1 //===-- llvm-objdump.cpp - Object file dumping utility for llvm -----------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This program is a utility that works like binutils "objdump", that is, it 10 // dumps out a plethora of information about an object file depending on the 11 // flags. 12 // 13 // The flags and output of this program should be near identical to those of 14 // binutils objdump. 15 // 16 //===----------------------------------------------------------------------===// 17 18 #include "llvm-objdump.h" 19 #include "COFFDump.h" 20 #include "ELFDump.h" 21 #include "MachODump.h" 22 #include "ObjdumpOptID.h" 23 #include "OffloadDump.h" 24 #include "SourcePrinter.h" 25 #include "WasmDump.h" 26 #include "XCOFFDump.h" 27 #include "llvm/ADT/IndexedMap.h" 28 #include "llvm/ADT/Optional.h" 29 #include "llvm/ADT/STLExtras.h" 30 #include "llvm/ADT/SetOperations.h" 31 #include "llvm/ADT/SmallSet.h" 32 #include "llvm/ADT/StringExtras.h" 33 #include "llvm/ADT/StringSet.h" 34 #include "llvm/ADT/Triple.h" 35 #include "llvm/ADT/Twine.h" 36 #include "llvm/DebugInfo/DWARF/DWARFContext.h" 37 #include "llvm/DebugInfo/Symbolize/SymbolizableModule.h" 38 #include "llvm/DebugInfo/Symbolize/Symbolize.h" 39 #include "llvm/Demangle/Demangle.h" 40 #include "llvm/MC/MCAsmInfo.h" 41 #include "llvm/MC/MCContext.h" 42 #include "llvm/MC/MCDisassembler/MCDisassembler.h" 43 #include "llvm/MC/MCDisassembler/MCRelocationInfo.h" 44 #include "llvm/MC/MCInst.h" 45 #include "llvm/MC/MCInstPrinter.h" 46 #include "llvm/MC/MCInstrAnalysis.h" 47 #include "llvm/MC/MCInstrInfo.h" 48 #include "llvm/MC/MCObjectFileInfo.h" 49 #include "llvm/MC/MCRegisterInfo.h" 50 #include "llvm/MC/MCSubtargetInfo.h" 51 #include "llvm/MC/MCTargetOptions.h" 52 #include "llvm/MC/TargetRegistry.h" 53 #include "llvm/Object/Archive.h" 54 #include "llvm/Object/COFF.h" 55 #include "llvm/Object/COFFImportFile.h" 56 #include "llvm/Object/ELFObjectFile.h" 57 #include "llvm/Object/ELFTypes.h" 58 #include "llvm/Object/FaultMapParser.h" 59 #include "llvm/Object/MachO.h" 60 #include "llvm/Object/MachOUniversal.h" 61 #include "llvm/Object/ObjectFile.h" 62 #include "llvm/Object/OffloadBinary.h" 63 #include "llvm/Object/Wasm.h" 64 #include "llvm/Option/Arg.h" 65 #include "llvm/Option/ArgList.h" 66 #include "llvm/Option/Option.h" 67 #include "llvm/Support/Casting.h" 68 #include "llvm/Support/Debug.h" 69 #include "llvm/Support/Errc.h" 70 #include "llvm/Support/FileSystem.h" 71 #include "llvm/Support/Format.h" 72 #include "llvm/Support/FormatVariadic.h" 73 #include "llvm/Support/GraphWriter.h" 74 #include "llvm/Support/Host.h" 75 #include "llvm/Support/InitLLVM.h" 76 #include "llvm/Support/MemoryBuffer.h" 77 #include "llvm/Support/SourceMgr.h" 78 #include "llvm/Support/StringSaver.h" 79 #include "llvm/Support/TargetSelect.h" 80 #include "llvm/Support/WithColor.h" 81 #include "llvm/Support/raw_ostream.h" 82 #include <algorithm> 83 #include <cctype> 84 #include <cstring> 85 #include <system_error> 86 #include <unordered_map> 87 #include <utility> 88 89 using namespace llvm; 90 using namespace llvm::object; 91 using namespace llvm::objdump; 92 using namespace llvm::opt; 93 94 namespace { 95 96 class CommonOptTable : public opt::OptTable { 97 public: 98 CommonOptTable(ArrayRef<Info> OptionInfos, const char *Usage, 99 const char *Description) 100 : OptTable(OptionInfos), Usage(Usage), Description(Description) { 101 setGroupedShortOptions(true); 102 } 103 104 void printHelp(StringRef Argv0, bool ShowHidden = false) const { 105 Argv0 = sys::path::filename(Argv0); 106 opt::OptTable::printHelp(outs(), (Argv0 + Usage).str().c_str(), Description, 107 ShowHidden, ShowHidden); 108 // TODO Replace this with OptTable API once it adds extrahelp support. 109 outs() << "\nPass @FILE as argument to read options from FILE.\n"; 110 } 111 112 private: 113 const char *Usage; 114 const char *Description; 115 }; 116 117 // ObjdumpOptID is in ObjdumpOptID.h 118 119 #define PREFIX(NAME, VALUE) const char *const OBJDUMP_##NAME[] = VALUE; 120 #include "ObjdumpOpts.inc" 121 #undef PREFIX 122 123 static constexpr opt::OptTable::Info ObjdumpInfoTable[] = { 124 #define OBJDUMP_nullptr nullptr 125 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 126 HELPTEXT, METAVAR, VALUES) \ 127 {OBJDUMP_##PREFIX, NAME, HELPTEXT, \ 128 METAVAR, OBJDUMP_##ID, opt::Option::KIND##Class, \ 129 PARAM, FLAGS, OBJDUMP_##GROUP, \ 130 OBJDUMP_##ALIAS, ALIASARGS, VALUES}, 131 #include "ObjdumpOpts.inc" 132 #undef OPTION 133 #undef OBJDUMP_nullptr 134 }; 135 136 class ObjdumpOptTable : public CommonOptTable { 137 public: 138 ObjdumpOptTable() 139 : CommonOptTable(ObjdumpInfoTable, " [options] <input object files>", 140 "llvm object file dumper") {} 141 }; 142 143 enum OtoolOptID { 144 OTOOL_INVALID = 0, // This is not an option ID. 145 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 146 HELPTEXT, METAVAR, VALUES) \ 147 OTOOL_##ID, 148 #include "OtoolOpts.inc" 149 #undef OPTION 150 }; 151 152 #define PREFIX(NAME, VALUE) const char *const OTOOL_##NAME[] = VALUE; 153 #include "OtoolOpts.inc" 154 #undef PREFIX 155 156 static constexpr opt::OptTable::Info OtoolInfoTable[] = { 157 #define OTOOL_nullptr nullptr 158 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 159 HELPTEXT, METAVAR, VALUES) \ 160 {OTOOL_##PREFIX, NAME, HELPTEXT, \ 161 METAVAR, OTOOL_##ID, opt::Option::KIND##Class, \ 162 PARAM, FLAGS, OTOOL_##GROUP, \ 163 OTOOL_##ALIAS, ALIASARGS, VALUES}, 164 #include "OtoolOpts.inc" 165 #undef OPTION 166 #undef OTOOL_nullptr 167 }; 168 169 class OtoolOptTable : public CommonOptTable { 170 public: 171 OtoolOptTable() 172 : CommonOptTable(OtoolInfoTable, " [option...] [file...]", 173 "Mach-O object file displaying tool") {} 174 }; 175 176 } // namespace 177 178 #define DEBUG_TYPE "objdump" 179 180 static uint64_t AdjustVMA; 181 static bool AllHeaders; 182 static std::string ArchName; 183 bool objdump::ArchiveHeaders; 184 bool objdump::Demangle; 185 bool objdump::Disassemble; 186 bool objdump::DisassembleAll; 187 bool objdump::SymbolDescription; 188 static std::vector<std::string> DisassembleSymbols; 189 static bool DisassembleZeroes; 190 static std::vector<std::string> DisassemblerOptions; 191 DIDumpType objdump::DwarfDumpType; 192 static bool DynamicRelocations; 193 static bool FaultMapSection; 194 static bool FileHeaders; 195 bool objdump::SectionContents; 196 static std::vector<std::string> InputFilenames; 197 bool objdump::PrintLines; 198 static bool MachOOpt; 199 std::string objdump::MCPU; 200 std::vector<std::string> objdump::MAttrs; 201 bool objdump::ShowRawInsn; 202 bool objdump::LeadingAddr; 203 static bool Offloading; 204 static bool RawClangAST; 205 bool objdump::Relocations; 206 bool objdump::PrintImmHex; 207 bool objdump::PrivateHeaders; 208 std::vector<std::string> objdump::FilterSections; 209 bool objdump::SectionHeaders; 210 static bool ShowLMA; 211 bool objdump::PrintSource; 212 213 static uint64_t StartAddress; 214 static bool HasStartAddressFlag; 215 static uint64_t StopAddress = UINT64_MAX; 216 static bool HasStopAddressFlag; 217 218 bool objdump::SymbolTable; 219 static bool SymbolizeOperands; 220 static bool DynamicSymbolTable; 221 std::string objdump::TripleName; 222 bool objdump::UnwindInfo; 223 static bool Wide; 224 std::string objdump::Prefix; 225 uint32_t objdump::PrefixStrip; 226 227 DebugVarsFormat objdump::DbgVariables = DVDisabled; 228 229 int objdump::DbgIndent = 52; 230 231 static StringSet<> DisasmSymbolSet; 232 StringSet<> objdump::FoundSectionSet; 233 static StringRef ToolName; 234 235 namespace { 236 struct FilterResult { 237 // True if the section should not be skipped. 238 bool Keep; 239 240 // True if the index counter should be incremented, even if the section should 241 // be skipped. For example, sections may be skipped if they are not included 242 // in the --section flag, but we still want those to count toward the section 243 // count. 244 bool IncrementIndex; 245 }; 246 } // namespace 247 248 static FilterResult checkSectionFilter(object::SectionRef S) { 249 if (FilterSections.empty()) 250 return {/*Keep=*/true, /*IncrementIndex=*/true}; 251 252 Expected<StringRef> SecNameOrErr = S.getName(); 253 if (!SecNameOrErr) { 254 consumeError(SecNameOrErr.takeError()); 255 return {/*Keep=*/false, /*IncrementIndex=*/false}; 256 } 257 StringRef SecName = *SecNameOrErr; 258 259 // StringSet does not allow empty key so avoid adding sections with 260 // no name (such as the section with index 0) here. 261 if (!SecName.empty()) 262 FoundSectionSet.insert(SecName); 263 264 // Only show the section if it's in the FilterSections list, but always 265 // increment so the indexing is stable. 266 return {/*Keep=*/is_contained(FilterSections, SecName), 267 /*IncrementIndex=*/true}; 268 } 269 270 SectionFilter objdump::ToolSectionFilter(object::ObjectFile const &O, 271 uint64_t *Idx) { 272 // Start at UINT64_MAX so that the first index returned after an increment is 273 // zero (after the unsigned wrap). 274 if (Idx) 275 *Idx = UINT64_MAX; 276 return SectionFilter( 277 [Idx](object::SectionRef S) { 278 FilterResult Result = checkSectionFilter(S); 279 if (Idx != nullptr && Result.IncrementIndex) 280 *Idx += 1; 281 return Result.Keep; 282 }, 283 O); 284 } 285 286 std::string objdump::getFileNameForError(const object::Archive::Child &C, 287 unsigned Index) { 288 Expected<StringRef> NameOrErr = C.getName(); 289 if (NameOrErr) 290 return std::string(NameOrErr.get()); 291 // If we have an error getting the name then we print the index of the archive 292 // member. Since we are already in an error state, we just ignore this error. 293 consumeError(NameOrErr.takeError()); 294 return "<file index: " + std::to_string(Index) + ">"; 295 } 296 297 void objdump::reportWarning(const Twine &Message, StringRef File) { 298 // Output order between errs() and outs() matters especially for archive 299 // files where the output is per member object. 300 outs().flush(); 301 WithColor::warning(errs(), ToolName) 302 << "'" << File << "': " << Message << "\n"; 303 } 304 305 [[noreturn]] void objdump::reportError(StringRef File, const Twine &Message) { 306 outs().flush(); 307 WithColor::error(errs(), ToolName) << "'" << File << "': " << Message << "\n"; 308 exit(1); 309 } 310 311 [[noreturn]] void objdump::reportError(Error E, StringRef FileName, 312 StringRef ArchiveName, 313 StringRef ArchitectureName) { 314 assert(E); 315 outs().flush(); 316 WithColor::error(errs(), ToolName); 317 if (ArchiveName != "") 318 errs() << ArchiveName << "(" << FileName << ")"; 319 else 320 errs() << "'" << FileName << "'"; 321 if (!ArchitectureName.empty()) 322 errs() << " (for architecture " << ArchitectureName << ")"; 323 errs() << ": "; 324 logAllUnhandledErrors(std::move(E), errs()); 325 exit(1); 326 } 327 328 static void reportCmdLineWarning(const Twine &Message) { 329 WithColor::warning(errs(), ToolName) << Message << "\n"; 330 } 331 332 [[noreturn]] static void reportCmdLineError(const Twine &Message) { 333 WithColor::error(errs(), ToolName) << Message << "\n"; 334 exit(1); 335 } 336 337 static void warnOnNoMatchForSections() { 338 SetVector<StringRef> MissingSections; 339 for (StringRef S : FilterSections) { 340 if (FoundSectionSet.count(S)) 341 return; 342 // User may specify a unnamed section. Don't warn for it. 343 if (!S.empty()) 344 MissingSections.insert(S); 345 } 346 347 // Warn only if no section in FilterSections is matched. 348 for (StringRef S : MissingSections) 349 reportCmdLineWarning("section '" + S + 350 "' mentioned in a -j/--section option, but not " 351 "found in any input file"); 352 } 353 354 static const Target *getTarget(const ObjectFile *Obj) { 355 // Figure out the target triple. 356 Triple TheTriple("unknown-unknown-unknown"); 357 if (TripleName.empty()) { 358 TheTriple = Obj->makeTriple(); 359 } else { 360 TheTriple.setTriple(Triple::normalize(TripleName)); 361 auto Arch = Obj->getArch(); 362 if (Arch == Triple::arm || Arch == Triple::armeb) 363 Obj->setARMSubArch(TheTriple); 364 } 365 366 // Get the target specific parser. 367 std::string Error; 368 const Target *TheTarget = TargetRegistry::lookupTarget(ArchName, TheTriple, 369 Error); 370 if (!TheTarget) 371 reportError(Obj->getFileName(), "can't find target: " + Error); 372 373 // Update the triple name and return the found target. 374 TripleName = TheTriple.getTriple(); 375 return TheTarget; 376 } 377 378 bool objdump::isRelocAddressLess(RelocationRef A, RelocationRef B) { 379 return A.getOffset() < B.getOffset(); 380 } 381 382 static Error getRelocationValueString(const RelocationRef &Rel, 383 SmallVectorImpl<char> &Result) { 384 const ObjectFile *Obj = Rel.getObject(); 385 if (auto *ELF = dyn_cast<ELFObjectFileBase>(Obj)) 386 return getELFRelocationValueString(ELF, Rel, Result); 387 if (auto *COFF = dyn_cast<COFFObjectFile>(Obj)) 388 return getCOFFRelocationValueString(COFF, Rel, Result); 389 if (auto *Wasm = dyn_cast<WasmObjectFile>(Obj)) 390 return getWasmRelocationValueString(Wasm, Rel, Result); 391 if (auto *MachO = dyn_cast<MachOObjectFile>(Obj)) 392 return getMachORelocationValueString(MachO, Rel, Result); 393 if (auto *XCOFF = dyn_cast<XCOFFObjectFile>(Obj)) 394 return getXCOFFRelocationValueString(*XCOFF, Rel, Result); 395 llvm_unreachable("unknown object file format"); 396 } 397 398 /// Indicates whether this relocation should hidden when listing 399 /// relocations, usually because it is the trailing part of a multipart 400 /// relocation that will be printed as part of the leading relocation. 