1 //===-- llvm-objdump.cpp - Object file dumping utility for llvm -----------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This program is a utility that works like binutils "objdump", that is, it 10 // dumps out a plethora of information about an object file depending on the 11 // flags. 12 // 13 // The flags and output of this program should be near identical to those of 14 // binutils objdump. 15 // 16 //===----------------------------------------------------------------------===// 17 18 #include "llvm-objdump.h" 19 #include "COFFDump.h" 20 #include "ELFDump.h" 21 #include "MachODump.h" 22 #include "ObjdumpOptID.h" 23 #include "OffloadDump.h" 24 #include "SourcePrinter.h" 25 #include "WasmDump.h" 26 #include "XCOFFDump.h" 27 #include "llvm/ADT/IndexedMap.h" 28 #include "llvm/ADT/Optional.h" 29 #include "llvm/ADT/STLExtras.h" 30 #include "llvm/ADT/SetOperations.h" 31 #include "llvm/ADT/SmallSet.h" 32 #include "llvm/ADT/StringExtras.h" 33 #include "llvm/ADT/StringSet.h" 34 #include "llvm/ADT/Triple.h" 35 #include "llvm/ADT/Twine.h" 36 #include "llvm/DebugInfo/DWARF/DWARFContext.h" 37 #include "llvm/DebugInfo/Symbolize/SymbolizableModule.h" 38 #include "llvm/DebugInfo/Symbolize/Symbolize.h" 39 #include "llvm/Demangle/Demangle.h" 40 #include "llvm/MC/MCAsmInfo.h" 41 #include "llvm/MC/MCContext.h" 42 #include "llvm/MC/MCDisassembler/MCDisassembler.h" 43 #include "llvm/MC/MCDisassembler/MCRelocationInfo.h" 44 #include "llvm/MC/MCInst.h" 45 #include "llvm/MC/MCInstPrinter.h" 46 #include "llvm/MC/MCInstrAnalysis.h" 47 #include "llvm/MC/MCInstrInfo.h" 48 #include "llvm/MC/MCObjectFileInfo.h" 49 #include "llvm/MC/MCRegisterInfo.h" 50 #include "llvm/MC/MCSubtargetInfo.h" 51 #include "llvm/MC/MCTargetOptions.h" 52 #include "llvm/MC/TargetRegistry.h" 53 #include "llvm/Object/Archive.h" 54 #include "llvm/Object/COFF.h" 55 #include "llvm/Object/COFFImportFile.h" 56 #include "llvm/Object/ELFObjectFile.h" 57 #include "llvm/Object/ELFTypes.h" 58 #include "llvm/Object/FaultMapParser.h" 59 #include "llvm/Object/MachO.h" 60 #include "llvm/Object/MachOUniversal.h" 61 #include "llvm/Object/ObjectFile.h" 62 #include "llvm/Object/OffloadBinary.h" 63 #include "llvm/Object/Wasm.h" 64 #include "llvm/Option/Arg.h" 65 #include "llvm/Option/ArgList.h" 66 #include "llvm/Option/Option.h" 67 #include "llvm/Support/Casting.h" 68 #include "llvm/Support/Debug.h" 69 #include "llvm/Support/Errc.h" 70 #include "llvm/Support/FileSystem.h" 71 #include "llvm/Support/Format.h" 72 #include "llvm/Support/FormatVariadic.h" 73 #include "llvm/Support/GraphWriter.h" 74 #include "llvm/Support/Host.h" 75 #include "llvm/Support/InitLLVM.h" 76 #include "llvm/Support/MemoryBuffer.h" 77 #include "llvm/Support/SourceMgr.h" 78 #include "llvm/Support/StringSaver.h" 79 #include "llvm/Support/TargetSelect.h" 80 #include "llvm/Support/WithColor.h" 81 #include "llvm/Support/raw_ostream.h" 82 #include <algorithm> 83 #include <cctype> 84 #include <cstring> 85 #include <system_error> 86 #include <unordered_map> 87 #include <utility> 88 89 using namespace llvm; 90 using namespace llvm::object; 91 using namespace llvm::objdump; 92 using namespace llvm::opt; 93 94 namespace { 95 96 class CommonOptTable : public opt::OptTable { 97 public: 98 CommonOptTable(ArrayRef<Info> OptionInfos, const char *Usage, 99 const char *Description) 100 : OptTable(OptionInfos), Usage(Usage), Description(Description) { 101 setGroupedShortOptions(true); 102 } 103 104 void printHelp(StringRef Argv0, bool ShowHidden = false) const { 105 Argv0 = sys::path::filename(Argv0); 106 opt::OptTable::printHelp(outs(), (Argv0 + Usage).str().c_str(), Description, 107 ShowHidden, ShowHidden); 108 // TODO Replace this with OptTable API once it adds extrahelp support. 109 outs() << "\nPass @FILE as argument to read options from FILE.\n"; 110 } 111 112 private: 113 const char *Usage; 114 const char *Description; 115 }; 116 117 // ObjdumpOptID is in ObjdumpOptID.h 118 119 #define PREFIX(NAME, VALUE) const char *const OBJDUMP_##NAME[] = VALUE; 120 #include "ObjdumpOpts.inc" 121 #undef PREFIX 122 123 static constexpr opt::OptTable::Info ObjdumpInfoTable[] = { 124 #define OBJDUMP_nullptr nullptr 125 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 126 HELPTEXT, METAVAR, VALUES) \ 127 {OBJDUMP_##PREFIX, NAME, HELPTEXT, \ 128 METAVAR, OBJDUMP_##ID, opt::Option::KIND##Class, \ 129 PARAM, FLAGS, OBJDUMP_##GROUP, \ 130 OBJDUMP_##ALIAS, ALIASARGS, VALUES}, 131 #include "ObjdumpOpts.inc" 132 #undef OPTION 133 #undef OBJDUMP_nullptr 134 }; 135 136 class ObjdumpOptTable : public CommonOptTable { 137 public: 138 ObjdumpOptTable() 139 : CommonOptTable(ObjdumpInfoTable, " [options] <input object files>", 140 "llvm object file dumper") {} 141 }; 142 143 enum OtoolOptID { 144 OTOOL_INVALID = 0, // This is not an option ID. 145 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 146 HELPTEXT, METAVAR, VALUES) \ 147 OTOOL_##ID, 148 #include "OtoolOpts.inc" 149 #undef OPTION 150 }; 151 152 #define PREFIX(NAME, VALUE) const char *const OTOOL_##NAME[] = VALUE; 153 #include "OtoolOpts.inc" 154 #undef PREFIX 155 156 static constexpr opt::OptTable::Info OtoolInfoTable[] = { 157 #define OTOOL_nullptr nullptr 158 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 159 HELPTEXT, METAVAR, VALUES) \ 160 {OTOOL_##PREFIX, NAME, HELPTEXT, \ 161 METAVAR, OTOOL_##ID, opt::Option::KIND##Class, \ 162 PARAM, FLAGS, OTOOL_##GROUP, \ 163 OTOOL_##ALIAS, ALIASARGS, VALUES}, 164 #include "OtoolOpts.inc" 165 #undef OPTION 166 #undef OTOOL_nullptr 167 }; 168 169 class OtoolOptTable : public CommonOptTable { 170 public: 171 OtoolOptTable() 172 : CommonOptTable(OtoolInfoTable, " [option...] [file...]", 173 "Mach-O object file displaying tool") {} 174 }; 175 176 } // namespace 177 178 #define DEBUG_TYPE "objdump" 179 180 static uint64_t AdjustVMA; 181 static bool AllHeaders; 182 static std::string ArchName; 183 bool objdump::ArchiveHeaders; 184 bool objdump::Demangle; 185 bool objdump::Disassemble; 186 bool objdump::DisassembleAll; 187 bool objdump::SymbolDescription; 188 static std::vector<std::string> DisassembleSymbols; 189 static bool DisassembleZeroes; 190 static std::vector<std::string> DisassemblerOptions; 191 DIDumpType objdump::DwarfDumpType; 192 static bool DynamicRelocations; 193 static bool FaultMapSection; 194 static bool FileHeaders; 195 bool objdump::SectionContents; 196 static std::vector<std::string> InputFilenames; 197 bool objdump::PrintLines; 198 static bool MachOOpt; 199 std::string objdump::MCPU; 200 std::vector<std::string> objdump::MAttrs; 201 bool objdump::ShowRawInsn; 202 bool objdump::LeadingAddr; 203 static bool Offloading; 204 static bool RawClangAST; 205 bool objdump::Relocations; 206 bool objdump::PrintImmHex; 207 bool objdump::PrivateHeaders; 208 std::vector<std::string> objdump::FilterSections; 209 bool objdump::SectionHeaders; 210 static bool ShowAllSymbols; 211 static bool ShowLMA; 212 bool objdump::PrintSource; 213 214 static uint64_t StartAddress; 215 static bool HasStartAddressFlag; 216 static uint64_t StopAddress = UINT64_MAX; 217 static bool HasStopAddressFlag; 218 219 bool objdump::SymbolTable; 220 static bool SymbolizeOperands; 221 static bool DynamicSymbolTable; 222 std::string objdump::TripleName; 223 bool objdump::UnwindInfo; 224 static bool Wide; 225 std::string objdump::Prefix; 226 uint32_t objdump::PrefixStrip; 227 228 DebugVarsFormat objdump::DbgVariables = DVDisabled; 229 230 int objdump::DbgIndent = 52; 231 232 static StringSet<> DisasmSymbolSet; 233 StringSet<> objdump::FoundSectionSet; 234 static StringRef ToolName; 235 236 namespace { 237 struct FilterResult { 238 // True if the section should not be skipped. 239 bool Keep; 240 241 // True if the index counter should be incremented, even if the section should 242 // be skipped. For example, sections may be skipped if they are not included 243 // in the --section flag, but we still want those to count toward the section 244 // count. 245 bool IncrementIndex; 246 }; 247 } // namespace 248 249 static FilterResult checkSectionFilter(object::SectionRef S) { 250 if (FilterSections.empty()) 251 return {/*Keep=*/true, /*IncrementIndex=*/true}; 252 253 Expected<StringRef> SecNameOrErr = S.getName(); 254 if (!SecNameOrErr) { 255 consumeError(SecNameOrErr.takeError()); 256 return {/*Keep=*/false, /*IncrementIndex=*/false}; 257 } 258 StringRef SecName = *SecNameOrErr; 259 260 // StringSet does not allow empty key so avoid adding sections with 261 // no name (such as the section with index 0) here. 262 if (!SecName.empty()) 263 FoundSectionSet.insert(SecName); 264 265 // Only show the section if it's in the FilterSections list, but always 266 // increment so the indexing is stable. 267 return {/*Keep=*/is_contained(FilterSections, SecName), 268 /*IncrementIndex=*/true}; 269 } 270 271 SectionFilter objdump::ToolSectionFilter(object::ObjectFile const &O, 272 uint64_t *Idx) { 273 // Start at UINT64_MAX so that the first index returned after an increment is 274 // zero (after the unsigned wrap). 275 if (Idx) 276 *Idx = UINT64_MAX; 277 return SectionFilter( 278 [Idx](object::SectionRef S) { 279 FilterResult Result = checkSectionFilter(S); 280 if (Idx != nullptr && Result.IncrementIndex) 281 *Idx += 1; 282 return Result.Keep; 283 }, 284 O); 285 } 286 287 std::string objdump::getFileNameForError(const object::Archive::Child &C, 288 unsigned Index) { 289 Expected<StringRef> NameOrErr = C.getName(); 290 if (NameOrErr) 291 return std::string(NameOrErr.get()); 292 // If we have an error getting the name then we print the index of the archive 293 // member. Since we are already in an error state, we just ignore this error. 294 consumeError(NameOrErr.takeError()); 295 return "<file index: " + std::to_string(Index) + ">"; 296 } 297 298 void objdump::reportWarning(const Twine &Message, StringRef File) { 299 // Output order between errs() and outs() matters especially for archive 300 // files where the output is per member object. 301 outs().flush(); 302 WithColor::warning(errs(), ToolName) 303 << "'" << File << "': " << Message << "\n"; 304 } 305 306 [[noreturn]] void objdump::reportError(StringRef File, const Twine &Message) { 307 outs().flush(); 308 WithColor::error(errs(), ToolName) << "'" << File << "': " << Message << "\n"; 309 exit(1); 310 } 311 312 [[noreturn]] void objdump::reportError(Error E, StringRef FileName, 313 StringRef ArchiveName, 314 StringRef ArchitectureName) { 315 assert(E); 316 outs().flush(); 317 WithColor::error(errs(), ToolName); 318 if (ArchiveName != "") 319 errs() << ArchiveName << "(" << FileName << ")"; 320 else 321 errs() << "'" << FileName << "'"; 322 if (!ArchitectureName.empty()) 323 errs() << " (for architecture " << ArchitectureName << ")"; 324 errs() << ": "; 325 logAllUnhandledErrors(std::move(E), errs()); 326 exit(1); 327 } 328 329 static void reportCmdLineWarning(const Twine &Message) { 330 WithColor::warning(errs(), ToolName) << Message << "\n"; 331 } 332 333 [[noreturn]] static void reportCmdLineError(const Twine &Message) { 334 WithColor::error(errs(), ToolName) << Message << "\n"; 335 exit(1); 336 } 337 338 static void warnOnNoMatchForSections() { 339 SetVector<StringRef> MissingSections; 340 for (StringRef S : FilterSections) { 341 if (FoundSectionSet.count(S)) 342 return; 343 // User may specify a unnamed section. Don't warn for it. 344 if (!S.empty()) 345 MissingSections.insert(S); 346 } 347 348 // Warn only if no section in FilterSections is matched. 349 for (StringRef S : MissingSections) 350 reportCmdLineWarning("section '" + S + 351 "' mentioned in a -j/--section option, but not " 352 "found in any input file"); 353 } 354 355 static const Target *getTarget(const ObjectFile *Obj) { 356 // Figure out the target triple. 357 Triple TheTriple("unknown-unknown-unknown"); 358 if (TripleName.empty()) { 359 TheTriple = Obj->makeTriple(); 360 } else { 361 TheTriple.setTriple(Triple::normalize(TripleName)); 362 auto Arch = Obj->getArch(); 363 if (Arch == Triple::arm || Arch == Triple::armeb) 364 Obj->setARMSubArch(TheTriple); 365 } 366 367 // Get the target specific parser. 368 std::string Error; 369 const Target *TheTarget = TargetRegistry::lookupTarget(ArchName, TheTriple, 370 Error); 371 if (!TheTarget) 372 reportError(Obj->getFileName(), "can't find target: " + Error); 373 374 // Update the triple name and return the found target. 375 TripleName = TheTriple.getTriple(); 376 return TheTarget; 377 } 378 379 bool objdump::isRelocAddressLess(RelocationRef A, RelocationRef B) { 380 return A.getOffset() < B.getOffset(); 381 } 382 383 static Error getRelocationValueString(const RelocationRef &Rel, 384 SmallVectorImpl<char> &Result) { 385 const ObjectFile *Obj = Rel.getObject(); 386 if (auto *ELF = dyn_cast<ELFObjectFileBase>(Obj)) 387 return getELFRelocationValueString(ELF, Rel, Result); 388 if (auto *COFF = dyn_cast<COFFObjectFile>(Obj)) 389 return getCOFFRelocationValueString(COFF, Rel, Result); 390 if (auto *Wasm = dyn_cast<WasmObjectFile>(Obj)) 391 return getWasmRelocationValueString(Wasm, Rel, Result); 392 if (auto *MachO = dyn_cast<MachOObjectFile>(Obj)) 393 return getMachORelocationValueString(MachO, Rel, Result); 394 if (auto *XCOFF = dyn_cast<XCOFFObjectFile>(Obj)) 395 return getXCOFFRelocationValueString(*XCOFF, Rel, Result); 396 llvm_unreachable("unknown object file format"); 397 } 398 399 /// Indicates whether this relocation should hidden when listing 400 /// relocations, usually because it is the trailing part of a multipart 401 /// relocation that will be printed as part of the leading relocation. 402 static bool getHidden(RelocationRef RelRef) { 403 auto *MachO = dyn_cast<MachOObjectFile>(RelRef.getObject()); 404 if (!MachO) 405 return false; 406 407 unsigned Arch = MachO->getArch(); 408 DataRefImpl Rel = RelRef.getRawDataRefImpl(); 409 uint64_t Type = MachO->getRelocationType(Rel); 410 411 // On arches that use the generic relocations, GENERIC_RELOC_PAIR 412 // is always hidden. 413 if (Arch == Triple::x86 || Arch == Triple::arm || Arch == Triple::ppc) 414 return Type == MachO::GENERIC_RELOC_PAIR; 415 416 if (Arch == Triple::x86_64) { 417 // On x86_64, X86_64_RELOC_UNSIGNED is hidden only when it follows 418 // an X86_64_RELOC_SUBTRACTOR. 