter->second.Addr; 1123 if (BBAddress >= EndAddress) 1124 continue; 1125 Labels[BBAddress].push_back(("BB" + Twine(I)).str()); 1126 } 1127 } 1128 1129 static void collectLocalBranchTargets( 1130 ArrayRef<uint8_t> Bytes, const MCInstrAnalysis *MIA, MCDisassembler *DisAsm, 1131 MCInstPrinter *IP, const MCSubtargetInfo *STI, uint64_t SectionAddr, 1132 uint64_t Start, uint64_t End, std::unordered_map<uint64_t, std::string> &Labels) { 1133 // So far only supports PowerPC and X86. 1134 if (!STI->getTargetTriple().isPPC() && !STI->getTargetTriple().isX86()) 1135 return; 1136 1137 Labels.clear(); 1138 unsigned LabelCount = 0; 1139 Start += SectionAddr; 1140 End += SectionAddr; 1141 uint64_t Index = Start; 1142 while (Index < End) { 1143 // Disassemble a real instruction and record function-local branch labels. 1144 MCInst Inst; 1145 uint64_t Size; 1146 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index - SectionAddr); 1147 bool Disassembled = 1148 DisAsm->getInstruction(Inst, Size, ThisBytes, Index, nulls()); 1149 if (Size == 0) 1150 Size = std::min<uint64_t>(ThisBytes.size(), 1151 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1152 1153 if (Disassembled && MIA) { 1154 uint64_t Target; 1155 bool TargetKnown = MIA->evaluateBranch(Inst, Index, Size, Target); 1156 // On PowerPC, if the address of a branch is the same as the target, it 1157 // means that it's a function call. Do not mark the label for this case. 1158 if (TargetKnown && (Target >= Start && Target < End) && 1159 !Labels.count(Target) && 1160 !(STI->getTargetTriple().isPPC() && Target == Index)) 1161 Labels[Target] = ("L" + Twine(LabelCount++)).str(); 1162 } 1163 Index += Size; 1164 } 1165 } 1166 1167 // Create an MCSymbolizer for the target and add it to the MCDisassembler. 1168 // This is currently only used on AMDGPU, and assumes the format of the 1169 // void * argument passed to AMDGPU's createMCSymbolizer. 1170 static void addSymbolizer( 1171 MCContext &Ctx, const Target *Target, StringRef TripleName, 1172 MCDisassembler *DisAsm, uint64_t SectionAddr, ArrayRef<uint8_t> Bytes, 1173 SectionSymbolsTy &Symbols, 1174 std::vector<std::unique_ptr<std::string>> &SynthesizedLabelNames) { 1175 1176 std::unique_ptr<MCRelocationInfo> RelInfo( 1177 Target->createMCRelocationInfo(TripleName, Ctx)); 1178 if (!RelInfo) 1179 return; 1180 std::unique_ptr<MCSymbolizer> Symbolizer(Target->createMCSymbolizer( 1181 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1182 MCSymbolizer *SymbolizerPtr = &*Symbolizer; 1183 DisAsm->setSymbolizer(std::move(Symbolizer)); 1184 1185 if (!SymbolizeOperands) 1186 return; 1187 1188 // Synthesize labels referenced by branch instructions by 1189 // disassembling, discarding the output, and collecting the referenced 1190 // addresses from the symbolizer. 1191 for (size_t Index = 0; Index != Bytes.size();) { 1192 MCInst Inst; 1193 uint64_t Size; 1194 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index); 1195 const uint64_t ThisAddr = SectionAddr + Index; 1196 DisAsm->getInstruction(Inst, Size, ThisBytes, ThisAddr, nulls()); 1197 if (Size == 0) 1198 Size = std::min<uint64_t>(ThisBytes.size(), 1199 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1200 Index += Size; 1201 } 1202 ArrayRef<uint64_t> LabelAddrsRef = SymbolizerPtr->getReferencedAddresses(); 1203 // Copy and sort to remove duplicates. 1204 std::vector<uint64_t> LabelAddrs; 1205 LabelAddrs.insert(LabelAddrs.end(), LabelAddrsRef.begin(), 1206 LabelAddrsRef.end()); 1207 llvm::sort(LabelAddrs); 1208 LabelAddrs.resize(std::unique(LabelAddrs.begin(), LabelAddrs.end()) - 1209 LabelAddrs.begin()); 1210 // Add the labels. 1211 for (unsigned LabelNum = 0; LabelNum != LabelAddrs.size(); ++LabelNum) { 1212 auto Name = std::make_unique<std::string>(); 1213 *Name = (Twine("L") + Twine(LabelNum)).str(); 1214 SynthesizedLabelNames.push_back(std::move(Name)); 1215 Symbols.push_back(SymbolInfoTy( 1216 LabelAddrs[LabelNum], *SynthesizedLabelNames.back(), ELF::STT_NOTYPE)); 1217 } 1218 llvm::stable_sort(Symbols); 1219 // Recreate the symbolizer with the new symbols list. 1220 RelInfo.reset(Target->createMCRelocationInfo(TripleName, Ctx)); 1221 Symbolizer.reset(Target->createMCSymbolizer( 1222 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1223 DisAsm->setSymbolizer(std::move(Symbolizer)); 1224 } 1225 1226 static StringRef getSegmentName(const MachOObjectFile *MachO, 1227 const SectionRef &Section) { 1228 if (MachO) { 1229 DataRefImpl DR = Section.getRawDataRefImpl(); 1230 StringRef SegmentName = MachO->getSectionFinalSegmentName(DR); 1231 return SegmentName; 1232 } 1233 return ""; 1234 } 1235 1236 static void emitPostInstructionInfo(formatted_raw_ostream &FOS, 1237 const MCAsmInfo &MAI, 1238 const MCSubtargetInfo &STI, 1239 StringRef Comments, 1240 LiveVariablePrinter &LVP) { 1241 do { 1242 if (!Comments.empty()) { 1243 // Emit a line of comments. 1244 StringRef Comment; 1245 std::tie(Comment, Comments) = Comments.split('\n'); 1246 // MAI.getCommentColumn() assumes that instructions are printed at the 1247 // position of 8, while getInstStartColumn() returns the actual position. 1248 unsigned CommentColumn = 1249 MAI.getCommentColumn() - 8 + getInstStartColumn(STI); 1250 FOS.PadToColumn(CommentColumn); 1251 FOS << MAI.getCommentString() << ' ' << Comment; 1252 } 1253 LVP.printAfterInst(FOS); 1254 FOS << '\n'; 1255 } while (!Comments.empty()); 1256 FOS.flush(); 1257 } 1258 1259 static void createFakeELFSections(ObjectFile &Obj) { 1260 assert(Obj.isELF()); 1261 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 1262 Elf32LEObj->createFakeSections(); 1263 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 1264 Elf64LEObj->createFakeSections(); 1265 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 1266 Elf32BEObj->createFakeSections(); 1267 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 1268 Elf64BEObj->createFakeSections(); 1269 else 1270 llvm_unreachable("Unsupported binary format"); 1271 } 1272 1273 // Tries to fetch a more complete version of the given object file using its 1274 // Build ID. Returns None if nothing was found. 1275 static Optional<OwningBinary<Binary>> 1276 fetchBinaryByBuildID(const ObjectFile &Obj) { 1277 Optional<object::BuildIDRef> BuildID = getBuildID(&Obj); 1278 if (!BuildID) 1279 return std::nullopt; 1280 Optional<std::string> Path = BIDFetcher->fetch(*BuildID); 1281 if (!Path) 1282 return std::nullopt; 1283 Expected<OwningBinary<Binary>> DebugBinary = createBinary(*Path); 1284 if (!DebugBinary) { 1285 reportWarning(toString(DebugBinary.takeError()), *Path); 1286 return std::nullopt; 1287 } 1288 return std::move(*DebugBinary); 1289 } 1290 1291 static void disassembleObject(const Target *TheTarget, ObjectFile &Obj, 1292 const ObjectFile &DbgObj, MCContext &Ctx, 1293 MCDisassembler *PrimaryDisAsm, 1294 MCDisassembler *SecondaryDisAsm, 1295 const MCInstrAnalysis *MIA, MCInstPrinter *IP, 1296 const MCSubtargetInfo *PrimarySTI, 1297 const MCSubtargetInfo *SecondarySTI, 1298 PrettyPrinter &PIP, SourcePrinter &SP, 1299 bool InlineRelocs) { 1300 const MCSubtargetInfo *STI = PrimarySTI; 1301 MCDisassembler *DisAsm = PrimaryDisAsm; 1302 bool PrimaryIsThumb = false; 1303 if (isArmElf(Obj)) 1304 PrimaryIsThumb = STI->checkFeatures("+thumb-mode"); 1305 1306 std::map<SectionRef, std::vector<RelocationRef>> RelocMap; 1307 if (InlineRelocs) 1308 RelocMap = getRelocsMap(Obj); 1309 bool Is64Bits = Obj.getBytesInAddress() > 4; 1310 1311 // Create a mapping from virtual address to symbol name. This is used to 1312 // pretty print the symbols while disassembling. 1313 std::map<SectionRef, SectionSymbolsTy> AllSymbols; 1314 SectionSymbolsTy AbsoluteSymbols; 1315 const StringRef FileName = Obj.getFileName(); 1316 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&Obj); 1317 for (const SymbolRef &Symbol : Obj.symbols()) { 1318 Expected<StringRef> NameOrErr = Symbol.getName(); 1319 if (!NameOrErr) { 1320 reportWarning(toString(NameOrErr.takeError()), FileName); 1321 continue; 1322 } 1323 if (NameOrErr->empty() && !(Obj.isXCOFF() && SymbolDescription)) 1324 continue; 1325 1326 if (Obj.isELF() && getElfSymbolType(Obj, Symbol) == ELF::STT_SECTION) 1327 continue; 1328 1329 if (MachO) { 1330 // __mh_(execute|dylib|dylinker|bundle|preload|object)_header are special 1331 // symbols that support MachO header introspection. They do not bind to 1332 // code locations and are irrelevant for disassembly. 1333 if (NameOrErr->startswith("__mh_") && NameOrErr->endswith("_header")) 1334 continue; 1335 // Don't ask a Mach-O STAB symbol for its section unless you know that 1336 // STAB symbol's section field refers to a valid section index. Otherwise 1337 // the symbol may error trying to load a section that does not exist. 1338 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 1339 uint8_t NType = (MachO->is64Bit() ? 1340 MachO->getSymbol64TableEntry(SymDRI).n_type: 1341 MachO->getSymbolTableEntry(SymDRI).n_type); 1342 if (NType & MachO::N_STAB) 1343 continue; 1344 } 1345 1346 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1347 if (SecI != Obj.section_end()) 1348 AllSymbols[*SecI].push_back(createSymbolInfo(Obj, Symbol)); 1349 else 1350 AbsoluteSymbols.push_back(createSymbolInfo(Obj, Symbol)); 1351 } 1352 1353 if (AllSymbols.empty() && Obj.isELF()) 1354 addDynamicElfSymbols(cast<ELFObjectFileBase>(Obj), AllSymbols); 1355 1356 if (Obj.isWasm()) 1357 addMissingWasmCodeSymbols(cast<WasmObjectFile>(Obj), AllSymbols); 1358 1359 if (Obj.isELF() && Obj.sections().empty()) 1360 createFakeELFSections(Obj); 1361 1362 BumpPtrAllocator A; 1363 StringSaver Saver(A); 1364 addPltEntries(Obj, AllSymbols, Saver); 1365 1366 // Create a mapping from virtual address to section. An empty section can 1367 // cause more than one section at the same address. Sort such sections to be 1368 // before same-addressed non-empty sections so that symbol lookups prefer the 1369 // non-empty section. 