401 static bool getHidden(RelocationRef RelRef) { 402 auto *MachO = dyn_cast<MachOObjectFile>(RelRef.getObject()); 403 if (!MachO) 404 return false; 405 406 unsigned Arch = MachO->getArch(); 407 DataRefImpl Rel = RelRef.getRawDataRefImpl(); 408 uint64_t Type = MachO->getRelocationType(Rel); 409 410 // On arches that use the generic relocations, GENERIC_RELOC_PAIR 411 // is always hidden. 412 if (Arch == Triple::x86 || Arch == Triple::arm || Arch == Triple::ppc) 413 return Type == MachO::GENERIC_RELOC_PAIR; 414 415 if (Arch == Triple::x86_64) { 416 // On x86_64, X86_64_RELOC_UNSIGNED is hidden only when it follows 417 // an X86_64_RELOC_SUBTRACTOR. 418 if (Type == MachO::X86_64_RELOC_UNSIGNED && Rel.d.a > 0) { 419 DataRefImpl RelPrev = Rel; 420 RelPrev.d.a--; 421 uint64_t PrevType = MachO->getRelocationType(RelPrev); 422 if (PrevType == MachO::X86_64_RELOC_SUBTRACTOR) 423 return true; 424 } 425 } 426 427 return false; 428 } 429 430 namespace { 431 432 /// Get the column at which we want to start printing the instruction 433 /// disassembly, taking into account anything which appears to the left of it. 434 unsigned getInstStartColumn(const MCSubtargetInfo &STI) { 435 return !ShowRawInsn ? 16 : STI.getTargetTriple().isX86() ? 40 : 24; 436 } 437 438 static bool isAArch64Elf(const ObjectFile &Obj) { 439 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 440 return Elf && Elf->getEMachine() == ELF::EM_AARCH64; 441 } 442 443 static bool isArmElf(const ObjectFile &Obj) { 444 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 445 return Elf && Elf->getEMachine() == ELF::EM_ARM; 446 } 447 448 static bool isCSKYElf(const ObjectFile &Obj) { 449 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 450 return Elf && Elf->getEMachine() == ELF::EM_CSKY; 451 } 452 453 static bool hasMappingSymbols(const ObjectFile &Obj) { 454 return isArmElf(Obj) || isAArch64Elf(Obj) || isCSKYElf(Obj) ; 455 } 456 457 static void printRelocation(formatted_raw_ostream &OS, StringRef FileName, 458 const RelocationRef &Rel, uint64_t Address, 459 bool Is64Bits) { 460 StringRef Fmt = Is64Bits ? "\t\t%016" PRIx64 ": " : "\t\t\t%08" PRIx64 ": "; 461 SmallString<16> Name; 462 SmallString<32> Val; 463 Rel.getTypeName(Name); 464 if (Error E = getRelocationValueString(Rel, Val)) 465 reportError(std::move(E), FileName); 466 OS << format(Fmt.data(), Address) << Name << "\t" << Val; 467 } 468 469 class PrettyPrinter { 470 public: 471 virtual ~PrettyPrinter() = default; 472 virtual void 473 printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 474 object::SectionedAddress Address, formatted_raw_ostream &OS, 475 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 476 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 477 LiveVariablePrinter &LVP) { 478 if (SP && (PrintSource || PrintLines)) 479 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 480 LVP.printBetweenInsts(OS, false); 481 482 size_t Start = OS.tell(); 483 if (LeadingAddr) 484 OS << format("%8" PRIx64 ":", Address.Address); 485 if (ShowRawInsn) { 486 OS << ' '; 487 dumpBytes(Bytes, OS); 488 } 489 490 // The output of printInst starts with a tab. Print some spaces so that 491 // the tab has 1 column and advances to the target tab stop. 492 unsigned TabStop = getInstStartColumn(STI); 493 unsigned Column = OS.tell() - Start; 494 OS.indent(Column < TabStop - 1 ? TabStop - 1 - Column : 7 - Column % 8); 495 496 if (MI) { 497 // See MCInstPrinter::printInst. On targets where a PC relative immediate 498 // is relative to the next instruction and the length of a MCInst is 499 // difficult to measure (x86), this is the address of the next 500 // instruction. 501 uint64_t Addr = 502 Address.Address + (STI.getTargetTriple().isX86() ? Bytes.size() : 0); 503 IP.printInst(MI, Addr, "", STI, OS); 504 } else 505 OS << "\t<unknown>"; 506 } 507 }; 508 PrettyPrinter PrettyPrinterInst; 509 510 class HexagonPrettyPrinter : public PrettyPrinter { 511 public: 512 void printLead(ArrayRef<uint8_t> Bytes, uint64_t Address, 513 formatted_raw_ostream &OS) { 514 uint32_t opcode = 515 (Bytes[3] << 24) | (Bytes[2] << 16) | (Bytes[1] << 8) | Bytes[0]; 516 if (LeadingAddr) 517 OS << format("%8" PRIx64 ":", Address); 518 if (ShowRawInsn) { 519 OS << "\t"; 520 dumpBytes(Bytes.slice(0, 4), OS); 521 OS << format("\t%08" PRIx32, opcode); 522 } 523 } 524 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 525 object::SectionedAddress Address, formatted_raw_ostream &OS, 526 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 527 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 528 LiveVariablePrinter &LVP) override { 529 if (SP && (PrintSource || PrintLines)) 530 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 531 if (!MI) { 532 printLead(Bytes, Address.Address, OS); 533 OS << " <unknown>"; 534 return; 535 } 536 std::string Buffer; 537 { 538 raw_string_ostream TempStream(Buffer); 539 IP.printInst(MI, Address.Address, "", STI, TempStream); 540 } 541 StringRef Contents(Buffer); 542 // Split off bundle attributes 543 auto PacketBundle = Contents.rsplit('\n'); 544 // Split off first instruction from the rest 545 auto HeadTail = PacketBundle.first.split('\n'); 546 auto Preamble = " { "; 547 auto Separator = ""; 548 549 // Hexagon's packets require relocations to be inline rather than 550 // clustered at the end of the packet. 551 std::vector<RelocationRef>::const_iterator RelCur = Rels->begin(); 552 std::vector<RelocationRef>::const_iterator RelEnd = Rels->end(); 553 auto PrintReloc = [&]() -> void { 554 while ((RelCur != RelEnd) && (RelCur->getOffset() <= Address.Address)) { 555 if (RelCur->getOffset() == Address.Address) { 556 printRelocation(OS, ObjectFilename, *RelCur, Address.Address, false); 557 return; 558 } 559 ++RelCur; 560 } 561 }; 562 563 while (!HeadTail.first.empty()) { 564 OS << Separator; 565 Separator = "\n"; 566 if (SP && (PrintSource || PrintLines)) 567 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 568 printLead(Bytes, Address.Address, OS); 569 OS << Preamble; 570 Preamble = " "; 571 StringRef Inst; 572 auto Duplex = HeadTail.first.split('\v'); 573 if (!Duplex.second.empty()) { 574 OS << Duplex.first; 575 OS << "; "; 576 Inst = Duplex.second; 577 } 578 else 579 Inst = HeadTail.first; 580 OS << Inst; 581 HeadTail = HeadTail.second.split('\n'); 582 if (HeadTail.first.empty()) 583 OS << " } " << PacketBundle.second; 584 PrintReloc(); 585 Bytes = Bytes.slice(4); 586 Address.Address += 4; 587 } 588 } 589 }; 590 HexagonPrettyPrinter HexagonPrettyPrinterInst; 591 592 class AMDGCNPrettyPrinter : public PrettyPrinter { 593 public: 594 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 595 object::SectionedAddress Address, formatted_raw_ostream &OS, 596 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 597 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 598 LiveVariablePrinter &LVP) override { 599 if (SP && (PrintSource || PrintLines)) 600 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 601 602 if (MI) { 603 SmallString<40> InstStr; 604 raw_svector_ostream IS(InstStr); 605 606 IP.printInst(MI, Address.Address, "", STI, IS); 607 608 OS << left_justify(IS.str(), 60); 609 } else { 610 // an unrecognized encoding - this is probably data so represent it 611 // using the .long directive, or .byte directive if fewer than 4 bytes 612 // remaining 613 if (Bytes.size() >= 4) { 614 OS << format("\t.long 0x%08" PRIx32 " ", 615 support::endian::read32<support::little>(Bytes.data())); 616 OS.indent(42); 617 } else { 618 OS << format("\t.byte 0x%02" PRIx8, Bytes[0]); 619 for (unsigned int i = 1; i < Bytes.size(); i++) 620 OS << format(", 0x%02" PRIx8, Bytes[i]); 621 OS.indent(55 - (6 * Bytes.size())); 622 } 623 } 624 625 OS << format("// %012" PRIX64 ":", Address.Address); 626 if (Bytes.size() >= 4) { 627 // D should be casted to uint32_t here as it is passed by format to 628 // snprintf as vararg. 629 for (uint32_t D : makeArrayRef( 630 reinterpret_cast<const support::little32_t *>(Bytes.data()), 631 Bytes.size() / 4)) 632 OS << format(" %08" PRIX32, D); 633 } else { 634 for (unsigned char B : Bytes) 635 OS << format(" %02" PRIX8, B); 636 } 637 638 if (!Annot.empty()) 639 OS << " // " << Annot; 640 } 641 }; 642 AMDGCNPrettyPrinter AMDGCNPrettyPrinterInst; 643 644 class BPFPrettyPrinter : public PrettyPrinter { 645 public: 646 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 647 object::SectionedAddress Address, formatted_raw_ostream &OS, 648 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 649 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 650 LiveVariablePrinter &LVP) override { 651 if (SP && (PrintSource || PrintLines)) 652 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 653 if (LeadingAddr) 654 OS << format("%8" PRId64 ":", Address.Address / 8); 655 if (ShowRawInsn) { 656 OS << "\t"; 657 dumpBytes(Bytes, OS); 658 } 659 if (MI) 660 IP.printInst(MI, Address.Address, "", STI, OS); 661 else 662 OS << "\t<unknown>"; 663 } 664 }; 665 BPFPrettyPrinter BPFPrettyPrinterInst; 666 667 PrettyPrinter &selectPrettyPrinter(Triple const &Triple) { 668 switch(Triple.getArch()) { 669 default: 670 return PrettyPrinterInst; 671 case Triple::hexagon: 672 return HexagonPrettyPrinterInst; 673 case Triple::amdgcn: 674 return AMDGCNPrettyPrinterInst; 675 case Triple::bpfel: 676 case Triple::bpfeb: 677 return BPFPrettyPrinterInst; 678 } 679 } 680 } 681 682 static uint8_t getElfSymbolType(const ObjectFile &Obj, const SymbolRef &Sym) { 683 assert(Obj.isELF()); 684 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 685 return unwrapOrError(Elf32LEObj->getSymbol(Sym.getRawDataRefImpl()), 686 Obj.getFileName()) 687 ->getType(); 688 if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 689 return unwrapOrError(Elf64LEObj->getSymbol(Sym.getRawDataRefImpl()), 690 Obj.getFileName()) 691 ->getType(); 692 if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 693 return unwrapOrError(Elf32BEObj->getSymbol(Sym.getRawDataRefImpl()), 694 Obj.getFileName()) 695 ->getType(); 696 if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 697 return unwrapOrError(Elf64BEObj->getSymbol(Sym.getRawDataRefImpl()), 698 Obj.getFileName()) 699 ->getType(); 700 llvm_unreachable("Unsupported binary format"); 701 } 702 703 template <class ELFT> 704 static void 705 addDynamicElfSymbols(const ELFObjectFile<ELFT> &Obj, 706 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 707 for (auto Symbol : Obj.getDynamicSymbolIterators()) { 708 uint8_t SymbolType = Symbol.getELFType(); 709 if (SymbolType == ELF::STT_SECTION) 710 continue; 711 712 uint64_t Address = unwrapOrError(Symbol.getAddress(), Obj.getFileName()); 713 // ELFSymbolRef::getAddress() returns size instead of value for common 714 // symbols which is not desirable for disassembly output. Overriding. 715 if (SymbolType == ELF::STT_COMMON) 716 Address = unwrapOrError(Obj.getSymbol(Symbol.getRawDataRefImpl()), 717 Obj.getFileName()) 718 ->st_value; 719 720 StringRef Name = unwrapOrError(Symbol.getName(), Obj.getFileName()); 721 if (Name.empty()) 722 continue; 723 724 section_iterator SecI = 725 unwrapOrError(Symbol.getSection(), Obj.getFileName()); 726 if (SecI == Obj.section_end()) 727 continue; 728 729 AllSymbols[*SecI].emplace_back(Address, Name, SymbolType); 730 } 731 } 732 733 static void 734 addDynamicElfSymbols(const ELFObjectFileBase &Obj, 735 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 736 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 737 addDynamicElfSymbols(*Elf32LEObj, AllSymbols); 738 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 739 addDynamicElfSymbols(*Elf64LEObj, AllSymbols); 740 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 741 addDynamicElfSymbols(*Elf32BEObj, AllSymbols); 742 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 743 addDynamicElfSymbols(*Elf64BEObj, AllSymbols); 744 else 745 llvm_unreachable("Unsupported binary format"); 746 } 747 748 static Optional<SectionRef> getWasmCodeSection(const WasmObjectFile &Obj) { 749 for (auto SecI : Obj.sections()) { 750 const WasmSection &Section = Obj.getWasmSection(SecI); 751 if (Section.Type == wasm::WASM_SEC_CODE) 752 return SecI; 753 } 754 return None; 755 } 756 757 static void 758 addMissingWasmCodeSymbols(const WasmObjectFile &Obj, 759 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 760 Optional<SectionRef> Section = getWasmCodeSection(Obj); 761 if (!Section) 762 return; 763 SectionSymbolsTy &Symbols = AllSymbols[*Section]; 764 765 std::set<uint64_t> SymbolAddresses; 766 for (const auto &Sym : Symbols) 767 SymbolAddresses.insert(Sym.Addr); 768 769 for (const wasm::WasmFunction &Function : Obj.functions()) { 770 uint64_t Address = Function.CodeSectionOffset; 771 // Only add fallback symbols for functions not already present in the symbol 772 // table. 