419 if (Type == MachO::X86_64_RELOC_UNSIGNED && Rel.d.a > 0) { 420 DataRefImpl RelPrev = Rel; 421 RelPrev.d.a--; 422 uint64_t PrevType = MachO->getRelocationType(RelPrev); 423 if (PrevType == MachO::X86_64_RELOC_SUBTRACTOR) 424 return true; 425 } 426 } 427 428 return false; 429 } 430 431 namespace { 432 433 /// Get the column at which we want to start printing the instruction 434 /// disassembly, taking into account anything which appears to the left of it. 435 unsigned getInstStartColumn(const MCSubtargetInfo &STI) { 436 return !ShowRawInsn ? 16 : STI.getTargetTriple().isX86() ? 40 : 24; 437 } 438 439 static bool isAArch64Elf(const ObjectFile &Obj) { 440 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 441 return Elf && Elf->getEMachine() == ELF::EM_AARCH64; 442 } 443 444 static bool isArmElf(const ObjectFile &Obj) { 445 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 446 return Elf && Elf->getEMachine() == ELF::EM_ARM; 447 } 448 449 static bool isCSKYElf(const ObjectFile &Obj) { 450 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 451 return Elf && Elf->getEMachine() == ELF::EM_CSKY; 452 } 453 454 static bool hasMappingSymbols(const ObjectFile &Obj) { 455 return isArmElf(Obj) || isAArch64Elf(Obj) || isCSKYElf(Obj) ; 456 } 457 458 static void printRelocation(formatted_raw_ostream &OS, StringRef FileName, 459 const RelocationRef &Rel, uint64_t Address, 460 bool Is64Bits) { 461 StringRef Fmt = Is64Bits ? "\t\t%016" PRIx64 ": " : "\t\t\t%08" PRIx64 ": "; 462 SmallString<16> Name; 463 SmallString<32> Val; 464 Rel.getTypeName(Name); 465 if (Error E = getRelocationValueString(Rel, Val)) 466 reportError(std::move(E), FileName); 467 OS << format(Fmt.data(), Address) << Name << "\t" << Val; 468 } 469 470 static void AlignToInstStartColumn(size_t Start, const MCSubtargetInfo &STI, 471 raw_ostream &OS) { 472 // The output of printInst starts with a tab. Print some spaces so that 473 // the tab has 1 column and advances to the target tab stop. 474 unsigned TabStop = getInstStartColumn(STI); 475 unsigned Column = OS.tell() - Start; 476 OS.indent(Column < TabStop - 1 ? TabStop - 1 - Column : 7 - Column % 8); 477 } 478 479 class PrettyPrinter { 480 public: 481 virtual ~PrettyPrinter() = default; 482 virtual void 483 printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 484 object::SectionedAddress Address, formatted_raw_ostream &OS, 485 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 486 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 487 LiveVariablePrinter &LVP) { 488 if (SP && (PrintSource || PrintLines)) 489 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 490 LVP.printBetweenInsts(OS, false); 491 492 size_t Start = OS.tell(); 493 if (LeadingAddr) 494 OS << format("%8" PRIx64 ":", Address.Address); 495 if (ShowRawInsn) { 496 OS << ' '; 497 dumpBytes(Bytes, OS); 498 } 499 500 AlignToInstStartColumn(Start, STI, OS); 501 502 if (MI) { 503 // See MCInstPrinter::printInst. On targets where a PC relative immediate 504 // is relative to the next instruction and the length of a MCInst is 505 // difficult to measure (x86), this is the address of the next 506 // instruction. 507 uint64_t Addr = 508 Address.Address + (STI.getTargetTriple().isX86() ? Bytes.size() : 0); 509 IP.printInst(MI, Addr, "", STI, OS); 510 } else 511 OS << "\t<unknown>"; 512 } 513 }; 514 PrettyPrinter PrettyPrinterInst; 515 516 class HexagonPrettyPrinter : public PrettyPrinter { 517 public: 518 void printLead(ArrayRef<uint8_t> Bytes, uint64_t Address, 519 formatted_raw_ostream &OS) { 520 uint32_t opcode = 521 (Bytes[3] << 24) | (Bytes[2] << 16) | (Bytes[1] << 8) | Bytes[0]; 522 if (LeadingAddr) 523 OS << format("%8" PRIx64 ":", Address); 524 if (ShowRawInsn) { 525 OS << "\t"; 526 dumpBytes(Bytes.slice(0, 4), OS); 527 OS << format("\t%08" PRIx32, opcode); 528 } 529 } 530 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 531 object::SectionedAddress Address, formatted_raw_ostream &OS, 532 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 533 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 534 LiveVariablePrinter &LVP) override { 535 if (SP && (PrintSource || PrintLines)) 536 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 537 if (!MI) { 538 printLead(Bytes, Address.Address, OS); 539 OS << " <unknown>"; 540 return; 541 } 542 std::string Buffer; 543 { 544 raw_string_ostream TempStream(Buffer); 545 IP.printInst(MI, Address.Address, "", STI, TempStream); 546 } 547 StringRef Contents(Buffer); 548 // Split off bundle attributes 549 auto PacketBundle = Contents.rsplit('\n'); 550 // Split off first instruction from the rest 551 auto HeadTail = PacketBundle.first.split('\n'); 552 auto Preamble = " { "; 553 auto Separator = ""; 554 555 // Hexagon's packets require relocations to be inline rather than 556 // clustered at the end of the packet. 557 std::vector<RelocationRef>::const_iterator RelCur = Rels->begin(); 558 std::vector<RelocationRef>::const_iterator RelEnd = Rels->end(); 559 auto PrintReloc = [&]() -> void { 560 while ((RelCur != RelEnd) && (RelCur->getOffset() <= Address.Address)) { 561 if (RelCur->getOffset() == Address.Address) { 562 printRelocation(OS, ObjectFilename, *RelCur, Address.Address, false); 563 return; 564 } 565 ++RelCur; 566 } 567 }; 568 569 while (!HeadTail.first.empty()) { 570 OS << Separator; 571 Separator = "\n"; 572 if (SP && (PrintSource || PrintLines)) 573 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 574 printLead(Bytes, Address.Address, OS); 575 OS << Preamble; 576 Preamble = " "; 577 StringRef Inst; 578 auto Duplex = HeadTail.first.split('\v'); 579 if (!Duplex.second.empty()) { 580 OS << Duplex.first; 581 OS << "; "; 582 Inst = Duplex.second; 583 } 584 else 585 Inst = HeadTail.first; 586 OS << Inst; 587 HeadTail = HeadTail.second.split('\n'); 588 if (HeadTail.first.empty()) 589 OS << " } " << PacketBundle.second; 590 PrintReloc(); 591 Bytes = Bytes.slice(4); 592 Address.Address += 4; 593 } 594 } 595 }; 596 HexagonPrettyPrinter HexagonPrettyPrinterInst; 597 598 class AMDGCNPrettyPrinter : public PrettyPrinter { 599 public: 600 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 601 object::SectionedAddress Address, formatted_raw_ostream &OS, 602 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 603 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 604 LiveVariablePrinter &LVP) override { 605 if (SP && (PrintSource || PrintLines)) 606 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 607 608 if (MI) { 609 SmallString<40> InstStr; 610 raw_svector_ostream IS(InstStr); 611 612 IP.printInst(MI, Address.Address, "", STI, IS); 613 614 OS << left_justify(IS.str(), 60); 615 } else { 616 // an unrecognized encoding - this is probably data so represent it 617 // using the .long directive, or .byte directive if fewer than 4 bytes 618 // remaining 619 if (Bytes.size() >= 4) { 620 OS << format("\t.long 0x%08" PRIx32 " ", 621 support::endian::read32<support::little>(Bytes.data())); 622 OS.indent(42); 623 } else { 624 OS << format("\t.byte 0x%02" PRIx8, Bytes[0]); 625 for (unsigned int i = 1; i < Bytes.size(); i++) 626 OS << format(", 0x%02" PRIx8, Bytes[i]); 627 OS.indent(55 - (6 * Bytes.size())); 628 } 629 } 630 631 OS << format("// %012" PRIX64 ":", Address.Address); 632 if (Bytes.size() >= 4) { 633 // D should be casted to uint32_t here as it is passed by format to 634 // snprintf as vararg. 635 for (uint32_t D : makeArrayRef( 636 reinterpret_cast<const support::little32_t *>(Bytes.data()), 637 Bytes.size() / 4)) 638 OS << format(" %08" PRIX32, D); 639 } else { 640 for (unsigned char B : Bytes) 641 OS << format(" %02" PRIX8, B); 642 } 643 644 if (!Annot.empty()) 645 OS << " // " << Annot; 646 } 647 }; 648 AMDGCNPrettyPrinter AMDGCNPrettyPrinterInst; 649 650 class BPFPrettyPrinter : public PrettyPrinter { 651 public: 652 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 653 object::SectionedAddress Address, formatted_raw_ostream &OS, 654 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 655 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 656 LiveVariablePrinter &LVP) override { 657 if (SP && (PrintSource || PrintLines)) 658 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 659 if (LeadingAddr) 660 OS << format("%8" PRId64 ":", Address.Address / 8); 661 if (ShowRawInsn) { 662 OS << "\t"; 663 dumpBytes(Bytes, OS); 664 } 665 if (MI) 666 IP.printInst(MI, Address.Address, "", STI, OS); 667 else 668 OS << "\t<unknown>"; 669 } 670 }; 671 BPFPrettyPrinter BPFPrettyPrinterInst; 672 673 class ARMPrettyPrinter : public PrettyPrinter { 674 public: 675 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 676 object::SectionedAddress Address, formatted_raw_ostream &OS, 677 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 678 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 679 LiveVariablePrinter &LVP) override { 680 if (SP && (PrintSource || PrintLines)) 681 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 682 LVP.printBetweenInsts(OS, false); 683 684 size_t Start = OS.tell(); 685 if (LeadingAddr) 686 OS << format("%8" PRIx64 ":", Address.Address); 687 if (ShowRawInsn) { 688 size_t Pos = 0, End = Bytes.size(); 689 if (STI.checkFeatures("+thumb-mode")) { 690 for (; Pos + 2 <= End; Pos += 2) 691 OS << ' ' 692 << format_hex_no_prefix( 693 llvm::support::endian::read<uint16_t>( 694 Bytes.data() + Pos, InstructionEndianness), 695 4); 696 } else { 697 for (; Pos + 4 <= End; Pos += 4) 698 OS << ' ' 699 << format_hex_no_prefix( 700 llvm::support::endian::read<uint32_t>( 701 Bytes.data() + Pos, InstructionEndianness), 702 8); 703 } 704 if (Pos < End) { 705 OS << ' '; 706 dumpBytes(Bytes.slice(Pos), OS); 707 } 708 } 709 710 AlignToInstStartColumn(Start, STI, OS); 711 712 if (MI) { 713 IP.printInst(MI, Address.Address, "", STI, OS); 714 } else 715 OS << "\t<unknown>"; 716 } 717 718 void setInstructionEndianness(llvm::support::endianness Endianness) { 719 InstructionEndianness = Endianness; 720 } 721 722 private: 723 llvm::support::endianness InstructionEndianness = llvm::support::little; 724 }; 725 ARMPrettyPrinter ARMPrettyPrinterInst; 726 727 class AArch64PrettyPrinter : public PrettyPrinter { 728 public: 729 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 730 object::SectionedAddress Address, formatted_raw_ostream &OS, 731 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 732 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 733 LiveVariablePrinter &LVP) override { 734 if (SP && (PrintSource || PrintLines)) 735 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 736 LVP.printBetweenInsts(OS, false); 737 738 size_t Start = OS.tell(); 739 if (LeadingAddr) 740 OS << format("%8" PRIx64 ":", Address.Address); 741 if (ShowRawInsn) { 742 size_t Pos = 0, End = Bytes.size(); 743 for (; Pos + 4 <= End; Pos += 4) 744 OS << ' ' 745 << format_hex_no_prefix( 746 llvm::support::endian::read<uint32_t>(Bytes.data() + Pos, 747 llvm::support::little), 748 8); 749 if (Pos < End) { 750 OS << ' '; 751 dumpBytes(Bytes.slice(Pos), OS); 752 } 753 } 754 755 AlignToInstStartColumn(Start, STI, OS); 756 757 if (MI) { 758 IP.printInst(MI, Address.Address, "", STI, OS); 759 } else 760 OS << "\t<unknown>"; 761 } 762 }; 763 AArch64PrettyPrinter AArch64PrettyPrinterInst; 764 765 PrettyPrinter &selectPrettyPrinter(Triple const &Triple) { 766 switch(Triple.getArch()) { 767 default: 768 return PrettyPrinterInst; 769 case Triple::hexagon: 770 return HexagonPrettyPrinterInst; 771 case Triple::amdgcn: 772 return AMDGCNPrettyPrinterInst; 773 case Triple::bpfel: 774 case Triple::bpfeb: 775 return BPFPrettyPrinterInst; 776 case Triple::arm: 777 case Triple::armeb: 778 case Triple::thumb: 779 case Triple::thumbeb: 780 return ARMPrettyPrinterInst; 781 case Triple::aarch64: 782 case Triple::aarch64_be: 783 case Triple::aarch64_32: 784 return AArch64PrettyPrinterInst; 785 } 786 } 787 } 788 789 static uint8_t getElfSymbolType(const ObjectFile &Obj, const SymbolRef &Sym) { 790 assert(Obj.isELF()); 791 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 792 return unwrapOrError(Elf32LEObj->getSymbol(Sym.getRawDataRefImpl()), 793 Obj.getFileName()) 794 ->getType(); 795 if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 796 return unwrapOrError(Elf64LEObj->getSymbol(Sym.getRawDataRefImpl()), 797 Obj.getFileName()) 798 ->getType(); 799 if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 800 return unwrapOrError(Elf32BEObj->getSymbol(Sym.getRawDataRefImpl()), 801 Obj.getFileName()) 802 ->getType(); 803 if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 804 return unwrapOrError(Elf64BEObj->getSymbol(Sym.getRawDataRefImpl()), 805 Obj.getFileName()) 806 ->getType(); 807 llvm_unreachable("Unsupported binary format"); 808 } 809 810 template <class ELFT> 811 static void 812 addDynamicElfSymbols(const ELFObjectFile<ELFT> &Obj, 813 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 814 for (auto Symbol : Obj.getDynamicSymbolIterators()) { 815 uint8_t SymbolType = Symbol.getELFType(); 816 if (SymbolType == ELF::STT_SECTION) 817 continue; 818 819 uint64_t Address = unwrapOrError(Symbol.getAddress(), Obj.getFileName()); 820 // ELFSymbolRef::getAddress() returns size instead of value for common 821 // symbols which is not desirable for disassembly output. Overriding. 