1 //===-- llvm-objdump.cpp - Object file dumping utility for llvm -----------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This program is a utility that works like binutils "objdump", that is, it 11 // dumps out a plethora of information about an object file depending on the 12 // flags. 13 // 14 // The flags and output of this program should be near identical to those of 15 // binutils objdump. 16 // 17 //===----------------------------------------------------------------------===// 18 19 #include "llvm-objdump.h" 20 #include "llvm/ADT/Optional.h" 21 #include "llvm/ADT/STLExtras.h" 22 #include "llvm/ADT/StringExtras.h" 23 #include "llvm/ADT/StringSet.h" 24 #include "llvm/ADT/Triple.h" 25 #include "llvm/CodeGen/FaultMaps.h" 26 #include "llvm/DebugInfo/DWARF/DWARFContext.h" 27 #include "llvm/DebugInfo/Symbolize/Symbolize.h" 28 #include "llvm/MC/MCAsmInfo.h" 29 #include "llvm/MC/MCContext.h" 30 #include "llvm/MC/MCDisassembler/MCDisassembler.h" 31 #include "llvm/MC/MCDisassembler/MCRelocationInfo.h" 32 #include "llvm/MC/MCInst.h" 33 #include "llvm/MC/MCInstPrinter.h" 34 #include "llvm/MC/MCInstrAnalysis.h" 35 #include "llvm/MC/MCInstrInfo.h" 36 #include "llvm/MC/MCObjectFileInfo.h" 37 #include "llvm/MC/MCRegisterInfo.h" 38 #include "llvm/MC/MCSubtargetInfo.h" 39 #include "llvm/Object/Archive.h" 40 #include "llvm/Object/COFF.h" 41 #include "llvm/Object/COFFImportFile.h" 42 #include "llvm/Object/ELFObjectFile.h" 43 #include "llvm/Object/MachO.h" 44 #include "llvm/Object/ObjectFile.h" 45 #include "llvm/Object/Wasm.h" 46 #include "llvm/Support/Casting.h" 47 #include "llvm/Support/CommandLine.h" 48 #include "llvm/Support/Debug.h" 49 #include "llvm/Support/Errc.h" 50 #include "llvm/Support/FileSystem.h" 51 #include "llvm/Support/Format.h" 52 #include "llvm/Support/GraphWriter.h" 53 #include "llvm/Support/Host.h" 54 #include "llvm/Support/InitLLVM.h" 55 #include "llvm/Support/MemoryBuffer.h" 56 #include "llvm/Support/SourceMgr.h" 57 #include "llvm/Support/TargetRegistry.h" 58 #include "llvm/Support/TargetSelect.h" 59 #include "llvm/Support/raw_ostream.h" 60 #include <algorithm> 61 #include <cctype> 62 #include <cstring> 63 #include <system_error> 64 #include <unordered_map> 65 #include <utility> 66 67 using namespace llvm; 68 using namespace object; 69 70 static cl::list<std::string> 71 InputFilenames(cl::Positional, cl::desc("<input object files>"),cl::ZeroOrMore); 72 73 cl::opt<bool> 74 llvm::Disassemble("disassemble", 75 cl::desc("Display assembler mnemonics for the machine instructions")); 76 static cl::alias 77 Disassembled("d", cl::desc("Alias for --disassemble"), 78 cl::aliasopt(Disassemble)); 79 80 cl::opt<bool> 81 llvm::DisassembleAll("disassemble-all", 82 cl::desc("Display assembler mnemonics for the machine instructions")); 83 static cl::alias 84 DisassembleAlld("D", cl::desc("Alias for --disassemble-all"), 85 cl::aliasopt(DisassembleAll)); 86 87 static cl::list<std::string> 88 DisassembleFunctions("df", 89 cl::CommaSeparated, 90 cl::desc("List of functions to disassemble")); 91 static StringSet<> DisasmFuncsSet; 92 93 cl::opt<bool> 94 llvm::Relocations("r", cl::desc("Display the relocation entries in the file")); 95 96 cl::opt<bool> 97 llvm::DynamicRelocations("dynamic-reloc", 98 cl::desc("Display the dynamic relocation entries in the file")); 99 static cl::alias 100 DynamicRelocationsd("R", cl::desc("Alias for --dynamic-reloc"), 101 cl::aliasopt(DynamicRelocations)); 102 103 cl::opt<bool> 104 llvm::SectionContents("s", cl::desc("Display the content of each section")); 