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