822 if (SymbolType == ELF::STT_COMMON) 823 Address = unwrapOrError(Obj.getSymbol(Symbol.getRawDataRefImpl()), 824 Obj.getFileName()) 825 ->st_value; 826 827 StringRef Name = unwrapOrError(Symbol.getName(), Obj.getFileName()); 828 if (Name.empty()) 829 continue; 830 831 section_iterator SecI = 832 unwrapOrError(Symbol.getSection(), Obj.getFileName()); 833 if (SecI == Obj.section_end()) 834 continue; 835 836 AllSymbols[*SecI].emplace_back(Address, Name, SymbolType); 837 } 838 } 839 840 static void 841 addDynamicElfSymbols(const ELFObjectFileBase &Obj, 842 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 843 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 844 addDynamicElfSymbols(*Elf32LEObj, AllSymbols); 845 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 846 addDynamicElfSymbols(*Elf64LEObj, AllSymbols); 847 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 848 addDynamicElfSymbols(*Elf32BEObj, AllSymbols); 849 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 850 addDynamicElfSymbols(*Elf64BEObj, AllSymbols); 851 else 852 llvm_unreachable("Unsupported binary format"); 853 } 854 855 static Optional<SectionRef> getWasmCodeSection(const WasmObjectFile &Obj) { 856 for (auto SecI : Obj.sections()) { 857 const WasmSection &Section = Obj.getWasmSection(SecI); 858 if (Section.Type == wasm::WASM_SEC_CODE) 859 return SecI; 860 } 861 return None; 862 } 863 864 static void 865 addMissingWasmCodeSymbols(const WasmObjectFile &Obj, 866 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 867 Optional<SectionRef> Section = getWasmCodeSection(Obj); 868 if (!Section) 869 return; 870 SectionSymbolsTy &Symbols = AllSymbols[*Section]; 871 872 std::set<uint64_t> SymbolAddresses; 873 for (const auto &Sym : Symbols) 874 SymbolAddresses.insert(Sym.Addr); 875 876 for (const wasm::WasmFunction &Function : Obj.functions()) { 877 uint64_t Address = Function.CodeSectionOffset; 878 // Only add fallback symbols for functions not already present in the symbol 879 // table. 880 if (SymbolAddresses.count(Address)) 881 continue; 882 // This function has no symbol, so it should have no SymbolName. 883 assert(Function.SymbolName.empty()); 884 // We use DebugName for the name, though it may be empty if there is no 885 // "name" custom section, or that section is missing a name for this 886 // function. 887 StringRef Name = Function.DebugName; 888 Symbols.emplace_back(Address, Name, ELF::STT_NOTYPE); 889 } 890 } 891 892 static void addPltEntries(const ObjectFile &Obj, 893 std::map<SectionRef, SectionSymbolsTy> &AllSymbols, 894 StringSaver &Saver) { 895 Optional<SectionRef> Plt; 896 for (const SectionRef &Section : Obj.sections()) { 897 Expected<StringRef> SecNameOrErr = Section.getName(); 898 if (!SecNameOrErr) { 899 consumeError(SecNameOrErr.takeError()); 900 continue; 901 } 902 if (*SecNameOrErr == ".plt") 903 Plt = Section; 904 } 905 if (!Plt) 906 return; 907 if (auto *ElfObj = dyn_cast<ELFObjectFileBase>(&Obj)) { 908 for (auto PltEntry : ElfObj->getPltAddresses()) { 909 if (PltEntry.first) { 910 SymbolRef Symbol(*PltEntry.first, ElfObj); 911 uint8_t SymbolType = getElfSymbolType(Obj, Symbol); 912 if (Expected<StringRef> NameOrErr = Symbol.getName()) { 913 if (!NameOrErr->empty()) 914 AllSymbols[*Plt].emplace_back( 915 PltEntry.second, Saver.save((*NameOrErr + "@plt").str()), 916 SymbolType); 917 continue; 918 } else { 919 // The warning has been reported in disassembleObject(). 920 consumeError(NameOrErr.takeError()); 921 } 922 } 923 reportWarning("PLT entry at 0x" + Twine::utohexstr(PltEntry.second) + 924 " references an invalid symbol", 925 Obj.getFileName()); 926 } 927 } 928 } 929 930 // Normally the disassembly output will skip blocks of zeroes. This function 931 // returns the number of zero bytes that can be skipped when dumping the 932 // disassembly of the instructions in Buf. 933 static size_t countSkippableZeroBytes(ArrayRef<uint8_t> Buf) { 934 // Find the number of leading zeroes. 935 size_t N = 0; 936 while (N < Buf.size() && !Buf[N]) 937 ++N; 938 939 // We may want to skip blocks of zero bytes, but unless we see 940 // at least 8 of them in a row. 941 if (N < 8) 942 return 0; 943 944 // We skip zeroes in multiples of 4 because do not want to truncate an 945 // instruction if it starts with a zero byte. 946 return N & ~0x3; 947 } 948 949 // Returns a map from sections to their relocations. 950 static std::map<SectionRef, std::vector<RelocationRef>> 951 getRelocsMap(object::ObjectFile const &Obj) { 952 std::map<SectionRef, std::vector<RelocationRef>> Ret; 953 uint64_t I = (uint64_t)-1; 954 for (SectionRef Sec : Obj.sections()) { 955 ++I; 956 Expected<section_iterator> RelocatedOrErr = Sec.getRelocatedSection(); 957 if (!RelocatedOrErr) 958 reportError(Obj.getFileName(), 959 "section (" + Twine(I) + 960 "): failed to get a relocated section: " + 961 toString(RelocatedOrErr.takeError())); 962 963 section_iterator Relocated = *RelocatedOrErr; 964 if (Relocated == Obj.section_end() || !checkSectionFilter(*Relocated).Keep) 965 continue; 966 std::vector<RelocationRef> &V = Ret[*Relocated]; 967 append_range(V, Sec.relocations()); 968 // Sort relocations by address. 969 llvm::stable_sort(V, isRelocAddressLess); 970 } 971 return Ret; 972 } 973 974 // Used for --adjust-vma to check if address should be adjusted by the 975 // specified value for a given section. 976 // For ELF we do not adjust non-allocatable sections like debug ones, 977 // because they are not loadable. 978 // TODO: implement for other file formats. 979 static bool shouldAdjustVA(const SectionRef &Section) { 980 const ObjectFile *Obj = Section.getObject(); 981 if (Obj->isELF()) 982 return ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC; 983 return false; 984 } 985 986 987 typedef std::pair<uint64_t, char> MappingSymbolPair; 988 static char getMappingSymbolKind(ArrayRef<MappingSymbolPair> MappingSymbols, 989 uint64_t Address) { 990 auto It = 991 partition_point(MappingSymbols, [Address](const MappingSymbolPair &Val) { 992 return Val.first <= Address; 993 }); 994 // Return zero for any address before the first mapping symbol; this means 995 // we should use the default disassembly mode, depending on the target. 996 if (It == MappingSymbols.begin()) 997 return '\x00'; 998 return (It - 1)->second; 999 } 1000 1001 static uint64_t dumpARMELFData(uint64_t SectionAddr, uint64_t Index, 1002 uint64_t End, const ObjectFile &Obj, 1003 ArrayRef<uint8_t> Bytes, 1004 ArrayRef<MappingSymbolPair> MappingSymbols, 1005 const MCSubtargetInfo &STI, raw_ostream &OS) { 1006 support::endianness Endian = 1007 Obj.isLittleEndian() ? support::little : support::big; 1008 size_t Start = OS.tell(); 1009 OS << format("%8" PRIx64 ": ", SectionAddr + Index); 1010 if (Index + 4 <= End) { 1011 dumpBytes(Bytes.slice(Index, 4), OS); 1012 AlignToInstStartColumn(Start, STI, OS); 1013 OS << "\t.word\t" 1014 << format_hex(support::endian::read32(Bytes.data() + Index, Endian), 1015 10); 1016 return 4; 1017 } 1018 if (Index + 2 <= End) { 1019 dumpBytes(Bytes.slice(Index, 2), OS); 1020 AlignToInstStartColumn(Start, STI, OS); 1021 OS << "\t.short\t" 1022 << format_hex(support::endian::read16(Bytes.data() + Index, Endian), 6); 1023 return 2; 1024 } 1025 dumpBytes(Bytes.slice(Index, 1), OS); 1026 AlignToInstStartColumn(Start, STI, OS); 1027 OS << "\t.byte\t" << format_hex(Bytes[Index], 4); 1028 return 1; 1029 } 1030 1031 static void dumpELFData(uint64_t SectionAddr, uint64_t Index, uint64_t End, 1032 ArrayRef<uint8_t> Bytes) { 1033 // print out data up to 8 bytes at a time in hex and ascii 1034 uint8_t AsciiData[9] = {'\0'}; 1035 uint8_t Byte; 1036 int NumBytes = 0; 1037 1038 for (; Index < End; ++Index) { 1039 if (NumBytes == 0) 1040 outs() << format("%8" PRIx64 ":", SectionAddr + Index); 1041 Byte = Bytes.slice(Index)[0]; 1042 outs() << format(" %02x", Byte); 1043 AsciiData[NumBytes] = isPrint(Byte) ? Byte : '.'; 1044 1045 uint8_t IndentOffset = 0; 1046 NumBytes++; 1047 if (Index == End - 1 || NumBytes > 8) { 1048 // Indent the space for less than 8 bytes data. 1049 // 2 spaces for byte and one for space between bytes 1050 IndentOffset = 3 * (8 - NumBytes); 1051 for (int Excess = NumBytes; Excess < 8; Excess++) 1052 AsciiData[Excess] = '\0'; 1053 NumBytes = 8; 1054 } 1055 if (NumBytes == 8) { 1056 AsciiData[8] = '\0'; 1057 outs() << std::string(IndentOffset, ' ') << " "; 1058 outs() << reinterpret_cast<char *>(AsciiData); 1059 outs() << '\n'; 1060 NumBytes = 0; 1061 } 1062 } 1063 } 1064 1065 SymbolInfoTy objdump::createSymbolInfo(const ObjectFile &Obj, 1066 const SymbolRef &Symbol) { 1067 const StringRef FileName = Obj.getFileName(); 1068 const uint64_t Addr = unwrapOrError(Symbol.getAddress(), FileName); 1069 const StringRef Name = unwrapOrError(Symbol.getName(), FileName); 1070 1071 if (Obj.isXCOFF() && SymbolDescription) { 1072 const auto &XCOFFObj = cast<XCOFFObjectFile>(Obj); 1073 DataRefImpl SymbolDRI = Symbol.getRawDataRefImpl(); 1074 1075 const uint32_t SymbolIndex = XCOFFObj.getSymbolIndex(SymbolDRI.p); 1076 Optional<XCOFF::StorageMappingClass> Smc = 1077 getXCOFFSymbolCsectSMC(XCOFFObj, Symbol); 1078 return SymbolInfoTy(Addr, Name, Smc, SymbolIndex, 1079 isLabel(XCOFFObj, Symbol)); 1080 } else if (Obj.isXCOFF()) { 1081 const SymbolRef::Type SymType = unwrapOrError(Symbol.getType(), FileName); 1082 return SymbolInfoTy(Addr, Name, SymType, true); 1083 } else 1084 return SymbolInfoTy(Addr, Name, 1085 Obj.isELF() ? getElfSymbolType(Obj, Symbol) 1086 : (uint8_t)ELF::STT_NOTYPE); 1087 } 1088 1089 static SymbolInfoTy createDummySymbolInfo(const ObjectFile &Obj, 1090 const uint64_t Addr, StringRef &Name, 1091 uint8_t Type) { 1092 if (Obj.isXCOFF() && SymbolDescription) 1093 return SymbolInfoTy(Addr, Name, None, None, false); 1094 else 1095 return SymbolInfoTy(Addr, Name, Type); 1096 } 1097 1098 static void 1099 collectBBAddrMapLabels(const std::unordered_map<uint64_t, BBAddrMap> &AddrToBBAddrMap, 1100 uint64_t SectionAddr, uint64_t Start, uint64_t End, 1101 std::unordered_map<uint64_t, std::vector<std::string>> &Labels) { 1102 if (AddrToBBAddrMap.empty()) 1103 return; 1104 Labels.clear(); 1105 uint64_t StartAddress = SectionAddr + Start; 1106 uint64_t EndAddress = SectionAddr + End; 1107 auto Iter = AddrToBBAddrMap.find(StartAddress); 1108 if (Iter == AddrToBBAddrMap.end()) 1109 return; 1110 for (unsigned I = 0, Size = Iter->second.BBEntries.size(); I < Size; ++I) { 1111 uint64_t BBAddress = Iter->second.BBEntries[I].Offset + Iter->second.Addr; 1112 if (BBAddress >= EndAddress) 1113 continue; 1114 Labels[BBAddress].push_back(("BB" + Twine(I)).str()); 1115 } 1116 } 1117 1118 static void collectLocalBranchTargets( 1119 ArrayRef<uint8_t> Bytes, const MCInstrAnalysis *MIA, MCDisassembler *DisAsm, 1120 MCInstPrinter *IP, const MCSubtargetInfo *STI, uint64_t SectionAddr, 1121 uint64_t Start, uint64_t End, std::unordered_map<uint64_t, std::string> &Labels) { 1122 // So far only supports PowerPC and X86. 1123 if (!STI->getTargetTriple().isPPC() && !STI->getTargetTriple().isX86()) 1124 return; 1125 1126 Labels.clear(); 1127 unsigned LabelCount = 0; 1128 Start += SectionAddr; 1129 End += SectionAddr; 1130 uint64_t Index = Start; 1131 while (Index < End) { 1132 // Disassemble a real instruction and record function-local branch labels. 1133 MCInst Inst; 1134 uint64_t Size; 1135 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index - SectionAddr); 1136 bool Disassembled = 1137 DisAsm->getInstruction(Inst, Size, ThisBytes, Index, nulls()); 1138 if (Size == 0) 1139 Size = std::min<uint64_t>(ThisBytes.size(), 1140 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1141 1142 if (Disassembled && MIA) { 1143 uint64_t Target; 1144 bool TargetKnown = MIA->evaluateBranch(Inst, Index, Size, Target); 1145 // On PowerPC, if the address of a branch is the same as the target, it 1146 // means that it's a function call. Do not mark the label for this case. 1147 if (TargetKnown && (Target >= Start && Target < End) && 1148 !Labels.count(Target) && 1149 !(STI->getTargetTriple().isPPC() && Target == Index)) 1150 Labels[Target] = ("L" + Twine(LabelCount++)).str(); 1151 } 1152 Index += Size; 1153 } 1154 } 1155 1156 // Create an MCSymbolizer for the target and add it to the MCDisassembler. 1157 // This is currently only used on AMDGPU, and assumes the format of the 1158 // void * argument passed to AMDGPU's createMCSymbolizer. 1159 static void addSymbolizer( 1160 MCContext &Ctx, const Target *Target, StringRef TripleName, 1161 MCDisassembler *DisAsm, uint64_t SectionAddr, ArrayRef<uint8_t> Bytes, 1162 SectionSymbolsTy &Symbols, 1163 std::vector<std::unique_ptr<std::string>> &SynthesizedLabelNames) { 1164 1165 std::unique_ptr<MCRelocationInfo> RelInfo( 1166 Target->createMCRelocationInfo(TripleName, Ctx)); 1167 if (!RelInfo) 1168 return; 1169 std::unique_ptr<MCSymbolizer> Symbolizer(Target->createMCSymbolizer( 1170 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1171 MCSymbolizer *SymbolizerPtr = &*Symbolizer; 1172 DisAsm->setSymbolizer(std::move(Symbolizer)); 1173 1174 if (!SymbolizeOperands) 1175 return; 1176 1177 // Synthesize labels referenced by branch instructions by 1178 // disassembling, discarding the output, and collecting the referenced 1179 // addresses from the symbolizer. 1180 for (size_t Index = 0; Index != Bytes.size();) { 1181 MCInst Inst; 1182 uint64_t Size; 1183 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index - SectionAddr); 1184 DisAsm->getInstruction(Inst, Size, ThisBytes, Index, nulls()); 1185 if (Size == 0) 1186 Size = std::min<uint64_t>(ThisBytes.size(), 1187 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1188 Index += Size; 1189 } 1190 ArrayRef<uint64_t> LabelAddrsRef = SymbolizerPtr->getReferencedAddresses(); 1191 // Copy and sort to remove duplicates. 1192 std::vector<uint64_t> LabelAddrs; 1193 LabelAddrs.insert(LabelAddrs.end(), LabelAddrsRef.begin(), 1194 LabelAddrsRef.end()); 1195 llvm::sort(LabelAddrs); 1196 LabelAddrs.resize(std::unique(LabelAddrs.begin(), LabelAddrs.end()) - 1197 LabelAddrs.begin()); 1198 // Add the labels. 1199 for (unsigned LabelNum = 0; LabelNum != LabelAddrs.size(); ++LabelNum) { 1200 auto Name = std::make_unique<std::string>(); 1201 *Name = (Twine("L") + Twine(LabelNum)).str(); 1202 SynthesizedLabelNames.push_back(std::move(Name)); 1203 Symbols.push_back(SymbolInfoTy( 1204 LabelAddrs[LabelNum], *SynthesizedLabelNames.back(), ELF::STT_NOTYPE)); 1205 } 1206 llvm::stable_sort(Symbols); 1207 // Recreate the symbolizer with the new symbols list. 1208 RelInfo.reset(Target->createMCRelocationInfo(TripleName, Ctx)); 1209 Symbolizer.reset(Target->createMCSymbolizer( 1210 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1211 DisAsm->setSymbolizer(std::move(Symbolizer)); 1212 } 1213 1214 static StringRef getSegmentName(const MachOObjectFile *MachO, 1215 const SectionRef &Section) { 1216 if (MachO) { 1217 DataRefImpl DR = Section.getRawDataRefImpl(); 1218 StringRef SegmentName = MachO->getSectionFinalSegmentName(DR); 1219 return SegmentName; 1220 } 1221 return ""; 1222 } 1223 1224 static void emitPostInstructionInfo(formatted_raw_ostream &FOS, 1225 const MCAsmInfo &MAI, 1226 const MCSubtargetInfo &STI, 1227 StringRef Comments, 1228 LiveVariablePrinter &LVP) { 1229 do { 1230 if (!Comments.empty()) { 1231 // Emit a line of comments. 