773 if (SymbolAddresses.count(Address)) 774 continue; 775 // This function has no symbol, so it should have no SymbolName. 776 assert(Function.SymbolName.empty()); 777 // We use DebugName for the name, though it may be empty if there is no 778 // "name" custom section, or that section is missing a name for this 779 // function. 780 StringRef Name = Function.DebugName; 781 Symbols.emplace_back(Address, Name, ELF::STT_NOTYPE); 782 } 783 } 784 785 static void addPltEntries(const ObjectFile &Obj, 786 std::map<SectionRef, SectionSymbolsTy> &AllSymbols, 787 StringSaver &Saver) { 788 Optional<SectionRef> Plt = None; 789 for (const SectionRef &Section : Obj.sections()) { 790 Expected<StringRef> SecNameOrErr = Section.getName(); 791 if (!SecNameOrErr) { 792 consumeError(SecNameOrErr.takeError()); 793 continue; 794 } 795 if (*SecNameOrErr == ".plt") 796 Plt = Section; 797 } 798 if (!Plt) 799 return; 800 if (auto *ElfObj = dyn_cast<ELFObjectFileBase>(&Obj)) { 801 for (auto PltEntry : ElfObj->getPltAddresses()) { 802 if (PltEntry.first) { 803 SymbolRef Symbol(*PltEntry.first, ElfObj); 804 uint8_t SymbolType = getElfSymbolType(Obj, Symbol); 805 if (Expected<StringRef> NameOrErr = Symbol.getName()) { 806 if (!NameOrErr->empty()) 807 AllSymbols[*Plt].emplace_back( 808 PltEntry.second, Saver.save((*NameOrErr + "@plt").str()), 809 SymbolType); 810 continue; 811 } else { 812 // The warning has been reported in disassembleObject(). 813 consumeError(NameOrErr.takeError()); 814 } 815 } 816 reportWarning("PLT entry at 0x" + Twine::utohexstr(PltEntry.second) + 817 " references an invalid symbol", 818 Obj.getFileName()); 819 } 820 } 821 } 822 823 // Normally the disassembly output will skip blocks of zeroes. This function 824 // returns the number of zero bytes that can be skipped when dumping the 825 // disassembly of the instructions in Buf. 826 static size_t countSkippableZeroBytes(ArrayRef<uint8_t> Buf) { 827 // Find the number of leading zeroes. 828 size_t N = 0; 829 while (N < Buf.size() && !Buf[N]) 830 ++N; 831 832 // We may want to skip blocks of zero bytes, but unless we see 833 // at least 8 of them in a row. 834 if (N < 8) 835 return 0; 836 837 // We skip zeroes in multiples of 4 because do not want to truncate an 838 // instruction if it starts with a zero byte. 839 return N & ~0x3; 840 } 841 842 // Returns a map from sections to their relocations. 843 static std::map<SectionRef, std::vector<RelocationRef>> 844 getRelocsMap(object::ObjectFile const &Obj) { 845 std::map<SectionRef, std::vector<RelocationRef>> Ret; 846 uint64_t I = (uint64_t)-1; 847 for (SectionRef Sec : Obj.sections()) { 848 ++I; 849 Expected<section_iterator> RelocatedOrErr = Sec.getRelocatedSection(); 850 if (!RelocatedOrErr) 851 reportError(Obj.getFileName(), 852 "section (" + Twine(I) + 853 "): failed to get a relocated section: " + 854 toString(RelocatedOrErr.takeError())); 855 856 section_iterator Relocated = *RelocatedOrErr; 857 if (Relocated == Obj.section_end() || !checkSectionFilter(*Relocated).Keep) 858 continue; 859 std::vector<RelocationRef> &V = Ret[*Relocated]; 860 append_range(V, Sec.relocations()); 861 // Sort relocations by address. 862 llvm::stable_sort(V, isRelocAddressLess); 863 } 864 return Ret; 865 } 866 867 // Used for --adjust-vma to check if address should be adjusted by the 868 // specified value for a given section. 869 // For ELF we do not adjust non-allocatable sections like debug ones, 870 // because they are not loadable. 871 // TODO: implement for other file formats. 872 static bool shouldAdjustVA(const SectionRef &Section) { 873 const ObjectFile *Obj = Section.getObject(); 874 if (Obj->isELF()) 875 return ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC; 876 return false; 877 } 878 879 880 typedef std::pair<uint64_t, char> MappingSymbolPair; 881 static char getMappingSymbolKind(ArrayRef<MappingSymbolPair> MappingSymbols, 882 uint64_t Address) { 883 auto It = 884 partition_point(MappingSymbols, [Address](const MappingSymbolPair &Val) { 885 return Val.first <= Address; 886 }); 887 // Return zero for any address before the first mapping symbol; this means 888 // we should use the default disassembly mode, depending on the target. 889 if (It == MappingSymbols.begin()) 890 return '\x00'; 891 return (It - 1)->second; 892 } 893 894 static uint64_t dumpARMELFData(uint64_t SectionAddr, uint64_t Index, 895 uint64_t End, const ObjectFile &Obj, 896 ArrayRef<uint8_t> Bytes, 897 ArrayRef<MappingSymbolPair> MappingSymbols, 898 raw_ostream &OS) { 899 support::endianness Endian = 900 Obj.isLittleEndian() ? support::little : support::big; 901 OS << format("%8" PRIx64 ":\t", SectionAddr + Index); 902 if (Index + 4 <= End) { 903 dumpBytes(Bytes.slice(Index, 4), OS); 904 OS << "\t.word\t" 905 << format_hex(support::endian::read32(Bytes.data() + Index, Endian), 906 10); 907 return 4; 908 } 909 if (Index + 2 <= End) { 910 dumpBytes(Bytes.slice(Index, 2), OS); 911 OS << "\t\t.short\t" 912 << format_hex(support::endian::read16(Bytes.data() + Index, Endian), 913 6); 914 return 2; 915 } 916 dumpBytes(Bytes.slice(Index, 1), OS); 917 OS << "\t\t.byte\t" << format_hex(Bytes[Index], 4); 918 return 1; 919 } 920 921 static void dumpELFData(uint64_t SectionAddr, uint64_t Index, uint64_t End, 922 ArrayRef<uint8_t> Bytes) { 923 // print out data up to 8 bytes at a time in hex and ascii 924 uint8_t AsciiData[9] = {'\0'}; 925 uint8_t Byte; 926 int NumBytes = 0; 927 928 for (; Index < End; ++Index) { 929 if (NumBytes == 0) 930 outs() << format("%8" PRIx64 ":", SectionAddr + Index); 931 Byte = Bytes.slice(Index)[0]; 932 outs() << format(" %02x", Byte); 933 AsciiData[NumBytes] = isPrint(Byte) ? Byte : '.'; 934 935 uint8_t IndentOffset = 0; 936 NumBytes++; 937 if (Index == End - 1 || NumBytes > 8) { 938 // Indent the space for less than 8 bytes data. 939 // 2 spaces for byte and one for space between bytes 940 IndentOffset = 3 * (8 - NumBytes); 941 for (int Excess = NumBytes; Excess < 8; Excess++) 942 AsciiData[Excess] = '\0'; 943 NumBytes = 8; 944 } 945 if (NumBytes == 8) { 946 AsciiData[8] = '\0'; 947 outs() << std::string(IndentOffset, ' ') << " "; 948 outs() << reinterpret_cast<char *>(AsciiData); 949 outs() << '\n'; 950 NumBytes = 0; 951 } 952 } 953 } 954 955 SymbolInfoTy objdump::createSymbolInfo(const ObjectFile &Obj, 956 const SymbolRef &Symbol) { 957 const StringRef FileName = Obj.getFileName(); 958 const uint64_t Addr = unwrapOrError(Symbol.getAddress(), FileName); 959 const StringRef Name = unwrapOrError(Symbol.getName(), FileName); 960 961 if (Obj.isXCOFF() && SymbolDescription) { 962 const auto &XCOFFObj = cast<XCOFFObjectFile>(Obj); 963 DataRefImpl SymbolDRI = Symbol.getRawDataRefImpl(); 964 965 const uint32_t SymbolIndex = XCOFFObj.getSymbolIndex(SymbolDRI.p); 966 Optional<XCOFF::StorageMappingClass> Smc = 967 getXCOFFSymbolCsectSMC(XCOFFObj, Symbol); 968 return SymbolInfoTy(Addr, Name, Smc, SymbolIndex, 969 isLabel(XCOFFObj, Symbol)); 970 } else if (Obj.isXCOFF()) { 971 const SymbolRef::Type SymType = unwrapOrError(Symbol.getType(), FileName); 972 return SymbolInfoTy(Addr, Name, SymType, true); 973 } else 974 return SymbolInfoTy(Addr, Name, 975 Obj.isELF() ? getElfSymbolType(Obj, Symbol) 976 : (uint8_t)ELF::STT_NOTYPE); 977 } 978 979 static SymbolInfoTy createDummySymbolInfo(const ObjectFile &Obj, 980 const uint64_t Addr, StringRef &Name, 981 uint8_t Type) { 982 if (Obj.isXCOFF() && SymbolDescription) 983 return SymbolInfoTy(Addr, Name, None, None, false); 984 else 985 return SymbolInfoTy(Addr, Name, Type); 986 } 987 988 static void 989 collectBBAddrMapLabels(const std::unordered_map<uint64_t, BBAddrMap> &AddrToBBAddrMap, 990 uint64_t SectionAddr, uint64_t Start, uint64_t End, 991 std::unordered_map<uint64_t, std::vector<std::string>> &Labels) { 992 if (AddrToBBAddrMap.empty()) 993 return; 994 Labels.clear(); 995 uint64_t StartAddress = SectionAddr + Start; 996 uint64_t EndAddress = SectionAddr + End; 997 auto Iter = AddrToBBAddrMap.find(StartAddress); 998 if (Iter == AddrToBBAddrMap.end()) 999 return; 1000 for (unsigned I = 0, Size = Iter->second.BBEntries.size(); I < Size; ++I) { 1001 uint64_t BBAddress = Iter->second.BBEntries[I].Offset + Iter->second.Addr; 1002 if (BBAddress >= EndAddress) 1003 continue; 1004 Labels[BBAddress].push_back(("BB" + Twine(I)).str()); 1005 } 1006 } 1007 1008 static void collectLocalBranchTargets( 1009 ArrayRef<uint8_t> Bytes, const MCInstrAnalysis *MIA, MCDisassembler *DisAsm, 1010 MCInstPrinter *IP, const MCSubtargetInfo *STI, uint64_t SectionAddr, 1011 uint64_t Start, uint64_t End, std::unordered_map<uint64_t, std::string> &Labels) { 1012 // So far only supports PowerPC and X86. 1013 if (!STI->getTargetTriple().isPPC() && !STI->getTargetTriple().isX86()) 1014 return; 1015 1016 Labels.clear(); 1017 unsigned LabelCount = 0; 1018 Start += SectionAddr; 1019 End += SectionAddr; 1020 uint64_t Index = Start; 1021 while (Index < End) { 1022 // Disassemble a real instruction and record function-local branch labels. 1023 MCInst Inst; 1024 uint64_t Size; 1025 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index - SectionAddr); 1026 bool Disassembled = 1027 DisAsm->getInstruction(Inst, Size, ThisBytes, Index, nulls()); 1028 if (Size == 0) 1029 Size = std::min(ThisBytes.size(), 1030 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1031 1032 if (Disassembled && MIA) { 1033 uint64_t Target; 1034 bool TargetKnown = MIA->evaluateBranch(Inst, Index, Size, Target); 1035 // On PowerPC, if the address of a branch is the same as the target, it 1036 // means that it's a function call. Do not mark the label for this case. 1037 if (TargetKnown && (Target >= Start && Target < End) && 1038 !Labels.count(Target) && 1039 !(STI->getTargetTriple().isPPC() && Target == Index)) 1040 Labels[Target] = ("L" + Twine(LabelCount++)).str(); 1041 } 1042 Index += Size; 1043 } 1044 } 1045 1046 // Create an MCSymbolizer for the target and add it to the MCDisassembler. 1047 // This is currently only used on AMDGPU, and assumes the format of the 1048 // void * argument passed to AMDGPU's createMCSymbolizer. 1049 static void addSymbolizer( 1050 MCContext &Ctx, const Target *Target, StringRef TripleName, 1051 MCDisassembler *DisAsm, uint64_t SectionAddr, ArrayRef<uint8_t> Bytes, 1052 SectionSymbolsTy &Symbols, 1053 std::vector<std::unique_ptr<std::string>> &SynthesizedLabelNames) { 1054 1055 std::unique_ptr<MCRelocationInfo> RelInfo( 1056 Target->createMCRelocationInfo(TripleName, Ctx)); 1057 if (!RelInfo) 1058 return; 1059 std::unique_ptr<MCSymbolizer> Symbolizer(Target->createMCSymbolizer( 1060 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1061 MCSymbolizer *SymbolizerPtr = &*Symbolizer; 1062 DisAsm->setSymbolizer(std::move(Symbolizer)); 1063 1064 if (!SymbolizeOperands) 1065 return; 1066 1067 // Synthesize labels referenced by branch instructions by 1068 // disassembling, discarding the output, and collecting the referenced 1069 // addresses from the symbolizer. 1070 for (size_t Index = 0; Index != Bytes.size();) { 1071 MCInst Inst; 1072 uint64_t Size; 1073 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index - SectionAddr); 1074 DisAsm->getInstruction(Inst, Size, ThisBytes, Index, nulls()); 1075 if (Size == 0) 1076 Size = std::min(ThisBytes.size(), 1077 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1078 Index += Size; 1079 } 1080 ArrayRef<uint64_t> LabelAddrsRef = SymbolizerPtr->getReferencedAddresses(); 1081 // Copy and sort to remove duplicates. 1082 std::vector<uint64_t> LabelAddrs; 1083 LabelAddrs.insert(LabelAddrs.end(), LabelAddrsRef.begin(), 1084 LabelAddrsRef.end()); 1085 llvm::sort(LabelAddrs); 1086 LabelAddrs.resize(std::unique(LabelAddrs.begin(), LabelAddrs.end()) - 1087 LabelAddrs.begin()); 1088 // Add the labels. 1089 for (unsigned LabelNum = 0; LabelNum != LabelAddrs.size(); ++LabelNum) { 1090 auto Name = std::make_unique<std::string>(); 1091 *Name = (Twine("L") + Twine(LabelNum)).str(); 1092 SynthesizedLabelNames.push_back(std::move(Name)); 1093 Symbols.push_back(SymbolInfoTy( 1094 LabelAddrs[LabelNum], *SynthesizedLabelNames.back(), ELF::STT_NOTYPE)); 1095 } 1096 llvm::stable_sort(Symbols); 1097 // Recreate the symbolizer with the new symbols list. 1098 RelInfo.reset(Target->createMCRelocationInfo(TripleName, Ctx)); 1099 Symbolizer.reset(Target->createMCSymbolizer( 1100 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1101 DisAsm->setSymbolizer(std::move(Symbolizer)); 1102 } 1103 1104 static StringRef getSegmentName(const MachOObjectFile *MachO, 1105 const SectionRef &Section) { 1106 if (MachO) { 1107 DataRefImpl DR = Section.getRawDataRefImpl(); 1108 StringRef SegmentName = MachO->getSectionFinalSegmentName(DR); 1109 return SegmentName; 1110 } 1111 return ""; 1112 } 1113 1114 static void emitPostInstructionInfo(formatted_raw_ostream &FOS, 1115 const MCAsmInfo &MAI, 1116 const MCSubtargetInfo &STI, 1117 StringRef Comments, 1118 LiveVariablePrinter &LVP) { 1119 do { 1120 if (!Comments.empty()) { 1121 // Emit a line of comments. 