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/STLExtras.h" 28 #include "llvm/ADT/SetOperations.h" 29 #include "llvm/ADT/StringExtras.h" 30 #include "llvm/ADT/Twine.h" 31 #include "llvm/BinaryFormat/Wasm.h" 32 #include "llvm/DebugInfo/BTF/BTFParser.h" 33 #include "llvm/DebugInfo/DWARF/DWARFContext.h" 34 #include "llvm/DebugInfo/Symbolize/Symbolize.h" 35 #include "llvm/Debuginfod/BuildIDFetcher.h" 36 #include "llvm/Debuginfod/Debuginfod.h" 37 #include "llvm/Debuginfod/HTTPClient.h" 38 #include "llvm/Demangle/Demangle.h" 39 #include "llvm/MC/MCAsmInfo.h" 40 #include "llvm/MC/MCContext.h" 41 #include "llvm/MC/MCDisassembler/MCRelocationInfo.h" 42 #include "llvm/MC/MCInst.h" 43 #include "llvm/MC/MCInstPrinter.h" 44 #include "llvm/MC/MCInstrAnalysis.h" 45 #include "llvm/MC/MCInstrInfo.h" 46 #include "llvm/MC/MCObjectFileInfo.h" 47 #include "llvm/MC/MCRegisterInfo.h" 48 #include "llvm/MC/MCTargetOptions.h" 49 #include "llvm/MC/TargetRegistry.h" 50 #include "llvm/Object/BuildID.h" 51 #include "llvm/Object/COFF.h" 52 #include "llvm/Object/COFFImportFile.h" 53 #include "llvm/Object/ELFObjectFile.h" 54 #include "llvm/Object/ELFTypes.h" 55 #include "llvm/Object/FaultMapParser.h" 56 #include "llvm/Object/MachO.h" 57 #include "llvm/Object/MachOUniversal.h" 58 #include "llvm/Object/OffloadBinary.h" 59 #include "llvm/Object/Wasm.h" 60 #include "llvm/Option/Arg.h" 61 #include "llvm/Option/ArgList.h" 62 #include "llvm/Option/Option.h" 63 #include "llvm/Support/Casting.h" 64 #include "llvm/Support/Debug.h" 65 #include "llvm/Support/Errc.h" 66 #include "llvm/Support/FileSystem.h" 67 #include "llvm/Support/Format.h" 68 #include "llvm/Support/FormatVariadic.h" 69 #include "llvm/Support/GraphWriter.h" 70 #include "llvm/Support/LLVMDriver.h" 71 #include "llvm/Support/MemoryBuffer.h" 72 #include "llvm/Support/SourceMgr.h" 73 #include "llvm/Support/StringSaver.h" 74 #include "llvm/Support/TargetSelect.h" 75 #include "llvm/Support/WithColor.h" 76 #include "llvm/Support/raw_ostream.h" 77 #include "llvm/TargetParser/Host.h" 78 #include "llvm/TargetParser/Triple.h" 79 #include <algorithm> 80 #include <cctype> 81 #include <cstring> 82 #include <optional> 83 #include <set> 84 #include <system_error> 85 #include <unordered_map> 86 #include <utility> 87 88 using namespace llvm; 89 using namespace llvm::object; 90 using namespace llvm::objdump; 91 using namespace llvm::opt; 92 93 namespace { 94 95 class CommonOptTable : public opt::GenericOptTable { 96 public: 97 CommonOptTable(const StringTable &StrTable, 98 ArrayRef<StringTable::Offset> PrefixesTable, 99 ArrayRef<Info> OptionInfos, const char *Usage, 100 const char *Description) 101 : opt::GenericOptTable(StrTable, PrefixesTable, OptionInfos), 102 Usage(Usage), Description(Description) { 103 setGroupedShortOptions(true); 104 } 105 106 void printHelp(StringRef Argv0, bool ShowHidden = false) const { 107 Argv0 = sys::path::filename(Argv0); 108 opt::GenericOptTable::printHelp(outs(), (Argv0 + Usage).str().c_str(), 109 Description, ShowHidden, ShowHidden); 110 // TODO Replace this with OptTable API once it adds extrahelp support. 111 outs() << "\nPass @FILE as argument to read options from FILE.\n"; 112 } 113 114 private: 115 const char *Usage; 116 const char *Description; 117 }; 118 119 // ObjdumpOptID is in ObjdumpOptID.h 120 namespace objdump_opt { 121 #define OPTTABLE_STR_TABLE_CODE 122 #include "ObjdumpOpts.inc" 123 #undef OPTTABLE_STR_TABLE_CODE 124 125 #define OPTTABLE_PREFIXES_TABLE_CODE 126 #include "ObjdumpOpts.inc" 127 #undef OPTTABLE_PREFIXES_TABLE_CODE 128 129 static constexpr opt::OptTable::Info ObjdumpInfoTable[] = { 130 #define OPTION(...) \ 131 LLVM_CONSTRUCT_OPT_INFO_WITH_ID_PREFIX(OBJDUMP_, __VA_ARGS__), 132 #include "ObjdumpOpts.inc" 133 #undef OPTION 134 }; 135 } // namespace objdump_opt 136 137 class ObjdumpOptTable : public CommonOptTable { 138 public: 139 ObjdumpOptTable() 140 : CommonOptTable( 141 objdump_opt::OptionStrTable, objdump_opt::OptionPrefixesTable, 142 objdump_opt::ObjdumpInfoTable, " [options] <input object files>", 143 "llvm object file dumper") {} 144 }; 145 146 enum OtoolOptID { 147 OTOOL_INVALID = 0, // This is not an option ID. 148 #define OPTION(...) LLVM_MAKE_OPT_ID_WITH_ID_PREFIX(OTOOL_, __VA_ARGS__), 149 #include "OtoolOpts.inc" 150 #undef OPTION 151 }; 152 153 namespace otool { 154 #define OPTTABLE_STR_TABLE_CODE 155 #include "OtoolOpts.inc" 156 #undef OPTTABLE_STR_TABLE_CODE 157 158 #define OPTTABLE_PREFIXES_TABLE_CODE 159 #include "OtoolOpts.inc" 160 #undef OPTTABLE_PREFIXES_TABLE_CODE 161 162 static constexpr opt::OptTable::Info OtoolInfoTable[] = { 163 #define OPTION(...) LLVM_CONSTRUCT_OPT_INFO_WITH_ID_PREFIX(OTOOL_, __VA_ARGS__), 164 #include "OtoolOpts.inc" 165 #undef OPTION 166 }; 167 } // namespace otool 168 169 class OtoolOptTable : public CommonOptTable { 170 public: 171 OtoolOptTable() 172 : CommonOptTable(otool::OptionStrTable, otool::OptionPrefixesTable, 173 otool::OtoolInfoTable, " [option...] [file...]", 174 "Mach-O object file displaying tool") {} 175 }; 176 177 struct BBAddrMapLabel { 178 std::string BlockLabel; 179 std::string PGOAnalysis; 180 }; 181 182 // This class represents the BBAddrMap and PGOMap associated with a single 183 // function. 184 class BBAddrMapFunctionEntry { 185 public: 186 BBAddrMapFunctionEntry(BBAddrMap AddrMap, PGOAnalysisMap PGOMap) 187 : AddrMap(std::move(AddrMap)), PGOMap(std::move(PGOMap)) {} 188 189 const BBAddrMap &getAddrMap() const { return AddrMap; } 190 191 // Returns the PGO string associated with the entry of index `PGOBBEntryIndex` 192 // in `PGOMap`. If PrettyPGOAnalysis is true, prints BFI as relative frequency 193 // and BPI as percentage. Otherwise raw values are displayed. 194 std::string constructPGOLabelString(size_t PGOBBEntryIndex, 195 bool PrettyPGOAnalysis) const { 196 if (!PGOMap.FeatEnable.hasPGOAnalysis()) 197 return ""; 198 std::string PGOString; 199 raw_string_ostream PGOSS(PGOString); 200 201 PGOSS << " ("; 202 if (PGOMap.FeatEnable.FuncEntryCount && PGOBBEntryIndex == 0) { 203 PGOSS << "Entry count: " << Twine(PGOMap.FuncEntryCount); 204 if (PGOMap.FeatEnable.hasPGOAnalysisBBData()) { 205 PGOSS << ", "; 206 } 207 } 208 209 if (PGOMap.FeatEnable.hasPGOAnalysisBBData()) { 210 211 assert(PGOBBEntryIndex < PGOMap.BBEntries.size() && 212 "Expected PGOAnalysisMap and BBAddrMap to have the same entries"); 213 const PGOAnalysisMap::PGOBBEntry &PGOBBEntry = 214 PGOMap.BBEntries[PGOBBEntryIndex]; 215 216 if (PGOMap.FeatEnable.BBFreq) { 217 PGOSS << "Frequency: "; 218 if (PrettyPGOAnalysis) 219 printRelativeBlockFreq(PGOSS, PGOMap.BBEntries.front().BlockFreq, 220 PGOBBEntry.BlockFreq); 221 else 222 PGOSS << Twine(PGOBBEntry.BlockFreq.getFrequency()); 223 if (PGOMap.FeatEnable.BrProb && PGOBBEntry.Successors.size() > 0) { 224 PGOSS << ", "; 225 } 226 } 227 if (PGOMap.FeatEnable.BrProb && PGOBBEntry.Successors.size() > 0) { 228 PGOSS << "Successors: "; 229 interleaveComma( 230 PGOBBEntry.Successors, PGOSS, 231 [&](const PGOAnalysisMap::PGOBBEntry::SuccessorEntry &SE) { 232 PGOSS << "BB" << SE.ID << ":"; 233 if (PrettyPGOAnalysis) 234 PGOSS << "[" << SE.Prob << "]"; 235 else 236 PGOSS.write_hex(SE.Prob.getNumerator()); 237 }); 238 } 239 } 240 PGOSS << ")"; 241 242 return PGOString; 243 } 244 245 private: 246 const BBAddrMap AddrMap; 247 const PGOAnalysisMap PGOMap; 248 }; 249 250 // This class represents the BBAddrMap and PGOMap of potentially multiple 251 // functions in a section. 252 class BBAddrMapInfo { 253 public: 254 void clear() { 255 FunctionAddrToMap.clear(); 256 RangeBaseAddrToFunctionAddr.clear(); 257 } 258 259 bool empty() const { return FunctionAddrToMap.empty(); } 260 261 void AddFunctionEntry(BBAddrMap AddrMap, PGOAnalysisMap PGOMap) { 262 uint64_t FunctionAddr = AddrMap.getFunctionAddress(); 263 for (size_t I = 1; I < AddrMap.BBRanges.size(); ++I) 264 RangeBaseAddrToFunctionAddr.emplace(AddrMap.BBRanges[I].BaseAddress, 265 FunctionAddr); 266 [[maybe_unused]] auto R = FunctionAddrToMap.try_emplace( 267 FunctionAddr, std::move(AddrMap), std::move(PGOMap)); 268 assert(R.second && "duplicate function address"); 269 } 270 271 // Returns the BBAddrMap entry for the function associated with `BaseAddress`. 272 // `BaseAddress` could be the function address or the address of a range 273 // associated with that function. Returns `nullptr` if `BaseAddress` is not 274 // mapped to any entry. 275 const BBAddrMapFunctionEntry *getEntryForAddress(uint64_t BaseAddress) const { 276 uint64_t FunctionAddr = BaseAddress; 277 auto S = RangeBaseAddrToFunctionAddr.find(BaseAddress); 278 if (S != RangeBaseAddrToFunctionAddr.end()) 279 FunctionAddr = S->second; 280 auto R = FunctionAddrToMap.find(FunctionAddr); 281 if (R == FunctionAddrToMap.end()) 282 return nullptr; 283 return &R->second; 284 } 285 286 private: 287 std::unordered_map<uint64_t, BBAddrMapFunctionEntry> FunctionAddrToMap; 288 std::unordered_map<uint64_t, uint64_t> RangeBaseAddrToFunctionAddr; 289 }; 290 291 } // namespace 292 293 #define DEBUG_TYPE "objdump" 294 295 enum class ColorOutput { 296 Auto, 297 Enable, 298 Disable, 299 Invalid, 300 }; 301 302 static uint64_t AdjustVMA; 303 static bool AllHeaders; 304 static std::string ArchName; 305 bool objdump::ArchiveHeaders; 306 bool objdump::Demangle; 307 bool objdump::Disassemble; 308 bool objdump::DisassembleAll; 309 std::vector<std::string> objdump::DisassemblerOptions; 310 bool objdump::SymbolDescription; 311 bool objdump::TracebackTable; 312 static std::vector<std::string> DisassembleSymbols; 313 static bool DisassembleZeroes; 314 static ColorOutput DisassemblyColor; 315 DIDumpType objdump::DwarfDumpType; 316 static bool DynamicRelocations; 317 static bool FaultMapSection; 318 static bool FileHeaders; 319 bool objdump::SectionContents; 320 static std::vector<std::string> InputFilenames; 321 bool objdump::PrintLines; 322 static bool MachOOpt; 323 std::string objdump::MCPU; 324 std::vector<std::string> objdump::MAttrs; 325 bool objdump::ShowRawInsn; 326 bool objdump::LeadingAddr; 327 static bool Offloading; 328 static bool RawClangAST; 329 bool objdump::Relocations; 330 bool objdump::PrintImmHex; 331 bool objdump::PrivateHeaders; 332 std::vector<std::string> objdump::FilterSections; 333 bool objdump::SectionHeaders; 334 static bool ShowAllSymbols; 335 static bool ShowLMA; 336 bool objdump::PrintSource; 337 338 static uint64_t StartAddress; 339 static bool HasStartAddressFlag; 340 static uint64_t StopAddress = UINT64_MAX; 341 static bool HasStopAddressFlag; 342 343 bool objdump::SymbolTable; 344 static bool SymbolizeOperands; 345 static bool PrettyPGOAnalysisMap; 346 static bool DynamicSymbolTable; 347 std::string objdump::TripleName; 348 bool objdump::UnwindInfo; 349 static bool Wide; 350 std::string objdump::Prefix; 351 uint32_t objdump::PrefixStrip; 352 353 DebugVarsFormat objdump::DbgVariables = DVDisabled; 354 355 int objdump::DbgIndent = 52; 356 357 static StringSet<> DisasmSymbolSet; 358 StringSet<> objdump::FoundSectionSet; 359 static StringRef ToolName; 360 361 std::unique_ptr<BuildIDFetcher> BIDFetcher; 362 363 Dumper::Dumper(const object::ObjectFile &O) : O(O) { 364 WarningHandler = [this](const Twine &Msg) { 365 if (Warnings.insert(Msg.str()).second) 366 reportWarning(Msg, this->O.getFileName()); 367 return Error::success(); 368 }; 369 } 370 371 void Dumper::reportUniqueWarning(Error Err) { 372 reportUniqueWarning(toString(std::move(Err))); 373 } 374 375 void Dumper::reportUniqueWarning(const Twine &Msg) { 376 cantFail(WarningHandler(Msg)); 377 } 378 379 static Expected<std::unique_ptr<Dumper>> createDumper(const ObjectFile &Obj) { 380 if (const auto *O = dyn_cast<COFFObjectFile>(&Obj)) 381 return createCOFFDumper(*O); 382 if (const auto *O = dyn_cast<ELFObjectFileBase>(&Obj)) 383 return createELFDumper(*O); 384 if (const auto *O = dyn_cast<MachOObjectFile>(&Obj)) 385 return createMachODumper(*O); 386 if (const auto *O = dyn_cast<WasmObjectFile>(&Obj)) 387 return createWasmDumper(*O); 388 if (const auto *O = dyn_cast<XCOFFObjectFile>(&Obj)) 389 return createXCOFFDumper(*O); 390 391 return createStringError(errc::invalid_argument, 392 "unsupported object file format"); 393 } 394 395 namespace { 396 struct FilterResult { 397 // True if the section should not be skipped. 398 bool Keep; 399 400 // True if the index counter should be incremented, even if the section should 401 // be skipped. For example, sections may be skipped if they are not included 402 // in the --section flag, but we still want those to count toward the section 403 // count. 404 bool IncrementIndex; 405 }; 406 } // namespace 407 408 static FilterResult checkSectionFilter(object::SectionRef S) { 409 if (FilterSections.empty()) 410 return {/*Keep=*/true, /*IncrementIndex=*/true}; 411 412 Expected<StringRef> SecNameOrErr = S.getName(); 413 if (!SecNameOrErr) { 414 consumeError(SecNameOrErr.takeError()); 415 return {/*Keep=*/false, /*IncrementIndex=*/false}; 416 } 417 StringRef SecName = *SecNameOrErr; 418 419 // StringSet does not allow empty key so avoid adding sections with 420 // no name (such as the section with index 0) here. 421 if (!SecName.empty()) 422 FoundSectionSet.insert(SecName); 423 424 // Only show the section if it's in the FilterSections list, but always 425 // increment so the indexing is stable. 426 return {/*Keep=*/is_contained(FilterSections, SecName), 427 /*IncrementIndex=*/true}; 428 } 429 430 SectionFilter objdump::ToolSectionFilter(object::ObjectFile const &O, 431 uint64_t *Idx) { 432 // Start at UINT64_MAX so that the first index returned after an increment is 433 // zero (after the unsigned wrap). 434 if (Idx) 435 *Idx = UINT64_MAX; 436 return SectionFilter( 437 [Idx](object::SectionRef S) { 438 FilterResult Result = checkSectionFilter(S); 439 if (Idx != nullptr && Result.IncrementIndex) 440 *Idx += 1; 441 return Result.Keep; 442 }, 443 O); 444 } 445 446 std::string objdump::getFileNameForError(const object::Archive::Child &C, 447 unsigned Index) { 448 Expected<StringRef> NameOrErr = C.getName(); 449 if (NameOrErr) 450 return std::string(NameOrErr.get()); 451 // If we have an error getting the name then we print the index of the archive 452 // member. Since we are already in an error state, we just ignore this error. 453 consumeError(NameOrErr.takeError()); 454 return "<file index: " + std::to_string(Index) + ">"; 455 } 456 457 void objdump::reportWarning(const Twine &Message, StringRef File) { 458 // Output order between errs() and outs() matters especially for archive 459 // files where the output is per member object. 460 outs().flush(); 461 WithColor::warning(errs(), ToolName) 462 << "'" << File << "': " << Message << "\n"; 463 } 464 465 [[noreturn]] void objdump::reportError(StringRef File, const Twine &Message) { 466 outs().flush(); 467 WithColor::error(errs(), ToolName) << "'" << File << "': " << Message << "\n"; 468 exit(1); 469 } 470 471 [[noreturn]] void objdump::reportError(Error E, StringRef FileName, 472 StringRef ArchiveName, 473 StringRef ArchitectureName) { 474 assert(E); 475 outs().flush(); 476 WithColor::error(errs(), ToolName); 477 if (ArchiveName != "") 478 errs() << ArchiveName << "(" << FileName << ")"; 479 else 480 errs() << "'" << FileName << "'"; 481 if (!ArchitectureName.empty()) 482 errs() << " (for architecture " << ArchitectureName << ")"; 483 errs() << ": "; 484 logAllUnhandledErrors(std::move(E), errs()); 485 exit(1); 486 } 487 488 static void reportCmdLineWarning(const Twine &Message) { 489 WithColor::warning(errs(), ToolName) << Message << "\n"; 490 } 491 492 [[noreturn]] static void reportCmdLineError(const Twine &Message) { 493 WithColor::error(errs(), ToolName) << Message << "\n"; 494 exit(1); 495 } 496 497 static void warnOnNoMatchForSections() { 498 SetVector<StringRef> MissingSections; 499 for (StringRef S : FilterSections) { 500 if (FoundSectionSet.count(S)) 501 return; 502 // User may specify a unnamed section. Don't warn for it. 503 if (!S.empty()) 504 MissingSections.insert(S); 505 } 506 507 // Warn only if no section in FilterSections is matched. 508 for (StringRef S : MissingSections) 509 reportCmdLineWarning("section '" + S + 510 "' mentioned in a -j/--section option, but not " 511 "found in any input file"); 512 } 513 514 static const Target *getTarget(const ObjectFile *Obj) { 515 // Figure out the target triple. 516 Triple TheTriple("unknown-unknown-unknown"); 517 if (TripleName.empty()) { 518 TheTriple = Obj->makeTriple(); 519 } else { 520 TheTriple.setTriple(Triple::normalize(TripleName)); 521 auto Arch = Obj->getArch(); 522 if (Arch == Triple::arm || Arch == Triple::armeb) 523 Obj->setARMSubArch(TheTriple); 524 } 525 526 // Get the target specific parser. 527 std::string Error; 528 const Target *TheTarget = TargetRegistry::lookupTarget(ArchName, TheTriple, 529 Error); 530 if (!TheTarget) 531 reportError(Obj->getFileName(), "can't find target: " + Error); 532 533 // Update the triple name and return the found target. 