1232 StringRef Comment; 1233 std::tie(Comment, Comments) = Comments.split('\n'); 1234 // MAI.getCommentColumn() assumes that instructions are printed at the 1235 // position of 8, while getInstStartColumn() returns the actual position. 1236 unsigned CommentColumn = 1237 MAI.getCommentColumn() - 8 + getInstStartColumn(STI); 1238 FOS.PadToColumn(CommentColumn); 1239 FOS << MAI.getCommentString() << ' ' << Comment; 1240 } 1241 LVP.printAfterInst(FOS); 1242 FOS << '\n'; 1243 } while (!Comments.empty()); 1244 FOS.flush(); 1245 } 1246 1247 static void createFakeELFSections(ObjectFile &Obj) { 1248 assert(Obj.isELF()); 1249 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 1250 Elf32LEObj->createFakeSections(); 1251 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 1252 Elf64LEObj->createFakeSections(); 1253 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 1254 Elf32BEObj->createFakeSections(); 1255 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 1256 Elf64BEObj->createFakeSections(); 1257 else 1258 llvm_unreachable("Unsupported binary format"); 1259 } 1260 1261 static void disassembleObject(const Target *TheTarget, ObjectFile &Obj, 1262 MCContext &Ctx, MCDisassembler *PrimaryDisAsm, 1263 MCDisassembler *SecondaryDisAsm, 1264 const MCInstrAnalysis *MIA, MCInstPrinter *IP, 1265 const MCSubtargetInfo *PrimarySTI, 1266 const MCSubtargetInfo *SecondarySTI, 1267 PrettyPrinter &PIP, SourcePrinter &SP, 1268 bool InlineRelocs) { 1269 const MCSubtargetInfo *STI = PrimarySTI; 1270 MCDisassembler *DisAsm = PrimaryDisAsm; 1271 bool PrimaryIsThumb = false; 1272 if (isArmElf(Obj)) 1273 PrimaryIsThumb = STI->checkFeatures("+thumb-mode"); 1274 1275 std::map<SectionRef, std::vector<RelocationRef>> RelocMap; 1276 if (InlineRelocs) 1277 RelocMap = getRelocsMap(Obj); 1278 bool Is64Bits = Obj.getBytesInAddress() > 4; 1279 1280 // Create a mapping from virtual address to symbol name. This is used to 1281 // pretty print the symbols while disassembling. 1282 std::map<SectionRef, SectionSymbolsTy> AllSymbols; 1283 SectionSymbolsTy AbsoluteSymbols; 1284 const StringRef FileName = Obj.getFileName(); 1285 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&Obj); 1286 for (const SymbolRef &Symbol : Obj.symbols()) { 1287 Expected<StringRef> NameOrErr = Symbol.getName(); 1288 if (!NameOrErr) { 1289 reportWarning(toString(NameOrErr.takeError()), FileName); 1290 continue; 1291 } 1292 if (NameOrErr->empty() && !(Obj.isXCOFF() && SymbolDescription)) 1293 continue; 1294 1295 if (Obj.isELF() && getElfSymbolType(Obj, Symbol) == ELF::STT_SECTION) 1296 continue; 1297 1298 if (MachO) { 1299 // __mh_(execute|dylib|dylinker|bundle|preload|object)_header are special 1300 // symbols that support MachO header introspection. They do not bind to 1301 // code locations and are irrelevant for disassembly. 1302 if (NameOrErr->startswith("__mh_") && NameOrErr->endswith("_header")) 1303 continue; 1304 // Don't ask a Mach-O STAB symbol for its section unless you know that 1305 // STAB symbol's section field refers to a valid section index. Otherwise 1306 // the symbol may error trying to load a section that does not exist. 1307 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 1308 uint8_t NType = (MachO->is64Bit() ? 1309 MachO->getSymbol64TableEntry(SymDRI).n_type: 1310 MachO->getSymbolTableEntry(SymDRI).n_type); 1311 if (NType & MachO::N_STAB) 1312 continue; 1313 } 1314 1315 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1316 if (SecI != Obj.section_end()) 1317 AllSymbols[*SecI].push_back(createSymbolInfo(Obj, Symbol)); 1318 else 1319 AbsoluteSymbols.push_back(createSymbolInfo(Obj, Symbol)); 1320 } 1321 1322 if (AllSymbols.empty() && Obj.isELF()) 1323 addDynamicElfSymbols(cast<ELFObjectFileBase>(Obj), AllSymbols); 1324 1325 if (Obj.isWasm()) 1326 addMissingWasmCodeSymbols(cast<WasmObjectFile>(Obj), AllSymbols); 1327 1328 if (Obj.isELF() && Obj.sections().empty()) 1329 createFakeELFSections(Obj); 1330 1331 BumpPtrAllocator A; 1332 StringSaver Saver(A); 1333 addPltEntries(Obj, AllSymbols, Saver); 1334 1335 // Create a mapping from virtual address to section. An empty section can 1336 // cause more than one section at the same address. Sort such sections to be 1337 // before same-addressed non-empty sections so that symbol lookups prefer the 1338 // non-empty section. 1339 std::vector<std::pair<uint64_t, SectionRef>> SectionAddresses; 1340 for (SectionRef Sec : Obj.sections()) 1341 SectionAddresses.emplace_back(Sec.getAddress(), Sec); 1342 llvm::stable_sort(SectionAddresses, [](const auto &LHS, const auto &RHS) { 1343 if (LHS.first != RHS.first) 1344 return LHS.first < RHS.first; 1345 return LHS.second.getSize() < RHS.second.getSize(); 1346 }); 1347 1348 // Linked executables (.exe and .dll files) typically don't include a real 1349 // symbol table but they might contain an export table. 1350 if (const auto *COFFObj = dyn_cast<COFFObjectFile>(&Obj)) { 1351 for (const auto &ExportEntry : COFFObj->export_directories()) { 1352 StringRef Name; 1353 if (Error E = ExportEntry.getSymbolName(Name)) 1354 reportError(std::move(E), Obj.getFileName()); 1355 if (Name.empty()) 1356 continue; 1357 1358 uint32_t RVA; 1359 if (Error E = ExportEntry.getExportRVA(RVA)) 1360 reportError(std::move(E), Obj.getFileName()); 1361 1362 uint64_t VA = COFFObj->getImageBase() + RVA; 1363 auto Sec = partition_point( 1364 SectionAddresses, [VA](const std::pair<uint64_t, SectionRef> &O) { 1365 return O.first <= VA; 1366 }); 1367 if (Sec != SectionAddresses.begin()) { 1368 --Sec; 1369 AllSymbols[Sec->second].emplace_back(VA, Name, ELF::STT_NOTYPE); 1370 } else 1371 AbsoluteSymbols.emplace_back(VA, Name, ELF::STT_NOTYPE); 1372 } 1373 } 1374 1375 // Sort all the symbols, this allows us to use a simple binary search to find 1376 // Multiple symbols can have the same address. Use a stable sort to stabilize 1377 // the output. 1378 StringSet<> FoundDisasmSymbolSet; 1379 for (std::pair<const SectionRef, SectionSymbolsTy> &SecSyms : AllSymbols) 1380 llvm::stable_sort(SecSyms.second); 1381 llvm::stable_sort(AbsoluteSymbols); 1382 1383 std::unique_ptr<DWARFContext> DICtx; 1384 LiveVariablePrinter LVP(*Ctx.getRegisterInfo(), *STI); 1385 1386 if (DbgVariables != DVDisabled) { 1387 DICtx = DWARFContext::create(Obj); 1388 for (const std::unique_ptr<DWARFUnit> &CU : DICtx->compile_units()) 1389 LVP.addCompileUnit(CU->getUnitDIE(false)); 1390 } 1391 1392 LLVM_DEBUG(LVP.dump()); 1393 1394 std::unordered_map<uint64_t, BBAddrMap> AddrToBBAddrMap; 1395 auto ReadBBAddrMap = [&](Optional<unsigned> SectionIndex = None) { 1396 AddrToBBAddrMap.clear(); 1397 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj)) { 1398 auto BBAddrMapsOrErr = Elf->readBBAddrMap(SectionIndex); 1399 if (!BBAddrMapsOrErr) 1400 reportWarning(toString(BBAddrMapsOrErr.takeError()), 1401 Obj.getFileName()); 1402 for (auto &FunctionBBAddrMap : *BBAddrMapsOrErr) 1403 AddrToBBAddrMap.emplace(FunctionBBAddrMap.Addr, 1404 std::move(FunctionBBAddrMap)); 1405 } 1406 }; 1407 1408 // For non-relocatable objects, Read all LLVM_BB_ADDR_MAP sections into a 1409 // single mapping, since they don't have any conflicts. 1410 if (SymbolizeOperands && !Obj.isRelocatableObject()) 1411 ReadBBAddrMap(); 1412 1413 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 1414 if (FilterSections.empty() && !DisassembleAll && 1415 (!Section.isText() || Section.isVirtual())) 1416 continue; 1417 1418 uint64_t SectionAddr = Section.getAddress(); 1419 uint64_t SectSize = Section.getSize(); 1420 if (!SectSize) 1421 continue; 1422 1423 // For relocatable object files, read the LLVM_BB_ADDR_MAP section 1424 // corresponding to this section, if present. 1425 if (SymbolizeOperands && Obj.isRelocatableObject()) 1426 ReadBBAddrMap(Section.getIndex()); 1427 1428 // Get the list of all the symbols in this section. 1429 SectionSymbolsTy &Symbols = AllSymbols[Section]; 1430 std::vector<MappingSymbolPair> MappingSymbols; 1431 if (hasMappingSymbols(Obj)) { 1432 for (const auto &Symb : Symbols) { 1433 uint64_t Address = Symb.Addr; 1434 StringRef Name = Symb.Name; 1435 if (Name.startswith("$d")) 1436 MappingSymbols.emplace_back(Address - SectionAddr, 'd'); 1437 if (Name.startswith("$x")) 1438 MappingSymbols.emplace_back(Address - SectionAddr, 'x'); 1439 if (Name.startswith("$a")) 1440 MappingSymbols.emplace_back(Address - SectionAddr, 'a'); 1441 if (Name.startswith("$t")) 1442 MappingSymbols.emplace_back(Address - SectionAddr, 't'); 1443 } 1444 } 1445 1446 llvm::sort(MappingSymbols); 1447 1448 ArrayRef<uint8_t> Bytes = arrayRefFromStringRef( 1449 unwrapOrError(Section.getContents(), Obj.getFileName())); 1450 1451 std::vector<std::unique_ptr<std::string>> SynthesizedLabelNames; 1452 if (Obj.isELF() && Obj.getArch() == Triple::amdgcn) { 1453 // AMDGPU disassembler uses symbolizer for printing labels 1454 addSymbolizer(Ctx, TheTarget, TripleName, DisAsm, SectionAddr, Bytes, 1455 Symbols, SynthesizedLabelNames); 1456 } 1457 1458 StringRef SegmentName = getSegmentName(MachO, Section); 1459 StringRef SectionName = unwrapOrError(Section.getName(), Obj.getFileName()); 1460 // If the section has no symbol at the start, just insert a dummy one. 1461 if (Symbols.empty() || Symbols[0].Addr != 0) { 1462 Symbols.insert(Symbols.begin(), 1463 createDummySymbolInfo(Obj, SectionAddr, SectionName, 1464 Section.isText() ? ELF::STT_FUNC 1465 : ELF::STT_OBJECT)); 1466 } 1467 1468 SmallString<40> Comments; 1469 raw_svector_ostream CommentStream(Comments); 1470 1471 uint64_t VMAAdjustment = 0; 1472 if (shouldAdjustVA(Section)) 1473 VMAAdjustment = AdjustVMA; 1474 1475 // In executable and shared objects, r_offset holds a virtual address. 1476 // Subtract SectionAddr from the r_offset field of a relocation to get 1477 // the section offset. 1478 uint64_t RelAdjustment = Obj.isRelocatableObject() ? 0 : SectionAddr; 1479 uint64_t Size; 1480 uint64_t Index; 1481 bool PrintedSection = false; 1482 std::vector<RelocationRef> Rels = RelocMap[Section]; 1483 std::vector<RelocationRef>::const_iterator RelCur = Rels.begin(); 1484 std::vector<RelocationRef>::const_iterator RelEnd = Rels.end(); 1485 1486 // Loop over each chunk of code between two points where at least 1487 // one symbol is defined. 1488 for (size_t SI = 0, SE = Symbols.size(); SI != SE;) { 1489 // Advance SI past all the symbols starting at the same address, 1490 // and make an ArrayRef of them. 1491 unsigned FirstSI = SI; 1492 uint64_t Start = Symbols[SI].Addr; 1493 ArrayRef<SymbolInfoTy> SymbolsHere; 1494 while (SI != SE && Symbols[SI].Addr == Start) 1495 ++SI; 1496 SymbolsHere = ArrayRef<SymbolInfoTy>(&Symbols[FirstSI], SI - FirstSI); 1497 1498 // Get the demangled names of all those symbols. We end up with a vector 1499 // of StringRef that holds the names we're going to use, and a vector of 1500 // std::string that stores the new strings returned by demangle(), if 1501 // any. If we don't call demangle() then that vector can stay empty. 1502 std::vector<StringRef> SymNamesHere; 1503 std::vector<std::string> DemangledSymNamesHere; 1504 if (Demangle) { 1505 // Fetch the demangled names and store them locally. 1506 for (const SymbolInfoTy &Symbol : SymbolsHere) 1507 DemangledSymNamesHere.push_back(demangle(Symbol.Name.str())); 1508 // Now we've finished modifying that vector, it's safe to make 1509 // a vector of StringRefs pointing into it. 1510 SymNamesHere.insert(SymNamesHere.begin(), DemangledSymNamesHere.begin(), 1511 DemangledSymNamesHere.end()); 1512 } else { 1513 for (const SymbolInfoTy &Symbol : SymbolsHere) 1514 SymNamesHere.push_back(Symbol.Name); 1515 } 1516 1517 // Distinguish ELF data from code symbols, which will be used later on to 1518 // decide whether to 'disassemble' this chunk as a data declaration via 1519 // dumpELFData(), or whether to treat it as code. 1520 // 1521 // If data _and_ code symbols are defined at the same address, the code 1522 // takes priority, on the grounds that disassembling code is our main 1523 // purpose here, and it would be a worse failure to _not_ interpret 1524 // something that _was_ meaningful as code than vice versa. 1525 // 1526 // Any ELF symbol type that is not clearly data will be regarded as code. 1527 // In particular, one of the uses of STT_NOTYPE is for branch targets 1528 // inside functions, for which STT_FUNC would be inaccurate. 1529 // 1530 // So here, we spot whether there's any non-data symbol present at all, 1531 // and only set the DisassembleAsData flag if there isn't. Also, we use 1532 // this distinction to inform the decision of which symbol to print at 1533 // the head of the section, so that if we're printing code, we print a 1534 // code-related symbol name to go with it. 1535 bool DisassembleAsData = false; 1536 size_t DisplaySymIndex = SymbolsHere.size() - 1; 1537 if (Obj.isELF() && !DisassembleAll && Section.isText()) { 1538 DisassembleAsData = true; // unless we find a code symbol below 1539 1540 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 1541 uint8_t SymTy = SymbolsHere[i].Type; 1542 if (SymTy != ELF::STT_OBJECT && SymTy != ELF::STT_COMMON) { 1543 DisassembleAsData = false; 1544 DisplaySymIndex = i; 1545 } 1546 } 1547 } 1548 1549 // Decide which symbol(s) from this collection we're going to print. 1550 std::vector<bool> SymsToPrint(SymbolsHere.size(), false); 1551 // If the user has given the --disassemble-symbols option, then we must 1552 // display every symbol in that set, and no others. 1553 if (!DisasmSymbolSet.empty()) { 1554 bool FoundAny = false; 1555 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 1556 if (DisasmSymbolSet.count(SymNamesHere[i])) { 1557 SymsToPrint[i] = true; 1558 FoundAny = true; 1559 } 1560 } 1561 1562 // And if none of the symbols here is one that the user asked for, skip 1563 // disassembling this entire chunk of code. 1564 if (!FoundAny) 1565 continue; 1566 } else { 1567 // Otherwise, print whichever symbol at this location is last in the 1568 // Symbols array, because that array is pre-sorted in a way intended to 1569 // correlate with priority of which symbol to display. 