1370 std::vector<std::pair<uint64_t, SectionRef>> SectionAddresses; 1371 for (SectionRef Sec : Obj.sections()) 1372 SectionAddresses.emplace_back(Sec.getAddress(), Sec); 1373 llvm::stable_sort(SectionAddresses, [](const auto &LHS, const auto &RHS) { 1374 if (LHS.first != RHS.first) 1375 return LHS.first < RHS.first; 1376 return LHS.second.getSize() < RHS.second.getSize(); 1377 }); 1378 1379 // Linked executables (.exe and .dll files) typically don't include a real 1380 // symbol table but they might contain an export table. 1381 if (const auto *COFFObj = dyn_cast<COFFObjectFile>(&Obj)) { 1382 for (const auto &ExportEntry : COFFObj->export_directories()) { 1383 StringRef Name; 1384 if (Error E = ExportEntry.getSymbolName(Name)) 1385 reportError(std::move(E), Obj.getFileName()); 1386 if (Name.empty()) 1387 continue; 1388 1389 uint32_t RVA; 1390 if (Error E = ExportEntry.getExportRVA(RVA)) 1391 reportError(std::move(E), Obj.getFileName()); 1392 1393 uint64_t VA = COFFObj->getImageBase() + RVA; 1394 auto Sec = partition_point( 1395 SectionAddresses, [VA](const std::pair<uint64_t, SectionRef> &O) { 1396 return O.first <= VA; 1397 }); 1398 if (Sec != SectionAddresses.begin()) { 1399 --Sec; 1400 AllSymbols[Sec->second].emplace_back(VA, Name, ELF::STT_NOTYPE); 1401 } else 1402 AbsoluteSymbols.emplace_back(VA, Name, ELF::STT_NOTYPE); 1403 } 1404 } 1405 1406 // Sort all the symbols, this allows us to use a simple binary search to find 1407 // Multiple symbols can have the same address. Use a stable sort to stabilize 1408 // the output. 1409 StringSet<> FoundDisasmSymbolSet; 1410 for (std::pair<const SectionRef, SectionSymbolsTy> &SecSyms : AllSymbols) 1411 llvm::stable_sort(SecSyms.second); 1412 llvm::stable_sort(AbsoluteSymbols); 1413 1414 std::unique_ptr<DWARFContext> DICtx; 1415 LiveVariablePrinter LVP(*Ctx.getRegisterInfo(), *STI); 1416 1417 if (DbgVariables != DVDisabled) { 1418 DICtx = DWARFContext::create(DbgObj); 1419 for (const std::unique_ptr<DWARFUnit> &CU : DICtx->compile_units()) 1420 LVP.addCompileUnit(CU->getUnitDIE(false)); 1421 } 1422 1423 LLVM_DEBUG(LVP.dump()); 1424 1425 std::unordered_map<uint64_t, BBAddrMap> AddrToBBAddrMap; 1426 auto ReadBBAddrMap = [&](Optional<unsigned> SectionIndex = std::nullopt) { 1427 AddrToBBAddrMap.clear(); 1428 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj)) { 1429 auto BBAddrMapsOrErr = Elf->readBBAddrMap(SectionIndex); 1430 if (!BBAddrMapsOrErr) 1431 reportWarning(toString(BBAddrMapsOrErr.takeError()), Obj.getFileName()); 1432 for (auto &FunctionBBAddrMap : *BBAddrMapsOrErr) 1433 AddrToBBAddrMap.emplace(FunctionBBAddrMap.Addr, 1434 std::move(FunctionBBAddrMap)); 1435 } 1436 }; 1437 1438 // For non-relocatable objects, Read all LLVM_BB_ADDR_MAP sections into a 1439 // single mapping, since they don't have any conflicts. 1440 if (SymbolizeOperands && !Obj.isRelocatableObject()) 1441 ReadBBAddrMap(); 1442 1443 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 1444 if (FilterSections.empty() && !DisassembleAll && 1445 (!Section.isText() || Section.isVirtual())) 1446 continue; 1447 1448 uint64_t SectionAddr = Section.getAddress(); 1449 uint64_t SectSize = Section.getSize(); 1450 if (!SectSize) 1451 continue; 1452 1453 // For relocatable object files, read the LLVM_BB_ADDR_MAP section 1454 // corresponding to this section, if present. 1455 if (SymbolizeOperands && Obj.isRelocatableObject()) 1456 ReadBBAddrMap(Section.getIndex()); 1457 1458 // Get the list of all the symbols in this section. 1459 SectionSymbolsTy &Symbols = AllSymbols[Section]; 1460 std::vector<MappingSymbolPair> MappingSymbols; 1461 if (hasMappingSymbols(Obj)) { 1462 for (const auto &Symb : Symbols) { 1463 uint64_t Address = Symb.Addr; 1464 StringRef Name = Symb.Name; 1465 if (Name.startswith("$d")) 1466 MappingSymbols.emplace_back(Address - SectionAddr, 'd'); 1467 if (Name.startswith("$x")) 1468 MappingSymbols.emplace_back(Address - SectionAddr, 'x'); 1469 if (Name.startswith("$a")) 1470 MappingSymbols.emplace_back(Address - SectionAddr, 'a'); 1471 if (Name.startswith("$t")) 1472 MappingSymbols.emplace_back(Address - SectionAddr, 't'); 1473 } 1474 } 1475 1476 llvm::sort(MappingSymbols); 1477 1478 ArrayRef<uint8_t> Bytes = arrayRefFromStringRef( 1479 unwrapOrError(Section.getContents(), Obj.getFileName())); 1480 1481 std::vector<std::unique_ptr<std::string>> SynthesizedLabelNames; 1482 if (Obj.isELF() && Obj.getArch() == Triple::amdgcn) { 1483 // AMDGPU disassembler uses symbolizer for printing labels 1484 addSymbolizer(Ctx, TheTarget, TripleName, DisAsm, SectionAddr, Bytes, 1485 Symbols, SynthesizedLabelNames); 1486 } 1487 1488 StringRef SegmentName = getSegmentName(MachO, Section); 1489 StringRef SectionName = unwrapOrError(Section.getName(), Obj.getFileName()); 1490 // If the section has no symbol at the start, just insert a dummy one. 1491 if (Symbols.empty() || Symbols[0].Addr != 0) { 1492 Symbols.insert(Symbols.begin(), 1493 createDummySymbolInfo(Obj, SectionAddr, SectionName, 1494 Section.isText() ? ELF::STT_FUNC 1495 : ELF::STT_OBJECT)); 1496 } 1497 1498 SmallString<40> Comments; 1499 raw_svector_ostream CommentStream(Comments); 1500 1501 uint64_t VMAAdjustment = 0; 1502 if (shouldAdjustVA(Section)) 1503 VMAAdjustment = AdjustVMA; 1504 1505 // In executable and shared objects, r_offset holds a virtual address. 1506 // Subtract SectionAddr from the r_offset field of a relocation to get 1507 // the section offset. 1508 uint64_t RelAdjustment = Obj.isRelocatableObject() ? 0 : SectionAddr; 1509 uint64_t Size; 1510 uint64_t Index; 1511 bool PrintedSection = false; 1512 std::vector<RelocationRef> Rels = RelocMap[Section]; 1513 std::vector<RelocationRef>::const_iterator RelCur = Rels.begin(); 1514 std::vector<RelocationRef>::const_iterator RelEnd = Rels.end(); 1515 1516 // Loop over each chunk of code between two points where at least 1517 // one symbol is defined. 1518 for (size_t SI = 0, SE = Symbols.size(); SI != SE;) { 1519 // Advance SI past all the symbols starting at the same address, 1520 // and make an ArrayRef of them. 1521 unsigned FirstSI = SI; 1522 uint64_t Start = Symbols[SI].Addr; 1523 ArrayRef<SymbolInfoTy> SymbolsHere; 1524 while (SI != SE && Symbols[SI].Addr == Start) 1525 ++SI; 1526 SymbolsHere = ArrayRef<SymbolInfoTy>(&Symbols[FirstSI], SI - FirstSI); 1527 1528 // Get the demangled names of all those symbols. We end up with a vector 1529 // of StringRef that holds the names we're going to use, and a vector of 1530 // std::string that stores the new strings returned by demangle(), if 1531 // any. If we don't call demangle() then that vector can stay empty. 1532 std::vector<StringRef> SymNamesHere; 1533 std::vector<std::string> DemangledSymNamesHere; 1534 if (Demangle) { 1535 // Fetch the demangled names and store them locally. 1536 for (const SymbolInfoTy &Symbol : SymbolsHere) 1537 DemangledSymNamesHere.push_back(demangle(Symbol.Name.str())); 1538 // Now we've finished modifying that vector, it's safe to make 1539 // a vector of StringRefs pointing into it. 1540 SymNamesHere.insert(SymNamesHere.begin(), DemangledSymNamesHere.begin(), 1541 DemangledSymNamesHere.end()); 1542 } else { 1543 for (const SymbolInfoTy &Symbol : SymbolsHere) 1544 SymNamesHere.push_back(Symbol.Name); 1545 } 1546 1547 // Distinguish ELF data from code symbols, which will be used later on to 1548 // decide whether to 'disassemble' this chunk as a data declaration via 1549 // dumpELFData(), or whether to treat it as code. 1550 // 1551 // If data _and_ code symbols are defined at the same address, the code 1552 // takes priority, on the grounds that disassembling code is our main 1553 // purpose here, and it would be a worse failure to _not_ interpret 1554 // something that _was_ meaningful as code than vice versa. 1555 // 1556 // Any ELF symbol type that is not clearly data will be regarded as code. 1557 // In particular, one of the uses of STT_NOTYPE is for branch targets 1558 // inside functions, for which STT_FUNC would be inaccurate. 1559 // 1560 // So here, we spot whether there's any non-data symbol present at all, 1561 // and only set the DisassembleAsData flag if there isn't. Also, we use 1562 // this distinction to inform the decision of which symbol to print at 1563 // the head of the section, so that if we're printing code, we print a 1564 // code-related symbol name to go with it. 1565 bool DisassembleAsData = false; 1566 size_t DisplaySymIndex = SymbolsHere.size() - 1; 1567 if (Obj.isELF() && !DisassembleAll && Section.isText()) { 1568 DisassembleAsData = true; // unless we find a code symbol below 1569 1570 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 1571 uint8_t SymTy = SymbolsHere[i].Type; 1572 if (SymTy != ELF::STT_OBJECT && SymTy != ELF::STT_COMMON) { 1573 DisassembleAsData = false; 1574 DisplaySymIndex = i; 1575 } 1576 } 1577 } 1578 1579 // Decide which symbol(s) from this collection we're going to print. 1580 std::vector<bool> SymsToPrint(SymbolsHere.size(), false); 1581 // If the user has given the --disassemble-symbols option, then we must 1582 // display every symbol in that set, and no others. 1583 if (!DisasmSymbolSet.empty()) { 1584 bool FoundAny = false; 1585 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 1586 if (DisasmSymbolSet.count(SymNamesHere[i])) { 1587 SymsToPrint[i] = true; 1588 FoundAny = true; 1589 } 1590 } 1591 1592 // And if none of the symbols here is one that the user asked for, skip 1593 // disassembling this entire chunk of code. 1594 if (!FoundAny) 1595 continue; 1596 } else { 1597 // Otherwise, print whichever symbol at this location is last in the 1598 // Symbols array, because that array is pre-sorted in a way intended to 1599 // correlate with priority of which symbol to display. 1600 SymsToPrint[DisplaySymIndex] = true; 1601 } 1602 1603 // Now that we know we're disassembling this section, override the choice 1604 // of which symbols to display by printing _all_ of them at this address 1605 // if the user asked for all symbols. 