534 TripleName = TheTriple.getTriple(); 535 return TheTarget; 536 } 537 538 bool objdump::isRelocAddressLess(RelocationRef A, RelocationRef B) { 539 return A.getOffset() < B.getOffset(); 540 } 541 542 static Error getRelocationValueString(const RelocationRef &Rel, 543 bool SymbolDescription, 544 SmallVectorImpl<char> &Result) { 545 const ObjectFile *Obj = Rel.getObject(); 546 if (auto *ELF = dyn_cast<ELFObjectFileBase>(Obj)) 547 return getELFRelocationValueString(ELF, Rel, Result); 548 if (auto *COFF = dyn_cast<COFFObjectFile>(Obj)) 549 return getCOFFRelocationValueString(COFF, Rel, Result); 550 if (auto *Wasm = dyn_cast<WasmObjectFile>(Obj)) 551 return getWasmRelocationValueString(Wasm, Rel, Result); 552 if (auto *MachO = dyn_cast<MachOObjectFile>(Obj)) 553 return getMachORelocationValueString(MachO, Rel, Result); 554 if (auto *XCOFF = dyn_cast<XCOFFObjectFile>(Obj)) 555 return getXCOFFRelocationValueString(*XCOFF, Rel, SymbolDescription, 556 Result); 557 llvm_unreachable("unknown object file format"); 558 } 559 560 /// Indicates whether this relocation should hidden when listing 561 /// relocations, usually because it is the trailing part of a multipart 562 /// relocation that will be printed as part of the leading relocation. 563 static bool getHidden(RelocationRef RelRef) { 564 auto *MachO = dyn_cast<MachOObjectFile>(RelRef.getObject()); 565 if (!MachO) 566 return false; 567 568 unsigned Arch = MachO->getArch(); 569 DataRefImpl Rel = RelRef.getRawDataRefImpl(); 570 uint64_t Type = MachO->getRelocationType(Rel); 571 572 // On arches that use the generic relocations, GENERIC_RELOC_PAIR 573 // is always hidden. 574 if (Arch == Triple::x86 || Arch == Triple::arm || Arch == Triple::ppc) 575 return Type == MachO::GENERIC_RELOC_PAIR; 576 577 if (Arch == Triple::x86_64) { 578 // On x86_64, X86_64_RELOC_UNSIGNED is hidden only when it follows 579 // an X86_64_RELOC_SUBTRACTOR. 580 if (Type == MachO::X86_64_RELOC_UNSIGNED && Rel.d.a > 0) { 581 DataRefImpl RelPrev = Rel; 582 RelPrev.d.a--; 583 uint64_t PrevType = MachO->getRelocationType(RelPrev); 584 if (PrevType == MachO::X86_64_RELOC_SUBTRACTOR) 585 return true; 586 } 587 } 588 589 return false; 590 } 591 592 /// Get the column at which we want to start printing the instruction 593 /// disassembly, taking into account anything which appears to the left of it. 594 unsigned objdump::getInstStartColumn(const MCSubtargetInfo &STI) { 595 return !ShowRawInsn ? 16 : STI.getTargetTriple().isX86() ? 40 : 24; 596 } 597 598 static void AlignToInstStartColumn(size_t Start, const MCSubtargetInfo &STI, 599 raw_ostream &OS) { 600 // The output of printInst starts with a tab. Print some spaces so that 601 // the tab has 1 column and advances to the target tab stop. 602 unsigned TabStop = getInstStartColumn(STI); 603 unsigned Column = OS.tell() - Start; 604 OS.indent(Column < TabStop - 1 ? TabStop - 1 - Column : 7 - Column % 8); 605 } 606 607 void objdump::printRawData(ArrayRef<uint8_t> Bytes, uint64_t Address, 608 formatted_raw_ostream &OS, 609 MCSubtargetInfo const &STI) { 610 size_t Start = OS.tell(); 611 if (LeadingAddr) 612 OS << format("%8" PRIx64 ":", Address); 613 if (ShowRawInsn) { 614 OS << ' '; 615 dumpBytes(Bytes, OS); 616 } 617 AlignToInstStartColumn(Start, STI, OS); 618 } 619 620 namespace { 621 622 static bool isAArch64Elf(const ObjectFile &Obj) { 623 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 624 return Elf && Elf->getEMachine() == ELF::EM_AARCH64; 625 } 626 627 static bool isArmElf(const ObjectFile &Obj) { 628 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 629 return Elf && Elf->getEMachine() == ELF::EM_ARM; 630 } 631 632 static bool isCSKYElf(const ObjectFile &Obj) { 633 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 634 return Elf && Elf->getEMachine() == ELF::EM_CSKY; 635 } 636 637 static bool hasMappingSymbols(const ObjectFile &Obj) { 638 return isArmElf(Obj) || isAArch64Elf(Obj) || isCSKYElf(Obj) ; 639 } 640 641 static void printRelocation(formatted_raw_ostream &OS, StringRef FileName, 642 const RelocationRef &Rel, uint64_t Address, 643 bool Is64Bits) { 644 StringRef Fmt = Is64Bits ? "%016" PRIx64 ": " : "%08" PRIx64 ": "; 645 SmallString<16> Name; 646 SmallString<32> Val; 647 Rel.getTypeName(Name); 648 if (Error E = getRelocationValueString(Rel, SymbolDescription, Val)) 649 reportError(std::move(E), FileName); 650 OS << (Is64Bits || !LeadingAddr ? "\t\t" : "\t\t\t"); 651 if (LeadingAddr) 652 OS << format(Fmt.data(), Address); 653 OS << Name << "\t" << Val; 654 } 655 656 static void printBTFRelocation(formatted_raw_ostream &FOS, llvm::BTFParser &BTF, 657 object::SectionedAddress Address, 658 LiveVariablePrinter &LVP) { 659 const llvm::BTF::BPFFieldReloc *Reloc = BTF.findFieldReloc(Address); 660 if (!Reloc) 661 return; 662 663 SmallString<64> Val; 664 BTF.symbolize(Reloc, Val); 665 FOS << "\t\t"; 666 if (LeadingAddr) 667 FOS << format("%016" PRIx64 ": ", Address.Address + AdjustVMA); 668 FOS << "CO-RE " << Val; 669 LVP.printAfterOtherLine(FOS, true); 670 } 671 672 class PrettyPrinter { 673 public: 674 virtual ~PrettyPrinter() = default; 675 virtual void 676 printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 677 object::SectionedAddress Address, formatted_raw_ostream &OS, 678 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 679 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 680 LiveVariablePrinter &LVP) { 681 if (SP && (PrintSource || PrintLines)) 682 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 683 LVP.printBetweenInsts(OS, false); 684 685 printRawData(Bytes, Address.Address, OS, STI); 686 687 if (MI) { 688 // See MCInstPrinter::printInst. On targets where a PC relative immediate 689 // is relative to the next instruction and the length of a MCInst is 690 // difficult to measure (x86), this is the address of the next 691 // instruction. 692 uint64_t Addr = 693 Address.Address + (STI.getTargetTriple().isX86() ? Bytes.size() : 0); 694 IP.printInst(MI, Addr, "", STI, OS); 695 } else 696 OS << "\t<unknown>"; 697 } 698 }; 699 PrettyPrinter PrettyPrinterInst; 700 701 class HexagonPrettyPrinter : public PrettyPrinter { 702 public: 703 void printLead(ArrayRef<uint8_t> Bytes, uint64_t Address, 704 formatted_raw_ostream &OS) { 705 uint32_t opcode = 706 (Bytes[3] << 24) | (Bytes[2] << 16) | (Bytes[1] << 8) | Bytes[0]; 707 if (LeadingAddr) 708 OS << format("%8" PRIx64 ":", Address); 709 if (ShowRawInsn) { 710 OS << "\t"; 711 dumpBytes(Bytes.slice(0, 4), OS); 712 OS << format("\t%08" PRIx32, opcode); 713 } 714 } 715 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 716 object::SectionedAddress Address, formatted_raw_ostream &OS, 717 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 718 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 719 LiveVariablePrinter &LVP) override { 720 if (SP && (PrintSource || PrintLines)) 721 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 722 if (!MI) { 723 printLead(Bytes, Address.Address, OS); 724 OS << " <unknown>"; 725 return; 726 } 727 std::string Buffer; 728 { 729 raw_string_ostream TempStream(Buffer); 730 IP.printInst(MI, Address.Address, "", STI, TempStream); 731 } 732 StringRef Contents(Buffer); 733 // Split off bundle attributes 734 auto PacketBundle = Contents.rsplit('\n'); 735 // Split off first instruction from the rest 736 auto HeadTail = PacketBundle.first.split('\n'); 737 auto Preamble = " { "; 738 auto Separator = ""; 739 740 // Hexagon's packets require relocations to be inline rather than 741 // clustered at the end of the packet. 742 std::vector<RelocationRef>::const_iterator RelCur = Rels->begin(); 743 std::vector<RelocationRef>::const_iterator RelEnd = Rels->end(); 744 auto PrintReloc = [&]() -> void { 745 while ((RelCur != RelEnd) && (RelCur->getOffset() <= Address.Address)) { 746 if (RelCur->getOffset() == Address.Address) { 747 printRelocation(OS, ObjectFilename, *RelCur, Address.Address, false); 748 return; 749 } 750 ++RelCur; 751 } 752 }; 753 754 while (!HeadTail.first.empty()) { 755 OS << Separator; 756 Separator = "\n"; 757 if (SP && (PrintSource || PrintLines)) 758 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 759 printLead(Bytes, Address.Address, OS); 760 OS << Preamble; 761 Preamble = " "; 762 StringRef Inst; 763 auto Duplex = HeadTail.first.split('\v'); 764 if (!Duplex.second.empty()) { 765 OS << Duplex.first; 766 OS << "; "; 767 Inst = Duplex.second; 768 } 769 else 770 Inst = HeadTail.first; 771 OS << Inst; 772 HeadTail = HeadTail.second.split('\n'); 773 if (HeadTail.first.empty()) 774 OS << " } " << PacketBundle.second; 775 PrintReloc(); 776 Bytes = Bytes.slice(4); 777 Address.Address += 4; 778 } 779 } 780 }; 781 HexagonPrettyPrinter HexagonPrettyPrinterInst; 782 783 class AMDGCNPrettyPrinter : public PrettyPrinter { 784 public: 785 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 786 object::SectionedAddress Address, formatted_raw_ostream &OS, 787 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 788 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 789 LiveVariablePrinter &LVP) override { 790 if (SP && (PrintSource || PrintLines)) 791 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 792 793 if (MI) { 794 SmallString<40> InstStr; 795 raw_svector_ostream IS(InstStr); 796 797 IP.printInst(MI, Address.Address, "", STI, IS); 798 799 OS << left_justify(IS.str(), 60); 800 } else { 801 // an unrecognized encoding - this is probably data so represent it 802 // using the .long directive, or .byte directive if fewer than 4 bytes 803 // remaining 804 if (Bytes.size() >= 4) { 805 OS << format( 806 "\t.long 0x%08" PRIx32 " ", 807 support::endian::read32<llvm::endianness::little>(Bytes.data())); 808 OS.indent(42); 809 } else { 810 OS << format("\t.byte 0x%02" PRIx8, Bytes[0]); 811 for (unsigned int i = 1; i < Bytes.size(); i++) 812 OS << format(", 0x%02" PRIx8, Bytes[i]); 813 OS.indent(55 - (6 * Bytes.size())); 814 } 815 } 816 817 OS << format("// %012" PRIX64 ":", Address.Address); 818 if (Bytes.size() >= 4) { 819 // D should be casted to uint32_t here as it is passed by format to 820 // snprintf as vararg. 821 for (uint32_t D : 822 ArrayRef(reinterpret_cast<const support::little32_t *>(Bytes.data()), 823 Bytes.size() / 4)) 824 OS << format(" %08" PRIX32, D); 825 } else { 826 for (unsigned char B : Bytes) 827 OS << format(" %02" PRIX8, B); 828 } 829 830 if (!Annot.empty()) 831 OS << " // " << Annot; 832 } 833 }; 834 AMDGCNPrettyPrinter AMDGCNPrettyPrinterInst; 835 836 class BPFPrettyPrinter : public PrettyPrinter { 837 public: 838 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 839 object::SectionedAddress Address, formatted_raw_ostream &OS, 840 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 841 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 842 LiveVariablePrinter &LVP) override { 843 if (SP && (PrintSource || PrintLines)) 844 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 845 if (LeadingAddr) 846 OS << format("%8" PRId64 ":", Address.Address / 8); 847 if (ShowRawInsn) { 848 OS << "\t"; 849 dumpBytes(Bytes, OS); 850 } 851 if (MI) 852 IP.printInst(MI, Address.Address, "", STI, OS); 853 else 854 OS << "\t<unknown>"; 855 } 856 }; 857 BPFPrettyPrinter BPFPrettyPrinterInst; 858 859 class ARMPrettyPrinter : public PrettyPrinter { 860 public: 861 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 862 object::SectionedAddress Address, formatted_raw_ostream &OS, 863 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 864 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 865 LiveVariablePrinter &LVP) override { 866 if (SP && (PrintSource || PrintLines)) 867 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 868 LVP.printBetweenInsts(OS, false); 869 870 size_t Start = OS.tell(); 871 if (LeadingAddr) 872 OS << format("%8" PRIx64 ":", Address.Address); 873 if (ShowRawInsn) { 874 size_t Pos = 0, End = Bytes.size(); 875 if (STI.checkFeatures("+thumb-mode")) { 876 for (; Pos + 2 <= End; Pos += 2) 877 OS << ' ' 878 << format_hex_no_prefix( 879 llvm::support::endian::read<uint16_t>( 880 Bytes.data() + Pos, InstructionEndianness), 881 4); 882 } else { 883 for (; Pos + 4 <= End; Pos += 4) 884 OS << ' ' 885 << format_hex_no_prefix( 886 llvm::support::endian::read<uint32_t>( 887 Bytes.data() + Pos, InstructionEndianness), 888 8); 889 } 890 if (Pos < End) { 891 OS << ' '; 892 dumpBytes(Bytes.slice(Pos), OS); 893 } 894 } 895 896 AlignToInstStartColumn(Start, STI, OS); 897 898 if (MI) { 899 IP.printInst(MI, Address.Address, "", STI, OS); 900 } else 901 OS << "\t<unknown>"; 902 } 903 904 void setInstructionEndianness(llvm::endianness Endianness) { 905 InstructionEndianness = Endianness; 906 } 907 908 private: 909 llvm::endianness InstructionEndianness = llvm::endianness::little; 910 }; 911 ARMPrettyPrinter ARMPrettyPrinterInst; 912 913 class AArch64PrettyPrinter : public PrettyPrinter { 914 public: 915 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 916 object::SectionedAddress Address, formatted_raw_ostream &OS, 917 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 918 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 919 LiveVariablePrinter &LVP) override { 920 if (SP && (PrintSource || PrintLines)) 921 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 922 LVP.printBetweenInsts(OS, false); 923 924 size_t Start = OS.tell(); 925 if (LeadingAddr) 926 OS << format("%8" PRIx64 ":", Address.Address); 927 if (ShowRawInsn) { 928 size_t Pos = 0, End = Bytes.size(); 929 for (; Pos + 4 <= End; Pos += 4) 930 OS << ' ' 931 << format_hex_no_prefix( 932 llvm::support::endian::read<uint32_t>( 933 Bytes.data() + Pos, llvm::endianness::little), 934 8); 935 if (Pos < End) { 936 OS << ' '; 937 dumpBytes(Bytes.slice(Pos), OS); 938 } 939 } 940 941 AlignToInstStartColumn(Start, STI, OS); 942 943 if (MI) { 944 IP.printInst(MI, Address.Address, "", STI, OS); 945 } else 946 OS << "\t<unknown>"; 947 } 948 }; 949 AArch64PrettyPrinter AArch64PrettyPrinterInst; 950 951 class RISCVPrettyPrinter : public PrettyPrinter { 952 public: 953 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 954 object::SectionedAddress Address, formatted_raw_ostream &OS, 955 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 956 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 957 LiveVariablePrinter &LVP) override { 958 if (SP && (PrintSource || PrintLines)) 959 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 960 LVP.printBetweenInsts(OS, false); 961 962 size_t Start = OS.tell(); 963 if (LeadingAddr) 964 OS << format("%8" PRIx64 ":", Address.Address); 965 if (ShowRawInsn) { 966 size_t Pos = 0, End = Bytes.size(); 967 if (End % 4 == 0) { 968 // 32-bit and 64-bit instructions. 969 for (; Pos + 4 <= End; Pos += 4) 970 OS << ' ' 971 << format_hex_no_prefix( 972 llvm::support::endian::read<uint32_t>( 973 Bytes.data() + Pos, llvm::endianness::little), 974 8); 975 } else if (End % 2 == 0) { 976 // 16-bit and 48-bits instructions. 977 for (; Pos + 2 <= End; Pos += 2) 978 OS << ' ' 979 << format_hex_no_prefix( 980 llvm::support::endian::read<uint16_t>( 981 Bytes.data() + Pos, llvm::endianness::little), 982 4); 983 } 984 if (Pos < End) { 985 OS << ' '; 986 dumpBytes(Bytes.slice(Pos), OS); 987 } 988 } 989 990 AlignToInstStartColumn(Start, STI, OS); 991 992 if (MI) { 993 IP.printInst(MI, Address.Address, "", STI, OS); 994 } else 995 OS << "\t<unknown>"; 996 } 997 }; 998 RISCVPrettyPrinter RISCVPrettyPrinterInst; 999 1000 PrettyPrinter &selectPrettyPrinter(Triple const &Triple) { 1001 switch(Triple.getArch()) { 1002 default: 1003 return PrettyPrinterInst; 1004 case Triple::hexagon: 1005 return HexagonPrettyPrinterInst; 1006 case Triple::amdgcn: 1007 return AMDGCNPrettyPrinterInst; 1008 case Triple::bpfel: 1009 case Triple::bpfeb: 1010 return BPFPrettyPrinterInst; 1011 case Triple::arm: 1012 case Triple::armeb: 1013 case Triple::thumb: 1014 case Triple::thumbeb: 1015 return ARMPrettyPrinterInst; 1016 case Triple::aarch64: 1017 case Triple::aarch64_be: 1018 case Triple::aarch64_32: 1019 return AArch64PrettyPrinterInst; 1020 case Triple::riscv32: 1021 case Triple::riscv64: 1022 return RISCVPrettyPrinterInst; 1023 } 1024 } 1025 1026 class DisassemblerTarget { 1027 public: 1028 const Target *TheTarget; 1029 std::unique_ptr<const MCSubtargetInfo> SubtargetInfo; 1030 std::shared_ptr<MCContext> Context; 1031 std::unique_ptr<MCDisassembler> DisAsm; 1032 std::shared_ptr<MCInstrAnalysis> InstrAnalysis; 1033 std::shared_ptr<MCInstPrinter> InstPrinter; 1034 PrettyPrinter *Printer; 1035 1036 DisassemblerTarget(const Target *TheTarget, ObjectFile &Obj, 1037 StringRef TripleName, StringRef MCPU, 1038 SubtargetFeatures &Features); 1039 DisassemblerTarget(DisassemblerTarget &Other, SubtargetFeatures &Features); 1040 1041 private: 1042 MCTargetOptions Options; 1043 std::shared_ptr<const MCRegisterInfo> RegisterInfo; 1044 std::shared_ptr<const MCAsmInfo> AsmInfo; 1045 std::shared_ptr<const MCInstrInfo> InstrInfo; 1046 std::shared_ptr<MCObjectFileInfo> ObjectFileInfo; 1047 }; 1048 1049 DisassemblerTarget::DisassemblerTarget(const Target *TheTarget, ObjectFile &Obj, 1050 StringRef TripleName, StringRef MCPU, 1051 SubtargetFeatures &Features) 1052 : TheTarget(TheTarget), 1053 Printer(&selectPrettyPrinter(Triple(TripleName))), 1054 RegisterInfo(TheTarget->createMCRegInfo(TripleName)) { 1055 if (!RegisterInfo) 1056 reportError(Obj.getFileName(), "no register info for target " + TripleName); 1057 1058 // Set up disassembler. 