1570 SymsToPrint[DisplaySymIndex] = true; 1571 } 1572 1573 // Now that we know we're disassembling this section, override the choice 1574 // of which symbols to display by printing _all_ of them at this address 1575 // if the user asked for all symbols. 1576 // 1577 // That way, '--show-all-symbols --disassemble-symbol=foo' will print 1578 // only the chunk of code headed by 'foo', but also show any other 1579 // symbols defined at that address, such as aliases for 'foo', or the ARM 1580 // mapping symbol preceding its code. 1581 if (ShowAllSymbols) { 1582 for (size_t i = 0; i < SymbolsHere.size(); ++i) 1583 SymsToPrint[i] = true; 1584 } 1585 1586 if (Start < SectionAddr || StopAddress <= Start) 1587 continue; 1588 1589 for (size_t i = 0; i < SymbolsHere.size(); ++i) 1590 FoundDisasmSymbolSet.insert(SymNamesHere[i]); 1591 1592 // The end is the section end, the beginning of the next symbol, or 1593 // --stop-address. 1594 uint64_t End = std::min<uint64_t>(SectionAddr + SectSize, StopAddress); 1595 if (SI < SE) 1596 End = std::min(End, Symbols[SI].Addr); 1597 if (Start >= End || End <= StartAddress) 1598 continue; 1599 Start -= SectionAddr; 1600 End -= SectionAddr; 1601 1602 if (!PrintedSection) { 1603 PrintedSection = true; 1604 outs() << "\nDisassembly of section "; 1605 if (!SegmentName.empty()) 1606 outs() << SegmentName << ","; 1607 outs() << SectionName << ":\n"; 1608 } 1609 1610 outs() << '\n'; 1611 1612 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 1613 if (!SymsToPrint[i]) 1614 continue; 1615 1616 const SymbolInfoTy &Symbol = SymbolsHere[i]; 1617 const StringRef SymbolName = SymNamesHere[i]; 1618 1619 if (LeadingAddr) 1620 outs() << format(Is64Bits ? "%016" PRIx64 " " : "%08" PRIx64 " ", 1621 SectionAddr + Start + VMAAdjustment); 1622 if (Obj.isXCOFF() && SymbolDescription) { 1623 outs() << getXCOFFSymbolDescription(Symbol, SymbolName) << ":\n"; 1624 } else 1625 outs() << '<' << SymbolName << ">:\n"; 1626 } 1627 1628 // Don't print raw contents of a virtual section. A virtual section 1629 // doesn't have any contents in the file. 1630 if (Section.isVirtual()) { 1631 outs() << "...\n"; 1632 continue; 1633 } 1634 1635 // See if any of the symbols defined at this location triggers target- 1636 // specific disassembly behavior, e.g. of special descriptors or function 1637 // prelude information. 1638 // 1639 // We stop this loop at the first symbol that triggers some kind of 1640 // interesting behavior (if any), on the assumption that if two symbols 1641 // defined at the same address trigger two conflicting symbol handlers, 1642 // the object file is probably confused anyway, and it would make even 1643 // less sense to present the output of _both_ handlers, because that 1644 // would describe the same data twice. 1645 for (size_t SHI = 0; SHI < SymbolsHere.size(); ++SHI) { 1646 SymbolInfoTy Symbol = SymbolsHere[SHI]; 1647 1648 auto Status = 1649 DisAsm->onSymbolStart(Symbol, Size, Bytes.slice(Start, End - Start), 1650 SectionAddr + Start, CommentStream); 1651 1652 if (!Status) { 1653 // If onSymbolStart returns None, that means it didn't trigger any 1654 // interesting handling for this symbol. Try the other symbols 1655 // defined at this address. 1656 continue; 1657 } 1658 1659 if (Status.value() == MCDisassembler::Fail) { 1660 // If onSymbolStart returns Fail, that means it identified some kind 1661 // of special data at this address, but wasn't able to disassemble it 1662 // meaningfully. So we fall back to disassembling the failed region 1663 // as bytes, assuming that the target detected the failure before 1664 // printing anything. 1665 // 1666 // Return values Success or SoftFail (i.e no 'real' failure) are 1667 // expected to mean that the target has emitted its own output. 1668 // 1669 // Either way, 'Size' will have been set to the amount of data 1670 // covered by whatever prologue the target identified. So we advance 1671 // our own position to beyond that. Sometimes that will be the entire 1672 // distance to the next symbol, and sometimes it will be just a 1673 // prologue and we should start disassembling instructions from where 1674 // it left off. 1675 outs() << "// Error in decoding " << SymNamesHere[SHI] 1676 << " : Decoding failed region as bytes.\n"; 1677 for (uint64_t I = 0; I < Size; ++I) { 1678 outs() << "\t.byte\t " << format_hex(Bytes[I], 1, /*Upper=*/true) 1679 << "\n"; 1680 } 1681 } 1682 Start += Size; 1683 break; 1684 } 1685 1686 Index = Start; 1687 if (SectionAddr < StartAddress) 1688 Index = std::max<uint64_t>(Index, StartAddress - SectionAddr); 1689 1690 if (DisassembleAsData) { 1691 dumpELFData(SectionAddr, Index, End, Bytes); 1692 Index = End; 1693 continue; 1694 } 1695 1696 bool DumpARMELFData = false; 1697 formatted_raw_ostream FOS(outs()); 1698 1699 std::unordered_map<uint64_t, std::string> AllLabels; 1700 std::unordered_map<uint64_t, std::vector<std::string>> BBAddrMapLabels; 1701 if (SymbolizeOperands) { 1702 collectLocalBranchTargets(Bytes, MIA, DisAsm, IP, PrimarySTI, 1703 SectionAddr, Index, End, AllLabels); 1704 collectBBAddrMapLabels(AddrToBBAddrMap, SectionAddr, Index, End, 1705 BBAddrMapLabels); 1706 } 1707 1708 while (Index < End) { 1709 // ARM and AArch64 ELF binaries can interleave data and text in the 1710 // same section. We rely on the markers introduced to understand what 1711 // we need to dump. If the data marker is within a function, it is 1712 // denoted as a word/short etc. 1713 if (!MappingSymbols.empty()) { 1714 char Kind = getMappingSymbolKind(MappingSymbols, Index); 1715 DumpARMELFData = Kind == 'd'; 1716 if (SecondarySTI) { 1717 if (Kind == 'a') { 1718 STI = PrimaryIsThumb ? SecondarySTI : PrimarySTI; 1719 DisAsm = PrimaryIsThumb ? SecondaryDisAsm : PrimaryDisAsm; 1720 } else if (Kind == 't') { 1721 STI = PrimaryIsThumb ? PrimarySTI : SecondarySTI; 1722 DisAsm = PrimaryIsThumb ? PrimaryDisAsm : SecondaryDisAsm; 1723 } 1724 } 1725 } 1726 1727 if (DumpARMELFData) { 1728 Size = dumpARMELFData(SectionAddr, Index, End, Obj, Bytes, 1729 MappingSymbols, *STI, FOS); 1730 } else { 1731 // When -z or --disassemble-zeroes are given we always dissasemble 1732 // them. Otherwise we might want to skip zero bytes we see. 1733 if (!DisassembleZeroes) { 1734 uint64_t MaxOffset = End - Index; 1735 // For --reloc: print zero blocks patched by relocations, so that 1736 // relocations can be shown in the dump. 1737 if (RelCur != RelEnd) 1738 MaxOffset = std::min(RelCur->getOffset() - RelAdjustment - Index, 1739 MaxOffset); 1740 1741 if (size_t N = 1742 countSkippableZeroBytes(Bytes.slice(Index, MaxOffset))) { 1743 FOS << "\t\t..." << '\n'; 1744 Index += N; 1745 continue; 1746 } 1747 } 1748 1749 // Print local label if there's any. 1750 auto Iter1 = BBAddrMapLabels.find(SectionAddr + Index); 1751 if (Iter1 != BBAddrMapLabels.end()) { 1752 for (StringRef Label : Iter1->second) 1753 FOS << "<" << Label << ">:\n"; 1754 } else { 1755 auto Iter2 = AllLabels.find(SectionAddr + Index); 1756 if (Iter2 != AllLabels.end()) 1757 FOS << "<" << Iter2->second << ">:\n"; 1758 } 1759 1760 // Disassemble a real instruction or a data when disassemble all is 1761 // provided 1762 MCInst Inst; 1763 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index); 1764 uint64_t ThisAddr = SectionAddr + Index; 1765 bool Disassembled = DisAsm->getInstruction(Inst, Size, ThisBytes, 1766 ThisAddr, CommentStream); 1767 if (Size == 0) 1768 Size = std::min<uint64_t>( 1769 ThisBytes.size(), 1770 DisAsm->suggestBytesToSkip(ThisBytes, ThisAddr)); 1771 1772 LVP.update({Index, Section.getIndex()}, 1773 {Index + Size, Section.getIndex()}, Index + Size != End); 1774 1775 IP->setCommentStream(CommentStream); 1776 1777 PIP.printInst( 1778 *IP, Disassembled ? &Inst : nullptr, Bytes.slice(Index, Size), 1779 {SectionAddr + Index + VMAAdjustment, Section.getIndex()}, FOS, 1780 "", *STI, &SP, Obj.getFileName(), &Rels, LVP); 1781 1782 IP->setCommentStream(llvm::nulls()); 1783 1784 // If disassembly has failed, avoid analysing invalid/incomplete 1785 // instruction information. Otherwise, try to resolve the target 1786 // address (jump target or memory operand address) and print it on the 1787 // right of the instruction. 1788 if (Disassembled && MIA) { 1789 // Branch targets are printed just after the instructions. 1790 llvm::raw_ostream *TargetOS = &FOS; 1791 uint64_t Target; 1792 bool PrintTarget = 1793 MIA->evaluateBranch(Inst, SectionAddr + Index, Size, Target); 1794 if (!PrintTarget) 1795 if (Optional<uint64_t> MaybeTarget = 1796 MIA->evaluateMemoryOperandAddress( 1797 Inst, STI, SectionAddr + Index, Size)) { 1798 Target = *MaybeTarget; 1799 PrintTarget = true; 1800 // Do not print real address when symbolizing. 1801 if (!SymbolizeOperands) { 1802 // Memory operand addresses are printed as comments. 1803 TargetOS = &CommentStream; 1804 *TargetOS << "0x" << Twine::utohexstr(Target); 1805 } 1806 } 1807 if (PrintTarget) { 1808 // In a relocatable object, the target's section must reside in 1809 // the same section as the call instruction or it is accessed 1810 // through a relocation. 1811 // 1812 // In a non-relocatable object, the target may be in any section. 1813 // In that case, locate the section(s) containing the target 1814 // address and find the symbol in one of those, if possible. 1815 // 1816 // N.B. We don't walk the relocations in the relocatable case yet. 1817 std::vector<const SectionSymbolsTy *> TargetSectionSymbols; 1818 if (!Obj.isRelocatableObject()) { 1819 auto It = llvm::partition_point( 1820 SectionAddresses, 1821 [=](const std::pair<uint64_t, SectionRef> &O) { 1822 return O.first <= Target; 1823 }); 1824 uint64_t TargetSecAddr = 0; 1825 while (It != SectionAddresses.begin()) { 1826 --It; 1827 if (TargetSecAddr == 0) 1828 TargetSecAddr = It->first; 1829 if (It->first != TargetSecAddr) 1830 break; 1831 TargetSectionSymbols.push_back(&AllSymbols[It->second]); 1832 } 1833 } else { 1834 TargetSectionSymbols.push_back(&Symbols); 1835 } 1836 TargetSectionSymbols.push_back(&AbsoluteSymbols); 1837 1838 // Find the last symbol in the first candidate section whose 1839 // offset is less than or equal to the target. If there are no 1840 // such symbols, try in the next section and so on, before finally 1841 // using the nearest preceding absolute symbol (if any), if there 1842 // are no other valid symbols. 1843 const SymbolInfoTy *TargetSym = nullptr; 1844 for (const SectionSymbolsTy *TargetSymbols : 1845 TargetSectionSymbols) { 1846 auto It = llvm::partition_point( 1847 *TargetSymbols, 1848 [=](const SymbolInfoTy &O) { return O.Addr <= Target; }); 1849 if (It != TargetSymbols->begin()) { 1850 TargetSym = &*(It - 1); 1851 break; 1852 } 1853 } 1854 1855 // Print the labels corresponding to the target if there's any. 1856 bool BBAddrMapLabelAvailable = BBAddrMapLabels.count(Target); 1857 bool LabelAvailable = AllLabels.count(Target); 1858 if (TargetSym != nullptr) { 1859 uint64_t TargetAddress = TargetSym->Addr; 1860 uint64_t Disp = Target - TargetAddress; 1861 std::string TargetName = TargetSym->Name.str(); 1862 if (Demangle) 1863 TargetName = demangle(TargetName); 1864 1865 *TargetOS << " <"; 1866 if (!Disp) { 1867 // Always Print the binary symbol precisely corresponding to 1868 // the target address. 1869 *TargetOS << TargetName; 1870 } else if (BBAddrMapLabelAvailable) { 1871 *TargetOS << BBAddrMapLabels[Target].front(); 1872 } else if (LabelAvailable) { 1873 *TargetOS << AllLabels[Target]; 1874 } else { 1875 // Always Print the binary symbol plus an offset if there's no 1876 // local label corresponding to the target address. 1877 *TargetOS << TargetName << "+0x" << Twine::utohexstr(Disp); 1878 } 1879 *TargetOS << ">"; 1880 } else if (BBAddrMapLabelAvailable) { 1881 *TargetOS << " <" << BBAddrMapLabels[Target].front() << ">"; 1882 } else if (LabelAvailable) { 1883 *TargetOS << " <" << AllLabels[Target] << ">"; 1884 } 1885 // By convention, each record in the comment stream should be 1886 // terminated. 1887 if (TargetOS == &CommentStream) 1888 *TargetOS << "\n"; 1889 } 1890 } 1891 } 1892 1893 assert(Ctx.getAsmInfo()); 1894 emitPostInstructionInfo(FOS, *Ctx.getAsmInfo(), *STI, 1895 CommentStream.str(), LVP); 1896 Comments.clear(); 1897 1898 // Hexagon does this in pretty printer 1899 if (Obj.getArch() != Triple::hexagon) { 1900 // Print relocation for instruction and data. 1901 while (RelCur != RelEnd) { 1902 uint64_t Offset = RelCur->getOffset() - RelAdjustment; 1903 // If this relocation is hidden, skip it. 1904 if (getHidden(*RelCur) || SectionAddr + Offset < StartAddress) { 1905 ++RelCur; 1906 continue; 1907 } 1908 1909 // Stop when RelCur's offset is past the disassembled 1910 // instruction/data. Note that it's possible the disassembled data 1911 // is not the complete data: we might see the relocation printed in 1912 // the middle of the data, but this matches the binutils objdump 1913 // output. 1914 if (Offset >= Index + Size) 1915 break; 1916 1917 // When --adjust-vma is used, update the address printed. 1918 if (RelCur->getSymbol() != Obj.symbol_end()) { 1919 Expected<section_iterator> SymSI = 1920 RelCur->getSymbol()->getSection(); 1921 if (SymSI && *SymSI != Obj.section_end() && 1922 shouldAdjustVA(**SymSI)) 1923 Offset += AdjustVMA; 1924 } 1925 1926 printRelocation(FOS, Obj.getFileName(), *RelCur, 1927 SectionAddr + Offset, Is64Bits); 1928 LVP.printAfterOtherLine(FOS, true); 1929 ++RelCur; 1930 } 1931 } 1932 1933 Index += Size; 1934 } 1935 } 1936 } 1937 StringSet<> MissingDisasmSymbolSet = 1938 set_difference(DisasmSymbolSet, FoundDisasmSymbolSet); 1939 for (StringRef Sym : MissingDisasmSymbolSet.keys()) 1940 reportWarning("failed to disassemble missing symbol " + Sym, FileName); 1941 } 1942 1943 static void disassembleObject(ObjectFile *Obj, bool InlineRelocs) { 1944 const Target *TheTarget = getTarget(Obj); 1945 1946 // Package up features to be passed to target/subtarget 1947 SubtargetFeatures Features = Obj->getFeatures(); 1948 if (!MAttrs.empty()) { 1949 for (unsigned I = 0; I != MAttrs.size(); ++I) 1950 Features.AddFeature(MAttrs[I]); 1951 } else if (MCPU.empty() && Obj->getArch() == llvm::Triple::aarch64) { 1952 Features.AddFeature("+all"); 1953 } 1954 1955 std::unique_ptr<const MCRegisterInfo> MRI( 1956 TheTarget->createMCRegInfo(TripleName)); 1957 if (!MRI) 1958 reportError(Obj->getFileName(), 1959 "no register info for target " + TripleName); 1960 1961 // Set up disassembler. 