1606 // 1607 // That way, '--show-all-symbols --disassemble-symbol=foo' will print 1608 // only the chunk of code headed by 'foo', but also show any other 1609 // symbols defined at that address, such as aliases for 'foo', or the ARM 1610 // mapping symbol preceding its code. 1611 if (ShowAllSymbols) { 1612 for (size_t i = 0; i < SymbolsHere.size(); ++i) 1613 SymsToPrint[i] = true; 1614 } 1615 1616 if (Start < SectionAddr || StopAddress <= Start) 1617 continue; 1618 1619 for (size_t i = 0; i < SymbolsHere.size(); ++i) 1620 FoundDisasmSymbolSet.insert(SymNamesHere[i]); 1621 1622 // The end is the section end, the beginning of the next symbol, or 1623 // --stop-address. 1624 uint64_t End = std::min<uint64_t>(SectionAddr + SectSize, StopAddress); 1625 if (SI < SE) 1626 End = std::min(End, Symbols[SI].Addr); 1627 if (Start >= End || End <= StartAddress) 1628 continue; 1629 Start -= SectionAddr; 1630 End -= SectionAddr; 1631 1632 if (!PrintedSection) { 1633 PrintedSection = true; 1634 outs() << "\nDisassembly of section "; 1635 if (!SegmentName.empty()) 1636 outs() << SegmentName << ","; 1637 outs() << SectionName << ":\n"; 1638 } 1639 1640 outs() << '\n'; 1641 1642 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 1643 if (!SymsToPrint[i]) 1644 continue; 1645 1646 const SymbolInfoTy &Symbol = SymbolsHere[i]; 1647 const StringRef SymbolName = SymNamesHere[i]; 1648 1649 if (LeadingAddr) 1650 outs() << format(Is64Bits ? "%016" PRIx64 " " : "%08" PRIx64 " ", 1651 SectionAddr + Start + VMAAdjustment); 1652 if (Obj.isXCOFF() && SymbolDescription) { 1653 outs() << getXCOFFSymbolDescription(Symbol, SymbolName) << ":\n"; 1654 } else 1655 outs() << '<' << SymbolName << ">:\n"; 1656 } 1657 1658 // Don't print raw contents of a virtual section. A virtual section 1659 // doesn't have any contents in the file. 1660 if (Section.isVirtual()) { 1661 outs() << "...\n"; 1662 continue; 1663 } 1664 1665 // See if any of the symbols defined at this location triggers target- 1666 // specific disassembly behavior, e.g. of special descriptors or function 1667 // prelude information. 1668 // 1669 // We stop this loop at the first symbol that triggers some kind of 1670 // interesting behavior (if any), on the assumption that if two symbols 1671 // defined at the same address trigger two conflicting symbol handlers, 1672 // the object file is probably confused anyway, and it would make even 1673 // less sense to present the output of _both_ handlers, because that 1674 // would describe the same data twice. 1675 for (size_t SHI = 0; SHI < SymbolsHere.size(); ++SHI) { 1676 SymbolInfoTy Symbol = SymbolsHere[SHI]; 1677 1678 auto Status = 1679 DisAsm->onSymbolStart(Symbol, Size, Bytes.slice(Start, End - Start), 1680 SectionAddr + Start, CommentStream); 1681 1682 if (!Status) { 1683 // If onSymbolStart returns None, that means it didn't trigger any 1684 // interesting handling for this symbol. Try the other symbols 1685 // defined at this address. 1686 continue; 1687 } 1688 1689 if (Status.value() == MCDisassembler::Fail) { 1690 // If onSymbolStart returns Fail, that means it identified some kind 1691 // of special data at this address, but wasn't able to disassemble it 1692 // meaningfully. So we fall back to disassembling the failed region 1693 // as bytes, assuming that the target detected the failure before 1694 // printing anything. 1695 // 1696 // Return values Success or SoftFail (i.e no 'real' failure) are 1697 // expected to mean that the target has emitted its own output. 1698 // 1699 // Either way, 'Size' will have been set to the amount of data 1700 // covered by whatever prologue the target identified. So we advance 1701 // our own position to beyond that. Sometimes that will be the entire 1702 // distance to the next symbol, and sometimes it will be just a 1703 // prologue and we should start disassembling instructions from where 1704 // it left off. 1705 outs() << "// Error in decoding " << SymNamesHere[SHI] 1706 << " : Decoding failed region as bytes.\n"; 1707 for (uint64_t I = 0; I < Size; ++I) { 1708 outs() << "\t.byte\t " << format_hex(Bytes[I], 1, /*Upper=*/true) 1709 << "\n"; 1710 } 1711 } 1712 Start += Size; 1713 break; 1714 } 1715 1716 Index = Start; 1717 if (SectionAddr < StartAddress) 1718 Index = std::max<uint64_t>(Index, StartAddress - SectionAddr); 1719 1720 if (DisassembleAsData) { 1721 dumpELFData(SectionAddr, Index, End, Bytes); 1722 Index = End; 1723 continue; 1724 } 1725 1726 bool DumpARMELFData = false; 1727 formatted_raw_ostream FOS(outs()); 1728 1729 std::unordered_map<uint64_t, std::string> AllLabels; 1730 std::unordered_map<uint64_t, std::vector<std::string>> BBAddrMapLabels; 1731 if (SymbolizeOperands) { 1732 collectLocalBranchTargets(Bytes, MIA, DisAsm, IP, PrimarySTI, 1733 SectionAddr, Index, End, AllLabels); 1734 collectBBAddrMapLabels(AddrToBBAddrMap, SectionAddr, Index, End, 1735 BBAddrMapLabels); 1736 } 1737 1738 while (Index < End) { 1739 // ARM and AArch64 ELF binaries can interleave data and text in the 1740 // same section. We rely on the markers introduced to understand what 1741 // we need to dump. If the data marker is within a function, it is 1742 // denoted as a word/short etc. 1743 if (!MappingSymbols.empty()) { 1744 char Kind = getMappingSymbolKind(MappingSymbols, Index); 1745 DumpARMELFData = Kind == 'd'; 1746 if (SecondarySTI) { 1747 if (Kind == 'a') { 1748 STI = PrimaryIsThumb ? SecondarySTI : PrimarySTI; 1749 DisAsm = PrimaryIsThumb ? SecondaryDisAsm : PrimaryDisAsm; 1750 } else if (Kind == 't') { 1751 STI = PrimaryIsThumb ? PrimarySTI : SecondarySTI; 1752 DisAsm = PrimaryIsThumb ? PrimaryDisAsm : SecondaryDisAsm; 1753 } 1754 } 1755 } 1756 1757 if (DumpARMELFData) { 1758 Size = dumpARMELFData(SectionAddr, Index, End, Obj, Bytes, 1759 MappingSymbols, *STI, FOS); 1760 } else { 1761 // When -z or --disassemble-zeroes are given we always dissasemble 1762 // them. Otherwise we might want to skip zero bytes we see. 1763 if (!DisassembleZeroes) { 1764 uint64_t MaxOffset = End - Index; 1765 // For --reloc: print zero blocks patched by relocations, so that 1766 // relocations can be shown in the dump. 1767 if (RelCur != RelEnd) 1768 MaxOffset = std::min(RelCur->getOffset() - RelAdjustment - Index, 1769 MaxOffset); 1770 1771 if (size_t N = 1772 countSkippableZeroBytes(Bytes.slice(Index, MaxOffset))) { 1773 FOS << "\t\t..." << '\n'; 1774 Index += N; 1775 continue; 1776 } 1777 } 1778 1779 // Print local label if there's any. 1780 auto Iter1 = BBAddrMapLabels.find(SectionAddr + Index); 1781 if (Iter1 != BBAddrMapLabels.end()) { 1782 for (StringRef Label : Iter1->second) 1783 FOS << "<" << Label << ">:\n"; 1784 } else { 1785 auto Iter2 = AllLabels.find(SectionAddr + Index); 1786 if (Iter2 != AllLabels.end()) 1787 FOS << "<" << Iter2->second << ">:\n"; 1788 } 1789 1790 // Disassemble a real instruction or a data when disassemble all is 1791 // provided 1792 MCInst Inst; 1793 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index); 1794 uint64_t ThisAddr = SectionAddr + Index; 1795 bool Disassembled = DisAsm->getInstruction(Inst, Size, ThisBytes, 1796 ThisAddr, CommentStream); 1797 if (Size == 0) 1798 Size = std::min<uint64_t>( 1799 ThisBytes.size(), 1800 DisAsm->suggestBytesToSkip(ThisBytes, ThisAddr)); 1801 1802 LVP.update({Index, Section.getIndex()}, 1803 {Index + Size, Section.getIndex()}, Index + Size != End); 1804 1805 IP->setCommentStream(CommentStream); 1806 1807 PIP.printInst( 1808 *IP, Disassembled ? &Inst : nullptr, Bytes.slice(Index, Size), 1809 {SectionAddr + Index + VMAAdjustment, Section.getIndex()}, FOS, 1810 "", *STI, &SP, Obj.getFileName(), &Rels, LVP); 1811 1812 IP->setCommentStream(llvm::nulls()); 1813 1814 // If disassembly has failed, avoid analysing invalid/incomplete 1815 // instruction information. Otherwise, try to resolve the target 1816 // address (jump target or memory operand address) and print it on the 1817 // right of the instruction. 1818 if (Disassembled && MIA) { 1819 // Branch targets are printed just after the instructions. 1820 llvm::raw_ostream *TargetOS = &FOS; 1821 uint64_t Target; 1822 bool PrintTarget = 1823 MIA->evaluateBranch(Inst, SectionAddr + Index, Size, Target); 1824 if (!PrintTarget) 1825 if (Optional<uint64_t> MaybeTarget = 1826 MIA->evaluateMemoryOperandAddress( 1827 Inst, STI, SectionAddr + Index, Size)) { 1828 Target = *MaybeTarget; 1829 PrintTarget = true; 1830 // Do not print real address when symbolizing. 1831 if (!SymbolizeOperands) { 1832 // Memory operand addresses are printed as comments. 1833 TargetOS = &CommentStream; 1834 *TargetOS << "0x" << Twine::utohexstr(Target); 1835 } 1836 } 1837 if (PrintTarget) { 1838 // In a relocatable object, the target's section must reside in 1839 // the same section as the call instruction or it is accessed 1840 // through a relocation. 1841 // 1842 // In a non-relocatable object, the target may be in any section. 1843 // In that case, locate the section(s) containing the target 1844 // address and find the symbol in one of those, if possible. 1845 // 1846 // N.B. We don't walk the relocations in the relocatable case yet. 1847 std::vector<const SectionSymbolsTy *> TargetSectionSymbols; 1848 if (!Obj.isRelocatableObject()) { 1849 auto It = llvm::partition_point( 1850 SectionAddresses, 1851 [=](const std::pair<uint64_t, SectionRef> &O) { 1852 return O.first <= Target; 1853 }); 1854 uint64_t TargetSecAddr = 0; 1855 while (It != SectionAddresses.begin()) { 1856 --It; 1857 if (TargetSecAddr == 0) 1858 TargetSecAddr = It->first; 1859 if (It->first != TargetSecAddr) 1860 break; 1861 TargetSectionSymbols.push_back(&AllSymbols[It->second]); 1862 } 1863 } else { 1864 TargetSectionSymbols.push_back(&Symbols); 1865 } 1866 TargetSectionSymbols.push_back(&AbsoluteSymbols); 1867 1868 // Find the last symbol in the first candidate section whose 1869 // offset is less than or equal to the target. If there are no 1870 // such symbols, try in the next section and so on, before finally 1871 // using the nearest preceding absolute symbol (if any), if there 1872 // are no other valid symbols. 