1059 AsmInfo.reset(TheTarget->createMCAsmInfo(*RegisterInfo, TripleName, Options)); 1060 if (!AsmInfo) 1061 reportError(Obj.getFileName(), "no assembly info for target " + TripleName); 1062 1063 SubtargetInfo.reset( 1064 TheTarget->createMCSubtargetInfo(TripleName, MCPU, Features.getString())); 1065 if (!SubtargetInfo) 1066 reportError(Obj.getFileName(), 1067 "no subtarget info for target " + TripleName); 1068 InstrInfo.reset(TheTarget->createMCInstrInfo()); 1069 if (!InstrInfo) 1070 reportError(Obj.getFileName(), 1071 "no instruction info for target " + TripleName); 1072 Context = 1073 std::make_shared<MCContext>(Triple(TripleName), AsmInfo.get(), 1074 RegisterInfo.get(), SubtargetInfo.get()); 1075 1076 // FIXME: for now initialize MCObjectFileInfo with default values 1077 ObjectFileInfo.reset( 1078 TheTarget->createMCObjectFileInfo(*Context, /*PIC=*/false)); 1079 Context->setObjectFileInfo(ObjectFileInfo.get()); 1080 1081 DisAsm.reset(TheTarget->createMCDisassembler(*SubtargetInfo, *Context)); 1082 if (!DisAsm) 1083 reportError(Obj.getFileName(), "no disassembler for target " + TripleName); 1084 1085 if (auto *ELFObj = dyn_cast<ELFObjectFileBase>(&Obj)) 1086 DisAsm->setABIVersion(ELFObj->getEIdentABIVersion()); 1087 1088 InstrAnalysis.reset(TheTarget->createMCInstrAnalysis(InstrInfo.get())); 1089 1090 int AsmPrinterVariant = AsmInfo->getAssemblerDialect(); 1091 InstPrinter.reset(TheTarget->createMCInstPrinter(Triple(TripleName), 1092 AsmPrinterVariant, *AsmInfo, 1093 *InstrInfo, *RegisterInfo)); 1094 if (!InstPrinter) 1095 reportError(Obj.getFileName(), 1096 "no instruction printer for target " + TripleName); 1097 InstPrinter->setPrintImmHex(PrintImmHex); 1098 InstPrinter->setPrintBranchImmAsAddress(true); 1099 InstPrinter->setSymbolizeOperands(SymbolizeOperands); 1100 InstPrinter->setMCInstrAnalysis(InstrAnalysis.get()); 1101 1102 switch (DisassemblyColor) { 1103 case ColorOutput::Enable: 1104 InstPrinter->setUseColor(true); 1105 break; 1106 case ColorOutput::Auto: 1107 InstPrinter->setUseColor(outs().has_colors()); 1108 break; 1109 case ColorOutput::Disable: 1110 case ColorOutput::Invalid: 1111 InstPrinter->setUseColor(false); 1112 break; 1113 }; 1114 } 1115 1116 DisassemblerTarget::DisassemblerTarget(DisassemblerTarget &Other, 1117 SubtargetFeatures &Features) 1118 : TheTarget(Other.TheTarget), 1119 SubtargetInfo(TheTarget->createMCSubtargetInfo(TripleName, MCPU, 1120 Features.getString())), 1121 Context(Other.Context), 1122 DisAsm(TheTarget->createMCDisassembler(*SubtargetInfo, *Context)), 1123 InstrAnalysis(Other.InstrAnalysis), InstPrinter(Other.InstPrinter), 1124 Printer(Other.Printer), RegisterInfo(Other.RegisterInfo), 1125 AsmInfo(Other.AsmInfo), InstrInfo(Other.InstrInfo), 1126 ObjectFileInfo(Other.ObjectFileInfo) {} 1127 } // namespace 1128 1129 static uint8_t getElfSymbolType(const ObjectFile &Obj, const SymbolRef &Sym) { 1130 assert(Obj.isELF()); 1131 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 1132 return unwrapOrError(Elf32LEObj->getSymbol(Sym.getRawDataRefImpl()), 1133 Obj.getFileName()) 1134 ->getType(); 1135 if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 1136 return unwrapOrError(Elf64LEObj->getSymbol(Sym.getRawDataRefImpl()), 1137 Obj.getFileName()) 1138 ->getType(); 1139 if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 1140 return unwrapOrError(Elf32BEObj->getSymbol(Sym.getRawDataRefImpl()), 1141 Obj.getFileName()) 1142 ->getType(); 1143 if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 1144 return unwrapOrError(Elf64BEObj->getSymbol(Sym.getRawDataRefImpl()), 1145 Obj.getFileName()) 1146 ->getType(); 1147 llvm_unreachable("Unsupported binary format"); 1148 } 1149 1150 template <class ELFT> 1151 static void 1152 addDynamicElfSymbols(const ELFObjectFile<ELFT> &Obj, 1153 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 1154 for (auto Symbol : Obj.getDynamicSymbolIterators()) { 1155 uint8_t SymbolType = Symbol.getELFType(); 1156 if (SymbolType == ELF::STT_SECTION) 1157 continue; 1158 1159 uint64_t Address = unwrapOrError(Symbol.getAddress(), Obj.getFileName()); 1160 // ELFSymbolRef::getAddress() returns size instead of value for common 1161 // symbols which is not desirable for disassembly output. Overriding. 1162 if (SymbolType == ELF::STT_COMMON) 1163 Address = unwrapOrError(Obj.getSymbol(Symbol.getRawDataRefImpl()), 1164 Obj.getFileName()) 1165 ->st_value; 1166 1167 StringRef Name = unwrapOrError(Symbol.getName(), Obj.getFileName()); 1168 if (Name.empty()) 1169 continue; 1170 1171 section_iterator SecI = 1172 unwrapOrError(Symbol.getSection(), Obj.getFileName()); 1173 if (SecI == Obj.section_end()) 1174 continue; 1175 1176 AllSymbols[*SecI].emplace_back(Address, Name, SymbolType); 1177 } 1178 } 1179 1180 static void 1181 addDynamicElfSymbols(const ELFObjectFileBase &Obj, 1182 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 1183 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 1184 addDynamicElfSymbols(*Elf32LEObj, AllSymbols); 1185 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 1186 addDynamicElfSymbols(*Elf64LEObj, AllSymbols); 1187 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 1188 addDynamicElfSymbols(*Elf32BEObj, AllSymbols); 1189 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 1190 addDynamicElfSymbols(*Elf64BEObj, AllSymbols); 1191 else 1192 llvm_unreachable("Unsupported binary format"); 1193 } 1194 1195 static std::optional<SectionRef> getWasmCodeSection(const WasmObjectFile &Obj) { 1196 for (auto SecI : Obj.sections()) { 1197 const WasmSection &Section = Obj.getWasmSection(SecI); 1198 if (Section.Type == wasm::WASM_SEC_CODE) 1199 return SecI; 1200 } 1201 return std::nullopt; 1202 } 1203 1204 static void 1205 addMissingWasmCodeSymbols(const WasmObjectFile &Obj, 1206 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 1207 std::optional<SectionRef> Section = getWasmCodeSection(Obj); 1208 if (!Section) 1209 return; 1210 SectionSymbolsTy &Symbols = AllSymbols[*Section]; 1211 1212 std::set<uint64_t> SymbolAddresses; 1213 for (const auto &Sym : Symbols) 1214 SymbolAddresses.insert(Sym.Addr); 1215 1216 for (const wasm::WasmFunction &Function : Obj.functions()) { 1217 // This adjustment mirrors the one in WasmObjectFile::getSymbolAddress. 1218 uint32_t Adjustment = Obj.isRelocatableObject() || Obj.isSharedObject() 1219 ? 0 1220 : Section->getAddress(); 1221 uint64_t Address = Function.CodeSectionOffset + Adjustment; 1222 // Only add fallback symbols for functions not already present in the symbol 1223 // table. 1224 if (SymbolAddresses.count(Address)) 1225 continue; 1226 // This function has no symbol, so it should have no SymbolName. 1227 assert(Function.SymbolName.empty()); 1228 // We use DebugName for the name, though it may be empty if there is no 1229 // "name" custom section, or that section is missing a name for this 1230 // function. 1231 StringRef Name = Function.DebugName; 1232 Symbols.emplace_back(Address, Name, ELF::STT_NOTYPE); 1233 } 1234 } 1235 1236 static void addPltEntries(const ObjectFile &Obj, 1237 std::map<SectionRef, SectionSymbolsTy> &AllSymbols, 1238 StringSaver &Saver) { 1239 auto *ElfObj = dyn_cast<ELFObjectFileBase>(&Obj); 1240 if (!ElfObj) 1241 return; 1242 DenseMap<StringRef, SectionRef> Sections; 1243 for (SectionRef Section : Obj.sections()) { 1244 Expected<StringRef> SecNameOrErr = Section.getName(); 1245 if (!SecNameOrErr) { 1246 consumeError(SecNameOrErr.takeError()); 1247 continue; 1248 } 1249 Sections[*SecNameOrErr] = Section; 1250 } 1251 for (auto Plt : ElfObj->getPltEntries()) { 1252 if (Plt.Symbol) { 1253 SymbolRef Symbol(*Plt.Symbol, ElfObj); 1254 uint8_t SymbolType = getElfSymbolType(Obj, Symbol); 1255 if (Expected<StringRef> NameOrErr = Symbol.getName()) { 1256 if (!NameOrErr->empty()) 1257 AllSymbols[Sections[Plt.Section]].emplace_back( 1258 Plt.Address, Saver.save((*NameOrErr + "@plt").str()), SymbolType); 1259 continue; 1260 } else { 1261 // The warning has been reported in disassembleObject(). 1262 consumeError(NameOrErr.takeError()); 1263 } 1264 } 1265 reportWarning("PLT entry at 0x" + Twine::utohexstr(Plt.Address) + 1266 " references an invalid symbol", 1267 Obj.getFileName()); 1268 } 1269 } 1270 1271 // Normally the disassembly output will skip blocks of zeroes. This function 1272 // returns the number of zero bytes that can be skipped when dumping the 1273 // disassembly of the instructions in Buf. 1274 static size_t countSkippableZeroBytes(ArrayRef<uint8_t> Buf) { 1275 // Find the number of leading zeroes. 1276 size_t N = 0; 1277 while (N < Buf.size() && !Buf[N]) 1278 ++N; 1279 1280 // We may want to skip blocks of zero bytes, but unless we see 1281 // at least 8 of them in a row. 1282 if (N < 8) 1283 return 0; 1284 1285 // We skip zeroes in multiples of 4 because do not want to truncate an 1286 // instruction if it starts with a zero byte. 1287 return N & ~0x3; 1288 } 1289 1290 // Returns a map from sections to their relocations. 1291 static std::map<SectionRef, std::vector<RelocationRef>> 1292 getRelocsMap(object::ObjectFile const &Obj) { 1293 std::map<SectionRef, std::vector<RelocationRef>> Ret; 1294 uint64_t I = (uint64_t)-1; 1295 for (SectionRef Sec : Obj.sections()) { 1296 ++I; 1297 Expected<section_iterator> RelocatedOrErr = Sec.getRelocatedSection(); 1298 if (!RelocatedOrErr) 1299 reportError(Obj.getFileName(), 1300 "section (" + Twine(I) + 1301 "): failed to get a relocated section: " + 1302 toString(RelocatedOrErr.takeError())); 1303 1304 section_iterator Relocated = *RelocatedOrErr; 1305 if (Relocated == Obj.section_end() || !checkSectionFilter(*Relocated).Keep) 1306 continue; 1307 std::vector<RelocationRef> &V = Ret[*Relocated]; 1308 append_range(V, Sec.relocations()); 1309 // Sort relocations by address. 1310 llvm::stable_sort(V, isRelocAddressLess); 1311 } 1312 return Ret; 1313 } 1314 1315 // Used for --adjust-vma to check if address should be adjusted by the 1316 // specified value for a given section. 1317 // For ELF we do not adjust non-allocatable sections like debug ones, 1318 // because they are not loadable. 1319 // TODO: implement for other file formats. 1320 static bool shouldAdjustVA(const SectionRef &Section) { 1321 const ObjectFile *Obj = Section.getObject(); 1322 if (Obj->isELF()) 1323 return ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC; 1324 return false; 1325 } 1326 1327 1328 typedef std::pair<uint64_t, char> MappingSymbolPair; 1329 static char getMappingSymbolKind(ArrayRef<MappingSymbolPair> MappingSymbols, 1330 uint64_t Address) { 1331 auto It = 1332 partition_point(MappingSymbols, [Address](const MappingSymbolPair &Val) { 1333 return Val.first <= Address; 1334 }); 1335 // Return zero for any address before the first mapping symbol; this means 1336 // we should use the default disassembly mode, depending on the target. 1337 if (It == MappingSymbols.begin()) 1338 return '\x00'; 1339 return (It - 1)->second; 1340 } 1341 1342 static uint64_t dumpARMELFData(uint64_t SectionAddr, uint64_t Index, 1343 uint64_t End, const ObjectFile &Obj, 1344 ArrayRef<uint8_t> Bytes, 1345 ArrayRef<MappingSymbolPair> MappingSymbols, 1346 const MCSubtargetInfo &STI, raw_ostream &OS) { 1347 llvm::endianness Endian = 1348 Obj.isLittleEndian() ? llvm::endianness::little : llvm::endianness::big; 1349 size_t Start = OS.tell(); 1350 OS << format("%8" PRIx64 ": ", SectionAddr + Index); 1351 if (Index + 4 <= End) { 1352 dumpBytes(Bytes.slice(Index, 4), OS); 1353 AlignToInstStartColumn(Start, STI, OS); 1354 OS << "\t.word\t" 1355 << format_hex(support::endian::read32(Bytes.data() + Index, Endian), 1356 10); 1357 return 4; 1358 } 1359 if (Index + 2 <= End) { 1360 dumpBytes(Bytes.slice(Index, 2), OS); 1361 AlignToInstStartColumn(Start, STI, OS); 1362 OS << "\t.short\t" 1363 << format_hex(support::endian::read16(Bytes.data() + Index, Endian), 6); 1364 return 2; 1365 } 1366 dumpBytes(Bytes.slice(Index, 1), OS); 1367 AlignToInstStartColumn(Start, STI, OS); 1368 OS << "\t.byte\t" << format_hex(Bytes[Index], 4); 1369 return 1; 1370 } 1371 1372 static void dumpELFData(uint64_t SectionAddr, uint64_t Index, uint64_t End, 1373 ArrayRef<uint8_t> Bytes) { 1374 // print out data up to 8 bytes at a time in hex and ascii 1375 uint8_t AsciiData[9] = {'\0'}; 1376 uint8_t Byte; 1377 int NumBytes = 0; 1378 1379 for (; Index < End; ++Index) { 1380 if (NumBytes == 0) 1381 outs() << format("%8" PRIx64 ":", SectionAddr + Index); 1382 Byte = Bytes.slice(Index)[0]; 1383 outs() << format(" %02x", Byte); 1384 AsciiData[NumBytes] = isPrint(Byte) ? Byte : '.'; 1385 1386 uint8_t IndentOffset = 0; 1387 NumBytes++; 1388 if (Index == End - 1 || NumBytes > 8) { 1389 // Indent the space for less than 8 bytes data. 1390 // 2 spaces for byte and one for space between bytes 1391 IndentOffset = 3 * (8 - NumBytes); 1392 for (int Excess = NumBytes; Excess < 8; Excess++) 1393 AsciiData[Excess] = '\0'; 1394 NumBytes = 8; 1395 } 1396 if (NumBytes == 8) { 1397 AsciiData[8] = '\0'; 1398 outs() << std::string(IndentOffset, ' ') << " "; 1399 outs() << reinterpret_cast<char *>(AsciiData); 1400 outs() << '\n'; 1401 NumBytes = 0; 1402 } 1403 } 1404 } 1405 1406 SymbolInfoTy objdump::createSymbolInfo(const ObjectFile &Obj, 1407 const SymbolRef &Symbol, 1408 bool IsMappingSymbol) { 1409 const StringRef FileName = Obj.getFileName(); 1410 const uint64_t Addr = unwrapOrError(Symbol.getAddress(), FileName); 1411 const StringRef Name = unwrapOrError(Symbol.getName(), FileName); 1412 1413 if (Obj.isXCOFF() && (SymbolDescription || TracebackTable)) { 1414 const auto &XCOFFObj = cast<XCOFFObjectFile>(Obj); 1415 DataRefImpl SymbolDRI = Symbol.getRawDataRefImpl(); 1416 1417 const uint32_t SymbolIndex = XCOFFObj.getSymbolIndex(SymbolDRI.p); 1418 std::optional<XCOFF::StorageMappingClass> Smc = 1419 getXCOFFSymbolCsectSMC(XCOFFObj, Symbol); 1420 return SymbolInfoTy(Smc, Addr, Name, SymbolIndex, 1421 isLabel(XCOFFObj, Symbol)); 1422 } else if (Obj.isXCOFF()) { 1423 const SymbolRef::Type SymType = unwrapOrError(Symbol.getType(), FileName); 1424 return SymbolInfoTy(Addr, Name, SymType, /*IsMappingSymbol=*/false, 1425 /*IsXCOFF=*/true); 1426 } else if (Obj.isWasm()) { 1427 uint8_t SymType = 1428 cast<WasmObjectFile>(&Obj)->getWasmSymbol(Symbol).Info.Kind; 1429 return SymbolInfoTy(Addr, Name, SymType, false); 1430 } else { 1431 uint8_t Type = 1432 Obj.isELF() ? getElfSymbolType(Obj, Symbol) : (uint8_t)ELF::STT_NOTYPE; 1433 return SymbolInfoTy(Addr, Name, Type, IsMappingSymbol); 1434 } 1435 } 1436 1437 static SymbolInfoTy createDummySymbolInfo(const ObjectFile &Obj, 1438 const uint64_t Addr, StringRef &Name, 1439 uint8_t Type) { 1440 if (Obj.isXCOFF() && (SymbolDescription || TracebackTable)) 1441 return SymbolInfoTy(std::nullopt, Addr, Name, std::nullopt, false); 1442 if (Obj.isWasm()) 1443 return SymbolInfoTy(Addr, Name, wasm::WASM_SYMBOL_TYPE_SECTION); 1444 return SymbolInfoTy(Addr, Name, Type); 1445 } 1446 1447 static void collectBBAddrMapLabels( 1448 const BBAddrMapInfo &FullAddrMap, uint64_t SectionAddr, uint64_t Start, 1449 uint64_t End, 1450 std::unordered_map<uint64_t, std::vector<BBAddrMapLabel>> &Labels) { 1451 if (FullAddrMap.empty()) 1452 return; 1453 Labels.clear(); 1454 uint64_t StartAddress = SectionAddr + Start; 1455 uint64_t EndAddress = SectionAddr + End; 1456 const BBAddrMapFunctionEntry *FunctionMap = 1457 FullAddrMap.getEntryForAddress(StartAddress); 1458 if (!FunctionMap) 1459 return; 1460 std::optional<size_t> BBRangeIndex = 1461 FunctionMap->getAddrMap().getBBRangeIndexForBaseAddress(StartAddress); 1462 if (!BBRangeIndex) 1463 return; 1464 size_t NumBBEntriesBeforeRange = 0; 1465 for (size_t I = 0; I < *BBRangeIndex; ++I) 1466 NumBBEntriesBeforeRange += 1467 FunctionMap->getAddrMap().BBRanges[I].BBEntries.size(); 1468 const auto &BBRange = FunctionMap->getAddrMap().BBRanges[*BBRangeIndex]; 1469 for (size_t I = 0; I < BBRange.BBEntries.size(); ++I) { 1470 const BBAddrMap::BBEntry &BBEntry = BBRange.BBEntries[I]; 1471 uint64_t BBAddress = BBEntry.Offset + BBRange.BaseAddress; 1472 if (BBAddress >= EndAddress) 1473 continue; 1474 1475 std::string LabelString = ("BB" + Twine(BBEntry.ID)).str(); 1476 Labels[BBAddress].push_back( 1477 {LabelString, FunctionMap->constructPGOLabelString( 1478 NumBBEntriesBeforeRange + I, PrettyPGOAnalysisMap)}); 1479 } 1480 } 1481 1482 static void 1483 collectLocalBranchTargets(ArrayRef<uint8_t> Bytes, MCInstrAnalysis *MIA, 1484 MCDisassembler *DisAsm, MCInstPrinter *IP, 1485 const MCSubtargetInfo *STI, uint64_t SectionAddr, 1486 uint64_t Start, uint64_t End, 1487 std::unordered_map<uint64_t, std::string> &Labels) { 1488 // Supported by certain targets. 