105 106 cl::opt<bool> 107 llvm::SymbolTable("t", cl::desc("Display the symbol table")); 108 109 cl::opt<bool> 110 llvm::ExportsTrie("exports-trie", cl::desc("Display mach-o exported symbols")); 111 112 cl::opt<bool> 113 llvm::Rebase("rebase", cl::desc("Display mach-o rebasing info")); 114 115 cl::opt<bool> 116 llvm::Bind("bind", cl::desc("Display mach-o binding info")); 117 118 cl::opt<bool> 119 llvm::LazyBind("lazy-bind", cl::desc("Display mach-o lazy binding info")); 120 121 cl::opt<bool> 122 llvm::WeakBind("weak-bind", cl::desc("Display mach-o weak binding info")); 123 124 cl::opt<bool> 125 llvm::RawClangAST("raw-clang-ast", 126 cl::desc("Dump the raw binary contents of the clang AST section")); 127 128 static cl::opt<bool> 129 MachOOpt("macho", cl::desc("Use MachO specific object file parser")); 130 static cl::alias 131 MachOm("m", cl::desc("Alias for --macho"), cl::aliasopt(MachOOpt)); 132 133 cl::opt<std::string> 134 llvm::TripleName("triple", cl::desc("Target triple to disassemble for, " 135 "see -version for available targets")); 136 137 cl::opt<std::string> 138 llvm::MCPU("mcpu", 139 cl::desc("Target a specific cpu type (-mcpu=help for details)"), 140 cl::value_desc("cpu-name"), 141 cl::init("")); 142 143 cl::opt<std::string> 144 llvm::ArchName("arch-name", cl::desc("Target arch to disassemble for, " 145 "see -version for available targets")); 146 147 cl::opt<bool> 148 llvm::SectionHeaders("section-headers", cl::desc("Display summaries of the " 149 "headers for each section.")); 150 static cl::alias 151 SectionHeadersShort("headers", cl::desc("Alias for --section-headers"), 152 cl::aliasopt(SectionHeaders)); 153 static cl::alias 154 SectionHeadersShorter("h", cl::desc("Alias for --section-headers"), 155 cl::aliasopt(SectionHeaders)); 156 157 cl::list<std::string> 158 llvm::FilterSections("section", cl::desc("Operate on the specified sections only. " 159 "With -macho dump segment,section")); 160 cl::alias 161 static FilterSectionsj("j", cl::desc("Alias for --section"), 162 cl::aliasopt(llvm::FilterSections)); 163 164 cl::list<std::string> 165 llvm::MAttrs("mattr", 166 cl::CommaSeparated, 167 cl::desc("Target specific attributes"), 168 cl::value_desc("a1,+a2,-a3,...")); 169 170 cl::opt<bool> 171 llvm::NoShowRawInsn("no-show-raw-insn", cl::desc("When disassembling " 172 "instructions, do not print " 173 "the instruction bytes.")); 174 cl::opt<bool> 175 llvm::NoLeadingAddr("no-leading-addr", cl::desc("Print no leading address")); 176 177 cl::opt<bool> 178 llvm::UnwindInfo("unwind-info", cl::desc("Display unwind information")); 179 180 static cl::alias 181 UnwindInfoShort("u", cl::desc("Alias for --unwind-info"), 182 cl::aliasopt(UnwindInfo)); 183 184 cl::opt<bool> 185 llvm::PrivateHeaders("private-headers", 186 cl::desc("Display format specific file headers")); 187 188 cl::opt<bool> 189 llvm::FirstPrivateHeader("private-header", 190 cl::desc("Display only the first format specific file " 191 "header")); 192 193 static cl::alias 194 PrivateHeadersShort("p", cl::desc("Alias for --private-headers"), 195 cl::aliasopt(PrivateHeaders)); 196 197 cl::opt<bool> 198 llvm::PrintImmHex("print-imm-hex", 199 cl::desc("Use hex format for immediate values")); 200 201 cl::opt<bool> PrintFaultMaps("fault-map-section", 202 cl::desc("Display contents of faultmap section")); 203 204 cl::opt<DIDumpType> llvm::DwarfDumpType( 205 "dwarf", cl::init(DIDT_Null), cl::desc("Dump of dwarf debug sections:"), 206 cl::values(clEnumValN(DIDT_DebugFrame, "frames", ".debug_frame"))); 207 208 cl::opt<bool> PrintSource( 