1873 const SymbolInfoTy *TargetSym = nullptr; 1874 for (const SectionSymbolsTy *TargetSymbols : 1875 TargetSectionSymbols) { 1876 auto It = llvm::partition_point( 1877 *TargetSymbols, 1878 [=](const SymbolInfoTy &O) { return O.Addr <= Target; }); 1879 if (It != TargetSymbols->begin()) { 1880 TargetSym = &*(It - 1); 1881 break; 1882 } 1883 } 1884 1885 // Print the labels corresponding to the target if there's any. 1886 bool BBAddrMapLabelAvailable = BBAddrMapLabels.count(Target); 1887 bool LabelAvailable = AllLabels.count(Target); 1888 if (TargetSym != nullptr) { 1889 uint64_t TargetAddress = TargetSym->Addr; 1890 uint64_t Disp = Target - TargetAddress; 1891 std::string TargetName = TargetSym->Name.str(); 1892 if (Demangle) 1893 TargetName = demangle(TargetName); 1894 1895 *TargetOS << " <"; 1896 if (!Disp) { 1897 // Always Print the binary symbol precisely corresponding to 1898 // the target address. 1899 *TargetOS << TargetName; 1900 } else if (BBAddrMapLabelAvailable) { 1901 *TargetOS << BBAddrMapLabels[Target].front(); 1902 } else if (LabelAvailable) { 1903 *TargetOS << AllLabels[Target]; 1904 } else { 1905 // Always Print the binary symbol plus an offset if there's no 1906 // local label corresponding to the target address. 1907 *TargetOS << TargetName << "+0x" << Twine::utohexstr(Disp); 1908 } 1909 *TargetOS << ">"; 1910 } else if (BBAddrMapLabelAvailable) { 1911 *TargetOS << " <" << BBAddrMapLabels[Target].front() << ">"; 1912 } else if (LabelAvailable) { 1913 *TargetOS << " <" << AllLabels[Target] << ">"; 1914 } 1915 // By convention, each record in the comment stream should be 1916 // terminated. 1917 if (TargetOS == &CommentStream) 1918 *TargetOS << "\n"; 1919 } 1920 } 1921 } 1922 1923 assert(Ctx.getAsmInfo()); 1924 emitPostInstructionInfo(FOS, *Ctx.getAsmInfo(), *STI, 1925 CommentStream.str(), LVP); 1926 Comments.clear(); 1927 1928 // Hexagon does this in pretty printer 1929 if (Obj.getArch() != Triple::hexagon) { 1930 // Print relocation for instruction and data. 1931 while (RelCur != RelEnd) { 1932 uint64_t Offset = RelCur->getOffset() - RelAdjustment; 1933 // If this relocation is hidden, skip it. 1934 if (getHidden(*RelCur) || SectionAddr + Offset < StartAddress) { 1935 ++RelCur; 1936 continue; 1937 } 1938 1939 // Stop when RelCur's offset is past the disassembled 1940 // instruction/data. Note that it's possible the disassembled data 1941 // is not the complete data: we might see the relocation printed in 1942 // the middle of the data, but this matches the binutils objdump 1943 // output. 1944 if (Offset >= Index + Size) 1945 break; 1946 1947 // When --adjust-vma is used, update the address printed. 1948 if (RelCur->getSymbol() != Obj.symbol_end()) { 1949 Expected<section_iterator> SymSI = 1950 RelCur->getSymbol()->getSection(); 1951 if (SymSI && *SymSI != Obj.section_end() && 1952 shouldAdjustVA(**SymSI)) 1953 Offset += AdjustVMA; 1954 } 1955 1956 printRelocation(FOS, Obj.getFileName(), *RelCur, 1957 SectionAddr + Offset, Is64Bits); 1958 LVP.printAfterOtherLine(FOS, true); 1959 ++RelCur; 1960 } 1961 } 1962 1963 Index += Size; 1964 } 1965 } 1966 } 1967 StringSet<> MissingDisasmSymbolSet = 1968 set_difference(DisasmSymbolSet, FoundDisasmSymbolSet); 1969 for (StringRef Sym : MissingDisasmSymbolSet.keys()) 1970 reportWarning("failed to disassemble missing symbol " + Sym, FileName); 1971 } 1972 1973 static void disassembleObject(ObjectFile *Obj, bool InlineRelocs) { 1974 // If information useful for showing the disassembly is missing, try to find a 1975 // more complete binary and disassemble that instead. 1976 OwningBinary<Binary> FetchedBinary; 1977 if (Obj->symbols().empty()) { 1978 if (Optional<OwningBinary<Binary>> FetchedBinaryOpt = 1979 fetchBinaryByBuildID(*Obj)) { 1980 if (auto *O = dyn_cast<ObjectFile>(FetchedBinaryOpt->getBinary())) { 1981 if (!O->symbols().empty() || 1982 (!O->sections().empty() && Obj->sections().empty())) { 1983 FetchedBinary = std::move(*FetchedBinaryOpt); 1984 Obj = O; 1985 } 1986 } 1987 } 1988 } 1989 1990 const Target *TheTarget = getTarget(Obj); 1991 1992 // Package up features to be passed to target/subtarget 1993 SubtargetFeatures Features = Obj->getFeatures(); 1994 if (!MAttrs.empty()) { 1995 for (unsigned I = 0; I != MAttrs.size(); ++I) 1996 Features.AddFeature(MAttrs[I]); 1997 } else if (MCPU.empty() && Obj->getArch() == llvm::Triple::aarch64) { 1998 Features.AddFeature("+all"); 1999 } 2000 2001 std::unique_ptr<const MCRegisterInfo> MRI( 2002 TheTarget->createMCRegInfo(TripleName)); 2003 if (!MRI) 2004 reportError(Obj->getFileName(), 2005 "no register info for target " + TripleName); 2006 2007 // Set up disassembler. 2008 MCTargetOptions MCOptions; 2009 std::unique_ptr<const MCAsmInfo> AsmInfo( 2010 TheTarget->createMCAsmInfo(*MRI, TripleName, MCOptions)); 2011 if (!AsmInfo) 2012 reportError(Obj->getFileName(), 2013 "no assembly info for target " + TripleName); 2014 2015 if (MCPU.empty()) 2016 MCPU = Obj->tryGetCPUName().value_or("").str(); 2017 2018 if (isArmElf(*Obj)) { 2019 // When disassembling big-endian Arm ELF, the instruction endianness is 2020 // determined in a complex way. In relocatable objects, AAELF32 mandates 2021 // that instruction endianness matches the ELF file endianness; in 2022 // executable images, that's true unless the file header has the EF_ARM_BE8 2023 // flag, in which case instructions are little-endian regardless of data 2024 // endianness. 2025 // 2026 // We must set the big-endian-instructions SubtargetFeature to make the 2027 // disassembler read the instructions the right way round, and also tell 2028 // our own prettyprinter to retrieve the encodings the same way to print in 2029 // hex. 2030 const auto *Elf32BE = dyn_cast<ELF32BEObjectFile>(Obj); 2031 2032 if (Elf32BE && (Elf32BE->isRelocatableObject() || 2033 !(Elf32BE->getPlatformFlags() & ELF::EF_ARM_BE8))) { 2034 Features.AddFeature("+big-endian-instructions"); 2035 ARMPrettyPrinterInst.setInstructionEndianness(llvm::support::big); 2036 } else { 2037 ARMPrettyPrinterInst.setInstructionEndianness(llvm::support::little); 2038 } 2039 } 2040 2041 std::unique_ptr<const MCSubtargetInfo> STI( 2042 TheTarget->createMCSubtargetInfo(TripleName, MCPU, Features.getString())); 2043 if (!STI) 2044 reportError(Obj->getFileName(), 2045 "no subtarget info for target " + TripleName); 2046 std::unique_ptr<const MCInstrInfo> MII(TheTarget->createMCInstrInfo()); 2047 if (!MII) 2048 reportError(Obj->getFileName(), 2049 "no instruction info for target " + TripleName); 2050 MCContext Ctx(Triple(TripleName), AsmInfo.get(), MRI.get(), STI.get()); 2051 // FIXME: for now initialize MCObjectFileInfo with default values 2052 std::unique_ptr<MCObjectFileInfo> MOFI( 2053 TheTarget->createMCObjectFileInfo(Ctx, /*PIC=*/false)); 2054 Ctx.setObjectFileInfo(MOFI.get()); 2055 2056 std::unique_ptr<MCDisassembler> DisAsm( 2057 TheTarget->createMCDisassembler(*STI, Ctx)); 2058 if (!DisAsm) 2059 reportError(Obj->getFileName(), "no disassembler for target " + TripleName); 2060 2061 // If we have an ARM object file, we need a second disassembler, because 2062 // ARM CPUs have two different instruction sets: ARM mode, and Thumb mode. 2063 // We use mapping symbols to switch between the two assemblers, where 2064 // appropriate. 2065 std::unique_ptr<MCDisassembler> SecondaryDisAsm; 2066 std::unique_ptr<const MCSubtargetInfo> SecondarySTI; 2067 if (isArmElf(*Obj) && !STI->checkFeatures("+mclass")) { 2068 if (STI->checkFeatures("+thumb-mode")) 2069 Features.AddFeature("-thumb-mode"); 2070 else 2071 Features.AddFeature("+thumb-mode"); 2072 SecondarySTI.reset(TheTarget->createMCSubtargetInfo(TripleName, MCPU, 2073 Features.getString())); 2074 SecondaryDisAsm.reset(TheTarget->createMCDisassembler(*SecondarySTI, Ctx)); 2075 } 2076 2077 std::unique_ptr<const MCInstrAnalysis> MIA( 2078 TheTarget->createMCInstrAnalysis(MII.get())); 2079 2080 int AsmPrinterVariant = AsmInfo->getAssemblerDialect(); 2081 std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter( 2082 Triple(TripleName), AsmPrinterVariant, *AsmInfo, *MII, *MRI)); 2083 if (!IP) 2084 reportError(Obj->getFileName(), 2085 "no instruction printer for target " + TripleName); 2086 IP->setPrintImmHex(PrintImmHex); 2087 IP->setPrintBranchImmAsAddress(true); 2088 IP->setSymbolizeOperands(SymbolizeOperands); 2089 IP->setMCInstrAnalysis(MIA.get()); 2090 2091 PrettyPrinter &PIP = selectPrettyPrinter(Triple(TripleName)); 2092 2093 const ObjectFile *DbgObj = Obj; 2094 if (!FetchedBinary.getBinary() && !Obj->hasDebugInfo()) { 2095 if (Optional<OwningBinary<Binary>> DebugBinaryOpt = 2096 fetchBinaryByBuildID(*Obj)) { 2097 if (auto *FetchedObj = 2098 dyn_cast<const ObjectFile>(DebugBinaryOpt->getBinary())) { 2099 if (FetchedObj->hasDebugInfo()) { 2100 FetchedBinary = std::move(*DebugBinaryOpt); 2101 DbgObj = FetchedObj; 2102 } 2103 } 2104 } 2105 } 2106 2107 std::unique_ptr<object::Binary> DSYMBinary; 2108 std::unique_ptr<MemoryBuffer> DSYMBuf; 2109 if (!DbgObj->hasDebugInfo()) { 2110 if (const MachOObjectFile *MachOOF = dyn_cast<MachOObjectFile>(&*Obj)) { 2111 DbgObj = objdump::getMachODSymObject(MachOOF, Obj->getFileName(), 2112 DSYMBinary, DSYMBuf); 2113 if (!DbgObj) 2114 return; 2115 } 2116 } 2117 2118 SourcePrinter SP(DbgObj, TheTarget->getName()); 2119 2120 for (StringRef Opt : DisassemblerOptions) 2121 if (!IP->applyTargetSpecificCLOption(Opt)) 2122 reportError(Obj->getFileName(), 2123 "Unrecognized disassembler option: " + Opt); 2124 2125 disassembleObject(TheTarget, *Obj, *DbgObj, Ctx, DisAsm.get(), 2126 SecondaryDisAsm.get(), MIA.get(), IP.get(), STI.get(), 2127 SecondarySTI.get(), PIP, SP, InlineRelocs); 2128 } 2129 2130 void objdump::printRelocations(const ObjectFile *Obj) { 2131 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : 2132 "%08" PRIx64; 2133 2134 // Build a mapping from relocation target to a vector of relocation 2135 // sections. Usually, there is an only one relocation section for 2136 // each relocated section. 