1489 const bool isPPC = STI->getTargetTriple().isPPC(); 1490 const bool isX86 = STI->getTargetTriple().isX86(); 1491 const bool isBPF = STI->getTargetTriple().isBPF(); 1492 if (!isPPC && !isX86 && !isBPF) 1493 return; 1494 1495 if (MIA) 1496 MIA->resetState(); 1497 1498 Labels.clear(); 1499 unsigned LabelCount = 0; 1500 Start += SectionAddr; 1501 End += SectionAddr; 1502 const bool isXCOFF = STI->getTargetTriple().isOSBinFormatXCOFF(); 1503 for (uint64_t Index = Start; Index < End;) { 1504 // Disassemble a real instruction and record function-local branch labels. 1505 MCInst Inst; 1506 uint64_t Size; 1507 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index - SectionAddr); 1508 bool Disassembled = 1509 DisAsm->getInstruction(Inst, Size, ThisBytes, Index, nulls()); 1510 if (Size == 0) 1511 Size = std::min<uint64_t>(ThisBytes.size(), 1512 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1513 1514 if (MIA) { 1515 if (Disassembled) { 1516 uint64_t Target; 1517 bool TargetKnown = MIA->evaluateBranch(Inst, Index, Size, Target); 1518 if (TargetKnown && (Target >= Start && Target < End) && 1519 !Labels.count(Target)) { 1520 // On PowerPC and AIX, a function call is encoded as a branch to 0. 1521 // On other PowerPC platforms (ELF), a function call is encoded as 1522 // a branch to self. Do not add a label for these cases. 1523 if (!(isPPC && 1524 ((Target == 0 && isXCOFF) || (Target == Index && !isXCOFF)))) 1525 Labels[Target] = ("L" + Twine(LabelCount++)).str(); 1526 } 1527 MIA->updateState(Inst, Index); 1528 } else 1529 MIA->resetState(); 1530 } 1531 Index += Size; 1532 } 1533 } 1534 1535 // Create an MCSymbolizer for the target and add it to the MCDisassembler. 1536 // This is currently only used on AMDGPU, and assumes the format of the 1537 // void * argument passed to AMDGPU's createMCSymbolizer. 1538 static void addSymbolizer( 1539 MCContext &Ctx, const Target *Target, StringRef TripleName, 1540 MCDisassembler *DisAsm, uint64_t SectionAddr, ArrayRef<uint8_t> Bytes, 1541 SectionSymbolsTy &Symbols, 1542 std::vector<std::unique_ptr<std::string>> &SynthesizedLabelNames) { 1543 1544 std::unique_ptr<MCRelocationInfo> RelInfo( 1545 Target->createMCRelocationInfo(TripleName, Ctx)); 1546 if (!RelInfo) 1547 return; 1548 std::unique_ptr<MCSymbolizer> Symbolizer(Target->createMCSymbolizer( 1549 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1550 MCSymbolizer *SymbolizerPtr = &*Symbolizer; 1551 DisAsm->setSymbolizer(std::move(Symbolizer)); 1552 1553 if (!SymbolizeOperands) 1554 return; 1555 1556 // Synthesize labels referenced by branch instructions by 1557 // disassembling, discarding the output, and collecting the referenced 1558 // addresses from the symbolizer. 1559 for (size_t Index = 0; Index != Bytes.size();) { 1560 MCInst Inst; 1561 uint64_t Size; 1562 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index); 1563 const uint64_t ThisAddr = SectionAddr + Index; 1564 DisAsm->getInstruction(Inst, Size, ThisBytes, ThisAddr, nulls()); 1565 if (Size == 0) 1566 Size = std::min<uint64_t>(ThisBytes.size(), 1567 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1568 Index += Size; 1569 } 1570 ArrayRef<uint64_t> LabelAddrsRef = SymbolizerPtr->getReferencedAddresses(); 1571 // Copy and sort to remove duplicates. 1572 std::vector<uint64_t> LabelAddrs; 1573 LabelAddrs.insert(LabelAddrs.end(), LabelAddrsRef.begin(), 1574 LabelAddrsRef.end()); 1575 llvm::sort(LabelAddrs); 1576 LabelAddrs.resize(llvm::unique(LabelAddrs) - LabelAddrs.begin()); 1577 // Add the labels. 1578 for (unsigned LabelNum = 0; LabelNum != LabelAddrs.size(); ++LabelNum) { 1579 auto Name = std::make_unique<std::string>(); 1580 *Name = (Twine("L") + Twine(LabelNum)).str(); 1581 SynthesizedLabelNames.push_back(std::move(Name)); 1582 Symbols.push_back(SymbolInfoTy( 1583 LabelAddrs[LabelNum], *SynthesizedLabelNames.back(), ELF::STT_NOTYPE)); 1584 } 1585 llvm::stable_sort(Symbols); 1586 // Recreate the symbolizer with the new symbols list. 1587 RelInfo.reset(Target->createMCRelocationInfo(TripleName, Ctx)); 1588 Symbolizer.reset(Target->createMCSymbolizer( 1589 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1590 DisAsm->setSymbolizer(std::move(Symbolizer)); 1591 } 1592 1593 static StringRef getSegmentName(const MachOObjectFile *MachO, 1594 const SectionRef &Section) { 1595 if (MachO) { 1596 DataRefImpl DR = Section.getRawDataRefImpl(); 1597 StringRef SegmentName = MachO->getSectionFinalSegmentName(DR); 1598 return SegmentName; 1599 } 1600 return ""; 1601 } 1602 1603 static void emitPostInstructionInfo(formatted_raw_ostream &FOS, 1604 const MCAsmInfo &MAI, 1605 const MCSubtargetInfo &STI, 1606 StringRef Comments, 1607 LiveVariablePrinter &LVP) { 1608 do { 1609 if (!Comments.empty()) { 1610 // Emit a line of comments. 1611 StringRef Comment; 1612 std::tie(Comment, Comments) = Comments.split('\n'); 1613 // MAI.getCommentColumn() assumes that instructions are printed at the 1614 // position of 8, while getInstStartColumn() returns the actual position. 1615 unsigned CommentColumn = 1616 MAI.getCommentColumn() - 8 + getInstStartColumn(STI); 1617 FOS.PadToColumn(CommentColumn); 1618 FOS << MAI.getCommentString() << ' ' << Comment; 1619 } 1620 LVP.printAfterInst(FOS); 1621 FOS << '\n'; 1622 } while (!Comments.empty()); 1623 FOS.flush(); 1624 } 1625 1626 static void createFakeELFSections(ObjectFile &Obj) { 1627 assert(Obj.isELF()); 1628 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 1629 Elf32LEObj->createFakeSections(); 1630 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 1631 Elf64LEObj->createFakeSections(); 1632 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 1633 Elf32BEObj->createFakeSections(); 1634 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 1635 Elf64BEObj->createFakeSections(); 1636 else 1637 llvm_unreachable("Unsupported binary format"); 1638 } 1639 1640 // Tries to fetch a more complete version of the given object file using its 1641 // Build ID. Returns std::nullopt if nothing was found. 1642 static std::optional<OwningBinary<Binary>> 1643 fetchBinaryByBuildID(const ObjectFile &Obj) { 1644 object::BuildIDRef BuildID = getBuildID(&Obj); 1645 if (BuildID.empty()) 1646 return std::nullopt; 1647 std::optional<std::string> Path = BIDFetcher->fetch(BuildID); 1648 if (!Path) 1649 return std::nullopt; 1650 Expected<OwningBinary<Binary>> DebugBinary = createBinary(*Path); 1651 if (!DebugBinary) { 1652 reportWarning(toString(DebugBinary.takeError()), *Path); 1653 return std::nullopt; 1654 } 1655 return std::move(*DebugBinary); 1656 } 1657 1658 static void 1659 disassembleObject(ObjectFile &Obj, const ObjectFile &DbgObj, 1660 DisassemblerTarget &PrimaryTarget, 1661 std::optional<DisassemblerTarget> &SecondaryTarget, 1662 SourcePrinter &SP, bool InlineRelocs) { 1663 DisassemblerTarget *DT = &PrimaryTarget; 1664 bool PrimaryIsThumb = false; 1665 SmallVector<std::pair<uint64_t, uint64_t>, 0> CHPECodeMap; 1666 1667 if (SecondaryTarget) { 1668 if (isArmElf(Obj)) { 1669 PrimaryIsThumb = 1670 PrimaryTarget.SubtargetInfo->checkFeatures("+thumb-mode"); 1671 } else if (const auto *COFFObj = dyn_cast<COFFObjectFile>(&Obj)) { 1672 const chpe_metadata *CHPEMetadata = COFFObj->getCHPEMetadata(); 1673 if (CHPEMetadata && CHPEMetadata->CodeMapCount) { 1674 uintptr_t CodeMapInt; 1675 cantFail(COFFObj->getRvaPtr(CHPEMetadata->CodeMap, CodeMapInt)); 1676 auto CodeMap = reinterpret_cast<const chpe_range_entry *>(CodeMapInt); 1677 1678 for (uint32_t i = 0; i < CHPEMetadata->CodeMapCount; ++i) { 1679 if (CodeMap[i].getType() == chpe_range_type::Amd64 && 1680 CodeMap[i].Length) { 1681 // Store x86_64 CHPE code ranges. 1682 uint64_t Start = CodeMap[i].getStart() + COFFObj->getImageBase(); 1683 CHPECodeMap.emplace_back(Start, Start + CodeMap[i].Length); 1684 } 1685 } 1686 llvm::sort(CHPECodeMap); 1687 } 1688 } 1689 } 1690 1691 std::map<SectionRef, std::vector<RelocationRef>> RelocMap; 1692 if (InlineRelocs || Obj.isXCOFF()) 1693 RelocMap = getRelocsMap(Obj); 1694 bool Is64Bits = Obj.getBytesInAddress() > 4; 1695 1696 // Create a mapping from virtual address to symbol name. This is used to 1697 // pretty print the symbols while disassembling. 1698 std::map<SectionRef, SectionSymbolsTy> AllSymbols; 1699 std::map<SectionRef, SmallVector<MappingSymbolPair, 0>> AllMappingSymbols; 1700 SectionSymbolsTy AbsoluteSymbols; 1701 const StringRef FileName = Obj.getFileName(); 1702 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&Obj); 1703 for (const SymbolRef &Symbol : Obj.symbols()) { 1704 Expected<StringRef> NameOrErr = Symbol.getName(); 1705 if (!NameOrErr) { 1706 reportWarning(toString(NameOrErr.takeError()), FileName); 1707 continue; 1708 } 1709 if (NameOrErr->empty() && !(Obj.isXCOFF() && SymbolDescription)) 1710 continue; 1711 1712 if (Obj.isELF() && 1713 (cantFail(Symbol.getFlags()) & SymbolRef::SF_FormatSpecific)) { 1714 // Symbol is intended not to be displayed by default (STT_FILE, 1715 // STT_SECTION, or a mapping symbol). Ignore STT_SECTION symbols. We will 1716 // synthesize a section symbol if no symbol is defined at offset 0. 1717 // 1718 // For a mapping symbol, store it within both AllSymbols and 1719 // AllMappingSymbols. If --show-all-symbols is unspecified, its label will 1720 // not be printed in disassembly listing. 1721 if (getElfSymbolType(Obj, Symbol) != ELF::STT_SECTION && 1722 hasMappingSymbols(Obj)) { 1723 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1724 if (SecI != Obj.section_end()) { 1725 uint64_t SectionAddr = SecI->getAddress(); 1726 uint64_t Address = cantFail(Symbol.getAddress()); 1727 StringRef Name = *NameOrErr; 1728 if (Name.consume_front("$") && Name.size() && 1729 strchr("adtx", Name[0])) { 1730 AllMappingSymbols[*SecI].emplace_back(Address - SectionAddr, 1731 Name[0]); 1732 AllSymbols[*SecI].push_back( 1733 createSymbolInfo(Obj, Symbol, /*MappingSymbol=*/true)); 1734 } 1735 } 1736 } 1737 continue; 1738 } 1739 1740 if (MachO) { 1741 // __mh_(execute|dylib|dylinker|bundle|preload|object)_header are special 1742 // symbols that support MachO header introspection. They do not bind to 1743 // code locations and are irrelevant for disassembly. 1744 if (NameOrErr->starts_with("__mh_") && NameOrErr->ends_with("_header")) 1745 continue; 1746 // Don't ask a Mach-O STAB symbol for its section unless you know that 1747 // STAB symbol's section field refers to a valid section index. Otherwise 1748 // the symbol may error trying to load a section that does not exist. 1749 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 1750 uint8_t NType = (MachO->is64Bit() ? 1751 MachO->getSymbol64TableEntry(SymDRI).n_type: 1752 MachO->getSymbolTableEntry(SymDRI).n_type); 1753 if (NType & MachO::N_STAB) 1754 continue; 1755 } 1756 1757 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1758 if (SecI != Obj.section_end()) 1759 AllSymbols[*SecI].push_back(createSymbolInfo(Obj, Symbol)); 1760 else 1761 AbsoluteSymbols.push_back(createSymbolInfo(Obj, Symbol)); 1762 } 1763 1764 if (AllSymbols.empty() && Obj.isELF()) 1765 addDynamicElfSymbols(cast<ELFObjectFileBase>(Obj), AllSymbols); 1766 1767 if (Obj.isWasm()) 1768 addMissingWasmCodeSymbols(cast<WasmObjectFile>(Obj), AllSymbols); 1769 1770 if (Obj.isELF() && Obj.sections().empty()) 1771 createFakeELFSections(Obj); 1772 1773 BumpPtrAllocator A; 1774 StringSaver Saver(A); 1775 addPltEntries(Obj, AllSymbols, Saver); 1776 1777 // Create a mapping from virtual address to section. An empty section can 1778 // cause more than one section at the same address. Sort such sections to be 1779 // before same-addressed non-empty sections so that symbol lookups prefer the 1780 // non-empty section. 1781 std::vector<std::pair<uint64_t, SectionRef>> SectionAddresses; 1782 for (SectionRef Sec : Obj.sections()) 1783 SectionAddresses.emplace_back(Sec.getAddress(), Sec); 1784 llvm::stable_sort(SectionAddresses, [](const auto &LHS, const auto &RHS) { 1785 if (LHS.first != RHS.first) 1786 return LHS.first < RHS.first; 1787 return LHS.second.getSize() < RHS.second.getSize(); 1788 }); 1789 1790 // Linked executables (.exe and .dll files) typically don't include a real 1791 // symbol table but they might contain an export table. 1792 if (const auto *COFFObj = dyn_cast<COFFObjectFile>(&Obj)) { 1793 for (const auto &ExportEntry : COFFObj->export_directories()) { 1794 StringRef Name; 1795 if (Error E = ExportEntry.getSymbolName(Name)) 1796 reportError(std::move(E), Obj.getFileName()); 1797 if (Name.empty()) 1798 continue; 1799 1800 uint32_t RVA; 1801 if (Error E = ExportEntry.getExportRVA(RVA)) 1802 reportError(std::move(E), Obj.getFileName()); 1803 1804 uint64_t VA = COFFObj->getImageBase() + RVA; 1805 auto Sec = partition_point( 1806 SectionAddresses, [VA](const std::pair<uint64_t, SectionRef> &O) { 1807 return O.first <= VA; 1808 }); 1809 if (Sec != SectionAddresses.begin()) { 1810 --Sec; 1811 AllSymbols[Sec->second].emplace_back(VA, Name, ELF::STT_NOTYPE); 1812 } else 1813 AbsoluteSymbols.emplace_back(VA, Name, ELF::STT_NOTYPE); 1814 } 1815 } 1816 1817 // Sort all the symbols, this allows us to use a simple binary search to find 1818 // Multiple symbols can have the same address. Use a stable sort to stabilize 1819 // the output. 1820 StringSet<> FoundDisasmSymbolSet; 1821 for (std::pair<const SectionRef, SectionSymbolsTy> &SecSyms : AllSymbols) 1822 llvm::stable_sort(SecSyms.second); 1823 llvm::stable_sort(AbsoluteSymbols); 1824 1825 std::unique_ptr<DWARFContext> DICtx; 1826 LiveVariablePrinter LVP(*DT->Context->getRegisterInfo(), *DT->SubtargetInfo); 1827 1828 if (DbgVariables != DVDisabled) { 1829 DICtx = DWARFContext::create(DbgObj); 1830 for (const std::unique_ptr<DWARFUnit> &CU : DICtx->compile_units()) 1831 LVP.addCompileUnit(CU->getUnitDIE(false)); 1832 } 1833 1834 LLVM_DEBUG(LVP.dump()); 1835 1836 BBAddrMapInfo FullAddrMap; 1837 auto ReadBBAddrMap = [&](std::optional<unsigned> SectionIndex = 1838 std::nullopt) { 1839 FullAddrMap.clear(); 1840 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj)) { 1841 std::vector<PGOAnalysisMap> PGOAnalyses; 1842 auto BBAddrMapsOrErr = Elf->readBBAddrMap(SectionIndex, &PGOAnalyses); 1843 if (!BBAddrMapsOrErr) { 1844 reportWarning(toString(BBAddrMapsOrErr.takeError()), Obj.getFileName()); 1845 return; 1846 } 1847 for (auto &&[FunctionBBAddrMap, FunctionPGOAnalysis] : 1848 zip_equal(*std::move(BBAddrMapsOrErr), std::move(PGOAnalyses))) { 1849 FullAddrMap.AddFunctionEntry(std::move(FunctionBBAddrMap), 1850 std::move(FunctionPGOAnalysis)); 1851 } 1852 } 1853 }; 1854 1855 // For non-relocatable objects, Read all LLVM_BB_ADDR_MAP sections into a 1856 // single mapping, since they don't have any conflicts. 1857 if (SymbolizeOperands && !Obj.isRelocatableObject()) 1858 ReadBBAddrMap(); 1859 1860 std::optional<llvm::BTFParser> BTF; 1861 if (InlineRelocs && BTFParser::hasBTFSections(Obj)) { 1862 BTF.emplace(); 1863 BTFParser::ParseOptions Opts = {}; 1864 Opts.LoadTypes = true; 1865 Opts.LoadRelocs = true; 1866 if (Error E = BTF->parse(Obj, Opts)) 1867 WithColor::defaultErrorHandler(std::move(E)); 1868 } 1869 1870 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 1871 if (FilterSections.empty() && !DisassembleAll && 1872 (!Section.isText() || Section.isVirtual())) 1873 continue; 1874 1875 uint64_t SectionAddr = Section.getAddress(); 1876 uint64_t SectSize = Section.getSize(); 1877 if (!SectSize) 1878 continue; 1879 1880 // For relocatable object files, read the LLVM_BB_ADDR_MAP section 1881 // corresponding to this section, if present. 1882 if (SymbolizeOperands && Obj.isRelocatableObject()) 1883 ReadBBAddrMap(Section.getIndex()); 1884 1885 // Get the list of all the symbols in this section. 1886 SectionSymbolsTy &Symbols = AllSymbols[Section]; 1887 auto &MappingSymbols = AllMappingSymbols[Section]; 1888 llvm::sort(MappingSymbols); 1889 1890 ArrayRef<uint8_t> Bytes = arrayRefFromStringRef( 1891 unwrapOrError(Section.getContents(), Obj.getFileName())); 1892 1893 std::vector<std::unique_ptr<std::string>> SynthesizedLabelNames; 1894 if (Obj.isELF() && Obj.getArch() == Triple::amdgcn) { 1895 // AMDGPU disassembler uses symbolizer for printing labels 1896 addSymbolizer(*DT->Context, DT->TheTarget, TripleName, DT->DisAsm.get(), 1897 SectionAddr, Bytes, Symbols, SynthesizedLabelNames); 1898 } 1899 1900 StringRef SegmentName = getSegmentName(MachO, Section); 1901 StringRef SectionName = unwrapOrError(Section.getName(), Obj.getFileName()); 1902 // If the section has no symbol at the start, just insert a dummy one. 1903 // Without --show-all-symbols, also insert one if all symbols at the start 1904 // are mapping symbols. 