1962 MCTargetOptions MCOptions; 1963 std::unique_ptr<const MCAsmInfo> AsmInfo( 1964 TheTarget->createMCAsmInfo(*MRI, TripleName, MCOptions)); 1965 if (!AsmInfo) 1966 reportError(Obj->getFileName(), 1967 "no assembly info for target " + TripleName); 1968 1969 if (MCPU.empty()) 1970 MCPU = Obj->tryGetCPUName().value_or("").str(); 1971 1972 if (isArmElf(*Obj)) { 1973 // When disassembling big-endian Arm ELF, the instruction endianness is 1974 // determined in a complex way. In relocatable objects, AAELF32 mandates 1975 // that instruction endianness matches the ELF file endianness; in 1976 // executable images, that's true unless the file header has the EF_ARM_BE8 1977 // flag, in which case instructions are little-endian regardless of data 1978 // endianness. 1979 // 1980 // We must set the big-endian-instructions SubtargetFeature to make the 1981 // disassembler read the instructions the right way round, and also tell 1982 // our own prettyprinter to retrieve the encodings the same way to print in 1983 // hex. 1984 const auto *Elf32BE = dyn_cast<ELF32BEObjectFile>(Obj); 1985 1986 if (Elf32BE && (Elf32BE->isRelocatableObject() || 1987 !(Elf32BE->getPlatformFlags() & ELF::EF_ARM_BE8))) { 1988 Features.AddFeature("+big-endian-instructions"); 1989 ARMPrettyPrinterInst.setInstructionEndianness(llvm::support::big); 1990 } else { 1991 ARMPrettyPrinterInst.setInstructionEndianness(llvm::support::little); 1992 } 1993 } 1994 1995 std::unique_ptr<const MCSubtargetInfo> STI( 1996 TheTarget->createMCSubtargetInfo(TripleName, MCPU, Features.getString())); 1997 if (!STI) 1998 reportError(Obj->getFileName(), 1999 "no subtarget info for target " + TripleName); 2000 std::unique_ptr<const MCInstrInfo> MII(TheTarget->createMCInstrInfo()); 2001 if (!MII) 2002 reportError(Obj->getFileName(), 2003 "no instruction info for target " + TripleName); 2004 MCContext Ctx(Triple(TripleName), AsmInfo.get(), MRI.get(), STI.get()); 2005 // FIXME: for now initialize MCObjectFileInfo with default values 2006 std::unique_ptr<MCObjectFileInfo> MOFI( 2007 TheTarget->createMCObjectFileInfo(Ctx, /*PIC=*/false)); 2008 Ctx.setObjectFileInfo(MOFI.get()); 2009 2010 std::unique_ptr<MCDisassembler> DisAsm( 2011 TheTarget->createMCDisassembler(*STI, Ctx)); 2012 if (!DisAsm) 2013 reportError(Obj->getFileName(), "no disassembler for target " + TripleName); 2014 2015 // If we have an ARM object file, we need a second disassembler, because 2016 // ARM CPUs have two different instruction sets: ARM mode, and Thumb mode. 2017 // We use mapping symbols to switch between the two assemblers, where 2018 // appropriate. 2019 std::unique_ptr<MCDisassembler> SecondaryDisAsm; 2020 std::unique_ptr<const MCSubtargetInfo> SecondarySTI; 2021 if (isArmElf(*Obj) && !STI->checkFeatures("+mclass")) { 2022 if (STI->checkFeatures("+thumb-mode")) 2023 Features.AddFeature("-thumb-mode"); 2024 else 2025 Features.AddFeature("+thumb-mode"); 2026 SecondarySTI.reset(TheTarget->createMCSubtargetInfo(TripleName, MCPU, 2027 Features.getString())); 2028 SecondaryDisAsm.reset(TheTarget->createMCDisassembler(*SecondarySTI, Ctx)); 2029 } 2030 2031 std::unique_ptr<const MCInstrAnalysis> MIA( 2032 TheTarget->createMCInstrAnalysis(MII.get())); 2033 2034 int AsmPrinterVariant = AsmInfo->getAssemblerDialect(); 2035 std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter( 2036 Triple(TripleName), AsmPrinterVariant, *AsmInfo, *MII, *MRI)); 2037 if (!IP) 2038 reportError(Obj->getFileName(), 2039 "no instruction printer for target " + TripleName); 2040 IP->setPrintImmHex(PrintImmHex); 2041 IP->setPrintBranchImmAsAddress(true); 2042 IP->setSymbolizeOperands(SymbolizeOperands); 2043 IP->setMCInstrAnalysis(MIA.get()); 2044 2045 PrettyPrinter &PIP = selectPrettyPrinter(Triple(TripleName)); 2046 SourcePrinter SP(Obj, TheTarget->getName()); 2047 2048 for (StringRef Opt : DisassemblerOptions) 2049 if (!IP->applyTargetSpecificCLOption(Opt)) 2050 reportError(Obj->getFileName(), 2051 "Unrecognized disassembler option: " + Opt); 2052 2053 disassembleObject(TheTarget, *Obj, Ctx, DisAsm.get(), SecondaryDisAsm.get(), 2054 MIA.get(), IP.get(), STI.get(), SecondarySTI.get(), PIP, SP, 2055 InlineRelocs); 2056 } 2057 2058 void objdump::printRelocations(const ObjectFile *Obj) { 2059 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : 2060 "%08" PRIx64; 2061 2062 // Build a mapping from relocation target to a vector of relocation 2063 // sections. Usually, there is an only one relocation section for 2064 // each relocated section. 2065 MapVector<SectionRef, std::vector<SectionRef>> SecToRelSec; 2066 uint64_t Ndx; 2067 for (const SectionRef &Section : ToolSectionFilter(*Obj, &Ndx)) { 2068 if (Obj->isELF() && (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC)) 2069 continue; 2070 if (Section.relocation_begin() == Section.relocation_end()) 2071 continue; 2072 Expected<section_iterator> SecOrErr = Section.getRelocatedSection(); 2073 if (!SecOrErr) 2074 reportError(Obj->getFileName(), 2075 "section (" + Twine(Ndx) + 2076 "): unable to get a relocation target: " + 2077 toString(SecOrErr.takeError())); 2078 SecToRelSec[**SecOrErr].push_back(Section); 2079 } 2080 2081 for (std::pair<SectionRef, std::vector<SectionRef>> &P : SecToRelSec) { 2082 StringRef SecName = unwrapOrError(P.first.getName(), Obj->getFileName()); 2083 outs() << "\nRELOCATION RECORDS FOR [" << SecName << "]:\n"; 2084 uint32_t OffsetPadding = (Obj->getBytesInAddress() > 4 ? 16 : 8); 2085 uint32_t TypePadding = 24; 2086 outs() << left_justify("OFFSET", OffsetPadding) << " " 2087 << left_justify("TYPE", TypePadding) << " " 2088 << "VALUE\n"; 2089 2090 for (SectionRef Section : P.second) { 2091 for (const RelocationRef &Reloc : Section.relocations()) { 2092 uint64_t Address = Reloc.getOffset(); 2093 SmallString<32> RelocName; 2094 SmallString<32> ValueStr; 2095 if (Address < StartAddress || Address > StopAddress || getHidden(Reloc)) 2096 continue; 2097 Reloc.getTypeName(RelocName); 2098 if (Error E = getRelocationValueString(Reloc, ValueStr)) 2099 reportError(std::move(E), Obj->getFileName()); 2100 2101 outs() << format(Fmt.data(), Address) << " " 2102 << left_justify(RelocName, TypePadding) << " " << ValueStr 2103 << "\n"; 2104 } 2105 } 2106 } 2107 } 2108 2109 void objdump::printDynamicRelocations(const ObjectFile *Obj) { 2110 // For the moment, this option is for ELF only 2111 if (!Obj->isELF()) 2112 return; 2113 2114 const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj); 2115 if (!Elf || !any_of(Elf->sections(), [](const ELFSectionRef Sec) { 2116 return Sec.getType() == ELF::SHT_DYNAMIC; 2117 })) { 2118 reportError(Obj->getFileName(), "not a dynamic object"); 2119 return; 2120 } 2121 2122 std::vector<SectionRef> DynRelSec = Obj->dynamic_relocation_sections(); 2123 if (DynRelSec.empty()) 2124 return; 2125 2126 outs() << "\nDYNAMIC RELOCATION RECORDS\n"; 2127 const uint32_t OffsetPadding = (Obj->getBytesInAddress() > 4 ? 16 : 8); 2128 const uint32_t TypePadding = 24; 2129 outs() << left_justify("OFFSET", OffsetPadding) << ' ' 2130 << left_justify("TYPE", TypePadding) << " VALUE\n"; 2131 2132 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2133 for (const SectionRef &Section : DynRelSec) 2134 for (const RelocationRef &Reloc : Section.relocations()) { 2135 uint64_t Address = Reloc.getOffset(); 2136 SmallString<32> RelocName; 2137 SmallString<32> ValueStr; 2138 Reloc.getTypeName(RelocName); 2139 if (Error E = getRelocationValueString(Reloc, ValueStr)) 2140 reportError(std::move(E), Obj->getFileName()); 2141 outs() << format(Fmt.data(), Address) << ' ' 2142 << left_justify(RelocName, TypePadding) << ' ' << ValueStr << '\n'; 2143 } 2144 } 2145 2146 // Returns true if we need to show LMA column when dumping section headers. We 2147 // show it only when the platform is ELF and either we have at least one section 2148 // whose VMA and LMA are different and/or when --show-lma flag is used. 2149 static bool shouldDisplayLMA(const ObjectFile &Obj) { 2150 if (!Obj.isELF()) 2151 return false; 2152 for (const SectionRef &S : ToolSectionFilter(Obj)) 2153 if (S.getAddress() != getELFSectionLMA(S)) 2154 return true; 2155 return ShowLMA; 2156 } 2157 2158 static size_t getMaxSectionNameWidth(const ObjectFile &Obj) { 2159 // Default column width for names is 13 even if no names are that long. 2160 size_t MaxWidth = 13; 2161 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 2162 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 2163 MaxWidth = std::max(MaxWidth, Name.size()); 2164 } 2165 return MaxWidth; 2166 } 2167 2168 void objdump::printSectionHeaders(ObjectFile &Obj) { 2169 size_t NameWidth = getMaxSectionNameWidth(Obj); 2170 size_t AddressWidth = 2 * Obj.getBytesInAddress(); 2171 bool HasLMAColumn = shouldDisplayLMA(Obj); 2172 outs() << "\nSections:\n"; 2173 if (HasLMAColumn) 2174 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 2175 << left_justify("VMA", AddressWidth) << " " 2176 << left_justify("LMA", AddressWidth) << " Type\n"; 2177 else 2178 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 2179 << left_justify("VMA", AddressWidth) << " Type\n"; 2180 2181 if (Obj.isELF() && Obj.sections().empty()) 2182 createFakeELFSections(Obj); 2183 2184 uint64_t Idx; 2185 for (const SectionRef &Section : ToolSectionFilter(Obj, &Idx)) { 2186 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 2187 uint64_t VMA = Section.getAddress(); 2188 if (shouldAdjustVA(Section)) 2189 VMA += AdjustVMA; 2190 2191 uint64_t Size = Section.getSize(); 2192 2193 std::string Type = Section.isText() ? "TEXT" : ""; 2194 if (Section.isData()) 2195 Type += Type.empty() ? "DATA" : ", DATA"; 2196 if (Section.isBSS()) 2197 Type += Type.empty() ? "BSS" : ", BSS"; 2198 if (Section.isDebugSection()) 2199 Type += Type.empty() ? "DEBUG" : ", DEBUG"; 2200 2201 if (HasLMAColumn) 2202 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 2203 Name.str().c_str(), Size) 2204 << format_hex_no_prefix(VMA, AddressWidth) << " " 2205 << format_hex_no_prefix(getELFSectionLMA(Section), AddressWidth) 2206 << " " << Type << "\n"; 2207 else 2208 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 2209 Name.str().c_str(), Size) 2210 << format_hex_no_prefix(VMA, AddressWidth) << " " << Type << "\n"; 2211 } 2212 } 2213 2214 void objdump::printSectionContents(const ObjectFile *Obj) { 2215 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj); 2216 2217 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 2218 StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName()); 2219 uint64_t BaseAddr = Section.getAddress(); 2220 uint64_t Size = Section.getSize(); 2221 if (!Size) 2222 continue; 2223 2224 outs() << "Contents of section "; 2225 StringRef SegmentName = getSegmentName(MachO, Section); 2226 if (!SegmentName.empty()) 2227 outs() << SegmentName << ","; 2228 outs() << Name << ":\n"; 2229 if (Section.isBSS()) { 2230 outs() << format("<skipping contents of bss section at [%04" PRIx64 2231 ", %04" PRIx64 ")>\n", 2232 BaseAddr, BaseAddr + Size); 2233 continue; 2234 } 2235 2236 StringRef Contents = unwrapOrError(Section.getContents(), Obj->getFileName()); 2237 2238 // Dump out the content as hex and printable ascii characters. 2239 for (std::size_t Addr = 0, End = Contents.size(); Addr < End; Addr += 16) { 2240 outs() << format(" %04" PRIx64 " ", BaseAddr + Addr); 2241 // Dump line of hex. 2242 for (std::size_t I = 0; I < 16; ++I) { 2243 if (I != 0 && I % 4 == 0) 2244 outs() << ' '; 2245 if (Addr + I < End) 2246 outs() << hexdigit((Contents[Addr + I] >> 4) & 0xF, true) 2247 << hexdigit(Contents[Addr + I] & 0xF, true); 2248 else 2249 outs() << " "; 2250 } 2251 // Print ascii. 2252 outs() << " "; 2253 for (std::size_t I = 0; I < 16 && Addr + I < End; ++I) { 2254 if (isPrint(static_cast<unsigned char>(Contents[Addr + I]) & 0xFF)) 2255 outs() << Contents[Addr + I]; 2256 else 2257 outs() << "."; 2258 } 2259 outs() << "\n"; 2260 } 2261 } 2262 } 2263 2264 void objdump::printSymbolTable(const ObjectFile &O, StringRef ArchiveName, 2265 StringRef ArchitectureName, bool DumpDynamic) { 2266 if (O.isCOFF() && !DumpDynamic) { 2267 outs() << "\nSYMBOL TABLE:\n"; 2268 printCOFFSymbolTable(cast<const COFFObjectFile>(O)); 2269 return; 2270 } 2271 2272 const StringRef FileName = O.getFileName(); 2273 2274 if (!DumpDynamic) { 2275 outs() << "\nSYMBOL TABLE:\n"; 2276 for (auto I = O.symbol_begin(); I != O.symbol_end(); ++I) 2277 printSymbol(O, *I, {}, FileName, ArchiveName, ArchitectureName, 2278 DumpDynamic); 2279 return; 2280 } 2281 2282 outs() << "\nDYNAMIC SYMBOL TABLE:\n"; 2283 if (!O.isELF()) { 2284 reportWarning( 2285 "this operation is not currently supported for this file format", 2286 FileName); 2287 return; 2288 } 2289 2290 const ELFObjectFileBase *ELF = cast<const ELFObjectFileBase>(&O); 2291 auto Symbols = ELF->getDynamicSymbolIterators(); 2292 Expected<std::vector<VersionEntry>> SymbolVersionsOrErr = 2293 ELF->readDynsymVersions(); 2294 if (!SymbolVersionsOrErr) { 2295 reportWarning(toString(SymbolVersionsOrErr.takeError()), FileName); 2296 SymbolVersionsOrErr = std::vector<VersionEntry>(); 2297 (void)!SymbolVersionsOrErr; 2298 } 2299 for (auto &Sym : Symbols) 2300 printSymbol(O, Sym, *SymbolVersionsOrErr, FileName, ArchiveName, 2301 ArchitectureName, DumpDynamic); 2302 } 2303 2304 void objdump::printSymbol(const ObjectFile &O, const SymbolRef &Symbol, 2305 ArrayRef<VersionEntry> SymbolVersions, 2306 StringRef FileName, StringRef ArchiveName, 2307 StringRef ArchitectureName, bool DumpDynamic) { 2308 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&O); 2309 uint64_t Address = unwrapOrError(Symbol.getAddress(), FileName, ArchiveName, 2310 ArchitectureName); 2311 if ((Address < StartAddress) || (Address > StopAddress)) 2312 return; 2313 SymbolRef::Type Type = 2314 unwrapOrError(Symbol.getType(), FileName, ArchiveName, ArchitectureName); 2315 uint32_t Flags = 2316 unwrapOrError(Symbol.getFlags(), FileName, ArchiveName, ArchitectureName); 2317 2318 // Don't ask a Mach-O STAB symbol for its section unless you know that 2319 // STAB symbol's section field refers to a valid section index. Otherwise 2320 // the symbol may error trying to load a section that does not exist. 