2137 MapVector<SectionRef, std::vector<SectionRef>> SecToRelSec; 2138 uint64_t Ndx; 2139 for (const SectionRef &Section : ToolSectionFilter(*Obj, &Ndx)) { 2140 if (Obj->isELF() && (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC)) 2141 continue; 2142 if (Section.relocation_begin() == Section.relocation_end()) 2143 continue; 2144 Expected<section_iterator> SecOrErr = Section.getRelocatedSection(); 2145 if (!SecOrErr) 2146 reportError(Obj->getFileName(), 2147 "section (" + Twine(Ndx) + 2148 "): unable to get a relocation target: " + 2149 toString(SecOrErr.takeError())); 2150 SecToRelSec[**SecOrErr].push_back(Section); 2151 } 2152 2153 for (std::pair<SectionRef, std::vector<SectionRef>> &P : SecToRelSec) { 2154 StringRef SecName = unwrapOrError(P.first.getName(), Obj->getFileName()); 2155 outs() << "\nRELOCATION RECORDS FOR [" << SecName << "]:\n"; 2156 uint32_t OffsetPadding = (Obj->getBytesInAddress() > 4 ? 16 : 8); 2157 uint32_t TypePadding = 24; 2158 outs() << left_justify("OFFSET", OffsetPadding) << " " 2159 << left_justify("TYPE", TypePadding) << " " 2160 << "VALUE\n"; 2161 2162 for (SectionRef Section : P.second) { 2163 for (const RelocationRef &Reloc : Section.relocations()) { 2164 uint64_t Address = Reloc.getOffset(); 2165 SmallString<32> RelocName; 2166 SmallString<32> ValueStr; 2167 if (Address < StartAddress || Address > StopAddress || getHidden(Reloc)) 2168 continue; 2169 Reloc.getTypeName(RelocName); 2170 if (Error E = getRelocationValueString(Reloc, ValueStr)) 2171 reportError(std::move(E), Obj->getFileName()); 2172 2173 outs() << format(Fmt.data(), Address) << " " 2174 << left_justify(RelocName, TypePadding) << " " << ValueStr 2175 << "\n"; 2176 } 2177 } 2178 } 2179 } 2180 2181 void objdump::printDynamicRelocations(const ObjectFile *Obj) { 2182 // For the moment, this option is for ELF only 2183 if (!Obj->isELF()) 2184 return; 2185 2186 const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj); 2187 if (!Elf || !any_of(Elf->sections(), [](const ELFSectionRef Sec) { 2188 return Sec.getType() == ELF::SHT_DYNAMIC; 2189 })) { 2190 reportError(Obj->getFileName(), "not a dynamic object"); 2191 return; 2192 } 2193 2194 std::vector<SectionRef> DynRelSec = Obj->dynamic_relocation_sections(); 2195 if (DynRelSec.empty()) 2196 return; 2197 2198 outs() << "\nDYNAMIC RELOCATION RECORDS\n"; 2199 const uint32_t OffsetPadding = (Obj->getBytesInAddress() > 4 ? 16 : 8); 2200 const uint32_t TypePadding = 24; 2201 outs() << left_justify("OFFSET", OffsetPadding) << ' ' 2202 << left_justify("TYPE", TypePadding) << " VALUE\n"; 2203 2204 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2205 for (const SectionRef &Section : DynRelSec) 2206 for (const RelocationRef &Reloc : Section.relocations()) { 2207 uint64_t Address = Reloc.getOffset(); 2208 SmallString<32> RelocName; 2209 SmallString<32> ValueStr; 2210 Reloc.getTypeName(RelocName); 2211 if (Error E = getRelocationValueString(Reloc, ValueStr)) 2212 reportError(std::move(E), Obj->getFileName()); 2213 outs() << format(Fmt.data(), Address) << ' ' 2214 << left_justify(RelocName, TypePadding) << ' ' << ValueStr << '\n'; 2215 } 2216 } 2217 2218 // Returns true if we need to show LMA column when dumping section headers. We 2219 // show it only when the platform is ELF and either we have at least one section 2220 // whose VMA and LMA are different and/or when --show-lma flag is used. 2221 static bool shouldDisplayLMA(const ObjectFile &Obj) { 2222 if (!Obj.isELF()) 2223 return false; 2224 for (const SectionRef &S : ToolSectionFilter(Obj)) 2225 if (S.getAddress() != getELFSectionLMA(S)) 2226 return true; 2227 return ShowLMA; 2228 } 2229 2230 static size_t getMaxSectionNameWidth(const ObjectFile &Obj) { 2231 // Default column width for names is 13 even if no names are that long. 2232 size_t MaxWidth = 13; 2233 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 2234 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 2235 MaxWidth = std::max(MaxWidth, Name.size()); 2236 } 2237 return MaxWidth; 2238 } 2239 2240 void objdump::printSectionHeaders(ObjectFile &Obj) { 2241 if (Obj.isELF() && Obj.sections().empty()) 2242 createFakeELFSections(Obj); 2243 2244 size_t NameWidth = getMaxSectionNameWidth(Obj); 2245 size_t AddressWidth = 2 * Obj.getBytesInAddress(); 2246 bool HasLMAColumn = shouldDisplayLMA(Obj); 2247 outs() << "\nSections:\n"; 2248 if (HasLMAColumn) 2249 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 2250 << left_justify("VMA", AddressWidth) << " " 2251 << left_justify("LMA", AddressWidth) << " Type\n"; 2252 else 2253 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 2254 << left_justify("VMA", AddressWidth) << " Type\n"; 2255 2256 uint64_t Idx; 2257 for (const SectionRef &Section : ToolSectionFilter(Obj, &Idx)) { 2258 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 2259 uint64_t VMA = Section.getAddress(); 2260 if (shouldAdjustVA(Section)) 2261 VMA += AdjustVMA; 2262 2263 uint64_t Size = Section.getSize(); 2264 2265 std::string Type = Section.isText() ? "TEXT" : ""; 2266 if (Section.isData()) 2267 Type += Type.empty() ? "DATA" : ", DATA"; 2268 if (Section.isBSS()) 2269 Type += Type.empty() ? "BSS" : ", BSS"; 2270 if (Section.isDebugSection()) 2271 Type += Type.empty() ? "DEBUG" : ", DEBUG"; 2272 2273 if (HasLMAColumn) 2274 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 2275 Name.str().c_str(), Size) 2276 << format_hex_no_prefix(VMA, AddressWidth) << " " 2277 << format_hex_no_prefix(getELFSectionLMA(Section), AddressWidth) 2278 << " " << Type << "\n"; 2279 else 2280 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 2281 Name.str().c_str(), Size) 2282 << format_hex_no_prefix(VMA, AddressWidth) << " " << Type << "\n"; 2283 } 2284 } 2285 2286 void objdump::printSectionContents(const ObjectFile *Obj) { 2287 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj); 2288 2289 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 2290 StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName()); 2291 uint64_t BaseAddr = Section.getAddress(); 2292 uint64_t Size = Section.getSize(); 2293 if (!Size) 2294 continue; 2295 2296 outs() << "Contents of section "; 2297 StringRef SegmentName = getSegmentName(MachO, Section); 2298 if (!SegmentName.empty()) 2299 outs() << SegmentName << ","; 2300 outs() << Name << ":\n"; 2301 if (Section.isBSS()) { 2302 outs() << format("<skipping contents of bss section at [%04" PRIx64 2303 ", %04" PRIx64 ")>\n", 2304 BaseAddr, BaseAddr + Size); 2305 continue; 2306 } 2307 2308 StringRef Contents = unwrapOrError(Section.getContents(), Obj->getFileName()); 2309 2310 // Dump out the content as hex and printable ascii characters. 2311 for (std::size_t Addr = 0, End = Contents.size(); Addr < End; Addr += 16) { 2312 outs() << format(" %04" PRIx64 " ", BaseAddr + Addr); 2313 // Dump line of hex. 2314 for (std::size_t I = 0; I < 16; ++I) { 2315 if (I != 0 && I % 4 == 0) 2316 outs() << ' '; 2317 if (Addr + I < End) 2318 outs() << hexdigit((Contents[Addr + I] >> 4) & 0xF, true) 2319 << hexdigit(Contents[Addr + I] & 0xF, true); 2320 else 2321 outs() << " "; 2322 } 2323 // Print ascii. 2324 outs() << " "; 2325 for (std::size_t I = 0; I < 16 && Addr + I < End; ++I) { 2326 if (isPrint(static_cast<unsigned char>(Contents[Addr + I]) & 0xFF)) 2327 outs() << Contents[Addr + I]; 2328 else 2329 outs() << "."; 2330 } 2331 outs() << "\n"; 2332 } 2333 } 2334 } 2335 2336 void objdump::printSymbolTable(const ObjectFile &O, StringRef ArchiveName, 2337 StringRef ArchitectureName, bool DumpDynamic) { 2338 if (O.isCOFF() && !DumpDynamic) { 2339 outs() << "\nSYMBOL TABLE:\n"; 2340 printCOFFSymbolTable(cast<const COFFObjectFile>(O)); 2341 return; 2342 } 2343 2344 const StringRef FileName = O.getFileName(); 2345 2346 if (!DumpDynamic) { 2347 outs() << "\nSYMBOL TABLE:\n"; 2348 for (auto I = O.symbol_begin(); I != O.symbol_end(); ++I) 2349 printSymbol(O, *I, {}, FileName, ArchiveName, ArchitectureName, 2350 DumpDynamic); 2351 return; 2352 } 2353 2354 outs() << "\nDYNAMIC SYMBOL TABLE:\n"; 2355 if (!O.isELF()) { 2356 reportWarning( 2357 "this operation is not currently supported for this file format", 2358 FileName); 2359 return; 2360 } 2361 2362 const ELFObjectFileBase *ELF = cast<const ELFObjectFileBase>(&O); 2363 auto Symbols = ELF->getDynamicSymbolIterators(); 2364 Expected<std::vector<VersionEntry>> SymbolVersionsOrErr = 2365 ELF->readDynsymVersions(); 2366 if (!SymbolVersionsOrErr) { 2367 reportWarning(toString(SymbolVersionsOrErr.takeError()), FileName); 2368 SymbolVersionsOrErr = std::vector<VersionEntry>(); 2369 (void)!SymbolVersionsOrErr; 2370 } 2371 for (auto &Sym : Symbols) 2372 printSymbol(O, Sym, *SymbolVersionsOrErr, FileName, ArchiveName, 2373 ArchitectureName, DumpDynamic); 2374 } 2375 2376 void objdump::printSymbol(const ObjectFile &O, const SymbolRef &Symbol, 2377 ArrayRef<VersionEntry> SymbolVersions, 2378 StringRef FileName, StringRef ArchiveName, 2379 StringRef ArchitectureName, bool DumpDynamic) { 2380 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&O); 2381 uint64_t Address = unwrapOrError(Symbol.getAddress(), FileName, ArchiveName, 2382 ArchitectureName); 2383 if ((Address < StartAddress) || (Address > StopAddress)) 2384 return; 2385 SymbolRef::Type Type = 2386 unwrapOrError(Symbol.getType(), FileName, ArchiveName, ArchitectureName); 2387 uint32_t Flags = 2388 unwrapOrError(Symbol.getFlags(), FileName, ArchiveName, ArchitectureName); 2389 2390 // Don't ask a Mach-O STAB symbol for its section unless you know that 2391 // STAB symbol's section field refers to a valid section index. Otherwise 2392 // the symbol may error trying to load a section that does not exist. 