209 "source", 210 cl::desc( 211 "Display source inlined with disassembly. Implies disassemble object")); 212 213 cl::alias PrintSourceShort("S", cl::desc("Alias for -source"), 214 cl::aliasopt(PrintSource)); 215 216 cl::opt<bool> PrintLines("line-numbers", 217 cl::desc("Display source line numbers with " 218 "disassembly. Implies disassemble object")); 219 220 cl::alias PrintLinesShort("l", cl::desc("Alias for -line-numbers"), 221 cl::aliasopt(PrintLines)); 222 223 cl::opt<unsigned long long> 224 StartAddress("start-address", cl::desc("Disassemble beginning at address"), 225 cl::value_desc("address"), cl::init(0)); 226 cl::opt<unsigned long long> 227 StopAddress("stop-address", cl::desc("Stop disassembly at address"), 228 cl::value_desc("address"), cl::init(UINT64_MAX)); 229 static StringRef ToolName; 230 231 typedef std::vector<std::tuple<uint64_t, StringRef, uint8_t>> SectionSymbolsTy; 232 233 namespace { 234 typedef std::function<bool(llvm::object::SectionRef const &)> FilterPredicate; 235 236 class SectionFilterIterator { 237 public: 238 SectionFilterIterator(FilterPredicate P, 239 llvm::object::section_iterator const &I, 240 llvm::object::section_iterator const &E) 241 : Predicate(std::move(P)), Iterator(I), End(E) { 242 ScanPredicate(); 243 } 244 const llvm::object::SectionRef &operator*() const { return *Iterator; } 245 SectionFilterIterator &operator++() { 246 ++Iterator; 247 ScanPredicate(); 248 return *this; 249 } 250 bool operator!=(SectionFilterIterator const &Other) const { 251 return Iterator != Other.Iterator; 252 } 253 254 private: 255 void ScanPredicate() { 256 while (Iterator != End && !Predicate(*Iterator)) { 257 ++Iterator; 258 } 259 } 260 FilterPredicate Predicate; 261 llvm::object::section_iterator Iterator; 262 llvm::object::section_iterator End; 263 }; 264 265 class SectionFilter { 266 public: 267 SectionFilter(FilterPredicate P, llvm::object::ObjectFile const &O) 268 : Predicate(std::move(P)), Object(O) {} 269 SectionFilterIterator begin() { 270 return SectionFilterIterator(Predicate, Object.section_begin(), 271 Object.section_end()); 272 } 273 SectionFilterIterator end() { 274 return SectionFilterIterator(Predicate, Object.section_end(), 275 Object.section_end()); 276 } 277 278 private: 279 FilterPredicate Predicate; 280 llvm::object::ObjectFile const &Object; 281 }; 282 SectionFilter ToolSectionFilter(llvm::object::ObjectFile const &O) { 283 return SectionFilter( 284 [](llvm::object::SectionRef const &S) { 285 if (FilterSections.empty()) 286 return true; 287 llvm::StringRef String; 288 std::error_code error = S.getName(String); 289 if (error) 290 return false; 291 return is_contained(FilterSections, String); 292 }, 293 O); 294 } 295 } 296 297 void llvm::error(std::error_code EC) { 298 if (!EC) 299 return; 300 301 errs() << ToolName << ": error reading file: " << EC.message() << ".\n"; 302 errs().flush(); 303 exit(1); 304 } 305 306 LLVM_ATTRIBUTE_NORETURN void llvm::error(Twine Message) { 307 errs() << ToolName << ": " << Message << ".\n"; 308 errs().flush(); 309 exit(1); 310 } 311 312 LLVM_ATTRIBUTE_NORETURN void llvm::report_error(StringRef File, 313 Twine Message) { 314 errs() << ToolName << ": '" << File << "': " << Message << ".\n"; 315 exit(1); 316 } 317 318 LLVM_ATTRIBUTE_NORETURN void llvm::report_error(StringRef File, 319 std::error_code EC) { 320 assert(EC); 321 errs() << ToolName << ": '" << File << "': " << EC.message() << ".