1122 StringRef Comment; 1123 std::tie(Comment, Comments) = Comments.split('\n'); 1124 // MAI.getCommentColumn() assumes that instructions are printed at the 1125 // position of 8, while getInstStartColumn() returns the actual position. 1126 unsigned CommentColumn = 1127 MAI.getCommentColumn() - 8 + getInstStartColumn(STI); 1128 FOS.PadToColumn(CommentColumn); 1129 FOS << MAI.getCommentString() << ' ' << Comment; 1130 } 1131 LVP.printAfterInst(FOS); 1132 FOS << '\n'; 1133 } while (!Comments.empty()); 1134 FOS.flush(); 1135 } 1136 1137 static void createFakeELFSections(ObjectFile &Obj) { 1138 assert(Obj.isELF()); 1139 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 1140 Elf32LEObj->createFakeSections(); 1141 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 1142 Elf64LEObj->createFakeSections(); 1143 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 1144 Elf32BEObj->createFakeSections(); 1145 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 1146 Elf64BEObj->createFakeSections(); 1147 else 1148 llvm_unreachable("Unsupported binary format"); 1149 } 1150 1151 static void disassembleObject(const Target *TheTarget, ObjectFile &Obj, 1152 MCContext &Ctx, MCDisassembler *PrimaryDisAsm, 1153 MCDisassembler *SecondaryDisAsm, 1154 const MCInstrAnalysis *MIA, MCInstPrinter *IP, 1155 const MCSubtargetInfo *PrimarySTI, 1156 const MCSubtargetInfo *SecondarySTI, 1157 PrettyPrinter &PIP, SourcePrinter &SP, 1158 bool InlineRelocs) { 1159 const MCSubtargetInfo *STI = PrimarySTI; 1160 MCDisassembler *DisAsm = PrimaryDisAsm; 1161 bool PrimaryIsThumb = false; 1162 if (isArmElf(Obj)) 1163 PrimaryIsThumb = STI->checkFeatures("+thumb-mode"); 1164 1165 std::map<SectionRef, std::vector<RelocationRef>> RelocMap; 1166 if (InlineRelocs) 1167 RelocMap = getRelocsMap(Obj); 1168 bool Is64Bits = Obj.getBytesInAddress() > 4; 1169 1170 // Create a mapping from virtual address to symbol name. This is used to 1171 // pretty print the symbols while disassembling. 1172 std::map<SectionRef, SectionSymbolsTy> AllSymbols; 1173 SectionSymbolsTy AbsoluteSymbols; 1174 const StringRef FileName = Obj.getFileName(); 1175 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&Obj); 1176 for (const SymbolRef &Symbol : Obj.symbols()) { 1177 Expected<StringRef> NameOrErr = Symbol.getName(); 1178 if (!NameOrErr) { 1179 reportWarning(toString(NameOrErr.takeError()), FileName); 1180 continue; 1181 } 1182 if (NameOrErr->empty() && !(Obj.isXCOFF() && SymbolDescription)) 1183 continue; 1184 1185 if (Obj.isELF() && getElfSymbolType(Obj, Symbol) == ELF::STT_SECTION) 1186 continue; 1187 1188 if (MachO) { 1189 // __mh_(execute|dylib|dylinker|bundle|preload|object)_header are special 1190 // symbols that support MachO header introspection. They do not bind to 1191 // code locations and are irrelevant for disassembly. 1192 if (NameOrErr->startswith("__mh_") && NameOrErr->endswith("_header")) 1193 continue; 1194 // Don't ask a Mach-O STAB symbol for its section unless you know that 1195 // STAB symbol's section field refers to a valid section index. Otherwise 1196 // the symbol may error trying to load a section that does not exist. 1197 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 1198 uint8_t NType = (MachO->is64Bit() ? 1199 MachO->getSymbol64TableEntry(SymDRI).n_type: 1200 MachO->getSymbolTableEntry(SymDRI).n_type); 1201 if (NType & MachO::N_STAB) 1202 continue; 1203 } 1204 1205 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1206 if (SecI != Obj.section_end()) 1207 AllSymbols[*SecI].push_back(createSymbolInfo(Obj, Symbol)); 1208 else 1209 AbsoluteSymbols.push_back(createSymbolInfo(Obj, Symbol)); 1210 } 1211 1212 if (AllSymbols.empty() && Obj.isELF()) 1213 addDynamicElfSymbols(cast<ELFObjectFileBase>(Obj), AllSymbols); 1214 1215 if (Obj.isWasm()) 1216 addMissingWasmCodeSymbols(cast<WasmObjectFile>(Obj), AllSymbols); 1217 1218 if (Obj.isELF() && Obj.sections().empty()) 1219 createFakeELFSections(Obj); 1220 1221 BumpPtrAllocator A; 1222 StringSaver Saver(A); 1223 addPltEntries(Obj, AllSymbols, Saver); 1224 1225 // Create a mapping from virtual address to section. An empty section can 1226 // cause more than one section at the same address. Sort such sections to be 1227 // before same-addressed non-empty sections so that symbol lookups prefer the 1228 // non-empty section. 1229 std::vector<std::pair<uint64_t, SectionRef>> SectionAddresses; 1230 for (SectionRef Sec : Obj.sections()) 1231 SectionAddresses.emplace_back(Sec.getAddress(), Sec); 1232 llvm::stable_sort(SectionAddresses, [](const auto &LHS, const auto &RHS) { 1233 if (LHS.first != RHS.first) 1234 return LHS.first < RHS.first; 1235 return LHS.second.getSize() < RHS.second.getSize(); 1236 }); 1237 1238 // Linked executables (.exe and .dll files) typically don't include a real 1239 // symbol table but they might contain an export table. 1240 if (const auto *COFFObj = dyn_cast<COFFObjectFile>(&Obj)) { 1241 for (const auto &ExportEntry : COFFObj->export_directories()) { 1242 StringRef Name; 1243 if (Error E = ExportEntry.getSymbolName(Name)) 1244 reportError(std::move(E), Obj.getFileName()); 1245 if (Name.empty()) 1246 continue; 1247 1248 uint32_t RVA; 1249 if (Error E = ExportEntry.getExportRVA(RVA)) 1250 reportError(std::move(E), Obj.getFileName()); 1251 1252 uint64_t VA = COFFObj->getImageBase() + RVA; 1253 auto Sec = partition_point( 1254 SectionAddresses, [VA](const std::pair<uint64_t, SectionRef> &O) { 1255 return O.first <= VA; 1256 }); 1257 if (Sec != SectionAddresses.begin()) { 1258 --Sec; 1259 AllSymbols[Sec->second].emplace_back(VA, Name, ELF::STT_NOTYPE); 1260 } else 1261 AbsoluteSymbols.emplace_back(VA, Name, ELF::STT_NOTYPE); 1262 } 1263 } 1264 1265 // Sort all the symbols, this allows us to use a simple binary search to find 1266 // Multiple symbols can have the same address. Use a stable sort to stabilize 1267 // the output. 1268 StringSet<> FoundDisasmSymbolSet; 1269 for (std::pair<const SectionRef, SectionSymbolsTy> &SecSyms : AllSymbols) 1270 llvm::stable_sort(SecSyms.second); 1271 llvm::stable_sort(AbsoluteSymbols); 1272 1273 std::unique_ptr<DWARFContext> DICtx; 1274 LiveVariablePrinter LVP(*Ctx.getRegisterInfo(), *STI); 1275 1276 if (DbgVariables != DVDisabled) { 1277 DICtx = DWARFContext::create(Obj); 1278 for (const std::unique_ptr<DWARFUnit> &CU : DICtx->compile_units()) 1279 LVP.addCompileUnit(CU->getUnitDIE(false)); 1280 } 1281 1282 LLVM_DEBUG(LVP.dump()); 1283 1284 std::unordered_map<uint64_t, BBAddrMap> AddrToBBAddrMap; 1285 auto ReadBBAddrMap = [&](Optional<unsigned> SectionIndex = None) { 1286 AddrToBBAddrMap.clear(); 1287 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj)) { 1288 auto BBAddrMapsOrErr = Elf->readBBAddrMap(SectionIndex); 1289 if (!BBAddrMapsOrErr) 1290 reportWarning(toString(BBAddrMapsOrErr.takeError()), 1291 Obj.getFileName()); 1292 for (auto &FunctionBBAddrMap : *BBAddrMapsOrErr) 1293 AddrToBBAddrMap.emplace(FunctionBBAddrMap.Addr, 1294 std::move(FunctionBBAddrMap)); 1295 } 1296 }; 1297 1298 // For non-relocatable objects, Read all LLVM_BB_ADDR_MAP sections into a 1299 // single mapping, since they don't have any conflicts. 1300 if (SymbolizeOperands && !Obj.isRelocatableObject()) 1301 ReadBBAddrMap(); 1302 1303 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 1304 if (FilterSections.empty() && !DisassembleAll && 1305 (!Section.isText() || Section.isVirtual())) 1306 continue; 1307 1308 uint64_t SectionAddr = Section.getAddress(); 1309 uint64_t SectSize = Section.getSize(); 1310 if (!SectSize) 1311 continue; 1312 1313 // For relocatable object files, read the LLVM_BB_ADDR_MAP section 1314 // corresponding to this section, if present. 1315 if (SymbolizeOperands && Obj.isRelocatableObject()) 1316 ReadBBAddrMap(Section.getIndex()); 1317 1318 // Get the list of all the symbols in this section. 1319 SectionSymbolsTy &Symbols = AllSymbols[Section]; 1320 std::vector<MappingSymbolPair> MappingSymbols; 1321 if (hasMappingSymbols(Obj)) { 1322 for (const auto &Symb : Symbols) { 1323 uint64_t Address = Symb.Addr; 1324 StringRef Name = Symb.Name; 1325 if (Name.startswith("$d")) 1326 MappingSymbols.emplace_back(Address - SectionAddr, 'd'); 1327 if (Name.startswith("$x")) 1328 MappingSymbols.emplace_back(Address - SectionAddr, 'x'); 1329 if (Name.startswith("$a")) 1330 MappingSymbols.emplace_back(Address - SectionAddr, 'a'); 1331 if (Name.startswith("$t")) 1332 MappingSymbols.emplace_back(Address - SectionAddr, 't'); 1333 } 1334 } 1335 1336 llvm::sort(MappingSymbols); 1337 1338 ArrayRef<uint8_t> Bytes = arrayRefFromStringRef( 1339 unwrapOrError(Section.getContents(), Obj.getFileName())); 1340 1341 std::vector<std::unique_ptr<std::string>> SynthesizedLabelNames; 1342 if (Obj.isELF() && Obj.getArch() == Triple::amdgcn) { 1343 // AMDGPU disassembler uses symbolizer for printing labels 1344 addSymbolizer(Ctx, TheTarget, TripleName, DisAsm, SectionAddr, Bytes, 1345 Symbols, SynthesizedLabelNames); 1346 } 1347 1348 StringRef SegmentName = getSegmentName(MachO, Section); 1349 StringRef SectionName = unwrapOrError(Section.getName(), Obj.getFileName()); 1350 // If the section has no symbol at the start, just insert a dummy one. 1351 if (Symbols.empty() || Symbols[0].Addr != 0) { 1352 Symbols.insert(Symbols.begin(), 1353 createDummySymbolInfo(Obj, SectionAddr, SectionName, 1354 Section.isText() ? ELF::STT_FUNC 1355 : ELF::STT_OBJECT)); 1356 } 1357 1358 SmallString<40> Comments; 1359 raw_svector_ostream CommentStream(Comments); 1360 1361 uint64_t VMAAdjustment = 0; 1362 if (shouldAdjustVA(Section)) 1363 VMAAdjustment = AdjustVMA; 1364 1365 // In executable and shared objects, r_offset holds a virtual address. 1366 // Subtract SectionAddr from the r_offset field of a relocation to get 1367 // the section offset. 1368 uint64_t RelAdjustment = Obj.isRelocatableObject() ? 0 : SectionAddr; 1369 uint64_t Size; 1370 uint64_t Index; 1371 bool PrintedSection = false; 1372 std::vector<RelocationRef> Rels = RelocMap[Section]; 1373 std::vector<RelocationRef>::const_iterator RelCur = Rels.begin(); 1374 std::vector<RelocationRef>::const_iterator RelEnd = Rels.end(); 1375 // Disassemble symbol by symbol. 1376 for (unsigned SI = 0, SE = Symbols.size(); SI != SE; ++SI) { 1377 std::string SymbolName = Symbols[SI].Name.str(); 1378 if (Demangle) 1379 SymbolName = demangle(SymbolName); 1380 1381 // Skip if --disassemble-symbols is not empty and the symbol is not in 1382 // the list. 1383 if (!DisasmSymbolSet.empty() && !DisasmSymbolSet.count(SymbolName)) 1384 continue; 1385 1386 uint64_t Start = Symbols[SI].Addr; 1387 if (Start < SectionAddr || StopAddress <= Start) 1388 continue; 1389 else 1390 FoundDisasmSymbolSet.insert(SymbolName); 1391 1392 // The end is the section end, the beginning of the next symbol, or 1393 // --stop-address. 1394 uint64_t End = std::min<uint64_t>(SectionAddr + SectSize, StopAddress); 1395 if (SI + 1 < SE) 1396 End = std::min(End, Symbols[SI + 1].Addr); 1397 if (Start >= End || End <= StartAddress) 1398 continue; 1399 Start -= SectionAddr; 1400 End -= SectionAddr; 1401 1402 if (!PrintedSection) { 1403 PrintedSection = true; 1404 outs() << "\nDisassembly of section "; 1405 if (!SegmentName.empty()) 1406 outs() << SegmentName << ","; 1407 outs() << SectionName << ":\n"; 1408 } 1409 1410 outs() << '\n'; 1411 if (LeadingAddr) 1412 outs() << format(Is64Bits ? "%016" PRIx64 " " : "%08" PRIx64 " ", 1413 SectionAddr + Start + VMAAdjustment); 1414 if (Obj.isXCOFF() && SymbolDescription) { 1415 outs() << getXCOFFSymbolDescription(Symbols[SI], SymbolName) << ":\n"; 1416 } else 1417 outs() << '<' << SymbolName << ">:\n"; 1418 1419 // Don't print raw contents of a virtual section. A virtual section 1420 // doesn't have any contents in the file. 1421 if (Section.isVirtual()) { 1422 outs() << "...\n"; 1423 continue; 1424 } 1425 1426 auto Status = DisAsm->onSymbolStart(Symbols[SI], Size, 1427 Bytes.slice(Start, End - Start), 1428 SectionAddr + Start, CommentStream); 1429 // To have round trippable disassembly, we fall back to decoding the 1430 // remaining bytes as instructions. 1431 // 1432 // If there is a failure, we disassemble the failed region as bytes before 1433 // falling back. The target is expected to print nothing in this case. 1434 // 1435 // If there is Success or SoftFail i.e no 'real' failure, we go ahead by 1436 // Size bytes before falling back. 1437 // So if the entire symbol is 'eaten' by the target: 1438 // Start += Size // Now Start = End and we will never decode as 1439 // // instructions 1440 // 1441 // Right now, most targets return None i.e ignore to treat a symbol 1442 // separately. But WebAssembly decodes preludes for some symbols. 