2393 bool IsSTAB = false; 2394 if (MachO) { 2395 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 2396 uint8_t NType = 2397 (MachO->is64Bit() ? MachO->getSymbol64TableEntry(SymDRI).n_type 2398 : MachO->getSymbolTableEntry(SymDRI).n_type); 2399 if (NType & MachO::N_STAB) 2400 IsSTAB = true; 2401 } 2402 section_iterator Section = IsSTAB 2403 ? O.section_end() 2404 : unwrapOrError(Symbol.getSection(), FileName, 2405 ArchiveName, ArchitectureName); 2406 2407 StringRef Name; 2408 if (Type == SymbolRef::ST_Debug && Section != O.section_end()) { 2409 if (Expected<StringRef> NameOrErr = Section->getName()) 2410 Name = *NameOrErr; 2411 else 2412 consumeError(NameOrErr.takeError()); 2413 2414 } else { 2415 Name = unwrapOrError(Symbol.getName(), FileName, ArchiveName, 2416 ArchitectureName); 2417 } 2418 2419 bool Global = Flags & SymbolRef::SF_Global; 2420 bool Weak = Flags & SymbolRef::SF_Weak; 2421 bool Absolute = Flags & SymbolRef::SF_Absolute; 2422 bool Common = Flags & SymbolRef::SF_Common; 2423 bool Hidden = Flags & SymbolRef::SF_Hidden; 2424 2425 char GlobLoc = ' '; 2426 if ((Section != O.section_end() || Absolute) && !Weak) 2427 GlobLoc = Global ? 'g' : 'l'; 2428 char IFunc = ' '; 2429 if (O.isELF()) { 2430 if (ELFSymbolRef(Symbol).getELFType() == ELF::STT_GNU_IFUNC) 2431 IFunc = 'i'; 2432 if (ELFSymbolRef(Symbol).getBinding() == ELF::STB_GNU_UNIQUE) 2433 GlobLoc = 'u'; 2434 } 2435 2436 char Debug = ' '; 2437 if (DumpDynamic) 2438 Debug = 'D'; 2439 else if (Type == SymbolRef::ST_Debug || Type == SymbolRef::ST_File) 2440 Debug = 'd'; 2441 2442 char FileFunc = ' '; 2443 if (Type == SymbolRef::ST_File) 2444 FileFunc = 'f'; 2445 else if (Type == SymbolRef::ST_Function) 2446 FileFunc = 'F'; 2447 else if (Type == SymbolRef::ST_Data) 2448 FileFunc = 'O'; 2449 2450 const char *Fmt = O.getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2451 2452 outs() << format(Fmt, Address) << " " 2453 << GlobLoc // Local -> 'l', Global -> 'g', Neither -> ' ' 2454 << (Weak ? 'w' : ' ') // Weak? 2455 << ' ' // Constructor. Not supported yet. 2456 << ' ' // Warning. Not supported yet. 2457 << IFunc // Indirect reference to another symbol. 2458 << Debug // Debugging (d) or dynamic (D) symbol. 2459 << FileFunc // Name of function (F), file (f) or object (O). 2460 << ' '; 2461 if (Absolute) { 2462 outs() << "*ABS*"; 2463 } else if (Common) { 2464 outs() << "*COM*"; 2465 } else if (Section == O.section_end()) { 2466 if (O.isXCOFF()) { 2467 XCOFFSymbolRef XCOFFSym = cast<const XCOFFObjectFile>(O).toSymbolRef( 2468 Symbol.getRawDataRefImpl()); 2469 if (XCOFF::N_DEBUG == XCOFFSym.getSectionNumber()) 2470 outs() << "*DEBUG*"; 2471 else 2472 outs() << "*UND*"; 2473 } else 2474 outs() << "*UND*"; 2475 } else { 2476 StringRef SegmentName = getSegmentName(MachO, *Section); 2477 if (!SegmentName.empty()) 2478 outs() << SegmentName << ","; 2479 StringRef SectionName = unwrapOrError(Section->getName(), FileName); 2480 outs() << SectionName; 2481 if (O.isXCOFF()) { 2482 Optional<SymbolRef> SymRef = 2483 getXCOFFSymbolContainingSymbolRef(cast<XCOFFObjectFile>(O), Symbol); 2484 if (SymRef) { 2485 2486 Expected<StringRef> NameOrErr = SymRef->getName(); 2487 2488 if (NameOrErr) { 2489 outs() << " (csect:"; 2490 std::string SymName(NameOrErr.get()); 2491 2492 if (Demangle) 2493 SymName = demangle(SymName); 2494 2495 if (SymbolDescription) 2496 SymName = getXCOFFSymbolDescription( 2497 createSymbolInfo(O, SymRef.value()), SymName); 2498 2499 outs() << ' ' << SymName; 2500 outs() << ") "; 2501 } else 2502 reportWarning(toString(NameOrErr.takeError()), FileName); 2503 } 2504 } 2505 } 2506 2507 if (Common) 2508 outs() << '\t' << format(Fmt, static_cast<uint64_t>(Symbol.getAlignment())); 2509 else if (O.isXCOFF()) 2510 outs() << '\t' 2511 << format(Fmt, cast<XCOFFObjectFile>(O).getSymbolSize( 2512 Symbol.getRawDataRefImpl())); 2513 else if (O.isELF()) 2514 outs() << '\t' << format(Fmt, ELFSymbolRef(Symbol).getSize()); 2515 2516 if (O.isELF()) { 2517 if (!SymbolVersions.empty()) { 2518 const VersionEntry &Ver = 2519 SymbolVersions[Symbol.getRawDataRefImpl().d.b - 1]; 2520 std::string Str; 2521 if (!Ver.Name.empty()) 2522 Str = Ver.IsVerDef ? ' ' + Ver.Name : '(' + Ver.Name + ')'; 2523 outs() << ' ' << left_justify(Str, 12); 2524 } 2525 2526 uint8_t Other = ELFSymbolRef(Symbol).getOther(); 2527 switch (Other) { 2528 case ELF::STV_DEFAULT: 2529 break; 2530 case ELF::STV_INTERNAL: 2531 outs() << " .internal"; 2532 break; 2533 case ELF::STV_HIDDEN: 2534 outs() << " .hidden"; 2535 break; 2536 case ELF::STV_PROTECTED: 2537 outs() << " .protected"; 2538 break; 2539 default: 2540 outs() << format(" 0x%02x", Other); 2541 break; 2542 } 2543 } else if (Hidden) { 2544 outs() << " .hidden"; 2545 } 2546 2547 std::string SymName(Name); 2548 if (Demangle) 2549 SymName = demangle(SymName); 2550 2551 if (O.isXCOFF() && SymbolDescription) 2552 SymName = getXCOFFSymbolDescription(createSymbolInfo(O, Symbol), SymName); 2553 2554 outs() << ' ' << SymName << '\n'; 2555 } 2556 2557 static void printUnwindInfo(const ObjectFile *O) { 2558 outs() << "Unwind info:\n\n"; 2559 2560 if (const COFFObjectFile *Coff = dyn_cast<COFFObjectFile>(O)) 2561 printCOFFUnwindInfo(Coff); 2562 else if (const MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(O)) 2563 printMachOUnwindInfo(MachO); 2564 else 2565 // TODO: Extract DWARF dump tool to objdump. 2566 WithColor::error(errs(), ToolName) 2567 << "This operation is only currently supported " 2568 "for COFF and MachO object files.\n"; 2569 } 2570 2571 /// Dump the raw contents of the __clangast section so the output can be piped 2572 /// into llvm-bcanalyzer. 2573 static void printRawClangAST(const ObjectFile *Obj) { 2574 if (outs().is_displayed()) { 2575 WithColor::error(errs(), ToolName) 2576 << "The -raw-clang-ast option will dump the raw binary contents of " 2577 "the clang ast section.\n" 2578 "Please redirect the output to a file or another program such as " 2579 "llvm-bcanalyzer.\n"; 2580 return; 2581 } 2582 2583 StringRef ClangASTSectionName("__clangast"); 2584 if (Obj->isCOFF()) { 2585 ClangASTSectionName = "clangast"; 2586 } 2587 2588 std::optional<object::SectionRef> ClangASTSection; 2589 for (auto Sec : ToolSectionFilter(*Obj)) { 2590 StringRef Name; 2591 if (Expected<StringRef> NameOrErr = Sec.getName()) 2592 Name = *NameOrErr; 2593 else 2594 consumeError(NameOrErr.takeError()); 2595 2596 if (Name == ClangASTSectionName) { 2597 ClangASTSection = Sec; 2598 break; 2599 } 2600 } 2601 if (!ClangASTSection) 2602 return; 2603 2604 StringRef ClangASTContents = 2605 unwrapOrError(ClangASTSection.value().getContents(), Obj->getFileName()); 2606 outs().write(ClangASTContents.data(), ClangASTContents.size()); 2607 } 2608 2609 static void printFaultMaps(const ObjectFile *Obj) { 2610 StringRef FaultMapSectionName; 2611 2612 if (Obj->isELF()) { 2613 FaultMapSectionName = ".llvm_faultmaps"; 2614 } else if (Obj->isMachO()) { 2615 FaultMapSectionName = "__llvm_faultmaps"; 2616 } else { 2617 WithColor::error(errs(), ToolName) 2618 << "This operation is only currently supported " 2619 "for ELF and Mach-O executable files.\n"; 2620 return; 2621 } 2622 2623 std::optional<object::SectionRef> FaultMapSection; 2624 2625 for (auto Sec : ToolSectionFilter(*Obj)) { 2626 StringRef Name; 2627 if (Expected<StringRef> NameOrErr = Sec.getName()) 2628 Name = *NameOrErr; 2629 else 2630 consumeError(NameOrErr.takeError()); 2631 2632 if (Name == FaultMapSectionName) { 2633 FaultMapSection = Sec; 2634 break; 2635 } 2636 } 2637 2638 outs() << "FaultMap table:\n"; 2639 2640 if (!FaultMapSection) { 2641 outs() << "<not found>\n"; 2642 return; 2643 } 2644 2645 StringRef FaultMapContents = 2646 unwrapOrError(FaultMapSection->getContents(), Obj->getFileName()); 2647 FaultMapParser FMP(FaultMapContents.bytes_begin(), 2648 FaultMapContents.bytes_end()); 2649 2650 outs() << FMP; 2651 } 2652 2653 static void printPrivateFileHeaders(const ObjectFile *O, bool OnlyFirst) { 2654 if (O->isELF()) { 2655 printELFFileHeader(O); 2656 printELFDynamicSection(O); 2657 printELFSymbolVersionInfo(O); 2658 return; 2659 } 2660 if (O->isCOFF()) 2661 return printCOFFFileHeader(cast<object::COFFObjectFile>(*O)); 2662 if (O->isWasm()) 2663 return printWasmFileHeader(O); 2664 if (O->isMachO()) { 2665 printMachOFileHeader(O); 2666 if (!OnlyFirst) 2667 printMachOLoadCommands(O); 2668 return; 2669 } 2670 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2671 } 2672 2673 static void printFileHeaders(const ObjectFile *O) { 2674 if (!O->isELF() && !O->isCOFF()) 2675 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2676 2677 Triple::ArchType AT = O->getArch(); 2678 outs() << "architecture: " << Triple::getArchTypeName(AT) << "\n"; 2679 uint64_t Address = unwrapOrError(O->getStartAddress(), O->getFileName()); 2680 2681 StringRef Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2682 outs() << "start address: " 2683 << "0x" << format(Fmt.data(), Address) << "\n"; 2684 } 2685 2686 static void printArchiveChild(StringRef Filename, const Archive::Child &C) { 2687 Expected<sys::fs::perms> ModeOrErr = C.getAccessMode(); 2688 if (!ModeOrErr) { 2689 WithColor::error(errs(), ToolName) << "ill-formed archive entry.\n"; 2690 consumeError(ModeOrErr.takeError()); 2691 return; 2692 } 2693 sys::fs::perms Mode = ModeOrErr.get(); 2694 outs() << ((Mode & sys::fs::owner_read) ? "r" : "-"); 2695 outs() << ((Mode & sys::fs::owner_write) ? "w" : "-"); 2696 outs() << ((Mode & sys::fs::owner_exe) ? "x" : "-"); 2697 outs() << ((Mode & sys::fs::group_read) ? "r" : "-"); 2698 outs() << ((Mode & sys::fs::group_write) ? "w" : "-"); 2699 outs() << ((Mode & sys::fs::group_exe) ? "x" : "-"); 2700 outs() << ((Mode & sys::fs::others_read) ? "r" : "-"); 2701 outs() << ((Mode & sys::fs::others_write) ? "w" : "-"); 2702 outs() << ((Mode & sys::fs::others_exe) ? "x" : "-"); 2703 2704 outs() << " "; 2705 2706 outs() << format("%d/%d %6" PRId64 " ", unwrapOrError(C.getUID(), Filename), 2707 unwrapOrError(C.getGID(), Filename), 2708 unwrapOrError(C.getRawSize(), Filename)); 2709 2710 StringRef RawLastModified = C.getRawLastModified(); 2711 unsigned Seconds; 2712 if (RawLastModified.getAsInteger(10, Seconds)) 2713 outs() << "(date: \"" << RawLastModified 2714 << "\" contains non-decimal chars) "; 2715 else { 2716 // Since ctime(3) returns a 26 character string of the form: 2717 // "Sun Sep 16 01:03:52 1973\n\0" 2718 // just print 24 characters. 