1905 bool CreateDummy = Symbols.empty(); 1906 if (!CreateDummy) { 1907 CreateDummy = true; 1908 for (auto &Sym : Symbols) { 1909 if (Sym.Addr != SectionAddr) 1910 break; 1911 if (!Sym.IsMappingSymbol || ShowAllSymbols) 1912 CreateDummy = false; 1913 } 1914 } 1915 if (CreateDummy) { 1916 SymbolInfoTy Sym = createDummySymbolInfo( 1917 Obj, SectionAddr, SectionName, 1918 Section.isText() ? ELF::STT_FUNC : ELF::STT_OBJECT); 1919 if (Obj.isXCOFF()) 1920 Symbols.insert(Symbols.begin(), Sym); 1921 else 1922 Symbols.insert(llvm::lower_bound(Symbols, Sym), Sym); 1923 } 1924 1925 SmallString<40> Comments; 1926 raw_svector_ostream CommentStream(Comments); 1927 1928 uint64_t VMAAdjustment = 0; 1929 if (shouldAdjustVA(Section)) 1930 VMAAdjustment = AdjustVMA; 1931 1932 // In executable and shared objects, r_offset holds a virtual address. 1933 // Subtract SectionAddr from the r_offset field of a relocation to get 1934 // the section offset. 1935 uint64_t RelAdjustment = Obj.isRelocatableObject() ? 0 : SectionAddr; 1936 uint64_t Size; 1937 uint64_t Index; 1938 bool PrintedSection = false; 1939 std::vector<RelocationRef> Rels = RelocMap[Section]; 1940 std::vector<RelocationRef>::const_iterator RelCur = Rels.begin(); 1941 std::vector<RelocationRef>::const_iterator RelEnd = Rels.end(); 1942 1943 // Loop over each chunk of code between two points where at least 1944 // one symbol is defined. 1945 for (size_t SI = 0, SE = Symbols.size(); SI != SE;) { 1946 // Advance SI past all the symbols starting at the same address, 1947 // and make an ArrayRef of them. 1948 unsigned FirstSI = SI; 1949 uint64_t Start = Symbols[SI].Addr; 1950 ArrayRef<SymbolInfoTy> SymbolsHere; 1951 while (SI != SE && Symbols[SI].Addr == Start) 1952 ++SI; 1953 SymbolsHere = ArrayRef<SymbolInfoTy>(&Symbols[FirstSI], SI - FirstSI); 1954 1955 // Get the demangled names of all those symbols. We end up with a vector 1956 // of StringRef that holds the names we're going to use, and a vector of 1957 // std::string that stores the new strings returned by demangle(), if 1958 // any. If we don't call demangle() then that vector can stay empty. 1959 std::vector<StringRef> SymNamesHere; 1960 std::vector<std::string> DemangledSymNamesHere; 1961 if (Demangle) { 1962 // Fetch the demangled names and store them locally. 1963 for (const SymbolInfoTy &Symbol : SymbolsHere) 1964 DemangledSymNamesHere.push_back(demangle(Symbol.Name)); 1965 // Now we've finished modifying that vector, it's safe to make 1966 // a vector of StringRefs pointing into it. 1967 SymNamesHere.insert(SymNamesHere.begin(), DemangledSymNamesHere.begin(), 1968 DemangledSymNamesHere.end()); 1969 } else { 1970 for (const SymbolInfoTy &Symbol : SymbolsHere) 1971 SymNamesHere.push_back(Symbol.Name); 1972 } 1973 1974 // Distinguish ELF data from code symbols, which will be used later on to 1975 // decide whether to 'disassemble' this chunk as a data declaration via 1976 // dumpELFData(), or whether to treat it as code. 1977 // 1978 // If data _and_ code symbols are defined at the same address, the code 1979 // takes priority, on the grounds that disassembling code is our main 1980 // purpose here, and it would be a worse failure to _not_ interpret 1981 // something that _was_ meaningful as code than vice versa. 1982 // 1983 // Any ELF symbol type that is not clearly data will be regarded as code. 1984 // In particular, one of the uses of STT_NOTYPE is for branch targets 1985 // inside functions, for which STT_FUNC would be inaccurate. 1986 // 1987 // So here, we spot whether there's any non-data symbol present at all, 1988 // and only set the DisassembleAsELFData flag if there isn't. Also, we use 1989 // this distinction to inform the decision of which symbol to print at 1990 // the head of the section, so that if we're printing code, we print a 1991 // code-related symbol name to go with it. 1992 bool DisassembleAsELFData = false; 1993 size_t DisplaySymIndex = SymbolsHere.size() - 1; 1994 if (Obj.isELF() && !DisassembleAll && Section.isText()) { 1995 DisassembleAsELFData = true; // unless we find a code symbol below 1996 1997 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 1998 uint8_t SymTy = SymbolsHere[i].Type; 1999 if (SymTy != ELF::STT_OBJECT && SymTy != ELF::STT_COMMON) { 2000 DisassembleAsELFData = false; 2001 DisplaySymIndex = i; 2002 } 2003 } 2004 } 2005 2006 // Decide which symbol(s) from this collection we're going to print. 2007 std::vector<bool> SymsToPrint(SymbolsHere.size(), false); 2008 // If the user has given the --disassemble-symbols option, then we must 2009 // display every symbol in that set, and no others. 2010 if (!DisasmSymbolSet.empty()) { 2011 bool FoundAny = false; 2012 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 2013 if (DisasmSymbolSet.count(SymNamesHere[i])) { 2014 SymsToPrint[i] = true; 2015 FoundAny = true; 2016 } 2017 } 2018 2019 // And if none of the symbols here is one that the user asked for, skip 2020 // disassembling this entire chunk of code. 2021 if (!FoundAny) 2022 continue; 2023 } else if (!SymbolsHere[DisplaySymIndex].IsMappingSymbol) { 2024 // Otherwise, print whichever symbol at this location is last in the 2025 // Symbols array, because that array is pre-sorted in a way intended to 2026 // correlate with priority of which symbol to display. 2027 SymsToPrint[DisplaySymIndex] = true; 2028 } 2029 2030 // Now that we know we're disassembling this section, override the choice 2031 // of which symbols to display by printing _all_ of them at this address 2032 // if the user asked for all symbols. 2033 // 2034 // That way, '--show-all-symbols --disassemble-symbol=foo' will print 2035 // only the chunk of code headed by 'foo', but also show any other 2036 // symbols defined at that address, such as aliases for 'foo', or the ARM 2037 // mapping symbol preceding its code. 2038 if (ShowAllSymbols) { 2039 for (size_t i = 0; i < SymbolsHere.size(); ++i) 2040 SymsToPrint[i] = true; 2041 } 2042 2043 if (Start < SectionAddr || StopAddress <= Start) 2044 continue; 2045 2046 for (size_t i = 0; i < SymbolsHere.size(); ++i) 2047 FoundDisasmSymbolSet.insert(SymNamesHere[i]); 2048 2049 // The end is the section end, the beginning of the next symbol, or 2050 // --stop-address. 2051 uint64_t End = std::min<uint64_t>(SectionAddr + SectSize, StopAddress); 2052 if (SI < SE) 2053 End = std::min(End, Symbols[SI].Addr); 2054 if (Start >= End || End <= StartAddress) 2055 continue; 2056 Start -= SectionAddr; 2057 End -= SectionAddr; 2058 2059 if (!PrintedSection) { 2060 PrintedSection = true; 2061 outs() << "\nDisassembly of section "; 2062 if (!SegmentName.empty()) 2063 outs() << SegmentName << ","; 2064 outs() << SectionName << ":\n"; 2065 } 2066 2067 bool PrintedLabel = false; 2068 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 2069 if (!SymsToPrint[i]) 2070 continue; 2071 2072 const SymbolInfoTy &Symbol = SymbolsHere[i]; 2073 const StringRef SymbolName = SymNamesHere[i]; 2074 2075 if (!PrintedLabel) { 2076 outs() << '\n'; 2077 PrintedLabel = true; 2078 } 2079 if (LeadingAddr) 2080 outs() << format(Is64Bits ? "%016" PRIx64 " " : "%08" PRIx64 " ", 2081 SectionAddr + Start + VMAAdjustment); 2082 if (Obj.isXCOFF() && SymbolDescription) { 2083 outs() << getXCOFFSymbolDescription(Symbol, SymbolName) << ":\n"; 2084 } else 2085 outs() << '<' << SymbolName << ">:\n"; 2086 } 2087 2088 // Don't print raw contents of a virtual section. A virtual section 2089 // doesn't have any contents in the file. 2090 if (Section.isVirtual()) { 2091 outs() << "...\n"; 2092 continue; 2093 } 2094 2095 // See if any of the symbols defined at this location triggers target- 2096 // specific disassembly behavior, e.g. of special descriptors or function 2097 // prelude information. 2098 // 2099 // We stop this loop at the first symbol that triggers some kind of 2100 // interesting behavior (if any), on the assumption that if two symbols 2101 // defined at the same address trigger two conflicting symbol handlers, 2102 // the object file is probably confused anyway, and it would make even 2103 // less sense to present the output of _both_ handlers, because that 2104 // would describe the same data twice. 2105 for (size_t SHI = 0; SHI < SymbolsHere.size(); ++SHI) { 2106 SymbolInfoTy Symbol = SymbolsHere[SHI]; 2107 2108 Expected<bool> RespondedOrErr = DT->DisAsm->onSymbolStart( 2109 Symbol, Size, Bytes.slice(Start, End - Start), SectionAddr + Start); 2110 2111 if (RespondedOrErr && !*RespondedOrErr) { 2112 // This symbol didn't trigger any interesting handling. Try the other 2113 // symbols defined at this address. 2114 continue; 2115 } 2116 2117 // If onSymbolStart returned an Error, that means it identified some 2118 // kind of special data at this address, but wasn't able to disassemble 2119 // it meaningfully. So we fall back to printing the error out and 2120 // disassembling the failed region as bytes, assuming that the target 2121 // detected the failure before printing anything. 2122 if (!RespondedOrErr) { 2123 std::string ErrMsgStr = toString(RespondedOrErr.takeError()); 2124 StringRef ErrMsg = ErrMsgStr; 2125 do { 2126 StringRef Line; 2127 std::tie(Line, ErrMsg) = ErrMsg.split('\n'); 2128 outs() << DT->Context->getAsmInfo()->getCommentString() 2129 << " error decoding " << SymNamesHere[SHI] << ": " << Line 2130 << '\n'; 2131 } while (!ErrMsg.empty()); 2132 2133 if (Size) { 2134 outs() << DT->Context->getAsmInfo()->getCommentString() 2135 << " decoding failed region as bytes\n"; 2136 for (uint64_t I = 0; I < Size; ++I) 2137 outs() << "\t.byte\t " << format_hex(Bytes[I], 1, /*Upper=*/true) 2138 << '\n'; 2139 } 2140 } 2141 2142 // Regardless of whether onSymbolStart returned an Error or true, 'Size' 2143 // will have been set to the amount of data covered by whatever prologue 2144 // the target identified. So we advance our own position to beyond that. 2145 // Sometimes that will be the entire distance to the next symbol, and 2146 // sometimes it will be just a prologue and we should start 2147 // disassembling instructions from where it left off. 2148 Start += Size; 2149 break; 2150 } 2151 2152 Index = Start; 2153 if (SectionAddr < StartAddress) 2154 Index = std::max<uint64_t>(Index, StartAddress - SectionAddr); 2155 2156 if (DisassembleAsELFData) { 2157 dumpELFData(SectionAddr, Index, End, Bytes); 2158 Index = End; 2159 continue; 2160 } 2161 2162 // Skip relocations from symbols that are not dumped. 2163 for (; RelCur != RelEnd; ++RelCur) { 2164 uint64_t Offset = RelCur->getOffset() - RelAdjustment; 2165 if (Index <= Offset) 2166 break; 2167 } 2168 2169 bool DumpARMELFData = false; 2170 bool DumpTracebackTableForXCOFFFunction = 2171 Obj.isXCOFF() && Section.isText() && TracebackTable && 2172 Symbols[SI - 1].XCOFFSymInfo.StorageMappingClass && 2173 (*Symbols[SI - 1].XCOFFSymInfo.StorageMappingClass == XCOFF::XMC_PR); 2174 2175 formatted_raw_ostream FOS(outs()); 2176 2177 std::unordered_map<uint64_t, std::string> AllLabels; 2178 std::unordered_map<uint64_t, std::vector<BBAddrMapLabel>> BBAddrMapLabels; 2179 if (SymbolizeOperands) { 2180 collectLocalBranchTargets(Bytes, DT->InstrAnalysis.get(), 2181 DT->DisAsm.get(), DT->InstPrinter.get(), 2182 PrimaryTarget.SubtargetInfo.get(), 2183 SectionAddr, Index, End, AllLabels); 2184 collectBBAddrMapLabels(FullAddrMap, SectionAddr, Index, End, 2185 BBAddrMapLabels); 2186 } 2187 2188 if (DT->InstrAnalysis) 2189 DT->InstrAnalysis->resetState(); 2190 2191 while (Index < End) { 2192 uint64_t RelOffset; 2193 2194 // ARM and AArch64 ELF binaries can interleave data and text in the 2195 // same section. We rely on the markers introduced to understand what 2196 // we need to dump. If the data marker is within a function, it is 2197 // denoted as a word/short etc. 2198 if (!MappingSymbols.empty()) { 2199 char Kind = getMappingSymbolKind(MappingSymbols, Index); 2200 DumpARMELFData = Kind == 'd'; 2201 if (SecondaryTarget) { 2202 if (Kind == 'a') { 2203 DT = PrimaryIsThumb ? &*SecondaryTarget : &PrimaryTarget; 2204 } else if (Kind == 't') { 2205 DT = PrimaryIsThumb ? &PrimaryTarget : &*SecondaryTarget; 2206 } 2207 } 2208 } else if (!CHPECodeMap.empty()) { 2209 uint64_t Address = SectionAddr + Index; 2210 auto It = partition_point( 2211 CHPECodeMap, 2212 [Address](const std::pair<uint64_t, uint64_t> &Entry) { 2213 return Entry.first <= Address; 2214 }); 2215 if (It != CHPECodeMap.begin() && Address < (It - 1)->second) { 2216 DT = &*SecondaryTarget; 2217 } else { 2218 DT = &PrimaryTarget; 2219 // X64 disassembler range may have left Index unaligned, so 2220 // make sure that it's aligned when we switch back to ARM64 2221 // code. 2222 Index = llvm::alignTo(Index, 4); 2223 if (Index >= End) 2224 break; 2225 } 2226 } 2227 2228 auto findRel = [&]() { 2229 while (RelCur != RelEnd) { 2230 RelOffset = RelCur->getOffset() - RelAdjustment; 2231 // If this relocation is hidden, skip it. 2232 if (getHidden(*RelCur) || SectionAddr + RelOffset < StartAddress) { 2233 ++RelCur; 2234 continue; 2235 } 2236 2237 // Stop when RelCur's offset is past the disassembled 2238 // instruction/data. 2239 if (RelOffset >= Index + Size) 2240 return false; 2241 if (RelOffset >= Index) 2242 return true; 2243 ++RelCur; 2244 } 2245 return false; 2246 }; 2247 2248 // When -z or --disassemble-zeroes are given we always dissasemble 2249 // them. Otherwise we might want to skip zero bytes we see. 2250 if (!DisassembleZeroes) { 2251 uint64_t MaxOffset = End - Index; 2252 // For --reloc: print zero blocks patched by relocations, so that 2253 // relocations can be shown in the dump. 2254 if (InlineRelocs && RelCur != RelEnd) 2255 MaxOffset = std::min(RelCur->getOffset() - RelAdjustment - Index, 2256 MaxOffset); 2257 2258 if (size_t N = 2259 countSkippableZeroBytes(Bytes.slice(Index, MaxOffset))) { 2260 FOS << "\t\t..." << '\n'; 2261 Index += N; 2262 continue; 2263 } 2264 } 2265 2266 if (DumpARMELFData) { 2267 Size = dumpARMELFData(SectionAddr, Index, End, Obj, Bytes, 2268 MappingSymbols, *DT->SubtargetInfo, FOS); 2269 } else { 2270 2271 if (DumpTracebackTableForXCOFFFunction && 2272 doesXCOFFTracebackTableBegin(Bytes.slice(Index, 4))) { 2273 dumpTracebackTable(Bytes.slice(Index), 2274 SectionAddr + Index + VMAAdjustment, FOS, 2275 SectionAddr + End + VMAAdjustment, 2276 *DT->SubtargetInfo, cast<XCOFFObjectFile>(&Obj)); 2277 Index = End; 2278 continue; 2279 } 2280 2281 // Print local label if there's any. 2282 auto Iter1 = BBAddrMapLabels.find(SectionAddr + Index); 2283 if (Iter1 != BBAddrMapLabels.end()) { 2284 for (const auto &BBLabel : Iter1->second) 2285 FOS << "<" << BBLabel.BlockLabel << ">" << BBLabel.PGOAnalysis 2286 << ":\n"; 2287 } else { 2288 auto Iter2 = AllLabels.find(SectionAddr + Index); 2289 if (Iter2 != AllLabels.end()) 2290 FOS << "<" << Iter2->second << ">:\n"; 2291 } 2292 2293 // Disassemble a real instruction or a data when disassemble all is 2294 // provided 2295 MCInst Inst; 2296 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index); 2297 uint64_t ThisAddr = SectionAddr + Index; 2298 bool Disassembled = DT->DisAsm->getInstruction( 2299 Inst, Size, ThisBytes, ThisAddr, CommentStream); 2300 if (Size == 0) 2301 Size = std::min<uint64_t>( 2302 ThisBytes.size(), 2303 DT->DisAsm->suggestBytesToSkip(ThisBytes, ThisAddr)); 2304 2305 LVP.update({Index, Section.getIndex()}, 2306 {Index + Size, Section.getIndex()}, Index + Size != End); 2307 2308 DT->InstPrinter->setCommentStream(CommentStream); 2309 2310 DT->Printer->printInst( 2311 *DT->InstPrinter, Disassembled ? &Inst : nullptr, 2312 Bytes.slice(Index, Size), 2313 {SectionAddr + Index + VMAAdjustment, Section.getIndex()}, FOS, 2314 "", *DT->SubtargetInfo, &SP, Obj.getFileName(), &Rels, LVP); 2315 2316 DT->InstPrinter->setCommentStream(llvm::nulls()); 2317 2318 // If disassembly succeeds, we try to resolve the target address 2319 // (jump target or memory operand address) and print it to the 2320 // right of the instruction. 2321 // 2322 // Otherwise, we don't print anything else so that we avoid 2323 // analyzing invalid or incomplete instruction information. 2324 if (Disassembled && DT->InstrAnalysis) { 2325 llvm::raw_ostream *TargetOS = &FOS; 2326 uint64_t Target; 2327 bool PrintTarget = DT->InstrAnalysis->evaluateBranch( 2328 Inst, SectionAddr + Index, Size, Target); 2329 2330 if (!PrintTarget) { 2331 if (std::optional<uint64_t> MaybeTarget = 2332 DT->InstrAnalysis->evaluateMemoryOperandAddress( 2333 Inst, DT->SubtargetInfo.get(), SectionAddr + Index, 2334 Size)) { 2335 Target = *MaybeTarget; 2336 PrintTarget = true; 2337 // Do not print real address when symbolizing. 2338 if (!SymbolizeOperands) { 2339 // Memory operand addresses are printed as comments. 2340 TargetOS = &CommentStream; 2341 *TargetOS << "0x" << Twine::utohexstr(Target); 2342 } 2343 } 2344 } 2345 2346 if (PrintTarget) { 2347 // In a relocatable object, the target's section must reside in 2348 // the same section as the call instruction or it is accessed 2349 // through a relocation. 2350 // 2351 // In a non-relocatable object, the target may be in any section. 2352 // In that case, locate the section(s) containing the target 2353 // address and find the symbol in one of those, if possible. 