1443 // 1444 if (Status) { 1445 if (Status.value() == MCDisassembler::Fail) { 1446 outs() << "// Error in decoding " << SymbolName 1447 << " : Decoding failed region as bytes.\n"; 1448 for (uint64_t I = 0; I < Size; ++I) { 1449 outs() << "\t.byte\t " << format_hex(Bytes[I], 1, /*Upper=*/true) 1450 << "\n"; 1451 } 1452 } 1453 } else { 1454 Size = 0; 1455 } 1456 1457 Start += Size; 1458 1459 Index = Start; 1460 if (SectionAddr < StartAddress) 1461 Index = std::max<uint64_t>(Index, StartAddress - SectionAddr); 1462 1463 // If there is a data/common symbol inside an ELF text section and we are 1464 // only disassembling text (applicable all architectures), we are in a 1465 // situation where we must print the data and not disassemble it. 1466 if (Obj.isELF() && !DisassembleAll && Section.isText()) { 1467 uint8_t SymTy = Symbols[SI].Type; 1468 if (SymTy == ELF::STT_OBJECT || SymTy == ELF::STT_COMMON) { 1469 dumpELFData(SectionAddr, Index, End, Bytes); 1470 Index = End; 1471 } 1472 } 1473 1474 bool CheckARMELFData = hasMappingSymbols(Obj) && 1475 Symbols[SI].Type != ELF::STT_OBJECT && 1476 !DisassembleAll; 1477 bool DumpARMELFData = false; 1478 formatted_raw_ostream FOS(outs()); 1479 1480 std::unordered_map<uint64_t, std::string> AllLabels; 1481 std::unordered_map<uint64_t, std::vector<std::string>> BBAddrMapLabels; 1482 if (SymbolizeOperands) { 1483 collectLocalBranchTargets(Bytes, MIA, DisAsm, IP, PrimarySTI, 1484 SectionAddr, Index, End, AllLabels); 1485 collectBBAddrMapLabels(AddrToBBAddrMap, SectionAddr, Index, End, 1486 BBAddrMapLabels); 1487 } 1488 1489 while (Index < End) { 1490 // ARM and AArch64 ELF binaries can interleave data and text in the 1491 // same section. We rely on the markers introduced to understand what 1492 // we need to dump. If the data marker is within a function, it is 1493 // denoted as a word/short etc. 1494 if (CheckARMELFData) { 1495 char Kind = getMappingSymbolKind(MappingSymbols, Index); 1496 DumpARMELFData = Kind == 'd'; 1497 if (SecondarySTI) { 1498 if (Kind == 'a') { 1499 STI = PrimaryIsThumb ? SecondarySTI : PrimarySTI; 1500 DisAsm = PrimaryIsThumb ? SecondaryDisAsm : PrimaryDisAsm; 1501 } else if (Kind == 't') { 1502 STI = PrimaryIsThumb ? PrimarySTI : SecondarySTI; 1503 DisAsm = PrimaryIsThumb ? PrimaryDisAsm : SecondaryDisAsm; 1504 } 1505 } 1506 } 1507 1508 if (DumpARMELFData) { 1509 Size = dumpARMELFData(SectionAddr, Index, End, Obj, Bytes, 1510 MappingSymbols, FOS); 1511 } else { 1512 // When -z or --disassemble-zeroes are given we always dissasemble 1513 // them. Otherwise we might want to skip zero bytes we see. 1514 if (!DisassembleZeroes) { 1515 uint64_t MaxOffset = End - Index; 1516 // For --reloc: print zero blocks patched by relocations, so that 1517 // relocations can be shown in the dump. 1518 if (RelCur != RelEnd) 1519 MaxOffset = std::min(RelCur->getOffset() - RelAdjustment - Index, 1520 MaxOffset); 1521 1522 if (size_t N = 1523 countSkippableZeroBytes(Bytes.slice(Index, MaxOffset))) { 1524 FOS << "\t\t..." << '\n'; 1525 Index += N; 1526 continue; 1527 } 1528 } 1529 1530 // Print local label if there's any. 1531 auto Iter1 = BBAddrMapLabels.find(SectionAddr + Index); 1532 if (Iter1 != BBAddrMapLabels.end()) { 1533 for (StringRef Label : Iter1->second) 1534 FOS << "<" << Label << ">:\n"; 1535 } else { 1536 auto Iter2 = AllLabels.find(SectionAddr + Index); 1537 if (Iter2 != AllLabels.end()) 1538 FOS << "<" << Iter2->second << ">:\n"; 1539 } 1540 1541 // Disassemble a real instruction or a data when disassemble all is 1542 // provided 1543 MCInst Inst; 1544 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index); 1545 uint64_t ThisAddr = SectionAddr + Index; 1546 bool Disassembled = DisAsm->getInstruction(Inst, Size, ThisBytes, 1547 ThisAddr, CommentStream); 1548 if (Size == 0) 1549 Size = std::min(ThisBytes.size(), 1550 DisAsm->suggestBytesToSkip(ThisBytes, ThisAddr)); 1551 1552 LVP.update({Index, Section.getIndex()}, 1553 {Index + Size, Section.getIndex()}, Index + Size != End); 1554 1555 IP->setCommentStream(CommentStream); 1556 1557 PIP.printInst( 1558 *IP, Disassembled ? &Inst : nullptr, Bytes.slice(Index, Size), 1559 {SectionAddr + Index + VMAAdjustment, Section.getIndex()}, FOS, 1560 "", *STI, &SP, Obj.getFileName(), &Rels, LVP); 1561 1562 IP->setCommentStream(llvm::nulls()); 1563 1564 // If disassembly has failed, avoid analysing invalid/incomplete 1565 // instruction information. Otherwise, try to resolve the target 1566 // address (jump target or memory operand address) and print it on the 1567 // right of the instruction. 1568 if (Disassembled && MIA) { 1569 // Branch targets are printed just after the instructions. 1570 llvm::raw_ostream *TargetOS = &FOS; 1571 uint64_t Target; 1572 bool PrintTarget = 1573 MIA->evaluateBranch(Inst, SectionAddr + Index, Size, Target); 1574 if (!PrintTarget) 1575 if (Optional<uint64_t> MaybeTarget = 1576 MIA->evaluateMemoryOperandAddress( 1577 Inst, STI, SectionAddr + Index, Size)) { 1578 Target = *MaybeTarget; 1579 PrintTarget = true; 1580 // Do not print real address when symbolizing. 1581 if (!SymbolizeOperands) { 1582 // Memory operand addresses are printed as comments. 1583 TargetOS = &CommentStream; 1584 *TargetOS << "0x" << Twine::utohexstr(Target); 1585 } 1586 } 1587 if (PrintTarget) { 1588 // In a relocatable object, the target's section must reside in 1589 // the same section as the call instruction or it is accessed 1590 // through a relocation. 1591 // 1592 // In a non-relocatable object, the target may be in any section. 1593 // In that case, locate the section(s) containing the target 1594 // address and find the symbol in one of those, if possible. 1595 // 1596 // N.B. We don't walk the relocations in the relocatable case yet. 1597 std::vector<const SectionSymbolsTy *> TargetSectionSymbols; 1598 if (!Obj.isRelocatableObject()) { 1599 auto It = llvm::partition_point( 1600 SectionAddresses, 1601 [=](const std::pair<uint64_t, SectionRef> &O) { 1602 return O.first <= Target; 1603 }); 1604 uint64_t TargetSecAddr = 0; 1605 while (It != SectionAddresses.begin()) { 1606 --It; 1607 if (TargetSecAddr == 0) 1608 TargetSecAddr = It->first; 1609 if (It->first != TargetSecAddr) 1610 break; 1611 TargetSectionSymbols.push_back(&AllSymbols[It->second]); 1612 } 1613 } else { 1614 TargetSectionSymbols.push_back(&Symbols); 1615 } 1616 TargetSectionSymbols.push_back(&AbsoluteSymbols); 1617 1618 // Find the last symbol in the first candidate section whose 1619 // offset is less than or equal to the target. If there are no 1620 // such symbols, try in the next section and so on, before finally 1621 // using the nearest preceding absolute symbol (if any), if there 1622 // are no other valid symbols. 1623 const SymbolInfoTy *TargetSym = nullptr; 1624 for (const SectionSymbolsTy *TargetSymbols : 1625 TargetSectionSymbols) { 1626 auto It = llvm::partition_point( 1627 *TargetSymbols, 1628 [=](const SymbolInfoTy &O) { return O.Addr <= Target; }); 1629 if (It != TargetSymbols->begin()) { 1630 TargetSym = &*(It - 1); 1631 break; 1632 } 1633 } 1634 1635 // Print the labels corresponding to the target if there's any. 1636 bool BBAddrMapLabelAvailable = BBAddrMapLabels.count(Target); 1637 bool LabelAvailable = AllLabels.count(Target); 1638 if (TargetSym != nullptr) { 1639 uint64_t TargetAddress = TargetSym->Addr; 1640 uint64_t Disp = Target - TargetAddress; 1641 std::string TargetName = TargetSym->Name.str(); 1642 if (Demangle) 1643 TargetName = demangle(TargetName); 1644 1645 *TargetOS << " <"; 1646 if (!Disp) { 1647 // Always Print the binary symbol precisely corresponding to 1648 // the target address. 1649 *TargetOS << TargetName; 1650 } else if (BBAddrMapLabelAvailable) { 1651 *TargetOS << BBAddrMapLabels[Target].front(); 1652 } else if (LabelAvailable) { 1653 *TargetOS << AllLabels[Target]; 1654 } else { 1655 // Always Print the binary symbol plus an offset if there's no 1656 // local label corresponding to the target address. 1657 *TargetOS << TargetName << "+0x" << Twine::utohexstr(Disp); 1658 } 1659 *TargetOS << ">"; 1660 } else if (BBAddrMapLabelAvailable) { 1661 *TargetOS << " <" << BBAddrMapLabels[Target].front() << ">"; 1662 } else if (LabelAvailable) { 1663 *TargetOS << " <" << AllLabels[Target] << ">"; 1664 } 1665 // By convention, each record in the comment stream should be 1666 // terminated. 1667 if (TargetOS == &CommentStream) 1668 *TargetOS << "\n"; 1669 } 1670 } 1671 } 1672 1673 assert(Ctx.getAsmInfo()); 1674 emitPostInstructionInfo(FOS, *Ctx.getAsmInfo(), *STI, 1675 CommentStream.str(), LVP); 1676 Comments.clear(); 1677 1678 // Hexagon does this in pretty printer 1679 if (Obj.getArch() != Triple::hexagon) { 1680 // Print relocation for instruction and data. 1681 while (RelCur != RelEnd) { 1682 uint64_t Offset = RelCur->getOffset() - RelAdjustment; 1683 // If this relocation is hidden, skip it. 1684 if (getHidden(*RelCur) || SectionAddr + Offset < StartAddress) { 1685 ++RelCur; 1686 continue; 1687 } 1688 1689 // Stop when RelCur's offset is past the disassembled 1690 // instruction/data. Note that it's possible the disassembled data 1691 // is not the complete data: we might see the relocation printed in 1692 // the middle of the data, but this matches the binutils objdump 1693 // output. 1694 if (Offset >= Index + Size) 1695 break; 1696 1697 // When --adjust-vma is used, update the address printed. 1698 if (RelCur->getSymbol() != Obj.symbol_end()) { 1699 Expected<section_iterator> SymSI = 1700 RelCur->getSymbol()->getSection(); 1701 if (SymSI && *SymSI != Obj.section_end() && 1702 shouldAdjustVA(**SymSI)) 1703 Offset += AdjustVMA; 1704 } 1705 1706 printRelocation(FOS, Obj.getFileName(), *RelCur, 1707 SectionAddr + Offset, Is64Bits); 1708 LVP.printAfterOtherLine(FOS, true); 1709 ++RelCur; 1710 } 1711 } 1712 1713 Index += Size; 1714 } 1715 } 1716 } 1717 StringSet<> MissingDisasmSymbolSet = 1718 set_difference(DisasmSymbolSet, FoundDisasmSymbolSet); 1719 for (StringRef Sym : MissingDisasmSymbolSet.keys()) 1720 reportWarning("failed to disassemble missing symbol " + Sym, FileName); 1721 } 1722 1723 static void disassembleObject(ObjectFile *Obj, bool InlineRelocs) { 1724 const Target *TheTarget = getTarget(Obj); 1725 1726 // Package up features to be passed to target/subtarget 1727 SubtargetFeatures Features = Obj->getFeatures(); 1728 if (!MAttrs.empty()) { 1729 for (unsigned I = 0; I != MAttrs.size(); ++I) 1730 Features.AddFeature(MAttrs[I]); 1731 } else if (MCPU.empty() && Obj->getArch() == llvm::Triple::aarch64) { 1732 Features.AddFeature("+all"); 1733 } 1734 1735 std::unique_ptr<const MCRegisterInfo> MRI( 1736 TheTarget->createMCRegInfo(TripleName)); 1737 if (!MRI) 1738 reportError(Obj->getFileName(), 1739 "no register info for target " + TripleName); 1740 1741 // Set up disassembler. 1742 MCTargetOptions MCOptions; 1743 std::unique_ptr<const MCAsmInfo> AsmInfo( 1744 TheTarget->createMCAsmInfo(*MRI, TripleName, MCOptions)); 1745 if (!AsmInfo) 1746 reportError(Obj->getFileName(), 1747 "no assembly info for target " + TripleName); 1748 1749 if (MCPU.empty()) 1750 MCPU = Obj->tryGetCPUName().value_or("").str(); 1751 1752 std::unique_ptr<const MCSubtargetInfo> STI( 1753 TheTarget->createMCSubtargetInfo(TripleName, MCPU, Features.getString())); 1754 if (!STI) 1755 reportError(Obj->getFileName(), 1756 "no subtarget info for target " + TripleName); 1757 std::unique_ptr<const MCInstrInfo> MII(TheTarget->createMCInstrInfo()); 1758 if (!MII) 1759 reportError(Obj->getFileName(), 1760 "no instruction info for target " + TripleName); 1761 MCContext Ctx(Triple(TripleName), AsmInfo.get(), MRI.get(), STI.get()); 1762 // FIXME: for now initialize MCObjectFileInfo with default values 1763 std::unique_ptr<MCObjectFileInfo> MOFI( 1764 TheTarget->createMCObjectFileInfo(Ctx, /*PIC=*/false)); 1765 Ctx.setObjectFileInfo(MOFI.get()); 1766 1767 std::unique_ptr<MCDisassembler> DisAsm( 1768 TheTarget->createMCDisassembler(*STI, Ctx)); 1769 if (!DisAsm) 1770 reportError(Obj->getFileName(), "no disassembler for target " + TripleName); 1771 1772 // If we have an ARM object file, we need a second disassembler, because 1773 // ARM CPUs have two different instruction sets: ARM mode, and Thumb mode. 1774 // We use mapping symbols to switch between the two assemblers, where 1775 // appropriate. 