2719 time_t t = Seconds; 2720 outs() << format("%.24s ", ctime(&t)); 2721 } 2722 2723 StringRef Name = ""; 2724 Expected<StringRef> NameOrErr = C.getName(); 2725 if (!NameOrErr) { 2726 consumeError(NameOrErr.takeError()); 2727 Name = unwrapOrError(C.getRawName(), Filename); 2728 } else { 2729 Name = NameOrErr.get(); 2730 } 2731 outs() << Name << "\n"; 2732 } 2733 2734 // For ELF only now. 2735 static bool shouldWarnForInvalidStartStopAddress(ObjectFile *Obj) { 2736 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj)) { 2737 if (Elf->getEType() != ELF::ET_REL) 2738 return true; 2739 } 2740 return false; 2741 } 2742 2743 static void checkForInvalidStartStopAddress(ObjectFile *Obj, 2744 uint64_t Start, uint64_t Stop) { 2745 if (!shouldWarnForInvalidStartStopAddress(Obj)) 2746 return; 2747 2748 for (const SectionRef &Section : Obj->sections()) 2749 if (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC) { 2750 uint64_t BaseAddr = Section.getAddress(); 2751 uint64_t Size = Section.getSize(); 2752 if ((Start < BaseAddr + Size) && Stop > BaseAddr) 2753 return; 2754 } 2755 2756 if (!HasStartAddressFlag) 2757 reportWarning("no section has address less than 0x" + 2758 Twine::utohexstr(Stop) + " specified by --stop-address", 2759 Obj->getFileName()); 2760 else if (!HasStopAddressFlag) 2761 reportWarning("no section has address greater than or equal to 0x" + 2762 Twine::utohexstr(Start) + " specified by --start-address", 2763 Obj->getFileName()); 2764 else 2765 reportWarning("no section overlaps the range [0x" + 2766 Twine::utohexstr(Start) + ",0x" + Twine::utohexstr(Stop) + 2767 ") specified by --start-address/--stop-address", 2768 Obj->getFileName()); 2769 } 2770 2771 static void dumpObject(ObjectFile *O, const Archive *A = nullptr, 2772 const Archive::Child *C = nullptr) { 2773 // Avoid other output when using a raw option. 2774 if (!RawClangAST) { 2775 outs() << '\n'; 2776 if (A) 2777 outs() << A->getFileName() << "(" << O->getFileName() << ")"; 2778 else 2779 outs() << O->getFileName(); 2780 outs() << ":\tfile format " << O->getFileFormatName().lower() << "\n"; 2781 } 2782 2783 if (HasStartAddressFlag || HasStopAddressFlag) 2784 checkForInvalidStartStopAddress(O, StartAddress, StopAddress); 2785 2786 // Note: the order here matches GNU objdump for compatability. 2787 StringRef ArchiveName = A ? A->getFileName() : ""; 2788 if (ArchiveHeaders && !MachOOpt && C) 2789 printArchiveChild(ArchiveName, *C); 2790 if (FileHeaders) 2791 printFileHeaders(O); 2792 if (PrivateHeaders || FirstPrivateHeader) 2793 printPrivateFileHeaders(O, FirstPrivateHeader); 2794 if (SectionHeaders) 2795 printSectionHeaders(*O); 2796 if (SymbolTable) 2797 printSymbolTable(*O, ArchiveName); 2798 if (DynamicSymbolTable) 2799 printSymbolTable(*O, ArchiveName, /*ArchitectureName=*/"", 2800 /*DumpDynamic=*/true); 2801 if (DwarfDumpType != DIDT_Null) { 2802 std::unique_ptr<DIContext> DICtx = DWARFContext::create(*O); 2803 // Dump the complete DWARF structure. 2804 DIDumpOptions DumpOpts; 2805 DumpOpts.DumpType = DwarfDumpType; 2806 DICtx->dump(outs(), DumpOpts); 2807 } 2808 if (Relocations && !Disassemble) 2809 printRelocations(O); 2810 if (DynamicRelocations) 2811 printDynamicRelocations(O); 2812 if (SectionContents) 2813 printSectionContents(O); 2814 if (Disassemble) 2815 disassembleObject(O, Relocations); 2816 if (UnwindInfo) 2817 printUnwindInfo(O); 2818 2819 // Mach-O specific options: 2820 if (ExportsTrie) 2821 printExportsTrie(O); 2822 if (Rebase) 2823 printRebaseTable(O); 2824 if (Bind) 2825 printBindTable(O); 2826 if (LazyBind) 2827 printLazyBindTable(O); 2828 if (WeakBind) 2829 printWeakBindTable(O); 2830 2831 // Other special sections: 2832 if (RawClangAST) 2833 printRawClangAST(O); 2834 if (FaultMapSection) 2835 printFaultMaps(O); 2836 if (Offloading) 2837 dumpOffloadBinary(*O); 2838 } 2839 2840 static void dumpObject(const COFFImportFile *I, const Archive *A, 2841 const Archive::Child *C = nullptr) { 2842 StringRef ArchiveName = A ? A->getFileName() : ""; 2843 2844 // Avoid other output when using a raw option. 2845 if (!RawClangAST) 2846 outs() << '\n' 2847 << ArchiveName << "(" << I->getFileName() << ")" 2848 << ":\tfile format COFF-import-file" 2849 << "\n\n"; 2850 2851 if (ArchiveHeaders && !MachOOpt && C) 2852 printArchiveChild(ArchiveName, *C); 2853 if (SymbolTable) 2854 printCOFFSymbolTable(*I); 2855 } 2856 2857 /// Dump each object file in \a a; 2858 static void dumpArchive(const Archive *A) { 2859 Error Err = Error::success(); 2860 unsigned I = -1; 2861 for (auto &C : A->children(Err)) { 2862 ++I; 2863 Expected<std::unique_ptr<Binary>> ChildOrErr = C.getAsBinary(); 2864 if (!ChildOrErr) { 2865 if (auto E = isNotObjectErrorInvalidFileType(ChildOrErr.takeError())) 2866 reportError(std::move(E), getFileNameForError(C, I), A->getFileName()); 2867 continue; 2868 } 2869 if (ObjectFile *O = dyn_cast<ObjectFile>(&*ChildOrErr.get())) 2870 dumpObject(O, A, &C); 2871 else if (COFFImportFile *I = dyn_cast<COFFImportFile>(&*ChildOrErr.get())) 2872 dumpObject(I, A, &C); 2873 else 2874 reportError(errorCodeToError(object_error::invalid_file_type), 2875 A->getFileName()); 2876 } 2877 if (Err) 2878 reportError(std::move(Err), A->getFileName()); 2879 } 2880 2881 /// Open file and figure out how to dump it. 2882 static void dumpInput(StringRef file) { 2883 // If we are using the Mach-O specific object file parser, then let it parse 2884 // the file and process the command line options. So the -arch flags can 2885 // be used to select specific slices, etc. 2886 if (MachOOpt) { 2887 parseInputMachO(file); 2888 return; 2889 } 2890 2891 // Attempt to open the binary. 2892 OwningBinary<Binary> OBinary = unwrapOrError(createBinary(file), file); 2893 Binary &Binary = *OBinary.getBinary(); 2894 2895 if (Archive *A = dyn_cast<Archive>(&Binary)) 2896 dumpArchive(A); 2897 else if (ObjectFile *O = dyn_cast<ObjectFile>(&Binary)) 2898 dumpObject(O); 2899 else if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(&Binary)) 2900 parseInputMachO(UB); 2901 else if (OffloadBinary *OB = dyn_cast<OffloadBinary>(&Binary)) 2902 dumpOffloadSections(*OB); 2903 else 2904 reportError(errorCodeToError(object_error::invalid_file_type), file); 2905 } 2906 2907 template <typename T> 2908 static void parseIntArg(const llvm::opt::InputArgList &InputArgs, int ID, 2909 T &Value) { 2910 if (const opt::Arg *A = InputArgs.getLastArg(ID)) { 2911 StringRef V(A->getValue()); 2912 if (!llvm::to_integer(V, Value, 0)) { 2913 reportCmdLineError(A->getSpelling() + 2914 ": expected a non-negative integer, but got '" + V + 2915 "'"); 2916 } 2917 } 2918 } 2919 2920 static object::BuildID parseBuildIDArg(const opt::Arg *A) { 2921 StringRef V(A->getValue()); 2922 std::string Bytes; 2923 if (!tryGetFromHex(V, Bytes)) 2924 reportCmdLineError(A->getSpelling() + ": expected a build ID, but got '" + 2925 V + "'"); 2926 ArrayRef<uint8_t> BuildID(reinterpret_cast<const uint8_t *>(Bytes.data()), 2927 Bytes.size()); 2928 return object::BuildID(BuildID.begin(), BuildID.end()); 2929 } 2930 2931 void objdump::invalidArgValue(const opt::Arg *A) { 2932 reportCmdLineError("'" + StringRef(A->getValue()) + 2933 "' is not a valid value for '" + A->getSpelling() + "'"); 2934 } 2935 2936 static std::vector<std::string> 2937 commaSeparatedValues(const llvm::opt::InputArgList &InputArgs, int ID) { 2938 std::vector<std::string> Values; 2939 for (StringRef Value : InputArgs.getAllArgValues(ID)) { 2940 llvm::SmallVector<StringRef, 2> SplitValues; 2941 llvm::SplitString(Value, SplitValues, ","); 2942 for (StringRef SplitValue : SplitValues) 2943 Values.push_back(SplitValue.str()); 2944 } 2945 return Values; 2946 } 2947 2948 static void parseOtoolOptions(const llvm::opt::InputArgList &InputArgs) { 2949 MachOOpt = true; 2950 FullLeadingAddr = true; 2951 PrintImmHex = true; 2952 2953 ArchName = InputArgs.getLastArgValue(OTOOL_arch).str(); 2954 LinkOptHints = InputArgs.hasArg(OTOOL_C); 2955 if (InputArgs.hasArg(OTOOL_d)) 2956 FilterSections.push_back("__DATA,__data"); 2957 DylibId = InputArgs.hasArg(OTOOL_D); 2958 UniversalHeaders = InputArgs.hasArg(OTOOL_f); 2959 DataInCode = InputArgs.hasArg(OTOOL_G); 2960 FirstPrivateHeader = InputArgs.hasArg(OTOOL_h); 2961 IndirectSymbols = InputArgs.hasArg(OTOOL_I); 2962 ShowRawInsn = InputArgs.hasArg(OTOOL_j); 2963 PrivateHeaders = InputArgs.hasArg(OTOOL_l); 2964 DylibsUsed = InputArgs.hasArg(OTOOL_L); 2965 MCPU = InputArgs.getLastArgValue(OTOOL_mcpu_EQ).str(); 2966 ObjcMetaData = InputArgs.hasArg(OTOOL_o); 2967 DisSymName = InputArgs.getLastArgValue(OTOOL_p).str(); 2968 InfoPlist = InputArgs.hasArg(OTOOL_P); 2969 Relocations = InputArgs.hasArg(OTOOL_r); 2970 if (const Arg *A = InputArgs.getLastArg(OTOOL_s)) { 2971 auto Filter = (A->getValue(0) + StringRef(",") + A->getValue(1)).str(); 2972 FilterSections.push_back(Filter); 2973 } 2974 if (InputArgs.hasArg(OTOOL_t)) 2975 FilterSections.push_back("__TEXT,__text"); 2976 Verbose = InputArgs.hasArg(OTOOL_v) || InputArgs.hasArg(OTOOL_V) || 2977 InputArgs.hasArg(OTOOL_o); 2978 SymbolicOperands = InputArgs.hasArg(OTOOL_V); 2979 if (InputArgs.hasArg(OTOOL_x)) 2980 FilterSections.push_back(",__text"); 2981 LeadingAddr = LeadingHeaders = !InputArgs.hasArg(OTOOL_X); 2982 2983 ChainedFixups = InputArgs.hasArg(OTOOL_chained_fixups); 2984 DyldInfo = InputArgs.hasArg(OTOOL_dyld_info); 