2321 bool IsSTAB = false; 2322 if (MachO) { 2323 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 2324 uint8_t NType = 2325 (MachO->is64Bit() ? MachO->getSymbol64TableEntry(SymDRI).n_type 2326 : MachO->getSymbolTableEntry(SymDRI).n_type); 2327 if (NType & MachO::N_STAB) 2328 IsSTAB = true; 2329 } 2330 section_iterator Section = IsSTAB 2331 ? O.section_end() 2332 : unwrapOrError(Symbol.getSection(), FileName, 2333 ArchiveName, ArchitectureName); 2334 2335 StringRef Name; 2336 if (Type == SymbolRef::ST_Debug && Section != O.section_end()) { 2337 if (Expected<StringRef> NameOrErr = Section->getName()) 2338 Name = *NameOrErr; 2339 else 2340 consumeError(NameOrErr.takeError()); 2341 2342 } else { 2343 Name = unwrapOrError(Symbol.getName(), FileName, ArchiveName, 2344 ArchitectureName); 2345 } 2346 2347 bool Global = Flags & SymbolRef::SF_Global; 2348 bool Weak = Flags & SymbolRef::SF_Weak; 2349 bool Absolute = Flags & SymbolRef::SF_Absolute; 2350 bool Common = Flags & SymbolRef::SF_Common; 2351 bool Hidden = Flags & SymbolRef::SF_Hidden; 2352 2353 char GlobLoc = ' '; 2354 if ((Section != O.section_end() || Absolute) && !Weak) 2355 GlobLoc = Global ? 'g' : 'l'; 2356 char IFunc = ' '; 2357 if (O.isELF()) { 2358 if (ELFSymbolRef(Symbol).getELFType() == ELF::STT_GNU_IFUNC) 2359 IFunc = 'i'; 2360 if (ELFSymbolRef(Symbol).getBinding() == ELF::STB_GNU_UNIQUE) 2361 GlobLoc = 'u'; 2362 } 2363 2364 char Debug = ' '; 2365 if (DumpDynamic) 2366 Debug = 'D'; 2367 else if (Type == SymbolRef::ST_Debug || Type == SymbolRef::ST_File) 2368 Debug = 'd'; 2369 2370 char FileFunc = ' '; 2371 if (Type == SymbolRef::ST_File) 2372 FileFunc = 'f'; 2373 else if (Type == SymbolRef::ST_Function) 2374 FileFunc = 'F'; 2375 else if (Type == SymbolRef::ST_Data) 2376 FileFunc = 'O'; 2377 2378 const char *Fmt = O.getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2379 2380 outs() << format(Fmt, Address) << " " 2381 << GlobLoc // Local -> 'l', Global -> 'g', Neither -> ' ' 2382 << (Weak ? 'w' : ' ') // Weak? 2383 << ' ' // Constructor. Not supported yet. 2384 << ' ' // Warning. Not supported yet. 2385 << IFunc // Indirect reference to another symbol. 2386 << Debug // Debugging (d) or dynamic (D) symbol. 2387 << FileFunc // Name of function (F), file (f) or object (O). 2388 << ' '; 2389 if (Absolute) { 2390 outs() << "*ABS*"; 2391 } else if (Common) { 2392 outs() << "*COM*"; 2393 } else if (Section == O.section_end()) { 2394 if (O.isXCOFF()) { 2395 XCOFFSymbolRef XCOFFSym = cast<const XCOFFObjectFile>(O).toSymbolRef( 2396 Symbol.getRawDataRefImpl()); 2397 if (XCOFF::N_DEBUG == XCOFFSym.getSectionNumber()) 2398 outs() << "*DEBUG*"; 2399 else 2400 outs() << "*UND*"; 2401 } else 2402 outs() << "*UND*"; 2403 } else { 2404 StringRef SegmentName = getSegmentName(MachO, *Section); 2405 if (!SegmentName.empty()) 2406 outs() << SegmentName << ","; 2407 StringRef SectionName = unwrapOrError(Section->getName(), FileName); 2408 outs() << SectionName; 2409 if (O.isXCOFF()) { 2410 Optional<SymbolRef> SymRef = 2411 getXCOFFSymbolContainingSymbolRef(cast<XCOFFObjectFile>(O), Symbol); 2412 if (SymRef) { 2413 2414 Expected<StringRef> NameOrErr = SymRef->getName(); 2415 2416 if (NameOrErr) { 2417 outs() << " (csect:"; 2418 std::string SymName(NameOrErr.get()); 2419 2420 if (Demangle) 2421 SymName = demangle(SymName); 2422 2423 if (SymbolDescription) 2424 SymName = getXCOFFSymbolDescription( 2425 createSymbolInfo(O, SymRef.value()), SymName); 2426 2427 outs() << ' ' << SymName; 2428 outs() << ") "; 2429 } else 2430 reportWarning(toString(NameOrErr.takeError()), FileName); 2431 } 2432 } 2433 } 2434 2435 if (Common) 2436 outs() << '\t' << format(Fmt, static_cast<uint64_t>(Symbol.getAlignment())); 2437 else if (O.isXCOFF()) 2438 outs() << '\t' 2439 << format(Fmt, cast<XCOFFObjectFile>(O).getSymbolSize( 2440 Symbol.getRawDataRefImpl())); 2441 else if (O.isELF()) 2442 outs() << '\t' << format(Fmt, ELFSymbolRef(Symbol).getSize()); 2443 2444 if (O.isELF()) { 2445 if (!SymbolVersions.empty()) { 2446 const VersionEntry &Ver = 2447 SymbolVersions[Symbol.getRawDataRefImpl().d.b - 1]; 2448 std::string Str; 2449 if (!Ver.Name.empty()) 2450 Str = Ver.IsVerDef ? ' ' + Ver.Name : '(' + Ver.Name + ')'; 2451 outs() << ' ' << left_justify(Str, 12); 2452 } 2453 2454 uint8_t Other = ELFSymbolRef(Symbol).getOther(); 2455 switch (Other) { 2456 case ELF::STV_DEFAULT: 2457 break; 2458 case ELF::STV_INTERNAL: 2459 outs() << " .internal"; 2460 break; 2461 case ELF::STV_HIDDEN: 2462 outs() << " .hidden"; 2463 break; 2464 case ELF::STV_PROTECTED: 2465 outs() << " .protected"; 2466 break; 2467 default: 2468 outs() << format(" 0x%02x", Other); 2469 break; 2470 } 2471 } else if (Hidden) { 2472 outs() << " .hidden"; 2473 } 2474 2475 std::string SymName(Name); 2476 if (Demangle) 2477 SymName = demangle(SymName); 2478 2479 if (O.isXCOFF() && SymbolDescription) 2480 SymName = getXCOFFSymbolDescription(createSymbolInfo(O, Symbol), SymName); 2481 2482 outs() << ' ' << SymName << '\n'; 2483 } 2484 2485 static void printUnwindInfo(const ObjectFile *O) { 2486 outs() << "Unwind info:\n\n"; 2487 2488 if (const COFFObjectFile *Coff = dyn_cast<COFFObjectFile>(O)) 2489 printCOFFUnwindInfo(Coff); 2490 else if (const MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(O)) 2491 printMachOUnwindInfo(MachO); 2492 else 2493 // TODO: Extract DWARF dump tool to objdump. 2494 WithColor::error(errs(), ToolName) 2495 << "This operation is only currently supported " 2496 "for COFF and MachO object files.\n"; 2497 } 2498 2499 /// Dump the raw contents of the __clangast section so the output can be piped 2500 /// into llvm-bcanalyzer. 2501 static void printRawClangAST(const ObjectFile *Obj) { 2502 if (outs().is_displayed()) { 2503 WithColor::error(errs(), ToolName) 2504 << "The -raw-clang-ast option will dump the raw binary contents of " 2505 "the clang ast section.\n" 2506 "Please redirect the output to a file or another program such as " 2507 "llvm-bcanalyzer.\n"; 2508 return; 2509 } 2510 2511 StringRef ClangASTSectionName("__clangast"); 2512 if (Obj->isCOFF()) { 2513 ClangASTSectionName = "clangast"; 2514 } 2515 2516 Optional<object::SectionRef> ClangASTSection; 2517 for (auto Sec : ToolSectionFilter(*Obj)) { 2518 StringRef Name; 2519 if (Expected<StringRef> NameOrErr = Sec.getName()) 2520 Name = *NameOrErr; 2521 else 2522 consumeError(NameOrErr.takeError()); 2523 2524 if (Name == ClangASTSectionName) { 2525 ClangASTSection = Sec; 2526 break; 2527 } 2528 } 2529 if (!ClangASTSection) 2530 return; 2531 2532 StringRef ClangASTContents = 2533 unwrapOrError(ClangASTSection.value().getContents(), Obj->getFileName()); 2534 outs().write(ClangASTContents.data(), ClangASTContents.size()); 2535 } 2536 2537 static void printFaultMaps(const ObjectFile *Obj) { 2538 StringRef FaultMapSectionName; 2539 2540 if (Obj->isELF()) { 2541 FaultMapSectionName = ".llvm_faultmaps"; 2542 } else if (Obj->isMachO()) { 2543 FaultMapSectionName = "__llvm_faultmaps"; 2544 } else { 2545 WithColor::error(errs(), ToolName) 2546 << "This operation is only currently supported " 2547 "for ELF and Mach-O executable files.\n"; 2548 return; 2549 } 2550 2551 Optional<object::SectionRef> FaultMapSection; 2552 2553 for (auto Sec : ToolSectionFilter(*Obj)) { 2554 StringRef Name; 2555 if (Expected<StringRef> NameOrErr = Sec.getName()) 2556 Name = *NameOrErr; 2557 else 2558 consumeError(NameOrErr.takeError()); 2559 2560 if (Name == FaultMapSectionName) { 2561 FaultMapSection = Sec; 2562 break; 2563 } 2564 } 2565 2566 outs() << "FaultMap table:\n"; 2567 2568 if (!FaultMapSection) { 2569 outs() << "<not found>\n"; 2570 return; 2571 } 2572 2573 StringRef FaultMapContents = 2574 unwrapOrError(FaultMapSection->getContents(), Obj->getFileName()); 2575 FaultMapParser FMP(FaultMapContents.bytes_begin(), 2576 FaultMapContents.bytes_end()); 2577 2578 outs() << FMP; 2579 } 2580 2581 static void printPrivateFileHeaders(const ObjectFile *O, bool OnlyFirst) { 2582 if (O->isELF()) { 2583 printELFFileHeader(O); 2584 printELFDynamicSection(O); 2585 printELFSymbolVersionInfo(O); 2586 return; 2587 } 2588 if (O->isCOFF()) 2589 return printCOFFFileHeader(cast<object::COFFObjectFile>(*O)); 2590 if (O->isWasm()) 2591 return printWasmFileHeader(O); 2592 if (O->isMachO()) { 2593 printMachOFileHeader(O); 2594 if (!OnlyFirst) 2595 printMachOLoadCommands(O); 2596 return; 2597 } 2598 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2599 } 2600 2601 static void printFileHeaders(const ObjectFile *O) { 2602 if (!O->isELF() && !O->isCOFF()) 2603 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2604 2605 Triple::ArchType AT = O->getArch(); 2606 outs() << "architecture: " << Triple::getArchTypeName(AT) << "\n"; 2607 uint64_t Address = unwrapOrError(O->getStartAddress(), O->getFileName()); 2608 2609 StringRef Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2610 outs() << "start address: " 2611 << "0x" << format(Fmt.data(), Address) << "\n"; 2612 } 2613 2614 static void printArchiveChild(StringRef Filename, const Archive::Child &C) { 2615 Expected<sys::fs::perms> ModeOrErr = C.getAccessMode(); 2616 if (!ModeOrErr) { 2617 WithColor::error(errs(), ToolName) << "ill-formed archive entry.\n"; 2618 consumeError(ModeOrErr.takeError()); 2619 return; 2620 } 2621 sys::fs::perms Mode = ModeOrErr.get(); 2622 outs() << ((Mode & sys::fs::owner_read) ? "r" : "-"); 2623 outs() << ((Mode & sys::fs::owner_write) ? "w" : "-"); 2624 outs() << ((Mode & sys::fs::owner_exe) ? "x" : "-"); 2625 outs() << ((Mode & sys::fs::group_read) ? "r" : "-"); 2626 outs() << ((Mode & sys::fs::group_write) ? "w" : "-"); 2627 outs() << ((Mode & sys::fs::group_exe) ? "x" : "-"); 2628 outs() << ((Mode & sys::fs::others_read) ? "r" : "-"); 2629 outs() << ((Mode & sys::fs::others_write) ? "w" : "-"); 2630 outs() << ((Mode & sys::fs::others_exe) ? "x" : "-"); 2631 2632 outs() << " "; 2633 2634 outs() << format("%d/%d %6" PRId64 " ", unwrapOrError(C.getUID(), Filename), 2635 unwrapOrError(C.getGID(), Filename), 2636 unwrapOrError(C.getRawSize(), Filename)); 2637 2638 StringRef RawLastModified = C.getRawLastModified(); 2639 unsigned Seconds; 2640 if (RawLastModified.getAsInteger(10, Seconds)) 2641 outs() << "(date: \"" << RawLastModified 2642 << "\" contains non-decimal chars) "; 2643 else { 2644 // Since ctime(3) returns a 26 character string of the form: 2645 // "Sun Sep 16 01:03:52 1973\n\0" 2646 // just print 24 characters. 2647 time_t t = Seconds; 2648 outs() << format("%.24s ", ctime(&t)); 2649 } 2650 2651 StringRef Name = ""; 2652 Expected<StringRef> NameOrErr = C.getName(); 2653 if (!NameOrErr) { 2654 consumeError(NameOrErr.takeError()); 2655 Name = unwrapOrError(C.getRawName(), Filename); 2656 } else { 2657 Name = NameOrErr.get(); 2658 } 2659 outs() << Name << "\n"; 2660 } 2661 2662 // For ELF only now. 2663 static bool shouldWarnForInvalidStartStopAddress(ObjectFile *Obj) { 2664 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj)) { 2665 if (Elf->getEType() != ELF::ET_REL) 2666 return true; 2667 } 2668 return false; 2669 } 2670 2671 static void checkForInvalidStartStopAddress(ObjectFile *Obj, 2672 uint64_t Start, uint64_t Stop) { 2673 if (!shouldWarnForInvalidStartStopAddress(Obj)) 2674 return; 2675 2676 for (const SectionRef &Section : Obj->sections()) 2677 if (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC) { 2678 uint64_t BaseAddr = Section.getAddress(); 2679 uint64_t Size = Section.getSize(); 2680 if ((Start < BaseAddr + Size) && Stop > BaseAddr) 2681 return; 2682 } 2683 2684 if (!HasStartAddressFlag) 2685 reportWarning("no section has address less than 0x" + 2686 Twine::utohexstr(Stop) + " specified by --stop-address", 2687 Obj->getFileName()); 2688 else if (!HasStopAddressFlag) 2689 reportWarning("no section has address greater than or equal to 0x" + 2690 Twine::utohexstr(Start) + " specified by --start-address", 2691 Obj->getFileName()); 2692 else 2693 reportWarning("no section overlaps the range [0x" + 2694 Twine::utohexstr(Start) + ",0x" + Twine::utohexstr(Stop) + 2695 ") specified by --start-address/--stop-address", 2696 Obj->getFileName()); 2697 } 2698 2699 static void dumpObject(ObjectFile *O, const Archive *A = nullptr, 2700 const Archive::Child *C = nullptr) { 2701 // Avoid other output when using a raw option. 2702 if (!RawClangAST) { 2703 outs() << '\n'; 2704 if (A) 2705 outs() << A->getFileName() << "(" << O->getFileName() << ")"; 2706 else 2707 outs() << O->getFileName(); 2708 outs() << ":\tfile format " << O->getFileFormatName().lower() << "\n"; 2709 } 2710 2711 if (HasStartAddressFlag || HasStopAddressFlag) 2712 checkForInvalidStartStopAddress(O, StartAddress, StopAddress); 2713 2714 // Note: the order here matches GNU objdump for compatability. 2715 StringRef ArchiveName = A ? A->getFileName() : ""; 2716 if (ArchiveHeaders && !MachOOpt && C) 2717 printArchiveChild(ArchiveName, *C); 2718 if (FileHeaders) 2719 printFileHeaders(O); 2720 if (PrivateHeaders || FirstPrivateHeader) 2721 printPrivateFileHeaders(O, FirstPrivateHeader); 2722 if (SectionHeaders) 2723 printSectionHeaders(*O); 2724 if (SymbolTable) 2725 printSymbolTable(*O, ArchiveName); 2726 if (DynamicSymbolTable) 2727 printSymbolTable(*O, ArchiveName, /*ArchitectureName=*/"", 2728 /*DumpDynamic=*/true); 2729 if (DwarfDumpType != DIDT_Null) { 2730 std::unique_ptr<DIContext> DICtx = DWARFContext::create(*O); 2731 // Dump the complete DWARF structure. 