1776 std::unique_ptr<MCDisassembler> SecondaryDisAsm; 1777 std::unique_ptr<const MCSubtargetInfo> SecondarySTI; 1778 if (isArmElf(*Obj) && !STI->checkFeatures("+mclass")) { 1779 if (STI->checkFeatures("+thumb-mode")) 1780 Features.AddFeature("-thumb-mode"); 1781 else 1782 Features.AddFeature("+thumb-mode"); 1783 SecondarySTI.reset(TheTarget->createMCSubtargetInfo(TripleName, MCPU, 1784 Features.getString())); 1785 SecondaryDisAsm.reset(TheTarget->createMCDisassembler(*SecondarySTI, Ctx)); 1786 } 1787 1788 std::unique_ptr<const MCInstrAnalysis> MIA( 1789 TheTarget->createMCInstrAnalysis(MII.get())); 1790 1791 int AsmPrinterVariant = AsmInfo->getAssemblerDialect(); 1792 std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter( 1793 Triple(TripleName), AsmPrinterVariant, *AsmInfo, *MII, *MRI)); 1794 if (!IP) 1795 reportError(Obj->getFileName(), 1796 "no instruction printer for target " + TripleName); 1797 IP->setPrintImmHex(PrintImmHex); 1798 IP->setPrintBranchImmAsAddress(true); 1799 IP->setSymbolizeOperands(SymbolizeOperands); 1800 IP->setMCInstrAnalysis(MIA.get()); 1801 1802 PrettyPrinter &PIP = selectPrettyPrinter(Triple(TripleName)); 1803 SourcePrinter SP(Obj, TheTarget->getName()); 1804 1805 for (StringRef Opt : DisassemblerOptions) 1806 if (!IP->applyTargetSpecificCLOption(Opt)) 1807 reportError(Obj->getFileName(), 1808 "Unrecognized disassembler option: " + Opt); 1809 1810 disassembleObject(TheTarget, *Obj, Ctx, DisAsm.get(), SecondaryDisAsm.get(), 1811 MIA.get(), IP.get(), STI.get(), SecondarySTI.get(), PIP, SP, 1812 InlineRelocs); 1813 } 1814 1815 void objdump::printRelocations(const ObjectFile *Obj) { 1816 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : 1817 "%08" PRIx64; 1818 1819 // Build a mapping from relocation target to a vector of relocation 1820 // sections. Usually, there is an only one relocation section for 1821 // each relocated section. 1822 MapVector<SectionRef, std::vector<SectionRef>> SecToRelSec; 1823 uint64_t Ndx; 1824 for (const SectionRef &Section : ToolSectionFilter(*Obj, &Ndx)) { 1825 if (Obj->isELF() && (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC)) 1826 continue; 1827 if (Section.relocation_begin() == Section.relocation_end()) 1828 continue; 1829 Expected<section_iterator> SecOrErr = Section.getRelocatedSection(); 1830 if (!SecOrErr) 1831 reportError(Obj->getFileName(), 1832 "section (" + Twine(Ndx) + 1833 "): unable to get a relocation target: " + 1834 toString(SecOrErr.takeError())); 1835 SecToRelSec[**SecOrErr].push_back(Section); 1836 } 1837 1838 for (std::pair<SectionRef, std::vector<SectionRef>> &P : SecToRelSec) { 1839 StringRef SecName = unwrapOrError(P.first.getName(), Obj->getFileName()); 1840 outs() << "\nRELOCATION RECORDS FOR [" << SecName << "]:\n"; 1841 uint32_t OffsetPadding = (Obj->getBytesInAddress() > 4 ? 16 : 8); 1842 uint32_t TypePadding = 24; 1843 outs() << left_justify("OFFSET", OffsetPadding) << " " 1844 << left_justify("TYPE", TypePadding) << " " 1845 << "VALUE\n"; 1846 1847 for (SectionRef Section : P.second) { 1848 for (const RelocationRef &Reloc : Section.relocations()) { 1849 uint64_t Address = Reloc.getOffset(); 1850 SmallString<32> RelocName; 1851 SmallString<32> ValueStr; 1852 if (Address < StartAddress || Address > StopAddress || getHidden(Reloc)) 1853 continue; 1854 Reloc.getTypeName(RelocName); 1855 if (Error E = getRelocationValueString(Reloc, ValueStr)) 1856 reportError(std::move(E), Obj->getFileName()); 1857 1858 outs() << format(Fmt.data(), Address) << " " 1859 << left_justify(RelocName, TypePadding) << " " << ValueStr 1860 << "\n"; 1861 } 1862 } 1863 } 1864 } 1865 1866 void objdump::printDynamicRelocations(const ObjectFile *Obj) { 1867 // For the moment, this option is for ELF only 1868 if (!Obj->isELF()) 1869 return; 1870 1871 const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj); 1872 if (!Elf || !any_of(Elf->sections(), [](const ELFSectionRef Sec) { 1873 return Sec.getType() == ELF::SHT_DYNAMIC; 1874 })) { 1875 reportError(Obj->getFileName(), "not a dynamic object"); 1876 return; 1877 } 1878 1879 std::vector<SectionRef> DynRelSec = Obj->dynamic_relocation_sections(); 1880 if (DynRelSec.empty()) 1881 return; 1882 1883 outs() << "\nDYNAMIC RELOCATION RECORDS\n"; 1884 const uint32_t OffsetPadding = (Obj->getBytesInAddress() > 4 ? 16 : 8); 1885 const uint32_t TypePadding = 24; 1886 outs() << left_justify("OFFSET", OffsetPadding) << ' ' 1887 << left_justify("TYPE", TypePadding) << " VALUE\n"; 1888 1889 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 1890 for (const SectionRef &Section : DynRelSec) 1891 for (const RelocationRef &Reloc : Section.relocations()) { 1892 uint64_t Address = Reloc.getOffset(); 1893 SmallString<32> RelocName; 1894 SmallString<32> ValueStr; 1895 Reloc.getTypeName(RelocName); 1896 if (Error E = getRelocationValueString(Reloc, ValueStr)) 1897 reportError(std::move(E), Obj->getFileName()); 1898 outs() << format(Fmt.data(), Address) << ' ' 1899 << left_justify(RelocName, TypePadding) << ' ' << ValueStr << '\n'; 1900 } 1901 } 1902 1903 // Returns true if we need to show LMA column when dumping section headers. We 1904 // show it only when the platform is ELF and either we have at least one section 1905 // whose VMA and LMA are different and/or when --show-lma flag is used. 1906 static bool shouldDisplayLMA(const ObjectFile &Obj) { 1907 if (!Obj.isELF()) 1908 return false; 1909 for (const SectionRef &S : ToolSectionFilter(Obj)) 1910 if (S.getAddress() != getELFSectionLMA(S)) 1911 return true; 1912 return ShowLMA; 1913 } 1914 1915 static size_t getMaxSectionNameWidth(const ObjectFile &Obj) { 1916 // Default column width for names is 13 even if no names are that long. 1917 size_t MaxWidth = 13; 1918 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 1919 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 1920 MaxWidth = std::max(MaxWidth, Name.size()); 1921 } 1922 return MaxWidth; 1923 } 1924 1925 void objdump::printSectionHeaders(ObjectFile &Obj) { 1926 size_t NameWidth = getMaxSectionNameWidth(Obj); 1927 size_t AddressWidth = 2 * Obj.getBytesInAddress(); 1928 bool HasLMAColumn = shouldDisplayLMA(Obj); 1929 outs() << "\nSections:\n"; 1930 if (HasLMAColumn) 1931 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 1932 << left_justify("VMA", AddressWidth) << " " 1933 << left_justify("LMA", AddressWidth) << " Type\n"; 1934 else 1935 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 1936 << left_justify("VMA", AddressWidth) << " Type\n"; 1937 1938 if (Obj.isELF() && Obj.sections().empty()) 1939 createFakeELFSections(Obj); 1940 1941 uint64_t Idx; 1942 for (const SectionRef &Section : ToolSectionFilter(Obj, &Idx)) { 1943 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 1944 uint64_t VMA = Section.getAddress(); 1945 if (shouldAdjustVA(Section)) 1946 VMA += AdjustVMA; 1947 1948 uint64_t Size = Section.getSize(); 1949 1950 std::string Type = Section.isText() ? "TEXT" : ""; 1951 if (Section.isData()) 1952 Type += Type.empty() ? "DATA" : ", DATA"; 1953 if (Section.isBSS()) 1954 Type += Type.empty() ? "BSS" : ", BSS"; 1955 if (Section.isDebugSection()) 1956 Type += Type.empty() ? "DEBUG" : ", DEBUG"; 1957 1958 if (HasLMAColumn) 1959 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 1960 Name.str().c_str(), Size) 1961 << format_hex_no_prefix(VMA, AddressWidth) << " " 1962 << format_hex_no_prefix(getELFSectionLMA(Section), AddressWidth) 1963 << " " << Type << "\n"; 1964 else 1965 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 1966 Name.str().c_str(), Size) 1967 << format_hex_no_prefix(VMA, AddressWidth) << " " << Type << "\n"; 1968 } 1969 } 1970 1971 void objdump::printSectionContents(const ObjectFile *Obj) { 1972 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj); 1973 1974 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1975 StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName()); 1976 uint64_t BaseAddr = Section.getAddress(); 1977 uint64_t Size = Section.getSize(); 1978 if (!Size) 1979 continue; 1980 1981 outs() << "Contents of section "; 1982 StringRef SegmentName = getSegmentName(MachO, Section); 1983 if (!SegmentName.empty()) 1984 outs() << SegmentName << ","; 1985 outs() << Name << ":\n"; 1986 if (Section.isBSS()) { 1987 outs() << format("<skipping contents of bss section at [%04" PRIx64 1988 ", %04" PRIx64 ")>\n", 1989 BaseAddr, BaseAddr + Size); 1990 continue; 1991 } 1992 1993 StringRef Contents = unwrapOrError(Section.getContents(), Obj->getFileName()); 1994 1995 // Dump out the content as hex and printable ascii characters. 1996 for (std::size_t Addr = 0, End = Contents.size(); Addr < End; Addr += 16) { 1997 outs() << format(" %04" PRIx64 " ", BaseAddr + Addr); 1998 // Dump line of hex. 1999 for (std::size_t I = 0; I < 16; ++I) { 2000 if (I != 0 && I % 4 == 0) 2001 outs() << ' '; 2002 if (Addr + I < End) 2003 outs() << hexdigit((Contents[Addr + I] >> 4) & 0xF, true) 2004 << hexdigit(Contents[Addr + I] & 0xF, true); 2005 else 2006 outs() << " "; 2007 } 2008 // Print ascii. 2009 outs() << " "; 2010 for (std::size_t I = 0; I < 16 && Addr + I < End; ++I) { 2011 if (isPrint(static_cast<unsigned char>(Contents[Addr + I]) & 0xFF)) 2012 outs() << Contents[Addr + I]; 2013 else 2014 outs() << "."; 2015 } 2016 outs() << "\n"; 2017 } 2018 } 2019 } 2020 2021 void objdump::printSymbolTable(const ObjectFile &O, StringRef ArchiveName, 2022 StringRef ArchitectureName, bool DumpDynamic) { 2023 if (O.isCOFF() && !DumpDynamic) { 2024 outs() << "\nSYMBOL TABLE:\n"; 2025 printCOFFSymbolTable(cast<const COFFObjectFile>(O)); 2026 return; 2027 } 2028 2029 const StringRef FileName = O.getFileName(); 2030 2031 if (!DumpDynamic) { 2032 outs() << "\nSYMBOL TABLE:\n"; 2033 for (auto I = O.symbol_begin(); I != O.symbol_end(); ++I) 2034 printSymbol(O, *I, {}, FileName, ArchiveName, ArchitectureName, 2035 DumpDynamic); 2036 return; 2037 } 2038 2039 outs() << "\nDYNAMIC SYMBOL TABLE:\n"; 2040 if (!O.isELF()) { 2041 reportWarning( 2042 "this operation is not currently supported for this file format", 2043 FileName); 2044 return; 2045 } 2046 2047 const ELFObjectFileBase *ELF = cast<const ELFObjectFileBase>(&O); 2048 auto Symbols = ELF->getDynamicSymbolIterators(); 2049 Expected<std::vector<VersionEntry>> SymbolVersionsOrErr = 2050 ELF->readDynsymVersions(); 2051 if (!SymbolVersionsOrErr) { 2052 reportWarning(toString(SymbolVersionsOrErr.takeError()), FileName); 2053 SymbolVersionsOrErr = std::vector<VersionEntry>(); 2054 (void)!SymbolVersionsOrErr; 2055 } 2056 for (auto &Sym : Symbols) 2057 printSymbol(O, Sym, *SymbolVersionsOrErr, FileName, ArchiveName, 2058 ArchitectureName, DumpDynamic); 2059 } 2060 2061 void objdump::printSymbol(const ObjectFile &O, const SymbolRef &Symbol, 2062 ArrayRef<VersionEntry> SymbolVersions, 2063 StringRef FileName, StringRef ArchiveName, 2064 StringRef ArchitectureName, bool DumpDynamic) { 2065 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&O); 2066 uint64_t Address = unwrapOrError(Symbol.getAddress(), FileName, ArchiveName, 2067 ArchitectureName); 2068 if ((Address < StartAddress) || (Address > StopAddress)) 2069 return; 2070 SymbolRef::Type Type = 2071 unwrapOrError(Symbol.getType(), FileName, ArchiveName, ArchitectureName); 2072 uint32_t Flags = 2073 unwrapOrError(Symbol.getFlags(), FileName, ArchiveName, ArchitectureName); 2074 2075 // Don't ask a Mach-O STAB symbol for its section unless you know that 2076 // STAB symbol's section field refers to a valid section index. Otherwise 2077 // the symbol may error trying to load a section that does not exist. 2078 bool IsSTAB = false; 2079 if (MachO) { 2080 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 2081 uint8_t NType = 2082 (MachO->is64Bit() ? MachO->getSymbol64TableEntry(SymDRI).n_type 2083 : MachO->getSymbolTableEntry(SymDRI).n_type); 2084 if (NType & MachO::N_STAB) 2085 IsSTAB = true; 2086 } 2087 section_iterator Section = IsSTAB 2088 ? O.section_end() 2089 : unwrapOrError(Symbol.getSection(), FileName, 2090 ArchiveName, ArchitectureName); 2091 2092 StringRef Name; 2093 if (Type == SymbolRef::ST_Debug && Section != O.section_end()) { 2094 if (Expected<StringRef> NameOrErr = Section->getName()) 2095 Name = *NameOrErr; 2096 else 2097 consumeError(NameOrErr.takeError()); 2098 2099 } else { 2100 Name = unwrapOrError(Symbol.getName(), FileName, ArchiveName, 2101 ArchitectureName); 2102 } 2103 2104 bool Global = Flags & SymbolRef::SF_Global; 2105 bool Weak = Flags & SymbolRef::SF_Weak; 2106 bool Absolute = Flags & SymbolRef::SF_Absolute; 2107 bool Common = Flags & SymbolRef::SF_Common; 2108 bool Hidden = Flags & SymbolRef::SF_Hidden; 2109 2110 char GlobLoc = ' '; 2111 if ((Section != O.section_end() || Absolute) && !Weak) 2112 GlobLoc = Global ? 'g' : 'l'; 2113 char IFunc = ' '; 2114 if (O.isELF()) { 2115 if (ELFSymbolRef(Symbol).getELFType() == ELF::STT_GNU_IFUNC) 2116 IFunc = 'i'; 2117 if (ELFSymbolRef(Symbol).getBinding() == ELF::STB_GNU_UNIQUE) 2118 GlobLoc = 'u'; 2119 } 2120 2121 char Debug = ' '; 2122 if (DumpDynamic) 2123 Debug = 'D'; 2124 else if (Type == SymbolRef::ST_Debug || Type == SymbolRef::ST_File) 2125 Debug = 'd'; 2126 2127 char FileFunc = ' '; 2128 if (Type == SymbolRef::ST_File) 2129 FileFunc = 'f'; 2130 else if (Type == SymbolRef::ST_Function) 2131 FileFunc = 'F'; 2132 else if (Type == SymbolRef::ST_Data) 2133 FileFunc = 'O'; 2134 2135 const char *Fmt = O.getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2136 2137 outs() << format(Fmt, Address) << " " 2138 << GlobLoc // Local -> 'l', Global -> 'g', Neither -> ' ' 2139 << (Weak ? 