2354 // 2355 // N.B. Except for XCOFF, we don't walk the relocations in the 2356 // relocatable case yet. 2357 std::vector<const SectionSymbolsTy *> TargetSectionSymbols; 2358 if (!Obj.isRelocatableObject()) { 2359 auto It = llvm::partition_point( 2360 SectionAddresses, 2361 [=](const std::pair<uint64_t, SectionRef> &O) { 2362 return O.first <= Target; 2363 }); 2364 uint64_t TargetSecAddr = 0; 2365 while (It != SectionAddresses.begin()) { 2366 --It; 2367 if (TargetSecAddr == 0) 2368 TargetSecAddr = It->first; 2369 if (It->first != TargetSecAddr) 2370 break; 2371 TargetSectionSymbols.push_back(&AllSymbols[It->second]); 2372 } 2373 } else { 2374 TargetSectionSymbols.push_back(&Symbols); 2375 } 2376 TargetSectionSymbols.push_back(&AbsoluteSymbols); 2377 2378 // Find the last symbol in the first candidate section whose 2379 // offset is less than or equal to the target. If there are no 2380 // such symbols, try in the next section and so on, before finally 2381 // using the nearest preceding absolute symbol (if any), if there 2382 // are no other valid symbols. 2383 const SymbolInfoTy *TargetSym = nullptr; 2384 for (const SectionSymbolsTy *TargetSymbols : 2385 TargetSectionSymbols) { 2386 auto It = llvm::partition_point( 2387 *TargetSymbols, 2388 [=](const SymbolInfoTy &O) { return O.Addr <= Target; }); 2389 while (It != TargetSymbols->begin()) { 2390 --It; 2391 // Skip mapping symbols to avoid possible ambiguity as they 2392 // do not allow uniquely identifying the target address. 2393 if (!It->IsMappingSymbol) { 2394 TargetSym = &*It; 2395 break; 2396 } 2397 } 2398 if (TargetSym) 2399 break; 2400 } 2401 2402 // Branch targets are printed just after the instructions. 2403 // Print the labels corresponding to the target if there's any. 2404 bool BBAddrMapLabelAvailable = BBAddrMapLabels.count(Target); 2405 bool LabelAvailable = AllLabels.count(Target); 2406 2407 if (TargetSym != nullptr) { 2408 uint64_t TargetAddress = TargetSym->Addr; 2409 uint64_t Disp = Target - TargetAddress; 2410 std::string TargetName = Demangle ? demangle(TargetSym->Name) 2411 : TargetSym->Name.str(); 2412 bool RelFixedUp = false; 2413 SmallString<32> Val; 2414 2415 *TargetOS << " <"; 2416 // On XCOFF, we use relocations, even without -r, so we 2417 // can print the correct name for an extern function call. 2418 if (Obj.isXCOFF() && findRel()) { 2419 // Check for possible branch relocations and 2420 // branches to fixup code. 2421 bool BranchRelocationType = true; 2422 XCOFF::RelocationType RelocType; 2423 if (Obj.is64Bit()) { 2424 const XCOFFRelocation64 *Reloc = 2425 reinterpret_cast<XCOFFRelocation64 *>( 2426 RelCur->getRawDataRefImpl().p); 2427 RelFixedUp = Reloc->isFixupIndicated(); 2428 RelocType = Reloc->Type; 2429 } else { 2430 const XCOFFRelocation32 *Reloc = 2431 reinterpret_cast<XCOFFRelocation32 *>( 2432 RelCur->getRawDataRefImpl().p); 2433 RelFixedUp = Reloc->isFixupIndicated(); 2434 RelocType = Reloc->Type; 2435 } 2436 BranchRelocationType = 2437 RelocType == XCOFF::R_BA || RelocType == XCOFF::R_BR || 2438 RelocType == XCOFF::R_RBA || RelocType == XCOFF::R_RBR; 2439 2440 // If we have a valid relocation, try to print its 2441 // corresponding symbol name. Multiple relocations on the 2442 // same instruction are not handled. 2443 // Branches to fixup code will have the RelFixedUp flag set in 2444 // the RLD. For these instructions, we print the correct 2445 // branch target, but print the referenced symbol as a 2446 // comment. 2447 if (Error E = getRelocationValueString(*RelCur, false, Val)) { 2448 // If -r was used, this error will be printed later. 2449 // Otherwise, we ignore the error and print what 2450 // would have been printed without using relocations. 2451 consumeError(std::move(E)); 2452 *TargetOS << TargetName; 2453 RelFixedUp = false; // Suppress comment for RLD sym name 2454 } else if (BranchRelocationType && !RelFixedUp) 2455 *TargetOS << Val; 2456 else 2457 *TargetOS << TargetName; 2458 if (Disp) 2459 *TargetOS << "+0x" << Twine::utohexstr(Disp); 2460 } else if (!Disp) { 2461 *TargetOS << TargetName; 2462 } else if (BBAddrMapLabelAvailable) { 2463 *TargetOS << BBAddrMapLabels[Target].front().BlockLabel; 2464 } else if (LabelAvailable) { 2465 *TargetOS << AllLabels[Target]; 2466 } else { 2467 // Always Print the binary symbol plus an offset if there's no 2468 // local label corresponding to the target address. 2469 *TargetOS << TargetName << "+0x" << Twine::utohexstr(Disp); 2470 } 2471 *TargetOS << ">"; 2472 if (RelFixedUp && !InlineRelocs) { 2473 // We have fixup code for a relocation. We print the 2474 // referenced symbol as a comment. 2475 *TargetOS << "\t# " << Val; 2476 } 2477 2478 } else if (BBAddrMapLabelAvailable) { 2479 *TargetOS << " <" << BBAddrMapLabels[Target].front().BlockLabel 2480 << ">"; 2481 } else if (LabelAvailable) { 2482 *TargetOS << " <" << AllLabels[Target] << ">"; 2483 } 2484 // By convention, each record in the comment stream should be 2485 // terminated. 2486 if (TargetOS == &CommentStream) 2487 *TargetOS << "\n"; 2488 } 2489 2490 DT->InstrAnalysis->updateState(Inst, SectionAddr + Index); 2491 } else if (!Disassembled && DT->InstrAnalysis) { 2492 DT->InstrAnalysis->resetState(); 2493 } 2494 } 2495 2496 assert(DT->Context->getAsmInfo()); 2497 emitPostInstructionInfo(FOS, *DT->Context->getAsmInfo(), 2498 *DT->SubtargetInfo, CommentStream.str(), LVP); 2499 Comments.clear(); 2500 2501 if (BTF) 2502 printBTFRelocation(FOS, *BTF, {Index, Section.getIndex()}, LVP); 2503 2504 // Hexagon handles relocs in pretty printer 2505 if (InlineRelocs && Obj.getArch() != Triple::hexagon) { 2506 while (findRel()) { 2507 // When --adjust-vma is used, update the address printed. 2508 if (RelCur->getSymbol() != Obj.symbol_end()) { 2509 Expected<section_iterator> SymSI = 2510 RelCur->getSymbol()->getSection(); 2511 if (SymSI && *SymSI != Obj.section_end() && 2512 shouldAdjustVA(**SymSI)) 2513 RelOffset += AdjustVMA; 2514 } 2515 2516 printRelocation(FOS, Obj.getFileName(), *RelCur, 2517 SectionAddr + RelOffset, Is64Bits); 2518 LVP.printAfterOtherLine(FOS, true); 2519 ++RelCur; 2520 } 2521 } 2522 2523 Index += Size; 2524 } 2525 } 2526 } 2527 StringSet<> MissingDisasmSymbolSet = 2528 set_difference(DisasmSymbolSet, FoundDisasmSymbolSet); 2529 for (StringRef Sym : MissingDisasmSymbolSet.keys()) 2530 reportWarning("failed to disassemble missing symbol " + Sym, FileName); 2531 } 2532 2533 static void disassembleObject(ObjectFile *Obj, bool InlineRelocs) { 2534 // If information useful for showing the disassembly is missing, try to find a 2535 // more complete binary and disassemble that instead. 2536 OwningBinary<Binary> FetchedBinary; 2537 if (Obj->symbols().empty()) { 2538 if (std::optional<OwningBinary<Binary>> FetchedBinaryOpt = 2539 fetchBinaryByBuildID(*Obj)) { 2540 if (auto *O = dyn_cast<ObjectFile>(FetchedBinaryOpt->getBinary())) { 2541 if (!O->symbols().empty() || 2542 (!O->sections().empty() && Obj->sections().empty())) { 2543 FetchedBinary = std::move(*FetchedBinaryOpt); 2544 Obj = O; 2545 } 2546 } 2547 } 2548 } 2549 2550 const Target *TheTarget = getTarget(Obj); 2551 2552 // Package up features to be passed to target/subtarget 2553 Expected<SubtargetFeatures> FeaturesValue = Obj->getFeatures(); 2554 if (!FeaturesValue) 2555 reportError(FeaturesValue.takeError(), Obj->getFileName()); 2556 SubtargetFeatures Features = *FeaturesValue; 2557 if (!MAttrs.empty()) { 2558 for (unsigned I = 0; I != MAttrs.size(); ++I) 2559 Features.AddFeature(MAttrs[I]); 2560 } else if (MCPU.empty() && Obj->makeTriple().isAArch64()) { 2561 Features.AddFeature("+all"); 2562 } 2563 2564 if (MCPU.empty()) 2565 MCPU = Obj->tryGetCPUName().value_or("").str(); 2566 2567 if (isArmElf(*Obj)) { 2568 // When disassembling big-endian Arm ELF, the instruction endianness is 2569 // determined in a complex way. In relocatable objects, AAELF32 mandates 2570 // that instruction endianness matches the ELF file endianness; in 2571 // executable images, that's true unless the file header has the EF_ARM_BE8 2572 // flag, in which case instructions are little-endian regardless of data 2573 // endianness. 2574 // 2575 // We must set the big-endian-instructions SubtargetFeature to make the 2576 // disassembler read the instructions the right way round, and also tell 2577 // our own prettyprinter to retrieve the encodings the same way to print in 2578 // hex. 2579 const auto *Elf32BE = dyn_cast<ELF32BEObjectFile>(Obj); 2580 2581 if (Elf32BE && (Elf32BE->isRelocatableObject() || 2582 !(Elf32BE->getPlatformFlags() & ELF::EF_ARM_BE8))) { 2583 Features.AddFeature("+big-endian-instructions"); 2584 ARMPrettyPrinterInst.setInstructionEndianness(llvm::endianness::big); 2585 } else { 2586 ARMPrettyPrinterInst.setInstructionEndianness(llvm::endianness::little); 2587 } 2588 } 2589 2590 DisassemblerTarget PrimaryTarget(TheTarget, *Obj, TripleName, MCPU, Features); 2591 2592 // If we have an ARM object file, we need a second disassembler, because 2593 // ARM CPUs have two different instruction sets: ARM mode, and Thumb mode. 2594 // We use mapping symbols to switch between the two assemblers, where 2595 // appropriate. 2596 std::optional<DisassemblerTarget> SecondaryTarget; 2597 2598 if (isArmElf(*Obj)) { 2599 if (!PrimaryTarget.SubtargetInfo->checkFeatures("+mclass")) { 2600 if (PrimaryTarget.SubtargetInfo->checkFeatures("+thumb-mode")) 2601 Features.AddFeature("-thumb-mode"); 2602 else 2603 Features.AddFeature("+thumb-mode"); 2604 SecondaryTarget.emplace(PrimaryTarget, Features); 2605 } 2606 } else if (const auto *COFFObj = dyn_cast<COFFObjectFile>(Obj)) { 2607 const chpe_metadata *CHPEMetadata = COFFObj->getCHPEMetadata(); 2608 if (CHPEMetadata && CHPEMetadata->CodeMapCount) { 2609 // Set up x86_64 disassembler for ARM64EC binaries. 2610 Triple X64Triple(TripleName); 2611 X64Triple.setArch(Triple::ArchType::x86_64); 2612 2613 std::string Error; 2614 const Target *X64Target = 2615 TargetRegistry::lookupTarget("", X64Triple, Error); 2616 if (X64Target) { 2617 SubtargetFeatures X64Features; 2618 SecondaryTarget.emplace(X64Target, *Obj, X64Triple.getTriple(), "", 2619 X64Features); 2620 } else { 2621 reportWarning(Error, Obj->getFileName()); 2622 } 2623 } 2624 } 2625 2626 const ObjectFile *DbgObj = Obj; 2627 if (!FetchedBinary.getBinary() && !Obj->hasDebugInfo()) { 2628 if (std::optional<OwningBinary<Binary>> DebugBinaryOpt = 2629 fetchBinaryByBuildID(*Obj)) { 2630 if (auto *FetchedObj = 2631 dyn_cast<const ObjectFile>(DebugBinaryOpt->getBinary())) { 2632 if (FetchedObj->hasDebugInfo()) { 2633 FetchedBinary = std::move(*DebugBinaryOpt); 2634 DbgObj = FetchedObj; 2635 } 2636 } 2637 } 2638 } 2639 2640 std::unique_ptr<object::Binary> DSYMBinary; 2641 std::unique_ptr<MemoryBuffer> DSYMBuf; 2642 if (!DbgObj->hasDebugInfo()) { 2643 if (const MachOObjectFile *MachOOF = dyn_cast<MachOObjectFile>(&*Obj)) { 2644 DbgObj = objdump::getMachODSymObject(MachOOF, Obj->getFileName(), 2645 DSYMBinary, DSYMBuf); 2646 if (!DbgObj) 2647 return; 2648 } 2649 } 2650 2651 SourcePrinter SP(DbgObj, TheTarget->getName()); 2652 2653 for (StringRef Opt : DisassemblerOptions) 2654 if (!PrimaryTarget.InstPrinter->applyTargetSpecificCLOption(Opt)) 2655 reportError(Obj->getFileName(), 2656 "Unrecognized disassembler option: " + Opt); 2657 2658 disassembleObject(*Obj, *DbgObj, PrimaryTarget, SecondaryTarget, SP, 2659 InlineRelocs); 2660 } 2661 2662 void Dumper::printRelocations() { 2663 StringRef Fmt = O.getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2664 2665 // Build a mapping from relocation target to a vector of relocation 2666 // sections. Usually, there is an only one relocation section for 2667 // each relocated section. 2668 MapVector<SectionRef, std::vector<SectionRef>> SecToRelSec; 2669 uint64_t Ndx; 2670 for (const SectionRef &Section : ToolSectionFilter(O, &Ndx)) { 2671 if (O.isELF() && (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC)) 2672 continue; 2673 if (Section.relocation_begin() == Section.relocation_end()) 2674 continue; 2675 Expected<section_iterator> SecOrErr = Section.getRelocatedSection(); 2676 if (!SecOrErr) 2677 reportError(O.getFileName(), 2678 "section (" + Twine(Ndx) + 2679 "): unable to get a relocation target: " + 2680 toString(SecOrErr.takeError())); 2681 SecToRelSec[**SecOrErr].push_back(Section); 2682 } 2683 2684 for (std::pair<SectionRef, std::vector<SectionRef>> &P : SecToRelSec) { 2685 StringRef SecName = unwrapOrError(P.first.getName(), O.getFileName()); 2686 outs() << "\nRELOCATION RECORDS FOR [" << SecName << "]:\n"; 2687 uint32_t OffsetPadding = (O.getBytesInAddress() > 4 ? 16 : 8); 2688 uint32_t TypePadding = 24; 2689 outs() << left_justify("OFFSET", OffsetPadding) << " " 2690 << left_justify("TYPE", TypePadding) << " " 2691 << "VALUE\n"; 2692 2693 for (SectionRef Section : P.second) { 2694 // CREL sections require decoding, each section may have its own specific 2695 // decode problems. 2696 if (O.isELF() && ELFSectionRef(Section).getType() == ELF::SHT_CREL) { 2697 StringRef Err = 2698 cast<const ELFObjectFileBase>(O).getCrelDecodeProblem(Section); 2699 if (!Err.empty()) { 2700 reportUniqueWarning(Err); 2701 continue; 2702 } 2703 } 2704 for (const RelocationRef &Reloc : Section.relocations()) { 2705 uint64_t Address = Reloc.getOffset(); 2706 SmallString<32> RelocName; 2707 SmallString<32> ValueStr; 2708 if (Address < StartAddress || Address > StopAddress || getHidden(Reloc)) 2709 continue; 2710 Reloc.getTypeName(RelocName); 2711 if (Error E = 2712 getRelocationValueString(Reloc, SymbolDescription, ValueStr)) 2713 reportUniqueWarning(std::move(E)); 2714 2715 outs() << format(Fmt.data(), Address) << " " 2716 << left_justify(RelocName, TypePadding) << " " << ValueStr 2717 << "\n"; 2718 } 2719 } 2720 } 2721 } 2722 2723 // Returns true if we need to show LMA column when dumping section headers. We 2724 // show it only when the platform is ELF and either we have at least one section 2725 // whose VMA and LMA are different and/or when --show-lma flag is used. 2726 static bool shouldDisplayLMA(const ObjectFile &Obj) { 2727 if (!Obj.isELF()) 2728 return false; 2729 for (const SectionRef &S : ToolSectionFilter(Obj)) 2730 if (S.getAddress() != getELFSectionLMA(S)) 2731 return true; 2732 return ShowLMA; 2733 } 2734 2735 static size_t getMaxSectionNameWidth(const ObjectFile &Obj) { 2736 // Default column width for names is 13 even if no names are that long. 2737 size_t MaxWidth = 13; 2738 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 2739 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 2740 MaxWidth = std::max(MaxWidth, Name.size()); 2741 } 2742 return MaxWidth; 2743 } 2744 2745 void objdump::printSectionHeaders(ObjectFile &Obj) { 2746 if (Obj.isELF() && Obj.sections().empty()) 2747 createFakeELFSections(Obj); 2748 2749 size_t NameWidth = getMaxSectionNameWidth(Obj); 2750 size_t AddressWidth = 2 * Obj.getBytesInAddress(); 2751 bool HasLMAColumn = shouldDisplayLMA(Obj); 2752 outs() << "\nSections:\n"; 2753 if (HasLMAColumn) 2754 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 2755 << left_justify("VMA", AddressWidth) << " " 2756 << left_justify("LMA", AddressWidth) << " Type\n"; 2757 else 2758 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 2759 << left_justify("VMA", AddressWidth) << " Type\n"; 2760 2761 uint64_t Idx; 2762 for (const SectionRef &Section : ToolSectionFilter(Obj, &Idx)) { 2763 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 2764 uint64_t VMA = Section.getAddress(); 2765 if (shouldAdjustVA(Section)) 2766 VMA += AdjustVMA; 2767 2768 uint64_t Size = Section.getSize(); 2769 2770 std::string Type = Section.isText() ? "TEXT" : ""; 2771 if (Section.isData()) 2772 Type += Type.empty() ? "DATA" : ", DATA"; 2773 if (Section.isBSS()) 2774 Type += Type.empty() ? "BSS" : ", BSS"; 2775 if (Section.isDebugSection()) 2776 Type += Type.empty() ? "DEBUG" : ", DEBUG"; 2777 2778 if (HasLMAColumn) 2779 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 2780 Name.str().c_str(), Size) 2781 << format_hex_no_prefix(VMA, AddressWidth) << " " 2782 << format_hex_no_prefix(getELFSectionLMA(Section), AddressWidth) 2783 << " " << Type << "\n"; 2784 else 2785 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 2786 Name.str().c_str(), Size) 2787 << format_hex_no_prefix(VMA, AddressWidth) << " " << Type << "\n"; 2788 } 2789 } 2790 2791 void objdump::printSectionContents(const ObjectFile *Obj) { 2792 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj); 2793 2794 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 2795 StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName()); 2796 uint64_t BaseAddr = Section.getAddress(); 2797 uint64_t Size = Section.getSize(); 2798 if (!Size) 2799 continue; 2800 2801 outs() << "Contents of section "; 2802 StringRef SegmentName = getSegmentName(MachO, Section); 2803 if (!SegmentName.empty()) 2804 outs() << SegmentName << ","; 2805 outs() << Name << ":\n"; 2806 if (Section.isBSS()) { 2807 outs() << format("<skipping contents of bss section at [%04" PRIx64 2808 ", %04" PRIx64 ")>\n", 2809 BaseAddr, BaseAddr + Size); 2810 continue; 2811 } 2812 2813 StringRef Contents = unwrapOrError(Section.getContents(), Obj->getFileName()); 2814 2815 // Dump out the content as hex and printable ascii characters. 