2985 2986 InputFilenames = InputArgs.getAllArgValues(OTOOL_INPUT); 2987 if (InputFilenames.empty()) 2988 reportCmdLineError("no input file"); 2989 2990 for (const Arg *A : InputArgs) { 2991 const Option &O = A->getOption(); 2992 if (O.getGroup().isValid() && O.getGroup().getID() == OTOOL_grp_obsolete) { 2993 reportCmdLineWarning(O.getPrefixedName() + 2994 " is obsolete and not implemented"); 2995 } 2996 } 2997 } 2998 2999 static void parseObjdumpOptions(const llvm::opt::InputArgList &InputArgs) { 3000 parseIntArg(InputArgs, OBJDUMP_adjust_vma_EQ, AdjustVMA); 3001 AllHeaders = InputArgs.hasArg(OBJDUMP_all_headers); 3002 ArchName = InputArgs.getLastArgValue(OBJDUMP_arch_name_EQ).str(); 3003 ArchiveHeaders = InputArgs.hasArg(OBJDUMP_archive_headers); 3004 Demangle = InputArgs.hasArg(OBJDUMP_demangle); 3005 Disassemble = InputArgs.hasArg(OBJDUMP_disassemble); 3006 DisassembleAll = InputArgs.hasArg(OBJDUMP_disassemble_all); 3007 SymbolDescription = InputArgs.hasArg(OBJDUMP_symbol_description); 3008 DisassembleSymbols = 3009 commaSeparatedValues(InputArgs, OBJDUMP_disassemble_symbols_EQ); 3010 DisassembleZeroes = InputArgs.hasArg(OBJDUMP_disassemble_zeroes); 3011 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_dwarf_EQ)) { 3012 DwarfDumpType = StringSwitch<DIDumpType>(A->getValue()) 3013 .Case("frames", DIDT_DebugFrame) 3014 .Default(DIDT_Null); 3015 if (DwarfDumpType == DIDT_Null) 3016 invalidArgValue(A); 3017 } 3018 DynamicRelocations = InputArgs.hasArg(OBJDUMP_dynamic_reloc); 3019 FaultMapSection = InputArgs.hasArg(OBJDUMP_fault_map_section); 3020 Offloading = InputArgs.hasArg(OBJDUMP_offloading); 3021 FileHeaders = InputArgs.hasArg(OBJDUMP_file_headers); 3022 SectionContents = InputArgs.hasArg(OBJDUMP_full_contents); 3023 PrintLines = InputArgs.hasArg(OBJDUMP_line_numbers); 3024 InputFilenames = InputArgs.getAllArgValues(OBJDUMP_INPUT); 3025 MachOOpt = InputArgs.hasArg(OBJDUMP_macho); 3026 MCPU = InputArgs.getLastArgValue(OBJDUMP_mcpu_EQ).str(); 3027 MAttrs = commaSeparatedValues(InputArgs, OBJDUMP_mattr_EQ); 3028 ShowRawInsn = !InputArgs.hasArg(OBJDUMP_no_show_raw_insn); 3029 LeadingAddr = !InputArgs.hasArg(OBJDUMP_no_leading_addr); 3030 RawClangAST = InputArgs.hasArg(OBJDUMP_raw_clang_ast); 3031 Relocations = InputArgs.hasArg(OBJDUMP_reloc); 3032 PrintImmHex = 3033 InputArgs.hasFlag(OBJDUMP_print_imm_hex, OBJDUMP_no_print_imm_hex, true); 3034 PrivateHeaders = InputArgs.hasArg(OBJDUMP_private_headers); 3035 FilterSections = InputArgs.getAllArgValues(OBJDUMP_section_EQ); 3036 SectionHeaders = InputArgs.hasArg(OBJDUMP_section_headers); 3037 ShowAllSymbols = InputArgs.hasArg(OBJDUMP_show_all_symbols); 3038 ShowLMA = InputArgs.hasArg(OBJDUMP_show_lma); 3039 PrintSource = InputArgs.hasArg(OBJDUMP_source); 3040 parseIntArg(InputArgs, OBJDUMP_start_address_EQ, StartAddress); 3041 HasStartAddressFlag = InputArgs.hasArg(OBJDUMP_start_address_EQ); 3042 parseIntArg(InputArgs, OBJDUMP_stop_address_EQ, StopAddress); 3043 HasStopAddressFlag = InputArgs.hasArg(OBJDUMP_stop_address_EQ); 3044 SymbolTable = InputArgs.hasArg(OBJDUMP_syms); 3045 SymbolizeOperands = InputArgs.hasArg(OBJDUMP_symbolize_operands); 3046 DynamicSymbolTable = InputArgs.hasArg(OBJDUMP_dynamic_syms); 3047 TripleName = InputArgs.getLastArgValue(OBJDUMP_triple_EQ).str(); 3048 UnwindInfo = InputArgs.hasArg(OBJDUMP_unwind_info); 3049 Wide = InputArgs.hasArg(OBJDUMP_wide); 3050 Prefix = InputArgs.getLastArgValue(OBJDUMP_prefix).str(); 3051 parseIntArg(InputArgs, OBJDUMP_prefix_strip, PrefixStrip); 3052 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_debug_vars_EQ)) { 3053 DbgVariables = StringSwitch<DebugVarsFormat>(A->getValue()) 3054 .Case("ascii", DVASCII) 3055 .Case("unicode", DVUnicode) 3056 .Default(DVInvalid); 3057 if (DbgVariables == DVInvalid) 3058 invalidArgValue(A); 3059 } 3060 parseIntArg(InputArgs, OBJDUMP_debug_vars_indent_EQ, DbgIndent); 3061 3062 parseMachOOptions(InputArgs); 3063 3064 // Parse -M (--disassembler-options) and deprecated 3065 // --x86-asm-syntax={att,intel}. 3066 // 3067 // Note, for x86, the asm dialect (AssemblerDialect) is initialized when the 3068 // MCAsmInfo is constructed. MCInstPrinter::applyTargetSpecificCLOption is 3069 // called too late. For now we have to use the internal cl::opt option. 3070 const char *AsmSyntax = nullptr; 3071 for (const auto *A : InputArgs.filtered(OBJDUMP_disassembler_options_EQ, 3072 OBJDUMP_x86_asm_syntax_att, 3073 OBJDUMP_x86_asm_syntax_intel)) { 3074 switch (A->getOption().getID()) { 3075 case OBJDUMP_x86_asm_syntax_att: 3076 AsmSyntax = "--x86-asm-syntax=att"; 3077 continue; 3078 case OBJDUMP_x86_asm_syntax_intel: 3079 AsmSyntax = "--x86-asm-syntax=intel"; 3080 continue; 3081 } 3082 3083 SmallVector<StringRef, 2> Values; 3084 llvm::SplitString(A->getValue(), Values, ","); 3085 for (StringRef V : Values) { 3086 if (V == "att") 3087 AsmSyntax = "--x86-asm-syntax=att"; 3088 else if (V == "intel") 3089 AsmSyntax = "--x86-asm-syntax=intel"; 3090 else 3091 DisassemblerOptions.push_back(V.str()); 3092 } 3093 } 3094 if (AsmSyntax) { 3095 const char *Argv[] = {"llvm-objdump", AsmSyntax}; 3096 llvm::cl::ParseCommandLineOptions(2, Argv); 3097 } 3098 3099 // Look up any provided build IDs, then append them to the input filenames. 3100 for (const opt::Arg *A : InputArgs.filtered(OBJDUMP_build_id)) { 3101 object::BuildID BuildID = parseBuildIDArg(A); 3102 Optional<std::string> Path = BIDFetcher->fetch(BuildID); 3103 if (!Path) { 3104 reportCmdLineError(A->getSpelling() + ": could not find build ID '" + 3105 A->getValue() + "'"); 3106 } 3107 InputFilenames.push_back(std::move(*Path)); 3108 } 3109 3110 // objdump defaults to a.out if no filenames specified. 3111 if (InputFilenames.empty()) 3112 InputFilenames.push_back("a.out"); 3113 } 3114 3115 int main(int argc, char **argv) { 3116 using namespace llvm; 3117 InitLLVM X(argc, argv); 3118 3119 ToolName = argv[0]; 3120 std::unique_ptr<CommonOptTable> T; 3121 OptSpecifier Unknown, HelpFlag, HelpHiddenFlag, VersionFlag; 3122 3123 StringRef Stem = sys::path::stem(ToolName); 3124 auto Is = [=](StringRef Tool) { 3125 // We need to recognize the following filenames: 3126 // 3127 // llvm-objdump -> objdump 3128 // llvm-otool-10.exe -> otool 3129 // powerpc64-unknown-freebsd13-objdump -> objdump 3130 auto I = Stem.rfind_insensitive(Tool); 3131 return I != StringRef::npos && 3132 (I + Tool.size() == Stem.size() || !isAlnum(Stem[I + Tool.size()])); 3133 }; 3134 if (Is("otool")) { 3135 T = std::make_unique<OtoolOptTable>(); 3136 Unknown = OTOOL_UNKNOWN; 3137 HelpFlag = OTOOL_help; 3138 HelpHiddenFlag = OTOOL_help_hidden; 3139 VersionFlag = OTOOL_version; 3140 } else { 3141 T = std::make_unique<ObjdumpOptTable>(); 3142 Unknown = OBJDUMP_UNKNOWN; 3143 HelpFlag = OBJDUMP_help; 3144 HelpHiddenFlag = OBJDUMP_help_hidden; 3145 VersionFlag = OBJDUMP_version; 3146 } 3147 3148 BumpPtrAllocator A; 3149 StringSaver Saver(A); 3150 opt::InputArgList InputArgs = 3151 T->parseArgs(argc, argv, Unknown, Saver, 3152 [&](StringRef Msg) { reportCmdLineError(Msg); }); 3153 3154 if (InputArgs.size() == 0 || InputArgs.hasArg(HelpFlag)) { 3155 T->printHelp(ToolName); 3156 return 0; 3157 } 3158 if (InputArgs.hasArg(HelpHiddenFlag)) { 3159 T->printHelp(ToolName, /*ShowHidden=*/true); 3160 return 0; 3161 } 3162 3163 // Initialize targets and assembly printers/parsers. 3164 InitializeAllTargetInfos(); 3165 InitializeAllTargetMCs(); 3166 InitializeAllDisassemblers(); 3167 3168 if (InputArgs.hasArg(VersionFlag)) { 3169 cl::PrintVersionMessage(); 3170 if (!Is("otool")) { 3171 outs() << '\n'; 3172 TargetRegistry::printRegisteredTargetsForVersion(outs()); 3173 } 3174 return 0; 3175 } 3176 3177 // Initialize debuginfod. 3178 const bool ShouldUseDebuginfodByDefault = 3179 InputArgs.hasArg(OBJDUMP_build_id) || 3180 (HTTPClient::isAvailable() && 3181 !ExitOnErr(getDefaultDebuginfodUrls()).empty()); 3182 std::vector<std::string> DebugFileDirectories = 3183 InputArgs.getAllArgValues(OBJDUMP_debug_file_directory); 3184 if (InputArgs.hasFlag(OBJDUMP_debuginfod, OBJDUMP_no_debuginfod, 3185 ShouldUseDebuginfodByDefault)) { 3186 HTTPClient::initialize(); 3187 BIDFetcher = 3188 std::make_unique<DebuginfodFetcher>(std::move(DebugFileDirectories)); 3189 } else { 3190 BIDFetcher = 3191 std::make_unique<BuildIDFetcher>(std::move(DebugFileDirectories)); 3192 } 3193 3194 if (Is("otool")) 3195 parseOtoolOptions(InputArgs); 3196 else 3197 parseObjdumpOptions(InputArgs); 3198 3199 if (StartAddress >= StopAddress) 3200 reportCmdLineError("start address should be less than stop address"); 3201 3202 // Removes trailing separators from prefix. 3203 while (!Prefix.empty() && sys::path::is_separator(Prefix.back())) 3204 Prefix.pop_back(); 3205 3206 if (AllHeaders) 3207 ArchiveHeaders = FileHeaders = PrivateHeaders = Relocations = 3208 SectionHeaders = SymbolTable = true; 3209 3210 if (DisassembleAll || PrintSource || PrintLines || 3211 !DisassembleSymbols.empty()) 3212 Disassemble = true; 3213 3214 if (!ArchiveHeaders && !Disassemble && DwarfDumpType == DIDT_Null && 3215 !DynamicRelocations && !FileHeaders && !PrivateHeaders && !RawClangAST && 3216 !Relocations && !SectionHeaders && !SectionContents && !SymbolTable && 3217 !DynamicSymbolTable && !UnwindInfo && !FaultMapSection && !Offloading && 3218 !(MachOOpt && 3219 (Bind || DataInCode || ChainedFixups || DyldInfo || DylibId || 3220 DylibsUsed || ExportsTrie || FirstPrivateHeader || 3221 FunctionStartsType != FunctionStartsMode::None || IndirectSymbols || 3222 InfoPlist || LazyBind || LinkOptHints || ObjcMetaData || Rebase || 3223 Rpaths || UniversalHeaders || WeakBind || !FilterSections.empty()))) { 3224 T->printHelp(ToolName); 3225 return 2; 3226 } 3227 3228 DisasmSymbolSet.insert(DisassembleSymbols.begin(), DisassembleSymbols.end()); 3229 3230 llvm::for_each(InputFilenames, dumpInput); 3231 3232 warnOnNoMatchForSections(); 3233 3234 return EXIT_SUCCESS; 3235 } 3236