'w' : ' ') // Weak? 2140 << ' ' // Constructor. Not supported yet. 2141 << ' ' // Warning. Not supported yet. 2142 << IFunc // Indirect reference to another symbol. 2143 << Debug // Debugging (d) or dynamic (D) symbol. 2144 << FileFunc // Name of function (F), file (f) or object (O). 2145 << ' '; 2146 if (Absolute) { 2147 outs() << "*ABS*"; 2148 } else if (Common) { 2149 outs() << "*COM*"; 2150 } else if (Section == O.section_end()) { 2151 if (O.isXCOFF()) { 2152 XCOFFSymbolRef XCOFFSym = cast<const XCOFFObjectFile>(O).toSymbolRef( 2153 Symbol.getRawDataRefImpl()); 2154 if (XCOFF::N_DEBUG == XCOFFSym.getSectionNumber()) 2155 outs() << "*DEBUG*"; 2156 else 2157 outs() << "*UND*"; 2158 } else 2159 outs() << "*UND*"; 2160 } else { 2161 StringRef SegmentName = getSegmentName(MachO, *Section); 2162 if (!SegmentName.empty()) 2163 outs() << SegmentName << ","; 2164 StringRef SectionName = unwrapOrError(Section->getName(), FileName); 2165 outs() << SectionName; 2166 if (O.isXCOFF()) { 2167 Optional<SymbolRef> SymRef = 2168 getXCOFFSymbolContainingSymbolRef(cast<XCOFFObjectFile>(O), Symbol); 2169 if (SymRef) { 2170 2171 Expected<StringRef> NameOrErr = SymRef->getName(); 2172 2173 if (NameOrErr) { 2174 outs() << " (csect:"; 2175 std::string SymName(NameOrErr.get()); 2176 2177 if (Demangle) 2178 SymName = demangle(SymName); 2179 2180 if (SymbolDescription) 2181 SymName = getXCOFFSymbolDescription( 2182 createSymbolInfo(O, SymRef.value()), SymName); 2183 2184 outs() << ' ' << SymName; 2185 outs() << ") "; 2186 } else 2187 reportWarning(toString(NameOrErr.takeError()), FileName); 2188 } 2189 } 2190 } 2191 2192 if (Common) 2193 outs() << '\t' << format(Fmt, static_cast<uint64_t>(Symbol.getAlignment())); 2194 else if (O.isXCOFF()) 2195 outs() << '\t' 2196 << format(Fmt, cast<XCOFFObjectFile>(O).getSymbolSize( 2197 Symbol.getRawDataRefImpl())); 2198 else if (O.isELF()) 2199 outs() << '\t' << format(Fmt, ELFSymbolRef(Symbol).getSize()); 2200 2201 if (O.isELF()) { 2202 if (!SymbolVersions.empty()) { 2203 const VersionEntry &Ver = 2204 SymbolVersions[Symbol.getRawDataRefImpl().d.b - 1]; 2205 std::string Str; 2206 if (!Ver.Name.empty()) 2207 Str = Ver.IsVerDef ? ' ' + Ver.Name : '(' + Ver.Name + ')'; 2208 outs() << ' ' << left_justify(Str, 12); 2209 } 2210 2211 uint8_t Other = ELFSymbolRef(Symbol).getOther(); 2212 switch (Other) { 2213 case ELF::STV_DEFAULT: 2214 break; 2215 case ELF::STV_INTERNAL: 2216 outs() << " .internal"; 2217 break; 2218 case ELF::STV_HIDDEN: 2219 outs() << " .hidden"; 2220 break; 2221 case ELF::STV_PROTECTED: 2222 outs() << " .protected"; 2223 break; 2224 default: 2225 outs() << format(" 0x%02x", Other); 2226 break; 2227 } 2228 } else if (Hidden) { 2229 outs() << " .hidden"; 2230 } 2231 2232 std::string SymName(Name); 2233 if (Demangle) 2234 SymName = demangle(SymName); 2235 2236 if (O.isXCOFF() && SymbolDescription) 2237 SymName = getXCOFFSymbolDescription(createSymbolInfo(O, Symbol), SymName); 2238 2239 outs() << ' ' << SymName << '\n'; 2240 } 2241 2242 static void printUnwindInfo(const ObjectFile *O) { 2243 outs() << "Unwind info:\n\n"; 2244 2245 if (const COFFObjectFile *Coff = dyn_cast<COFFObjectFile>(O)) 2246 printCOFFUnwindInfo(Coff); 2247 else if (const MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(O)) 2248 printMachOUnwindInfo(MachO); 2249 else 2250 // TODO: Extract DWARF dump tool to objdump. 2251 WithColor::error(errs(), ToolName) 2252 << "This operation is only currently supported " 2253 "for COFF and MachO object files.\n"; 2254 } 2255 2256 /// Dump the raw contents of the __clangast section so the output can be piped 2257 /// into llvm-bcanalyzer. 2258 static void printRawClangAST(const ObjectFile *Obj) { 2259 if (outs().is_displayed()) { 2260 WithColor::error(errs(), ToolName) 2261 << "The -raw-clang-ast option will dump the raw binary contents of " 2262 "the clang ast section.\n" 2263 "Please redirect the output to a file or another program such as " 2264 "llvm-bcanalyzer.\n"; 2265 return; 2266 } 2267 2268 StringRef ClangASTSectionName("__clangast"); 2269 if (Obj->isCOFF()) { 2270 ClangASTSectionName = "clangast"; 2271 } 2272 2273 Optional<object::SectionRef> ClangASTSection; 2274 for (auto Sec : ToolSectionFilter(*Obj)) { 2275 StringRef Name; 2276 if (Expected<StringRef> NameOrErr = Sec.getName()) 2277 Name = *NameOrErr; 2278 else 2279 consumeError(NameOrErr.takeError()); 2280 2281 if (Name == ClangASTSectionName) { 2282 ClangASTSection = Sec; 2283 break; 2284 } 2285 } 2286 if (!ClangASTSection) 2287 return; 2288 2289 StringRef ClangASTContents = 2290 unwrapOrError(ClangASTSection.value().getContents(), Obj->getFileName()); 2291 outs().write(ClangASTContents.data(), ClangASTContents.size()); 2292 } 2293 2294 static void printFaultMaps(const ObjectFile *Obj) { 2295 StringRef FaultMapSectionName; 2296 2297 if (Obj->isELF()) { 2298 FaultMapSectionName = ".llvm_faultmaps"; 2299 } else if (Obj->isMachO()) { 2300 FaultMapSectionName = "__llvm_faultmaps"; 2301 } else { 2302 WithColor::error(errs(), ToolName) 2303 << "This operation is only currently supported " 2304 "for ELF and Mach-O executable files.\n"; 2305 return; 2306 } 2307 2308 Optional<object::SectionRef> FaultMapSection; 2309 2310 for (auto Sec : ToolSectionFilter(*Obj)) { 2311 StringRef Name; 2312 if (Expected<StringRef> NameOrErr = Sec.getName()) 2313 Name = *NameOrErr; 2314 else 2315 consumeError(NameOrErr.takeError()); 2316 2317 if (Name == FaultMapSectionName) { 2318 FaultMapSection = Sec; 2319 break; 2320 } 2321 } 2322 2323 outs() << "FaultMap table:\n"; 2324 2325 if (!FaultMapSection) { 2326 outs() << "<not found>\n"; 2327 return; 2328 } 2329 2330 StringRef FaultMapContents = 2331 unwrapOrError(FaultMapSection->getContents(), Obj->getFileName()); 2332 FaultMapParser FMP(FaultMapContents.bytes_begin(), 2333 FaultMapContents.bytes_end()); 2334 2335 outs() << FMP; 2336 } 2337 2338 static void printPrivateFileHeaders(const ObjectFile *O, bool OnlyFirst) { 2339 if (O->isELF()) { 2340 printELFFileHeader(O); 2341 printELFDynamicSection(O); 2342 printELFSymbolVersionInfo(O); 2343 return; 2344 } 2345 if (O->isCOFF()) 2346 return printCOFFFileHeader(cast<object::COFFObjectFile>(*O)); 2347 if (O->isWasm()) 2348 return printWasmFileHeader(O); 2349 if (O->isMachO()) { 2350 printMachOFileHeader(O); 2351 if (!OnlyFirst) 2352 printMachOLoadCommands(O); 2353 return; 2354 } 2355 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2356 } 2357 2358 static void printFileHeaders(const ObjectFile *O) { 2359 if (!O->isELF() && !O->isCOFF()) 2360 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2361 2362 Triple::ArchType AT = O->getArch(); 2363 outs() << "architecture: " << Triple::getArchTypeName(AT) << "\n"; 2364 uint64_t Address = unwrapOrError(O->getStartAddress(), O->getFileName()); 2365 2366 StringRef Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2367 outs() << "start address: " 2368 << "0x" << format(Fmt.data(), Address) << "\n"; 2369 } 2370 2371 static void printArchiveChild(StringRef Filename, const Archive::Child &C) { 2372 Expected<sys::fs::perms> ModeOrErr = C.getAccessMode(); 2373 if (!ModeOrErr) { 2374 WithColor::error(errs(), ToolName) << "ill-formed archive entry.\n"; 2375 consumeError(ModeOrErr.takeError()); 2376 return; 2377 } 2378 sys::fs::perms Mode = ModeOrErr.get(); 2379 outs() << ((Mode & sys::fs::owner_read) ? "r" : "-"); 2380 outs() << ((Mode & sys::fs::owner_write) ? "w" : "-"); 2381 outs() << ((Mode & sys::fs::owner_exe) ? "x" : "-"); 2382 outs() << ((Mode & sys::fs::group_read) ? "r" : "-"); 2383 outs() << ((Mode & sys::fs::group_write) ? "w" : "-"); 2384 outs() << ((Mode & sys::fs::group_exe) ? "x" : "-"); 2385 outs() << ((Mode & sys::fs::others_read) ? "r" : "-"); 2386 outs() << ((Mode & sys::fs::others_write) ? "w" : "-"); 2387 outs() << ((Mode & sys::fs::others_exe) ? "x" : "-"); 2388 2389 outs() << " "; 2390 2391 outs() << format("%d/%d %6" PRId64 " ", unwrapOrError(C.getUID(), Filename), 2392 unwrapOrError(C.getGID(), Filename), 2393 unwrapOrError(C.getRawSize(), Filename)); 2394 2395 StringRef RawLastModified = C.getRawLastModified(); 2396 unsigned Seconds; 2397 if (RawLastModified.getAsInteger(10, Seconds)) 2398 outs() << "(date: \"" << RawLastModified 2399 << "\" contains non-decimal chars) "; 2400 else { 2401 // Since ctime(3) returns a 26 character string of the form: 2402 // "Sun Sep 16 01:03:52 1973\n\0" 2403 // just print 24 characters. 2404 time_t t = Seconds; 2405 outs() << format("%.24s ", ctime(&t)); 2406 } 2407 2408 StringRef Name = ""; 2409 Expected<StringRef> NameOrErr = C.getName(); 2410 if (!NameOrErr) { 2411 consumeError(NameOrErr.takeError()); 2412 Name = unwrapOrError(C.getRawName(), Filename); 2413 } else { 2414 Name = NameOrErr.get(); 2415 } 2416 outs() << Name << "\n"; 2417 } 2418 2419 // For ELF only now. 2420 static bool shouldWarnForInvalidStartStopAddress(ObjectFile *Obj) { 2421 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj)) { 2422 if (Elf->getEType() != ELF::ET_REL) 2423 return true; 2424 } 2425 return false; 2426 } 2427 2428 static void checkForInvalidStartStopAddress(ObjectFile *Obj, 2429 uint64_t Start, uint64_t Stop) { 2430 if (!shouldWarnForInvalidStartStopAddress(Obj)) 2431 return; 2432 2433 for (const SectionRef &Section : Obj->sections()) 2434 if (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC) { 2435 uint64_t BaseAddr = Section.getAddress(); 2436 uint64_t Size = Section.getSize(); 2437 if ((Start < BaseAddr + Size) && Stop > BaseAddr) 2438 return; 2439 } 2440 2441 if (!HasStartAddressFlag) 2442 reportWarning("no section has address less than 0x" + 2443 Twine::utohexstr(Stop) + " specified by --stop-address", 2444 Obj->getFileName()); 2445 else if (!HasStopAddressFlag) 2446 reportWarning("no section has address greater than or equal to 0x" + 2447 Twine::utohexstr(Start) + " specified by --start-address", 2448 Obj->getFileName()); 2449 else 2450 reportWarning("no section overlaps the range [0x" + 2451 Twine::utohexstr(Start) + ",0x" + Twine::utohexstr(Stop) + 2452 ") specified by --start-address/--stop-address", 2453 Obj->getFileName()); 2454 } 2455 2456 static void dumpObject(ObjectFile *O, const Archive *A = nullptr, 2457 const Archive::Child *C = nullptr) { 2458 // Avoid other output when using a raw option. 2459 if (!RawClangAST) { 2460 outs() << '\n'; 2461 if (A) 2462 outs() << A->getFileName() << "(" << O->getFileName() << ")"; 2463 else 2464 outs() << O->getFileName(); 2465 outs() << ":\tfile format " << O->getFileFormatName().lower() << "\n"; 2466 } 2467 2468 if (HasStartAddressFlag || HasStopAddressFlag) 2469 checkForInvalidStartStopAddress(O, StartAddress, StopAddress); 2470 2471 // Note: the order here matches GNU objdump for compatability. 2472 StringRef ArchiveName = A ? A->getFileName() : ""; 2473 if (ArchiveHeaders && !MachOOpt && C) 2474 printArchiveChild(ArchiveName, *C); 2475 if (FileHeaders) 2476 printFileHeaders(O); 2477 if (PrivateHeaders || FirstPrivateHeader) 2478 printPrivateFileHeaders(O, FirstPrivateHeader); 2479 if (SectionHeaders) 2480 printSectionHeaders(*O); 2481 if (SymbolTable) 2482 printSymbolTable(*O, ArchiveName); 2483 if (DynamicSymbolTable) 2484 printSymbolTable(*O, ArchiveName, /*ArchitectureName=*/"", 2485 /*DumpDynamic=*/true); 2486 if (DwarfDumpType != DIDT_Null) { 2487 std::unique_ptr<DIContext> DICtx = DWARFContext::create(*O); 2488 // Dump the complete DWARF structure. 2489 DIDumpOptions DumpOpts; 2490 DumpOpts.DumpType = DwarfDumpType; 2491 DICtx->dump(outs(), DumpOpts); 2492 } 2493 if (Relocations && !Disassemble) 2494 printRelocations(O); 2495 if (DynamicRelocations) 2496 printDynamicRelocations(O); 2497 if (SectionContents) 2498 printSectionContents(O); 2499 if (Disassemble) 2500 disassembleObject(O, Relocations); 2501 if (UnwindInfo) 2502 printUnwindInfo(O); 2503 2504 // Mach-O specific options: 2505 if (ExportsTrie) 2506 printExportsTrie(O); 2507 if (Rebase) 2508 printRebaseTable(O); 2509 if (Bind) 2510 printBindTable(O); 2511 if (LazyBind) 2512 printLazyBindTable(O); 2513 if (WeakBind) 2514 printWeakBindTable(O); 2515 2516 // Other special sections: 2517 if (RawClangAST) 2518 printRawClangAST(O); 2519 if (FaultMapSection) 2520 printFaultMaps(O); 2521 if (Offloading) 2522 dumpOffloadBinary(*O); 2523 } 2524 2525 static void dumpObject(const COFFImportFile *I, const Archive *A, 2526 const Archive::Child *C = nullptr) { 2527 StringRef ArchiveName = A ? A->getFileName() : ""; 2528 2529 // Avoid other output when using a raw option. 