2816 for (std::size_t Addr = 0, End = Contents.size(); Addr < End; Addr += 16) { 2817 outs() << format(" %04" PRIx64 " ", BaseAddr + Addr); 2818 // Dump line of hex. 2819 for (std::size_t I = 0; I < 16; ++I) { 2820 if (I != 0 && I % 4 == 0) 2821 outs() << ' '; 2822 if (Addr + I < End) 2823 outs() << hexdigit((Contents[Addr + I] >> 4) & 0xF, true) 2824 << hexdigit(Contents[Addr + I] & 0xF, true); 2825 else 2826 outs() << " "; 2827 } 2828 // Print ascii. 2829 outs() << " "; 2830 for (std::size_t I = 0; I < 16 && Addr + I < End; ++I) { 2831 if (isPrint(static_cast<unsigned char>(Contents[Addr + I]) & 0xFF)) 2832 outs() << Contents[Addr + I]; 2833 else 2834 outs() << "."; 2835 } 2836 outs() << "\n"; 2837 } 2838 } 2839 } 2840 2841 void Dumper::printSymbolTable(StringRef ArchiveName, StringRef ArchitectureName, 2842 bool DumpDynamic) { 2843 if (O.isCOFF() && !DumpDynamic) { 2844 outs() << "\nSYMBOL TABLE:\n"; 2845 printCOFFSymbolTable(cast<const COFFObjectFile>(O)); 2846 return; 2847 } 2848 2849 const StringRef FileName = O.getFileName(); 2850 2851 if (!DumpDynamic) { 2852 outs() << "\nSYMBOL TABLE:\n"; 2853 for (auto I = O.symbol_begin(); I != O.symbol_end(); ++I) 2854 printSymbol(*I, {}, FileName, ArchiveName, ArchitectureName, DumpDynamic); 2855 return; 2856 } 2857 2858 outs() << "\nDYNAMIC SYMBOL TABLE:\n"; 2859 if (!O.isELF()) { 2860 reportWarning( 2861 "this operation is not currently supported for this file format", 2862 FileName); 2863 return; 2864 } 2865 2866 const ELFObjectFileBase *ELF = cast<const ELFObjectFileBase>(&O); 2867 auto Symbols = ELF->getDynamicSymbolIterators(); 2868 Expected<std::vector<VersionEntry>> SymbolVersionsOrErr = 2869 ELF->readDynsymVersions(); 2870 if (!SymbolVersionsOrErr) { 2871 reportWarning(toString(SymbolVersionsOrErr.takeError()), FileName); 2872 SymbolVersionsOrErr = std::vector<VersionEntry>(); 2873 (void)!SymbolVersionsOrErr; 2874 } 2875 for (auto &Sym : Symbols) 2876 printSymbol(Sym, *SymbolVersionsOrErr, FileName, ArchiveName, 2877 ArchitectureName, DumpDynamic); 2878 } 2879 2880 void Dumper::printSymbol(const SymbolRef &Symbol, 2881 ArrayRef<VersionEntry> SymbolVersions, 2882 StringRef FileName, StringRef ArchiveName, 2883 StringRef ArchitectureName, bool DumpDynamic) { 2884 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&O); 2885 Expected<uint64_t> AddrOrErr = Symbol.getAddress(); 2886 if (!AddrOrErr) { 2887 reportUniqueWarning(AddrOrErr.takeError()); 2888 return; 2889 } 2890 2891 // Don't ask a Mach-O STAB symbol for its section unless you know that 2892 // STAB symbol's section field refers to a valid section index. Otherwise 2893 // the symbol may error trying to load a section that does not exist. 2894 bool IsSTAB = false; 2895 if (MachO) { 2896 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 2897 uint8_t NType = 2898 (MachO->is64Bit() ? MachO->getSymbol64TableEntry(SymDRI).n_type 2899 : MachO->getSymbolTableEntry(SymDRI).n_type); 2900 if (NType & MachO::N_STAB) 2901 IsSTAB = true; 2902 } 2903 section_iterator Section = IsSTAB 2904 ? O.section_end() 2905 : unwrapOrError(Symbol.getSection(), FileName, 2906 ArchiveName, ArchitectureName); 2907 2908 uint64_t Address = *AddrOrErr; 2909 if (Section != O.section_end() && shouldAdjustVA(*Section)) 2910 Address += AdjustVMA; 2911 if ((Address < StartAddress) || (Address > StopAddress)) 2912 return; 2913 SymbolRef::Type Type = 2914 unwrapOrError(Symbol.getType(), FileName, ArchiveName, ArchitectureName); 2915 uint32_t Flags = 2916 unwrapOrError(Symbol.getFlags(), FileName, ArchiveName, ArchitectureName); 2917 2918 StringRef Name; 2919 if (Type == SymbolRef::ST_Debug && Section != O.section_end()) { 2920 if (Expected<StringRef> NameOrErr = Section->getName()) 2921 Name = *NameOrErr; 2922 else 2923 consumeError(NameOrErr.takeError()); 2924 2925 } else { 2926 Name = unwrapOrError(Symbol.getName(), FileName, ArchiveName, 2927 ArchitectureName); 2928 } 2929 2930 bool Global = Flags & SymbolRef::SF_Global; 2931 bool Weak = Flags & SymbolRef::SF_Weak; 2932 bool Absolute = Flags & SymbolRef::SF_Absolute; 2933 bool Common = Flags & SymbolRef::SF_Common; 2934 bool Hidden = Flags & SymbolRef::SF_Hidden; 2935 2936 char GlobLoc = ' '; 2937 if ((Section != O.section_end() || Absolute) && !Weak) 2938 GlobLoc = Global ? 'g' : 'l'; 2939 char IFunc = ' '; 2940 if (O.isELF()) { 2941 if (ELFSymbolRef(Symbol).getELFType() == ELF::STT_GNU_IFUNC) 2942 IFunc = 'i'; 2943 if (ELFSymbolRef(Symbol).getBinding() == ELF::STB_GNU_UNIQUE) 2944 GlobLoc = 'u'; 2945 } 2946 2947 char Debug = ' '; 2948 if (DumpDynamic) 2949 Debug = 'D'; 2950 else if (Type == SymbolRef::ST_Debug || Type == SymbolRef::ST_File) 2951 Debug = 'd'; 2952 2953 char FileFunc = ' '; 2954 if (Type == SymbolRef::ST_File) 2955 FileFunc = 'f'; 2956 else if (Type == SymbolRef::ST_Function) 2957 FileFunc = 'F'; 2958 else if (Type == SymbolRef::ST_Data) 2959 FileFunc = 'O'; 2960 2961 const char *Fmt = O.getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2962 2963 outs() << format(Fmt, Address) << " " 2964 << GlobLoc // Local -> 'l', Global -> 'g', Neither -> ' ' 2965 << (Weak ? 'w' : ' ') // Weak? 2966 << ' ' // Constructor. Not supported yet. 2967 << ' ' // Warning. Not supported yet. 2968 << IFunc // Indirect reference to another symbol. 2969 << Debug // Debugging (d) or dynamic (D) symbol. 2970 << FileFunc // Name of function (F), file (f) or object (O). 2971 << ' '; 2972 if (Absolute) { 2973 outs() << "*ABS*"; 2974 } else if (Common) { 2975 outs() << "*COM*"; 2976 } else if (Section == O.section_end()) { 2977 if (O.isXCOFF()) { 2978 XCOFFSymbolRef XCOFFSym = cast<const XCOFFObjectFile>(O).toSymbolRef( 2979 Symbol.getRawDataRefImpl()); 2980 if (XCOFF::N_DEBUG == XCOFFSym.getSectionNumber()) 2981 outs() << "*DEBUG*"; 2982 else 2983 outs() << "*UND*"; 2984 } else 2985 outs() << "*UND*"; 2986 } else { 2987 StringRef SegmentName = getSegmentName(MachO, *Section); 2988 if (!SegmentName.empty()) 2989 outs() << SegmentName << ","; 2990 StringRef SectionName = unwrapOrError(Section->getName(), FileName); 2991 outs() << SectionName; 2992 if (O.isXCOFF()) { 2993 std::optional<SymbolRef> SymRef = 2994 getXCOFFSymbolContainingSymbolRef(cast<XCOFFObjectFile>(O), Symbol); 2995 if (SymRef) { 2996 2997 Expected<StringRef> NameOrErr = SymRef->getName(); 2998 2999 if (NameOrErr) { 3000 outs() << " (csect:"; 3001 std::string SymName = 3002 Demangle ? demangle(*NameOrErr) : NameOrErr->str(); 3003 3004 if (SymbolDescription) 3005 SymName = getXCOFFSymbolDescription(createSymbolInfo(O, *SymRef), 3006 SymName); 3007 3008 outs() << ' ' << SymName; 3009 outs() << ") "; 3010 } else 3011 reportWarning(toString(NameOrErr.takeError()), FileName); 3012 } 3013 } 3014 } 3015 3016 if (Common) 3017 outs() << '\t' << format(Fmt, static_cast<uint64_t>(Symbol.getAlignment())); 3018 else if (O.isXCOFF()) 3019 outs() << '\t' 3020 << format(Fmt, cast<XCOFFObjectFile>(O).getSymbolSize( 3021 Symbol.getRawDataRefImpl())); 3022 else if (O.isELF()) 3023 outs() << '\t' << format(Fmt, ELFSymbolRef(Symbol).getSize()); 3024 else if (O.isWasm()) 3025 outs() << '\t' 3026 << format(Fmt, static_cast<uint64_t>( 3027 cast<WasmObjectFile>(O).getSymbolSize(Symbol))); 3028 3029 if (O.isELF()) { 3030 if (!SymbolVersions.empty()) { 3031 const VersionEntry &Ver = 3032 SymbolVersions[Symbol.getRawDataRefImpl().d.b - 1]; 3033 std::string Str; 3034 if (!Ver.Name.empty()) 3035 Str = Ver.IsVerDef ? ' ' + Ver.Name : '(' + Ver.Name + ')'; 3036 outs() << ' ' << left_justify(Str, 12); 3037 } 3038 3039 uint8_t Other = ELFSymbolRef(Symbol).getOther(); 3040 switch (Other) { 3041 case ELF::STV_DEFAULT: 3042 break; 3043 case ELF::STV_INTERNAL: 3044 outs() << " .internal"; 3045 break; 3046 case ELF::STV_HIDDEN: 3047 outs() << " .hidden"; 3048 break; 3049 case ELF::STV_PROTECTED: 3050 outs() << " .protected"; 3051 break; 3052 default: 3053 outs() << format(" 0x%02x", Other); 3054 break; 3055 } 3056 } else if (Hidden) { 3057 outs() << " .hidden"; 3058 } 3059 3060 std::string SymName = Demangle ? demangle(Name) : Name.str(); 3061 if (O.isXCOFF() && SymbolDescription) 3062 SymName = getXCOFFSymbolDescription(createSymbolInfo(O, Symbol), SymName); 3063 3064 outs() << ' ' << SymName << '\n'; 3065 } 3066 3067 static void printUnwindInfo(const ObjectFile *O) { 3068 outs() << "Unwind info:\n\n"; 3069 3070 if (const COFFObjectFile *Coff = dyn_cast<COFFObjectFile>(O)) 3071 printCOFFUnwindInfo(Coff); 3072 else if (const MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(O)) 3073 printMachOUnwindInfo(MachO); 3074 else 3075 // TODO: Extract DWARF dump tool to objdump. 3076 WithColor::error(errs(), ToolName) 3077 << "This operation is only currently supported " 3078 "for COFF and MachO object files.\n"; 3079 } 3080 3081 /// Dump the raw contents of the __clangast section so the output can be piped 3082 /// into llvm-bcanalyzer. 3083 static void printRawClangAST(const ObjectFile *Obj) { 3084 if (outs().is_displayed()) { 3085 WithColor::error(errs(), ToolName) 3086 << "The -raw-clang-ast option will dump the raw binary contents of " 3087 "the clang ast section.\n" 3088 "Please redirect the output to a file or another program such as " 3089 "llvm-bcanalyzer.\n"; 3090 return; 3091 } 3092 3093 StringRef ClangASTSectionName("__clangast"); 3094 if (Obj->isCOFF()) { 3095 ClangASTSectionName = "clangast"; 3096 } 3097 3098 std::optional<object::SectionRef> ClangASTSection; 3099 for (auto Sec : ToolSectionFilter(*Obj)) { 3100 StringRef Name; 3101 if (Expected<StringRef> NameOrErr = Sec.getName()) 3102 Name = *NameOrErr; 3103 else 3104 consumeError(NameOrErr.takeError()); 3105 3106 if (Name == ClangASTSectionName) { 3107 ClangASTSection = Sec; 3108 break; 3109 } 3110 } 3111 if (!ClangASTSection) 3112 return; 3113 3114 StringRef ClangASTContents = 3115 unwrapOrError(ClangASTSection->getContents(), Obj->getFileName()); 3116 outs().write(ClangASTContents.data(), ClangASTContents.size()); 3117 } 3118 3119 static void printFaultMaps(const ObjectFile *Obj) { 3120 StringRef FaultMapSectionName; 3121 3122 if (Obj->isELF()) { 3123 FaultMapSectionName = ".llvm_faultmaps"; 3124 } else if (Obj->isMachO()) { 3125 FaultMapSectionName = "__llvm_faultmaps"; 3126 } else { 3127 WithColor::error(errs(), ToolName) 3128 << "This operation is only currently supported " 3129 "for ELF and Mach-O executable files.\n"; 3130 return; 3131 } 3132 3133 std::optional<object::SectionRef> FaultMapSection; 3134 3135 for (auto Sec : ToolSectionFilter(*Obj)) { 3136 StringRef Name; 3137 if (Expected<StringRef> NameOrErr = Sec.getName()) 3138 Name = *NameOrErr; 3139 else 3140 consumeError(NameOrErr.takeError()); 3141 3142 if (Name == FaultMapSectionName) { 3143 FaultMapSection = Sec; 3144 break; 3145 } 3146 } 3147 3148 outs() << "FaultMap table:\n"; 3149 3150 if (!FaultMapSection) { 3151 outs() << "<not found>\n"; 3152 return; 3153 } 3154 3155 StringRef FaultMapContents = 3156 unwrapOrError(FaultMapSection->getContents(), Obj->getFileName()); 3157 FaultMapParser FMP(FaultMapContents.bytes_begin(), 3158 FaultMapContents.bytes_end()); 3159 3160 outs() << FMP; 3161 } 3162 3163 void Dumper::printPrivateHeaders() { 3164 reportError(O.getFileName(), "Invalid/Unsupported object file format"); 3165 } 3166 3167 static void printFileHeaders(const ObjectFile *O) { 3168 if (!O->isELF() && !O->isCOFF() && !O->isXCOFF()) 3169 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 3170 3171 Triple::ArchType AT = O->getArch(); 3172 outs() << "architecture: " << Triple::getArchTypeName(AT) << "\n"; 3173 uint64_t Address = unwrapOrError(O->getStartAddress(), O->getFileName()); 3174 3175 StringRef Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 3176 outs() << "start address: " 3177 << "0x" << format(Fmt.data(), Address) << "\n"; 3178 } 3179 3180 static void printArchiveChild(StringRef Filename, const Archive::Child &C) { 3181 Expected<sys::fs::perms> ModeOrErr = C.getAccessMode(); 3182 if (!ModeOrErr) { 3183 WithColor::error(errs(), ToolName) << "ill-formed archive entry.\n"; 3184 consumeError(ModeOrErr.takeError()); 3185 return; 3186 } 3187 sys::fs::perms Mode = ModeOrErr.get(); 3188 outs() << ((Mode & sys::fs::owner_read) ? "r" : "-"); 3189 outs() << ((Mode & sys::fs::owner_write) ? "w" : "-"); 3190 outs() << ((Mode & sys::fs::owner_exe) ? "x" : "-"); 3191 outs() << ((Mode & sys::fs::group_read) ? "r" : "-"); 3192 outs() << ((Mode & sys::fs::group_write) ? "w" : "-"); 3193 outs() << ((Mode & sys::fs::group_exe) ? "x" : "-"); 3194 outs() << ((Mode & sys::fs::others_read) ? "r" : "-"); 3195 outs() << ((Mode & sys::fs::others_write) ? "w" : "-"); 3196 outs() << ((Mode & sys::fs::others_exe) ? "x" : "-"); 3197 3198 outs() << " "; 3199 3200 outs() << format("%d/%d %6" PRId64 " ", unwrapOrError(C.getUID(), Filename), 3201 unwrapOrError(C.getGID(), Filename), 3202 unwrapOrError(C.getRawSize(), Filename)); 3203 3204 StringRef RawLastModified = C.getRawLastModified(); 3205 unsigned Seconds; 3206 if (RawLastModified.getAsInteger(10, Seconds)) 3207 outs() << "(date: \"" << RawLastModified 3208 << "\" contains non-decimal chars) "; 3209 else { 3210 // Since ctime(3) returns a 26 character string of the form: 3211 // "Sun Sep 16 01:03:52 1973\n\0" 3212 // just print 24 characters. 3213 time_t t = Seconds; 3214 outs() << format("%.24s ", ctime(&t)); 3215 } 3216 3217 StringRef Name = ""; 3218 Expected<StringRef> NameOrErr = C.getName(); 3219 if (!NameOrErr) { 3220 consumeError(NameOrErr.takeError()); 3221 Name = unwrapOrError(C.getRawName(), Filename); 3222 } else { 3223 Name = NameOrErr.get(); 3224 } 3225 outs() << Name << "\n"; 3226 } 3227 3228 // For ELF only now. 3229 static bool shouldWarnForInvalidStartStopAddress(ObjectFile *Obj) { 3230 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj)) { 3231 if (Elf->getEType() != ELF::ET_REL) 3232 return true; 3233 } 3234 return false; 3235 } 3236 3237 static void checkForInvalidStartStopAddress(ObjectFile *Obj, 3238 uint64_t Start, uint64_t Stop) { 3239 if (!shouldWarnForInvalidStartStopAddress(Obj)) 3240 return; 3241 3242 for (const SectionRef &Section : Obj->sections()) 3243 if (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC) { 3244 uint64_t BaseAddr = Section.getAddress(); 3245 uint64_t Size = Section.getSize(); 3246 if ((Start < BaseAddr + Size) && Stop > BaseAddr) 3247 return; 3248 } 3249 3250 if (!HasStartAddressFlag) 3251 reportWarning("no section has address less than 0x" + 3252 Twine::utohexstr(Stop) + " specified by --stop-address", 3253 Obj->getFileName()); 3254 else if (!HasStopAddressFlag) 3255 reportWarning("no section has address greater than or equal to 0x" + 3256 Twine::utohexstr(Start) + " specified by --start-address", 3257 Obj->getFileName()); 3258 else 3259 reportWarning("no section overlaps the range [0x" + 3260 Twine::utohexstr(Start) + ",0x" + Twine::utohexstr(Stop) + 3261 ") specified by --start-address/--stop-address", 3262 Obj->getFileName()); 3263 } 3264 3265 static void dumpObject(ObjectFile *O, const Archive *A = nullptr, 3266 const Archive::Child *C = nullptr) { 3267 Expected<std::unique_ptr<Dumper>> DumperOrErr = createDumper(*O); 3268 if (!DumperOrErr) { 3269 reportError(DumperOrErr.takeError(), O->getFileName(), 3270 A ? A->getFileName() : ""); 3271 return; 3272 } 3273 Dumper &D = **DumperOrErr; 3274 3275 // Avoid other output when using a raw option. 3276 if (!RawClangAST) { 3277 outs() << '\n'; 3278 if (A) 3279 outs() << A->getFileName() << "(" << O->getFileName() << ")"; 3280 else 3281 outs() << O->getFileName(); 3282 outs() << ":\tfile format " << O->getFileFormatName().lower() << "\n"; 3283 } 3284 3285 if (HasStartAddressFlag || HasStopAddressFlag) 3286 checkForInvalidStartStopAddress(O, StartAddress, StopAddress); 3287 3288 // TODO: Change print* free functions to Dumper member functions to utilitize 3289 // stateful functions like reportUniqueWarning. 3290 3291 // Note: the order here matches GNU objdump for compatability. 3292 StringRef ArchiveName = A ? A->getFileName() : ""; 3293 if (ArchiveHeaders && !MachOOpt && C) 3294 printArchiveChild(ArchiveName, *C); 3295 if (FileHeaders) 3296 printFileHeaders(O); 3297 if (PrivateHeaders || FirstPrivateHeader) 3298 D.printPrivateHeaders(); 3299 if (SectionHeaders) 3300 printSectionHeaders(*O); 3301 if (SymbolTable) 3302 D.printSymbolTable(ArchiveName); 3303 if (DynamicSymbolTable) 3304 D.printSymbolTable(ArchiveName, /*ArchitectureName=*/"", 3305 /*DumpDynamic=*/true); 3306 if (DwarfDumpType != DIDT_Null) { 3307 std::unique_ptr<DIContext> DICtx = DWARFContext::create(*O); 3308 // Dump the complete DWARF structure. 