2732 DIDumpOptions DumpOpts; 2733 DumpOpts.DumpType = DwarfDumpType; 2734 DICtx->dump(outs(), DumpOpts); 2735 } 2736 if (Relocations && !Disassemble) 2737 printRelocations(O); 2738 if (DynamicRelocations) 2739 printDynamicRelocations(O); 2740 if (SectionContents) 2741 printSectionContents(O); 2742 if (Disassemble) 2743 disassembleObject(O, Relocations); 2744 if (UnwindInfo) 2745 printUnwindInfo(O); 2746 2747 // Mach-O specific options: 2748 if (ExportsTrie) 2749 printExportsTrie(O); 2750 if (Rebase) 2751 printRebaseTable(O); 2752 if (Bind) 2753 printBindTable(O); 2754 if (LazyBind) 2755 printLazyBindTable(O); 2756 if (WeakBind) 2757 printWeakBindTable(O); 2758 2759 // Other special sections: 2760 if (RawClangAST) 2761 printRawClangAST(O); 2762 if (FaultMapSection) 2763 printFaultMaps(O); 2764 if (Offloading) 2765 dumpOffloadBinary(*O); 2766 } 2767 2768 static void dumpObject(const COFFImportFile *I, const Archive *A, 2769 const Archive::Child *C = nullptr) { 2770 StringRef ArchiveName = A ? A->getFileName() : ""; 2771 2772 // Avoid other output when using a raw option. 2773 if (!RawClangAST) 2774 outs() << '\n' 2775 << ArchiveName << "(" << I->getFileName() << ")" 2776 << ":\tfile format COFF-import-file" 2777 << "\n\n"; 2778 2779 if (ArchiveHeaders && !MachOOpt && C) 2780 printArchiveChild(ArchiveName, *C); 2781 if (SymbolTable) 2782 printCOFFSymbolTable(*I); 2783 } 2784 2785 /// Dump each object file in \a a; 2786 static void dumpArchive(const Archive *A) { 2787 Error Err = Error::success(); 2788 unsigned I = -1; 2789 for (auto &C : A->children(Err)) { 2790 ++I; 2791 Expected<std::unique_ptr<Binary>> ChildOrErr = C.getAsBinary(); 2792 if (!ChildOrErr) { 2793 if (auto E = isNotObjectErrorInvalidFileType(ChildOrErr.takeError())) 2794 reportError(std::move(E), getFileNameForError(C, I), A->getFileName()); 2795 continue; 2796 } 2797 if (ObjectFile *O = dyn_cast<ObjectFile>(&*ChildOrErr.get())) 2798 dumpObject(O, A, &C); 2799 else if (COFFImportFile *I = dyn_cast<COFFImportFile>(&*ChildOrErr.get())) 2800 dumpObject(I, A, &C); 2801 else 2802 reportError(errorCodeToError(object_error::invalid_file_type), 2803 A->getFileName()); 2804 } 2805 if (Err) 2806 reportError(std::move(Err), A->getFileName()); 2807 } 2808 2809 /// Open file and figure out how to dump it. 2810 static void dumpInput(StringRef file) { 2811 // If we are using the Mach-O specific object file parser, then let it parse 2812 // the file and process the command line options. So the -arch flags can 2813 // be used to select specific slices, etc. 2814 if (MachOOpt) { 2815 parseInputMachO(file); 2816 return; 2817 } 2818 2819 // Attempt to open the binary. 2820 OwningBinary<Binary> OBinary = unwrapOrError(createBinary(file), file); 2821 Binary &Binary = *OBinary.getBinary(); 2822 2823 if (Archive *A = dyn_cast<Archive>(&Binary)) 2824 dumpArchive(A); 2825 else if (ObjectFile *O = dyn_cast<ObjectFile>(&Binary)) 2826 dumpObject(O); 2827 else if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(&Binary)) 2828 parseInputMachO(UB); 2829 else if (OffloadBinary *OB = dyn_cast<OffloadBinary>(&Binary)) 2830 dumpOffloadSections(*OB); 2831 else 2832 reportError(errorCodeToError(object_error::invalid_file_type), file); 2833 } 2834 2835 template <typename T> 2836 static void parseIntArg(const llvm::opt::InputArgList &InputArgs, int ID, 2837 T &Value) { 2838 if (const opt::Arg *A = InputArgs.getLastArg(ID)) { 2839 StringRef V(A->getValue()); 2840 if (!llvm::to_integer(V, Value, 0)) { 2841 reportCmdLineError(A->getSpelling() + 2842 ": expected a non-negative integer, but got '" + V + 2843 "'"); 2844 } 2845 } 2846 } 2847 2848 static void invalidArgValue(const opt::Arg *A) { 2849 reportCmdLineError("'" + StringRef(A->getValue()) + 2850 "' is not a valid value for '" + A->getSpelling() + "'"); 2851 } 2852 2853 static std::vector<std::string> 2854 commaSeparatedValues(const llvm::opt::InputArgList &InputArgs, int ID) { 2855 std::vector<std::string> Values; 2856 for (StringRef Value : InputArgs.getAllArgValues(ID)) { 2857 llvm::SmallVector<StringRef, 2> SplitValues; 2858 llvm::SplitString(Value, SplitValues, ","); 2859 for (StringRef SplitValue : SplitValues) 2860 Values.push_back(SplitValue.str()); 2861 } 2862 return Values; 2863 } 2864 2865 static void parseOtoolOptions(const llvm::opt::InputArgList &InputArgs) { 2866 MachOOpt = true; 2867 FullLeadingAddr = true; 2868 PrintImmHex = true; 2869 2870 ArchName = InputArgs.getLastArgValue(OTOOL_arch).str(); 2871 LinkOptHints = InputArgs.hasArg(OTOOL_C); 2872 if (InputArgs.hasArg(OTOOL_d)) 2873 FilterSections.push_back("__DATA,__data"); 2874 DylibId = InputArgs.hasArg(OTOOL_D); 2875 UniversalHeaders = InputArgs.hasArg(OTOOL_f); 2876 DataInCode = InputArgs.hasArg(OTOOL_G); 2877 FirstPrivateHeader = InputArgs.hasArg(OTOOL_h); 2878 IndirectSymbols = InputArgs.hasArg(OTOOL_I); 2879 ShowRawInsn = InputArgs.hasArg(OTOOL_j); 2880 PrivateHeaders = InputArgs.hasArg(OTOOL_l); 2881 DylibsUsed = InputArgs.hasArg(OTOOL_L); 2882 MCPU = InputArgs.getLastArgValue(OTOOL_mcpu_EQ).str(); 2883 ObjcMetaData = InputArgs.hasArg(OTOOL_o); 2884 DisSymName = InputArgs.getLastArgValue(OTOOL_p).str(); 2885 InfoPlist = InputArgs.hasArg(OTOOL_P); 2886 Relocations = InputArgs.hasArg(OTOOL_r); 2887 if (const Arg *A = InputArgs.getLastArg(OTOOL_s)) { 2888 auto Filter = (A->getValue(0) + StringRef(",") + A->getValue(1)).str(); 2889 FilterSections.push_back(Filter); 2890 } 2891 if (InputArgs.hasArg(OTOOL_t)) 2892 FilterSections.push_back("__TEXT,__text"); 2893 Verbose = InputArgs.hasArg(OTOOL_v) || InputArgs.hasArg(OTOOL_V) || 2894 InputArgs.hasArg(OTOOL_o); 2895 SymbolicOperands = InputArgs.hasArg(OTOOL_V); 2896 if (InputArgs.hasArg(OTOOL_x)) 2897 FilterSections.push_back(",__text"); 2898 LeadingAddr = LeadingHeaders = !InputArgs.hasArg(OTOOL_X); 2899 2900 ChainedFixups = InputArgs.hasArg(OTOOL_chained_fixups); 2901 DyldInfo = InputArgs.hasArg(OTOOL_dyld_info); 2902 2903 InputFilenames = InputArgs.getAllArgValues(OTOOL_INPUT); 2904 if (InputFilenames.empty()) 2905 reportCmdLineError("no input file"); 2906 2907 for (const Arg *A : InputArgs) { 2908 const Option &O = A->getOption(); 2909 if (O.getGroup().isValid() && O.getGroup().getID() == OTOOL_grp_obsolete) { 2910 reportCmdLineWarning(O.getPrefixedName() + 2911 " is obsolete and not implemented"); 2912 } 2913 } 2914 } 2915 2916 static void parseObjdumpOptions(const llvm::opt::InputArgList &InputArgs) { 2917 parseIntArg(InputArgs, OBJDUMP_adjust_vma_EQ, AdjustVMA); 2918 AllHeaders = InputArgs.hasArg(OBJDUMP_all_headers); 2919 ArchName = InputArgs.getLastArgValue(OBJDUMP_arch_name_EQ).str(); 2920 ArchiveHeaders = InputArgs.hasArg(OBJDUMP_archive_headers); 2921 Demangle = InputArgs.hasArg(OBJDUMP_demangle); 2922 Disassemble = InputArgs.hasArg(OBJDUMP_disassemble); 2923 DisassembleAll = InputArgs.hasArg(OBJDUMP_disassemble_all); 2924 SymbolDescription = InputArgs.hasArg(OBJDUMP_symbol_description); 2925 DisassembleSymbols = 2926 commaSeparatedValues(InputArgs, OBJDUMP_disassemble_symbols_EQ); 2927 DisassembleZeroes = InputArgs.hasArg(OBJDUMP_disassemble_zeroes); 2928 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_dwarf_EQ)) { 2929 DwarfDumpType = StringSwitch<DIDumpType>(A->getValue()) 2930 .Case("frames", DIDT_DebugFrame) 2931 .Default(DIDT_Null); 2932 if (DwarfDumpType == DIDT_Null) 2933 invalidArgValue(A); 2934 } 2935 DynamicRelocations = InputArgs.hasArg(OBJDUMP_dynamic_reloc); 2936 FaultMapSection = InputArgs.hasArg(OBJDUMP_fault_map_section); 2937 Offloading = InputArgs.hasArg(OBJDUMP_offloading); 2938 FileHeaders = InputArgs.hasArg(OBJDUMP_file_headers); 2939 SectionContents = InputArgs.hasArg(OBJDUMP_full_contents); 2940 PrintLines = InputArgs.hasArg(OBJDUMP_line_numbers); 2941 InputFilenames = InputArgs.getAllArgValues(OBJDUMP_INPUT); 2942 MachOOpt = InputArgs.hasArg(OBJDUMP_macho); 2943 MCPU = InputArgs.getLastArgValue(OBJDUMP_mcpu_EQ).str(); 2944 MAttrs = commaSeparatedValues(InputArgs, OBJDUMP_mattr_EQ); 2945 ShowRawInsn = !InputArgs.hasArg(OBJDUMP_no_show_raw_insn); 2946 LeadingAddr = !InputArgs.hasArg(OBJDUMP_no_leading_addr); 2947 RawClangAST = InputArgs.hasArg(OBJDUMP_raw_clang_ast); 2948 Relocations = InputArgs.hasArg(OBJDUMP_reloc); 2949 PrintImmHex = 2950 InputArgs.hasFlag(OBJDUMP_print_imm_hex, OBJDUMP_no_print_imm_hex, false); 2951 PrivateHeaders = InputArgs.hasArg(OBJDUMP_private_headers); 2952 FilterSections = InputArgs.getAllArgValues(OBJDUMP_section_EQ); 2953 SectionHeaders = InputArgs.hasArg(OBJDUMP_section_headers); 2954 ShowAllSymbols = InputArgs.hasArg(OBJDUMP_show_all_symbols); 2955 ShowLMA = InputArgs.hasArg(OBJDUMP_show_lma); 2956 PrintSource = InputArgs.hasArg(OBJDUMP_source); 2957 parseIntArg(InputArgs, OBJDUMP_start_address_EQ, StartAddress); 2958 HasStartAddressFlag = InputArgs.hasArg(OBJDUMP_start_address_EQ); 2959 parseIntArg(InputArgs, OBJDUMP_stop_address_EQ, StopAddress); 2960 HasStopAddressFlag = InputArgs.hasArg(OBJDUMP_stop_address_EQ); 2961 SymbolTable = InputArgs.hasArg(OBJDUMP_syms); 2962 SymbolizeOperands = InputArgs.hasArg(OBJDUMP_symbolize_operands); 2963 DynamicSymbolTable = InputArgs.hasArg(OBJDUMP_dynamic_syms); 2964 TripleName = InputArgs.getLastArgValue(OBJDUMP_triple_EQ).str(); 2965 UnwindInfo = InputArgs.hasArg(OBJDUMP_unwind_info); 2966 Wide = InputArgs.hasArg(OBJDUMP_wide); 2967 Prefix = InputArgs.getLastArgValue(OBJDUMP_prefix).str(); 2968 parseIntArg(InputArgs, OBJDUMP_prefix_strip, PrefixStrip); 2969 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_debug_vars_EQ)) { 2970 DbgVariables = StringSwitch<DebugVarsFormat>(A->getValue()) 2971 .Case("ascii", DVASCII) 2972 .Case("unicode", DVUnicode) 2973 .Default(DVInvalid); 2974 if (DbgVariables == DVInvalid) 2975 invalidArgValue(A); 2976 } 2977 parseIntArg(InputArgs, OBJDUMP_debug_vars_indent_EQ, DbgIndent); 2978 2979 parseMachOOptions(InputArgs); 2980 2981 // Parse -M (--disassembler-options) and deprecated 2982 // --x86-asm-syntax={att,intel}. 2983 // 2984 // Note, for x86, the asm dialect (AssemblerDialect) is initialized when the 2985 // MCAsmInfo is constructed. MCInstPrinter::applyTargetSpecificCLOption is 2986 // called too late. For now we have to use the internal cl::opt option. 2987 const char *AsmSyntax = nullptr; 2988 for (const auto *A : InputArgs.filtered(OBJDUMP_disassembler_options_EQ, 2989 OBJDUMP_x86_asm_syntax_att, 2990 OBJDUMP_x86_asm_syntax_intel)) { 2991 switch (A->getOption().getID()) { 2992 case OBJDUMP_x86_asm_syntax_att: 2993 AsmSyntax = "--x86-asm-syntax=att"; 2994 continue; 2995 case OBJDUMP_x86_asm_syntax_intel: 2996 AsmSyntax = "--x86-asm-syntax=intel"; 2997 continue; 2998 } 2999 3000 SmallVector<StringRef, 2> Values; 3001 llvm::SplitString(A->getValue(), Values, ","); 3002 for (StringRef V : Values) { 3003 if (V == "att") 3004 AsmSyntax = "--x86-asm-syntax=att"; 3005 else if (V == "intel") 3006 AsmSyntax = "--x86-asm-syntax=intel"; 3007 else 3008 DisassemblerOptions.push_back(V.str()); 3009 } 3010 } 3011 if (AsmSyntax) { 3012 const char *Argv[] = {"llvm-objdump", AsmSyntax}; 3013 llvm::cl::ParseCommandLineOptions(2, Argv); 3014 } 3015 3016 // objdump defaults to a.out if no filenames specified. 3017 if (InputFilenames.empty()) 3018 InputFilenames.push_back("a.out"); 3019 } 3020 3021 int main(int argc, char **argv) { 3022 using namespace llvm; 3023 InitLLVM X(argc, argv); 3024 3025 ToolName = argv[0]; 3026 std::unique_ptr<CommonOptTable> T; 3027 OptSpecifier Unknown, HelpFlag, HelpHiddenFlag, VersionFlag; 3028 3029 StringRef Stem = sys::path::stem(ToolName); 3030 auto Is = [=](StringRef Tool) { 3031 // We need to recognize the following filenames: 3032 // 3033 // llvm-objdump -> objdump 3034 // llvm-otool-10.exe -> otool 3035 // powerpc64-unknown-freebsd13-objdump -> objdump 3036 auto I = Stem.rfind_insensitive(Tool); 3037 return I != StringRef::npos && 3038 (I + Tool.size() == Stem.size() || !isAlnum(Stem[I + Tool.size()])); 3039 }; 3040 if (Is("otool")) { 3041 T = std::make_unique<OtoolOptTable>(); 3042 Unknown = OTOOL_UNKNOWN; 3043 HelpFlag = OTOOL_help; 3044 HelpHiddenFlag = OTOOL_help_hidden; 3045 VersionFlag = OTOOL_version; 3046 } else { 3047 T = std::make_unique<ObjdumpOptTable>(); 3048 Unknown = OBJDUMP_UNKNOWN; 3049 HelpFlag = OBJDUMP_help; 3050 HelpHiddenFlag = OBJDUMP_help_hidden; 3051 VersionFlag = OBJDUMP_version; 3052 } 3053 3054 BumpPtrAllocator A; 3055 StringSaver Saver(A); 3056 opt::InputArgList InputArgs = 3057 T->parseArgs(argc, argv, Unknown, Saver, 3058 [&](StringRef Msg) { reportCmdLineError(Msg); }); 3059 3060 if (InputArgs.size() == 0 || InputArgs.hasArg(HelpFlag)) { 3061 T->printHelp(ToolName); 3062 return 0; 3063 } 3064 if (InputArgs.hasArg(HelpHiddenFlag)) { 3065 T->printHelp(ToolName, /*ShowHidden=*/true); 3066 return 0; 3067 } 3068 3069 // Initialize targets and assembly printers/parsers. 3070 InitializeAllTargetInfos(); 3071 InitializeAllTargetMCs(); 3072 InitializeAllDisassemblers(); 3073 3074 if (InputArgs.hasArg(VersionFlag)) { 3075 cl::PrintVersionMessage(); 3076 if (!Is("otool")) { 3077 outs() << '\n'; 3078 TargetRegistry::printRegisteredTargetsForVersion(outs()); 3079 } 3080 return 0; 3081 } 3082 3083 if (Is("otool")) 3084 parseOtoolOptions(InputArgs); 3085 else 3086 parseObjdumpOptions(InputArgs); 3087 3088 if (StartAddress >= StopAddress) 3089 reportCmdLineError("start address should be less than stop address"); 3090 3091 // Removes trailing separators from prefix. 3092 while (!Prefix.empty() && sys::path::is_separator(Prefix.back())) 3093 Prefix.pop_back(); 3094 3095 if (AllHeaders) 3096 ArchiveHeaders = FileHeaders = PrivateHeaders = Relocations = 3097 SectionHeaders = SymbolTable = true; 3098 3099 if (DisassembleAll || PrintSource || PrintLines || 3100 !DisassembleSymbols.empty()) 3101 Disassemble = true; 3102 3103 if (!ArchiveHeaders && !Disassemble && DwarfDumpType == DIDT_Null && 3104 !DynamicRelocations && !FileHeaders && !PrivateHeaders && !RawClangAST && 3105 !Relocations && !SectionHeaders && !SectionContents && !SymbolTable && 3106 !DynamicSymbolTable && !UnwindInfo && !FaultMapSection && !Offloading && 3107 !(MachOOpt && 3108 (Bind || DataInCode || ChainedFixups || DyldInfo || DylibId || 3109 DylibsUsed || ExportsTrie || FirstPrivateHeader || FunctionStarts || 3110 IndirectSymbols || InfoPlist || LazyBind || LinkOptHints || 3111 ObjcMetaData || Rebase || Rpaths || UniversalHeaders || WeakBind || 3112 !FilterSections.empty()))) { 3113 T->printHelp(ToolName); 3114 return 2; 3115 } 3116 3117 DisasmSymbolSet.insert(DisassembleSymbols.begin(), DisassembleSymbols.end()); 3118 3119 llvm::for_each(InputFilenames, dumpInput); 3120 3121 warnOnNoMatchForSections(); 3122 3123 return EXIT_SUCCESS; 3124 } 3125