2530 if (!RawClangAST) 2531 outs() << '\n' 2532 << ArchiveName << "(" << I->getFileName() << ")" 2533 << ":\tfile format COFF-import-file" 2534 << "\n\n"; 2535 2536 if (ArchiveHeaders && !MachOOpt && C) 2537 printArchiveChild(ArchiveName, *C); 2538 if (SymbolTable) 2539 printCOFFSymbolTable(*I); 2540 } 2541 2542 /// Dump each object file in \a a; 2543 static void dumpArchive(const Archive *A) { 2544 Error Err = Error::success(); 2545 unsigned I = -1; 2546 for (auto &C : A->children(Err)) { 2547 ++I; 2548 Expected<std::unique_ptr<Binary>> ChildOrErr = C.getAsBinary(); 2549 if (!ChildOrErr) { 2550 if (auto E = isNotObjectErrorInvalidFileType(ChildOrErr.takeError())) 2551 reportError(std::move(E), getFileNameForError(C, I), A->getFileName()); 2552 continue; 2553 } 2554 if (ObjectFile *O = dyn_cast<ObjectFile>(&*ChildOrErr.get())) 2555 dumpObject(O, A, &C); 2556 else if (COFFImportFile *I = dyn_cast<COFFImportFile>(&*ChildOrErr.get())) 2557 dumpObject(I, A, &C); 2558 else 2559 reportError(errorCodeToError(object_error::invalid_file_type), 2560 A->getFileName()); 2561 } 2562 if (Err) 2563 reportError(std::move(Err), A->getFileName()); 2564 } 2565 2566 /// Open file and figure out how to dump it. 2567 static void dumpInput(StringRef file) { 2568 // If we are using the Mach-O specific object file parser, then let it parse 2569 // the file and process the command line options. So the -arch flags can 2570 // be used to select specific slices, etc. 2571 if (MachOOpt) { 2572 parseInputMachO(file); 2573 return; 2574 } 2575 2576 // Attempt to open the binary. 2577 OwningBinary<Binary> OBinary = unwrapOrError(createBinary(file), file); 2578 Binary &Binary = *OBinary.getBinary(); 2579 2580 if (Archive *A = dyn_cast<Archive>(&Binary)) 2581 dumpArchive(A); 2582 else if (ObjectFile *O = dyn_cast<ObjectFile>(&Binary)) 2583 dumpObject(O); 2584 else if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(&Binary)) 2585 parseInputMachO(UB); 2586 else if (OffloadBinary *OB = dyn_cast<OffloadBinary>(&Binary)) 2587 dumpOffloadSections(*OB); 2588 else 2589 reportError(errorCodeToError(object_error::invalid_file_type), file); 2590 } 2591 2592 template <typename T> 2593 static void parseIntArg(const llvm::opt::InputArgList &InputArgs, int ID, 2594 T &Value) { 2595 if (const opt::Arg *A = InputArgs.getLastArg(ID)) { 2596 StringRef V(A->getValue()); 2597 if (!llvm::to_integer(V, Value, 0)) { 2598 reportCmdLineError(A->getSpelling() + 2599 ": expected a non-negative integer, but got '" + V + 2600 "'"); 2601 } 2602 } 2603 } 2604 2605 static void invalidArgValue(const opt::Arg *A) { 2606 reportCmdLineError("'" + StringRef(A->getValue()) + 2607 "' is not a valid value for '" + A->getSpelling() + "'"); 2608 } 2609 2610 static std::vector<std::string> 2611 commaSeparatedValues(const llvm::opt::InputArgList &InputArgs, int ID) { 2612 std::vector<std::string> Values; 2613 for (StringRef Value : InputArgs.getAllArgValues(ID)) { 2614 llvm::SmallVector<StringRef, 2> SplitValues; 2615 llvm::SplitString(Value, SplitValues, ","); 2616 for (StringRef SplitValue : SplitValues) 2617 Values.push_back(SplitValue.str()); 2618 } 2619 return Values; 2620 } 2621 2622 static void parseOtoolOptions(const llvm::opt::InputArgList &InputArgs) { 2623 MachOOpt = true; 2624 FullLeadingAddr = true; 2625 PrintImmHex = true; 2626 2627 ArchName = InputArgs.getLastArgValue(OTOOL_arch).str(); 2628 LinkOptHints = InputArgs.hasArg(OTOOL_C); 2629 if (InputArgs.hasArg(OTOOL_d)) 2630 FilterSections.push_back("__DATA,__data"); 2631 DylibId = InputArgs.hasArg(OTOOL_D); 2632 UniversalHeaders = InputArgs.hasArg(OTOOL_f); 2633 DataInCode = InputArgs.hasArg(OTOOL_G); 2634 FirstPrivateHeader = InputArgs.hasArg(OTOOL_h); 2635 IndirectSymbols = InputArgs.hasArg(OTOOL_I); 2636 ShowRawInsn = InputArgs.hasArg(OTOOL_j); 2637 PrivateHeaders = InputArgs.hasArg(OTOOL_l); 2638 DylibsUsed = InputArgs.hasArg(OTOOL_L); 2639 MCPU = InputArgs.getLastArgValue(OTOOL_mcpu_EQ).str(); 2640 ObjcMetaData = InputArgs.hasArg(OTOOL_o); 2641 DisSymName = InputArgs.getLastArgValue(OTOOL_p).str(); 2642 InfoPlist = InputArgs.hasArg(OTOOL_P); 2643 Relocations = InputArgs.hasArg(OTOOL_r); 2644 if (const Arg *A = InputArgs.getLastArg(OTOOL_s)) { 2645 auto Filter = (A->getValue(0) + StringRef(",") + A->getValue(1)).str(); 2646 FilterSections.push_back(Filter); 2647 } 2648 if (InputArgs.hasArg(OTOOL_t)) 2649 FilterSections.push_back("__TEXT,__text"); 2650 Verbose = InputArgs.hasArg(OTOOL_v) || InputArgs.hasArg(OTOOL_V) || 2651 InputArgs.hasArg(OTOOL_o); 2652 SymbolicOperands = InputArgs.hasArg(OTOOL_V); 2653 if (InputArgs.hasArg(OTOOL_x)) 2654 FilterSections.push_back(",__text"); 2655 LeadingAddr = LeadingHeaders = !InputArgs.hasArg(OTOOL_X); 2656 2657 InputFilenames = InputArgs.getAllArgValues(OTOOL_INPUT); 2658 if (InputFilenames.empty()) 2659 reportCmdLineError("no input file"); 2660 2661 for (const Arg *A : InputArgs) { 2662 const Option &O = A->getOption(); 2663 if (O.getGroup().isValid() && O.getGroup().getID() == OTOOL_grp_obsolete) { 2664 reportCmdLineWarning(O.getPrefixedName() + 2665 " is obsolete and not implemented"); 2666 } 2667 } 2668 } 2669 2670 static void parseObjdumpOptions(const llvm::opt::InputArgList &InputArgs) { 2671 parseIntArg(InputArgs, OBJDUMP_adjust_vma_EQ, AdjustVMA); 2672 AllHeaders = InputArgs.hasArg(OBJDUMP_all_headers); 2673 ArchName = InputArgs.getLastArgValue(OBJDUMP_arch_name_EQ).str(); 2674 ArchiveHeaders = InputArgs.hasArg(OBJDUMP_archive_headers); 2675 Demangle = InputArgs.hasArg(OBJDUMP_demangle); 2676 Disassemble = InputArgs.hasArg(OBJDUMP_disassemble); 2677 DisassembleAll = InputArgs.hasArg(OBJDUMP_disassemble_all); 2678 SymbolDescription = InputArgs.hasArg(OBJDUMP_symbol_description); 2679 DisassembleSymbols = 2680 commaSeparatedValues(InputArgs, OBJDUMP_disassemble_symbols_EQ); 2681 DisassembleZeroes = InputArgs.hasArg(OBJDUMP_disassemble_zeroes); 2682 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_dwarf_EQ)) { 2683 DwarfDumpType = StringSwitch<DIDumpType>(A->getValue()) 2684 .Case("frames", DIDT_DebugFrame) 2685 .Default(DIDT_Null); 2686 if (DwarfDumpType == DIDT_Null) 2687 invalidArgValue(A); 2688 } 2689 DynamicRelocations = InputArgs.hasArg(OBJDUMP_dynamic_reloc); 2690 FaultMapSection = InputArgs.hasArg(OBJDUMP_fault_map_section); 2691 Offloading = InputArgs.hasArg(OBJDUMP_offloading); 2692 FileHeaders = InputArgs.hasArg(OBJDUMP_file_headers); 2693 SectionContents = InputArgs.hasArg(OBJDUMP_full_contents); 2694 PrintLines = InputArgs.hasArg(OBJDUMP_line_numbers); 2695 InputFilenames = InputArgs.getAllArgValues(OBJDUMP_INPUT); 2696 MachOOpt = InputArgs.hasArg(OBJDUMP_macho); 2697 MCPU = InputArgs.getLastArgValue(OBJDUMP_mcpu_EQ).str(); 2698 MAttrs = commaSeparatedValues(InputArgs, OBJDUMP_mattr_EQ); 2699 ShowRawInsn = !InputArgs.hasArg(OBJDUMP_no_show_raw_insn); 2700 LeadingAddr = !InputArgs.hasArg(OBJDUMP_no_leading_addr); 2701 RawClangAST = InputArgs.hasArg(OBJDUMP_raw_clang_ast); 2702 Relocations = InputArgs.hasArg(OBJDUMP_reloc); 2703 PrintImmHex = 2704 InputArgs.hasFlag(OBJDUMP_print_imm_hex, OBJDUMP_no_print_imm_hex, false); 2705 PrivateHeaders = InputArgs.hasArg(OBJDUMP_private_headers); 2706 FilterSections = InputArgs.getAllArgValues(OBJDUMP_section_EQ); 2707 SectionHeaders = InputArgs.hasArg(OBJDUMP_section_headers); 2708 ShowLMA = InputArgs.hasArg(OBJDUMP_show_lma); 2709 PrintSource = InputArgs.hasArg(OBJDUMP_source); 2710 parseIntArg(InputArgs, OBJDUMP_start_address_EQ, StartAddress); 2711 HasStartAddressFlag = InputArgs.hasArg(OBJDUMP_start_address_EQ); 2712 parseIntArg(InputArgs, OBJDUMP_stop_address_EQ, StopAddress); 2713 HasStopAddressFlag = InputArgs.hasArg(OBJDUMP_stop_address_EQ); 2714 SymbolTable = InputArgs.hasArg(OBJDUMP_syms); 2715 SymbolizeOperands = InputArgs.hasArg(OBJDUMP_symbolize_operands); 2716 DynamicSymbolTable = InputArgs.hasArg(OBJDUMP_dynamic_syms); 2717 TripleName = InputArgs.getLastArgValue(OBJDUMP_triple_EQ).str(); 2718 UnwindInfo = InputArgs.hasArg(OBJDUMP_unwind_info); 2719 Wide = InputArgs.hasArg(OBJDUMP_wide); 2720 Prefix = InputArgs.getLastArgValue(OBJDUMP_prefix).str(); 2721 parseIntArg(InputArgs, OBJDUMP_prefix_strip, PrefixStrip); 2722 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_debug_vars_EQ)) { 2723 DbgVariables = StringSwitch<DebugVarsFormat>(A->getValue()) 2724 .Case("ascii", DVASCII) 2725 .Case("unicode", DVUnicode) 2726 .Default(DVInvalid); 2727 if (DbgVariables == DVInvalid) 2728 invalidArgValue(A); 2729 } 2730 parseIntArg(InputArgs, OBJDUMP_debug_vars_indent_EQ, DbgIndent); 2731 2732 parseMachOOptions(InputArgs); 2733 2734 // Parse -M (--disassembler-options) and deprecated 2735 // --x86-asm-syntax={att,intel}. 2736 // 2737 // Note, for x86, the asm dialect (AssemblerDialect) is initialized when the 2738 // MCAsmInfo is constructed. MCInstPrinter::applyTargetSpecificCLOption is 2739 // called too late. For now we have to use the internal cl::opt option. 2740 const char *AsmSyntax = nullptr; 2741 for (const auto *A : InputArgs.filtered(OBJDUMP_disassembler_options_EQ, 2742 OBJDUMP_x86_asm_syntax_att, 2743 OBJDUMP_x86_asm_syntax_intel)) { 2744 switch (A->getOption().getID()) { 2745 case OBJDUMP_x86_asm_syntax_att: 2746 AsmSyntax = "--x86-asm-syntax=att"; 2747 continue; 2748 case OBJDUMP_x86_asm_syntax_intel: 2749 AsmSyntax = "--x86-asm-syntax=intel"; 2750 continue; 2751 } 2752 2753 SmallVector<StringRef, 2> Values; 2754 llvm::SplitString(A->getValue(), Values, ","); 2755 for (StringRef V : Values) { 2756 if (V == "att") 2757 AsmSyntax = "--x86-asm-syntax=att"; 2758 else if (V == "intel") 2759 AsmSyntax = "--x86-asm-syntax=intel"; 2760 else 2761 DisassemblerOptions.push_back(V.str()); 2762 } 2763 } 2764 if (AsmSyntax) { 2765 const char *Argv[] = {"llvm-objdump", AsmSyntax}; 2766 llvm::cl::ParseCommandLineOptions(2, Argv); 2767 } 2768 2769 // objdump defaults to a.out if no filenames specified. 2770 if (InputFilenames.empty()) 2771 InputFilenames.push_back("a.out"); 2772 } 2773 2774 int main(int argc, char **argv) { 2775 using namespace llvm; 2776 InitLLVM X(argc, argv); 2777 2778 ToolName = argv[0]; 2779 std::unique_ptr<CommonOptTable> T; 2780 OptSpecifier Unknown, HelpFlag, HelpHiddenFlag, VersionFlag; 2781 2782 StringRef Stem = sys::path::stem(ToolName); 2783 auto Is = [=](StringRef Tool) { 2784 // We need to recognize the following filenames: 2785 // 2786 // llvm-objdump -> objdump 2787 // llvm-otool-10.exe -> otool 2788 // powerpc64-unknown-freebsd13-objdump -> objdump 2789 auto I = Stem.rfind_insensitive(Tool); 2790 return I != StringRef::npos && 2791 (I + Tool.size() == Stem.size() || !isAlnum(Stem[I + Tool.size()])); 2792 }; 2793 if (Is("otool")) { 2794 T = std::make_unique<OtoolOptTable>(); 2795 Unknown = OTOOL_UNKNOWN; 2796 HelpFlag = OTOOL_help; 2797 HelpHiddenFlag = OTOOL_help_hidden; 2798 VersionFlag = OTOOL_version; 2799 } else { 2800 T = std::make_unique<ObjdumpOptTable>(); 2801 Unknown = OBJDUMP_UNKNOWN; 2802 HelpFlag = OBJDUMP_help; 2803 HelpHiddenFlag = OBJDUMP_help_hidden; 2804 VersionFlag = OBJDUMP_version; 2805 } 2806 2807 BumpPtrAllocator A; 2808 StringSaver Saver(A); 2809 opt::InputArgList InputArgs = 2810 T->parseArgs(argc, argv, Unknown, Saver, 2811 [&](StringRef Msg) { reportCmdLineError(Msg); }); 2812 2813 if (InputArgs.size() == 0 || InputArgs.hasArg(HelpFlag)) { 2814 T->printHelp(ToolName); 2815 return 0; 2816 } 2817 if (InputArgs.hasArg(HelpHiddenFlag)) { 2818 T->printHelp(ToolName, /*ShowHidden=*/true); 2819 return 0; 2820 } 2821 2822 // Initialize targets and assembly printers/parsers. 2823 InitializeAllTargetInfos(); 2824 InitializeAllTargetMCs(); 2825 InitializeAllDisassemblers(); 2826 2827 if (InputArgs.hasArg(VersionFlag)) { 2828 cl::PrintVersionMessage(); 2829 if (!Is("otool")) { 2830 outs() << '\n'; 2831 TargetRegistry::printRegisteredTargetsForVersion(outs()); 2832 } 2833 return 0; 2834 } 2835 2836 if (Is("otool")) 2837 parseOtoolOptions(InputArgs); 2838 else 2839 parseObjdumpOptions(InputArgs); 2840 2841 if (StartAddress >= StopAddress) 2842 reportCmdLineError("start address should be less than stop address"); 2843 2844 // Removes trailing separators from prefix. 2845 while (!Prefix.empty() && sys::path::is_separator(Prefix.back())) 2846 Prefix.pop_back(); 2847 2848 if (AllHeaders) 2849 ArchiveHeaders = FileHeaders = PrivateHeaders = Relocations = 2850 SectionHeaders = SymbolTable = true; 2851 2852 if (DisassembleAll || PrintSource || PrintLines || 2853 !DisassembleSymbols.empty()) 2854 Disassemble = true; 2855 2856 if (!ArchiveHeaders && !Disassemble && DwarfDumpType == DIDT_Null && 2857 !DynamicRelocations && !FileHeaders && !PrivateHeaders && !RawClangAST && 2858 !Relocations && !SectionHeaders && !SectionContents && !SymbolTable && 2859 !DynamicSymbolTable && !UnwindInfo && !FaultMapSection && !Offloading && 2860 !(MachOOpt && (Bind || DataInCode || DyldInfo || DylibId || DylibsUsed || 2861 ExportsTrie || FirstPrivateHeader || FunctionStarts || 2862 IndirectSymbols || InfoPlist || LazyBind || LinkOptHints || 2863 ObjcMetaData || Rebase || Rpaths || UniversalHeaders || 2864 WeakBind || !FilterSections.empty()))) { 2865 T->printHelp(ToolName); 2866 return 2; 2867 } 2868 2869 DisasmSymbolSet.insert(DisassembleSymbols.begin(), DisassembleSymbols.end()); 2870 2871 llvm::for_each(InputFilenames, dumpInput); 2872 2873 warnOnNoMatchForSections(); 2874 2875 return EXIT_SUCCESS; 2876 } 2877