3309 DIDumpOptions DumpOpts; 3310 DumpOpts.DumpType = DwarfDumpType; 3311 DICtx->dump(outs(), DumpOpts); 3312 } 3313 if (Relocations && !Disassemble) 3314 D.printRelocations(); 3315 if (DynamicRelocations) 3316 D.printDynamicRelocations(); 3317 if (SectionContents) 3318 printSectionContents(O); 3319 if (Disassemble) 3320 disassembleObject(O, Relocations); 3321 if (UnwindInfo) 3322 printUnwindInfo(O); 3323 3324 // Mach-O specific options: 3325 if (ExportsTrie) 3326 printExportsTrie(O); 3327 if (Rebase) 3328 printRebaseTable(O); 3329 if (Bind) 3330 printBindTable(O); 3331 if (LazyBind) 3332 printLazyBindTable(O); 3333 if (WeakBind) 3334 printWeakBindTable(O); 3335 3336 // Other special sections: 3337 if (RawClangAST) 3338 printRawClangAST(O); 3339 if (FaultMapSection) 3340 printFaultMaps(O); 3341 if (Offloading) 3342 dumpOffloadBinary(*O); 3343 } 3344 3345 static void dumpObject(const COFFImportFile *I, const Archive *A, 3346 const Archive::Child *C = nullptr) { 3347 StringRef ArchiveName = A ? A->getFileName() : ""; 3348 3349 // Avoid other output when using a raw option. 3350 if (!RawClangAST) 3351 outs() << '\n' 3352 << ArchiveName << "(" << I->getFileName() << ")" 3353 << ":\tfile format COFF-import-file" 3354 << "\n\n"; 3355 3356 if (ArchiveHeaders && !MachOOpt && C) 3357 printArchiveChild(ArchiveName, *C); 3358 if (SymbolTable) 3359 printCOFFSymbolTable(*I); 3360 } 3361 3362 /// Dump each object file in \a a; 3363 static void dumpArchive(const Archive *A) { 3364 Error Err = Error::success(); 3365 unsigned I = -1; 3366 for (auto &C : A->children(Err)) { 3367 ++I; 3368 Expected<std::unique_ptr<Binary>> ChildOrErr = C.getAsBinary(); 3369 if (!ChildOrErr) { 3370 if (auto E = isNotObjectErrorInvalidFileType(ChildOrErr.takeError())) 3371 reportError(std::move(E), getFileNameForError(C, I), A->getFileName()); 3372 continue; 3373 } 3374 if (ObjectFile *O = dyn_cast<ObjectFile>(&*ChildOrErr.get())) 3375 dumpObject(O, A, &C); 3376 else if (COFFImportFile *I = dyn_cast<COFFImportFile>(&*ChildOrErr.get())) 3377 dumpObject(I, A, &C); 3378 else 3379 reportError(errorCodeToError(object_error::invalid_file_type), 3380 A->getFileName()); 3381 } 3382 if (Err) 3383 reportError(std::move(Err), A->getFileName()); 3384 } 3385 3386 /// Open file and figure out how to dump it. 3387 static void dumpInput(StringRef file) { 3388 // If we are using the Mach-O specific object file parser, then let it parse 3389 // the file and process the command line options. So the -arch flags can 3390 // be used to select specific slices, etc. 3391 if (MachOOpt) { 3392 parseInputMachO(file); 3393 return; 3394 } 3395 3396 // Attempt to open the binary. 3397 OwningBinary<Binary> OBinary = unwrapOrError(createBinary(file), file); 3398 Binary &Binary = *OBinary.getBinary(); 3399 3400 if (Archive *A = dyn_cast<Archive>(&Binary)) 3401 dumpArchive(A); 3402 else if (ObjectFile *O = dyn_cast<ObjectFile>(&Binary)) 3403 dumpObject(O); 3404 else if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(&Binary)) 3405 parseInputMachO(UB); 3406 else if (OffloadBinary *OB = dyn_cast<OffloadBinary>(&Binary)) 3407 dumpOffloadSections(*OB); 3408 else 3409 reportError(errorCodeToError(object_error::invalid_file_type), file); 3410 } 3411 3412 template <typename T> 3413 static void parseIntArg(const llvm::opt::InputArgList &InputArgs, int ID, 3414 T &Value) { 3415 if (const opt::Arg *A = InputArgs.getLastArg(ID)) { 3416 StringRef V(A->getValue()); 3417 if (!llvm::to_integer(V, Value, 0)) { 3418 reportCmdLineError(A->getSpelling() + 3419 ": expected a non-negative integer, but got '" + V + 3420 "'"); 3421 } 3422 } 3423 } 3424 3425 static object::BuildID parseBuildIDArg(const opt::Arg *A) { 3426 StringRef V(A->getValue()); 3427 object::BuildID BID = parseBuildID(V); 3428 if (BID.empty()) 3429 reportCmdLineError(A->getSpelling() + ": expected a build ID, but got '" + 3430 V + "'"); 3431 return BID; 3432 } 3433 3434 void objdump::invalidArgValue(const opt::Arg *A) { 3435 reportCmdLineError("'" + StringRef(A->getValue()) + 3436 "' is not a valid value for '" + A->getSpelling() + "'"); 3437 } 3438 3439 static std::vector<std::string> 3440 commaSeparatedValues(const llvm::opt::InputArgList &InputArgs, int ID) { 3441 std::vector<std::string> Values; 3442 for (StringRef Value : InputArgs.getAllArgValues(ID)) { 3443 llvm::SmallVector<StringRef, 2> SplitValues; 3444 llvm::SplitString(Value, SplitValues, ","); 3445 for (StringRef SplitValue : SplitValues) 3446 Values.push_back(SplitValue.str()); 3447 } 3448 return Values; 3449 } 3450 3451 static void parseOtoolOptions(const llvm::opt::InputArgList &InputArgs) { 3452 MachOOpt = true; 3453 FullLeadingAddr = true; 3454 PrintImmHex = true; 3455 3456 ArchName = InputArgs.getLastArgValue(OTOOL_arch).str(); 3457 LinkOptHints = InputArgs.hasArg(OTOOL_C); 3458 if (InputArgs.hasArg(OTOOL_d)) 3459 FilterSections.push_back("__DATA,__data"); 3460 DylibId = InputArgs.hasArg(OTOOL_D); 3461 UniversalHeaders = InputArgs.hasArg(OTOOL_f); 3462 DataInCode = InputArgs.hasArg(OTOOL_G); 3463 FirstPrivateHeader = InputArgs.hasArg(OTOOL_h); 3464 IndirectSymbols = InputArgs.hasArg(OTOOL_I); 3465 ShowRawInsn = InputArgs.hasArg(OTOOL_j); 3466 PrivateHeaders = InputArgs.hasArg(OTOOL_l); 3467 DylibsUsed = InputArgs.hasArg(OTOOL_L); 3468 MCPU = InputArgs.getLastArgValue(OTOOL_mcpu_EQ).str(); 3469 ObjcMetaData = InputArgs.hasArg(OTOOL_o); 3470 DisSymName = InputArgs.getLastArgValue(OTOOL_p).str(); 3471 InfoPlist = InputArgs.hasArg(OTOOL_P); 3472 Relocations = InputArgs.hasArg(OTOOL_r); 3473 if (const Arg *A = InputArgs.getLastArg(OTOOL_s)) { 3474 auto Filter = (A->getValue(0) + StringRef(",") + A->getValue(1)).str(); 3475 FilterSections.push_back(Filter); 3476 } 3477 if (InputArgs.hasArg(OTOOL_t)) 3478 FilterSections.push_back("__TEXT,__text"); 3479 Verbose = InputArgs.hasArg(OTOOL_v) || InputArgs.hasArg(OTOOL_V) || 3480 InputArgs.hasArg(OTOOL_o); 3481 SymbolicOperands = InputArgs.hasArg(OTOOL_V); 3482 if (InputArgs.hasArg(OTOOL_x)) 3483 FilterSections.push_back(",__text"); 3484 LeadingAddr = LeadingHeaders = !InputArgs.hasArg(OTOOL_X); 3485 3486 ChainedFixups = InputArgs.hasArg(OTOOL_chained_fixups); 3487 DyldInfo = InputArgs.hasArg(OTOOL_dyld_info); 3488 3489 InputFilenames = InputArgs.getAllArgValues(OTOOL_INPUT); 3490 if (InputFilenames.empty()) 3491 reportCmdLineError("no input file"); 3492 3493 for (const Arg *A : InputArgs) { 3494 const Option &O = A->getOption(); 3495 if (O.getGroup().isValid() && O.getGroup().getID() == OTOOL_grp_obsolete) { 3496 reportCmdLineWarning(O.getPrefixedName() + 3497 " is obsolete and not implemented"); 3498 } 3499 } 3500 } 3501 3502 static void parseObjdumpOptions(const llvm::opt::InputArgList &InputArgs) { 3503 parseIntArg(InputArgs, OBJDUMP_adjust_vma_EQ, AdjustVMA); 3504 AllHeaders = InputArgs.hasArg(OBJDUMP_all_headers); 3505 ArchName = InputArgs.getLastArgValue(OBJDUMP_arch_name_EQ).str(); 3506 ArchiveHeaders = InputArgs.hasArg(OBJDUMP_archive_headers); 3507 Demangle = InputArgs.hasArg(OBJDUMP_demangle); 3508 Disassemble = InputArgs.hasArg(OBJDUMP_disassemble); 3509 DisassembleAll = InputArgs.hasArg(OBJDUMP_disassemble_all); 3510 SymbolDescription = InputArgs.hasArg(OBJDUMP_symbol_description); 3511 TracebackTable = InputArgs.hasArg(OBJDUMP_traceback_table); 3512 DisassembleSymbols = 3513 commaSeparatedValues(InputArgs, OBJDUMP_disassemble_symbols_EQ); 3514 DisassembleZeroes = InputArgs.hasArg(OBJDUMP_disassemble_zeroes); 3515 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_dwarf_EQ)) { 3516 DwarfDumpType = StringSwitch<DIDumpType>(A->getValue()) 3517 .Case("frames", DIDT_DebugFrame) 3518 .Default(DIDT_Null); 3519 if (DwarfDumpType == DIDT_Null) 3520 invalidArgValue(A); 3521 } 3522 DynamicRelocations = InputArgs.hasArg(OBJDUMP_dynamic_reloc); 3523 FaultMapSection = InputArgs.hasArg(OBJDUMP_fault_map_section); 3524 Offloading = InputArgs.hasArg(OBJDUMP_offloading); 3525 FileHeaders = InputArgs.hasArg(OBJDUMP_file_headers); 3526 SectionContents = InputArgs.hasArg(OBJDUMP_full_contents); 3527 PrintLines = InputArgs.hasArg(OBJDUMP_line_numbers); 3528 InputFilenames = InputArgs.getAllArgValues(OBJDUMP_INPUT); 3529 MachOOpt = InputArgs.hasArg(OBJDUMP_macho); 3530 MCPU = InputArgs.getLastArgValue(OBJDUMP_mcpu_EQ).str(); 3531 MAttrs = commaSeparatedValues(InputArgs, OBJDUMP_mattr_EQ); 3532 ShowRawInsn = !InputArgs.hasArg(OBJDUMP_no_show_raw_insn); 3533 LeadingAddr = !InputArgs.hasArg(OBJDUMP_no_leading_addr); 3534 RawClangAST = InputArgs.hasArg(OBJDUMP_raw_clang_ast); 3535 Relocations = InputArgs.hasArg(OBJDUMP_reloc); 3536 PrintImmHex = 3537 InputArgs.hasFlag(OBJDUMP_print_imm_hex, OBJDUMP_no_print_imm_hex, true); 3538 PrivateHeaders = InputArgs.hasArg(OBJDUMP_private_headers); 3539 FilterSections = InputArgs.getAllArgValues(OBJDUMP_section_EQ); 3540 SectionHeaders = InputArgs.hasArg(OBJDUMP_section_headers); 3541 ShowAllSymbols = InputArgs.hasArg(OBJDUMP_show_all_symbols); 3542 ShowLMA = InputArgs.hasArg(OBJDUMP_show_lma); 3543 PrintSource = InputArgs.hasArg(OBJDUMP_source); 3544 parseIntArg(InputArgs, OBJDUMP_start_address_EQ, StartAddress); 3545 HasStartAddressFlag = InputArgs.hasArg(OBJDUMP_start_address_EQ); 3546 parseIntArg(InputArgs, OBJDUMP_stop_address_EQ, StopAddress); 3547 HasStopAddressFlag = InputArgs.hasArg(OBJDUMP_stop_address_EQ); 3548 SymbolTable = InputArgs.hasArg(OBJDUMP_syms); 3549 SymbolizeOperands = InputArgs.hasArg(OBJDUMP_symbolize_operands); 3550 PrettyPGOAnalysisMap = InputArgs.hasArg(OBJDUMP_pretty_pgo_analysis_map); 3551 if (PrettyPGOAnalysisMap && !SymbolizeOperands) 3552 reportCmdLineWarning("--symbolize-operands must be enabled for " 3553 "--pretty-pgo-analysis-map to have an effect"); 3554 DynamicSymbolTable = InputArgs.hasArg(OBJDUMP_dynamic_syms); 3555 TripleName = InputArgs.getLastArgValue(OBJDUMP_triple_EQ).str(); 3556 UnwindInfo = InputArgs.hasArg(OBJDUMP_unwind_info); 3557 Wide = InputArgs.hasArg(OBJDUMP_wide); 3558 Prefix = InputArgs.getLastArgValue(OBJDUMP_prefix).str(); 3559 parseIntArg(InputArgs, OBJDUMP_prefix_strip, PrefixStrip); 3560 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_debug_vars_EQ)) { 3561 DbgVariables = StringSwitch<DebugVarsFormat>(A->getValue()) 3562 .Case("ascii", DVASCII) 3563 .Case("unicode", DVUnicode) 3564 .Default(DVInvalid); 3565 if (DbgVariables == DVInvalid) 3566 invalidArgValue(A); 3567 } 3568 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_disassembler_color_EQ)) { 3569 DisassemblyColor = StringSwitch<ColorOutput>(A->getValue()) 3570 .Case("on", ColorOutput::Enable) 3571 .Case("off", ColorOutput::Disable) 3572 .Case("terminal", ColorOutput::Auto) 3573 .Default(ColorOutput::Invalid); 3574 if (DisassemblyColor == ColorOutput::Invalid) 3575 invalidArgValue(A); 3576 } 3577 3578 parseIntArg(InputArgs, OBJDUMP_debug_vars_indent_EQ, DbgIndent); 3579 3580 parseMachOOptions(InputArgs); 3581 3582 // Parse -M (--disassembler-options) and deprecated 3583 // --x86-asm-syntax={att,intel}. 3584 // 3585 // Note, for x86, the asm dialect (AssemblerDialect) is initialized when the 3586 // MCAsmInfo is constructed. MCInstPrinter::applyTargetSpecificCLOption is 3587 // called too late. For now we have to use the internal cl::opt option. 3588 const char *AsmSyntax = nullptr; 3589 for (const auto *A : InputArgs.filtered(OBJDUMP_disassembler_options_EQ, 3590 OBJDUMP_x86_asm_syntax_att, 3591 OBJDUMP_x86_asm_syntax_intel)) { 3592 switch (A->getOption().getID()) { 3593 case OBJDUMP_x86_asm_syntax_att: 3594 AsmSyntax = "--x86-asm-syntax=att"; 3595 continue; 3596 case OBJDUMP_x86_asm_syntax_intel: 3597 AsmSyntax = "--x86-asm-syntax=intel"; 3598 continue; 3599 } 3600 3601 SmallVector<StringRef, 2> Values; 3602 llvm::SplitString(A->getValue(), Values, ","); 3603 for (StringRef V : Values) { 3604 if (V == "att") 3605 AsmSyntax = "--x86-asm-syntax=att"; 3606 else if (V == "intel") 3607 AsmSyntax = "--x86-asm-syntax=intel"; 3608 else 3609 DisassemblerOptions.push_back(V.str()); 3610 } 3611 } 3612 SmallVector<const char *> Args = {"llvm-objdump"}; 3613 for (const opt::Arg *A : InputArgs.filtered(OBJDUMP_mllvm)) 3614 Args.push_back(A->getValue()); 3615 if (AsmSyntax) 3616 Args.push_back(AsmSyntax); 3617 if (Args.size() > 1) 3618 llvm::cl::ParseCommandLineOptions(Args.size(), Args.data()); 3619 3620 // Look up any provided build IDs, then append them to the input filenames. 3621 for (const opt::Arg *A : InputArgs.filtered(OBJDUMP_build_id)) { 3622 object::BuildID BuildID = parseBuildIDArg(A); 3623 std::optional<std::string> Path = BIDFetcher->fetch(BuildID); 3624 if (!Path) { 3625 reportCmdLineError(A->getSpelling() + ": could not find build ID '" + 3626 A->getValue() + "'"); 3627 } 3628 InputFilenames.push_back(std::move(*Path)); 3629 } 3630 3631 // objdump defaults to a.out if no filenames specified. 3632 if (InputFilenames.empty()) 3633 InputFilenames.push_back("a.out"); 3634 } 3635 3636 int llvm_objdump_main(int argc, char **argv, const llvm::ToolContext &) { 3637 using namespace llvm; 3638 3639 ToolName = argv[0]; 3640 std::unique_ptr<CommonOptTable> T; 3641 OptSpecifier Unknown, HelpFlag, HelpHiddenFlag, VersionFlag; 3642 3643 StringRef Stem = sys::path::stem(ToolName); 3644 auto Is = [=](StringRef Tool) { 3645 // We need to recognize the following filenames: 3646 // 3647 // llvm-objdump -> objdump 3648 // llvm-otool-10.exe -> otool 3649 // powerpc64-unknown-freebsd13-objdump -> objdump 3650 auto I = Stem.rfind_insensitive(Tool); 3651 return I != StringRef::npos && 3652 (I + Tool.size() == Stem.size() || !isAlnum(Stem[I + Tool.size()])); 3653 }; 3654 if (Is("otool")) { 3655 T = std::make_unique<OtoolOptTable>(); 3656 Unknown = OTOOL_UNKNOWN; 3657 HelpFlag = OTOOL_help; 3658 HelpHiddenFlag = OTOOL_help_hidden; 3659 VersionFlag = OTOOL_version; 3660 } else { 3661 T = std::make_unique<ObjdumpOptTable>(); 3662 Unknown = OBJDUMP_UNKNOWN; 3663 HelpFlag = OBJDUMP_help; 3664 HelpHiddenFlag = OBJDUMP_help_hidden; 3665 VersionFlag = OBJDUMP_version; 3666 } 3667 3668 BumpPtrAllocator A; 3669 StringSaver Saver(A); 3670 opt::InputArgList InputArgs = 3671 T->parseArgs(argc, argv, Unknown, Saver, 3672 [&](StringRef Msg) { reportCmdLineError(Msg); }); 3673 3674 if (InputArgs.size() == 0 || InputArgs.hasArg(HelpFlag)) { 3675 T->printHelp(ToolName); 3676 return 0; 3677 } 3678 if (InputArgs.hasArg(HelpHiddenFlag)) { 3679 T->printHelp(ToolName, /*ShowHidden=*/true); 3680 return 0; 3681 } 3682 3683 // Initialize targets and assembly printers/parsers. 3684 InitializeAllTargetInfos(); 3685 InitializeAllTargetMCs(); 3686 InitializeAllDisassemblers(); 3687 3688 if (InputArgs.hasArg(VersionFlag)) { 3689 cl::PrintVersionMessage(); 3690 if (!Is("otool")) { 3691 outs() << '\n'; 3692 TargetRegistry::printRegisteredTargetsForVersion(outs()); 3693 } 3694 return 0; 3695 } 3696 3697 // Initialize debuginfod. 3698 const bool ShouldUseDebuginfodByDefault = 3699 InputArgs.hasArg(OBJDUMP_build_id) || canUseDebuginfod(); 3700 std::vector<std::string> DebugFileDirectories = 3701 InputArgs.getAllArgValues(OBJDUMP_debug_file_directory); 3702 if (InputArgs.hasFlag(OBJDUMP_debuginfod, OBJDUMP_no_debuginfod, 3703 ShouldUseDebuginfodByDefault)) { 3704 HTTPClient::initialize(); 3705 BIDFetcher = 3706 std::make_unique<DebuginfodFetcher>(std::move(DebugFileDirectories)); 3707 } else { 3708 BIDFetcher = 3709 std::make_unique<BuildIDFetcher>(std::move(DebugFileDirectories)); 3710 } 3711 3712 if (Is("otool")) 3713 parseOtoolOptions(InputArgs); 3714 else 3715 parseObjdumpOptions(InputArgs); 3716 3717 if (StartAddress >= StopAddress) 3718 reportCmdLineError("start address should be less than stop address"); 3719 3720 // Removes trailing separators from prefix. 3721 while (!Prefix.empty() && sys::path::is_separator(Prefix.back())) 3722 Prefix.pop_back(); 3723 3724 if (AllHeaders) 3725 ArchiveHeaders = FileHeaders = PrivateHeaders = Relocations = 3726 SectionHeaders = SymbolTable = true; 3727 3728 if (DisassembleAll || PrintSource || PrintLines || TracebackTable || 3729 !DisassembleSymbols.empty()) 3730 Disassemble = true; 3731 3732 if (!ArchiveHeaders && !Disassemble && DwarfDumpType == DIDT_Null && 3733 !DynamicRelocations && !FileHeaders && !PrivateHeaders && !RawClangAST && 3734 !Relocations && !SectionHeaders && !SectionContents && !SymbolTable && 3735 !DynamicSymbolTable && !UnwindInfo && !FaultMapSection && !Offloading && 3736 !(MachOOpt && 3737 (Bind || DataInCode || ChainedFixups || DyldInfo || DylibId || 3738 DylibsUsed || ExportsTrie || FirstPrivateHeader || 3739 FunctionStartsType != FunctionStartsMode::None || IndirectSymbols || 3740 InfoPlist || LazyBind || LinkOptHints || ObjcMetaData || Rebase || 3741 Rpaths || UniversalHeaders || WeakBind || !FilterSections.empty()))) { 3742 T->printHelp(ToolName); 3743 return 2; 3744 } 3745 3746 DisasmSymbolSet.insert(DisassembleSymbols.begin(), DisassembleSymbols.end()); 3747 3748 llvm::for_each(InputFilenames, dumpInput); 3749 3750 warnOnNoMatchForSections(); 3751 3752 return EXIT_SUCCESS; 3753 } 3754