\n"; 322 exit(1); 323 } 324 325 LLVM_ATTRIBUTE_NORETURN void llvm::report_error(StringRef File, 326 llvm::Error E) { 327 assert(E); 328 std::string Buf; 329 raw_string_ostream OS(Buf); 330 logAllUnhandledErrors(std::move(E), OS, ""); 331 OS.flush(); 332 errs() << ToolName << ": '" << File << "': " << Buf; 333 exit(1); 334 } 335 336 LLVM_ATTRIBUTE_NORETURN void llvm::report_error(StringRef ArchiveName, 337 StringRef FileName, 338 llvm::Error E, 339 StringRef ArchitectureName) { 340 assert(E); 341 errs() << ToolName << ": "; 342 if (ArchiveName != "") 343 errs() << ArchiveName << "(" << FileName << ")"; 344 else 345 errs() << "'" << FileName << "'"; 346 if (!ArchitectureName.empty()) 347 errs() << " (for architecture " << ArchitectureName << ")"; 348 std::string Buf; 349 raw_string_ostream OS(Buf); 350 logAllUnhandledErrors(std::move(E), OS, ""); 351 OS.flush(); 352 errs() << ": " << Buf; 353 exit(1); 354 } 355 356 LLVM_ATTRIBUTE_NORETURN void llvm::report_error(StringRef ArchiveName, 357 const object::Archive::Child &C, 358 llvm::Error E, 359 StringRef ArchitectureName) { 360 Expected<StringRef> NameOrErr = C.getName(); 361 // TODO: if we have a error getting the name then it would be nice to print 362 // the index of which archive member this is and or its offset in the 363 // archive instead of "???" as the name. 364 if (!NameOrErr) { 365 consumeError(NameOrErr.takeError()); 366 llvm::report_error(ArchiveName, "???", std::move(E), ArchitectureName); 367 } else 368 llvm::report_error(ArchiveName, NameOrErr.get(), std::move(E), 369 ArchitectureName); 370 } 371 372 static const Target *getTarget(const ObjectFile *Obj = nullptr) { 373 // Figure out the target triple. 374 llvm::Triple TheTriple("unknown-unknown-unknown"); 375 if (TripleName.empty()) { 376 if (Obj) { 377 TheTriple = Obj->makeTriple(); 378 } 379 } else { 380 TheTriple.setTriple(Triple::normalize(TripleName)); 381 382 // Use the triple, but also try to combine with ARM build attributes. 383 if (Obj) { 384 auto Arch = Obj->getArch(); 385 if (Arch == Triple::arm || Arch == Triple::armeb) { 386 Obj->setARMSubArch(TheTriple); 387 } 388 } 389 } 390 391 // Get the target specific parser. 392 std::string Error; 393 const Target *TheTarget = TargetRegistry::lookupTarget(ArchName, TheTriple, 394 Error); 395 if (!TheTarget) { 396 if (Obj) 397 report_error(Obj->getFileName(), "can't find target: " + Error); 398 else 399 error("can't find target: " + Error); 400 } 401 402 // Update the triple name and return the found target. 403 TripleName = TheTriple.getTriple(); 404 return TheTarget; 405 } 406 407 bool llvm::RelocAddressLess(RelocationRef a, RelocationRef b) { 408 return a.getOffset() < b.getOffset(); 409 } 410 411 template <class ELFT> 412 static std::error_code getRelocationValueString(const ELFObjectFile<ELFT> *Obj, 413 const RelocationRef &RelRef, 414 SmallVectorImpl<char> &Result) { 415 DataRefImpl Rel = RelRef.getRawDataRefImpl(); 416 417 typedef typename ELFObjectFile<ELFT>::Elf_Sym Elf_Sym; 418 typedef typename ELFObjectFile<ELFT>::Elf_Shdr Elf_Shdr; 419 typedef typename ELFObjectFile<ELFT>::Elf_Rela Elf_Rela; 420 421 const ELFFile<ELFT> &EF = *Obj->getELFFile(); 422 423 auto SecOrErr = EF.getSection(Rel.d.a); 424 if (!SecOrErr) 425 return errorToErrorCode(SecOrErr.takeError()); 426 const Elf_Shdr *Sec = *SecOrErr; 427 auto SymTabOrErr = EF.getSection(Sec->sh_link); 428 if (!SymTabOrErr) 429 return errorToErrorCode(SymTabOrErr.takeError()); 430 const Elf_Shdr *SymTab = *SymTabOrErr; 431 assert(SymTab->sh_type == ELF::SHT_SYMTAB || 432 SymTab->sh_type == ELF::SHT_DYNSYM); 433 auto StrTabSec = EF.getSection(SymTab->sh_link); 434 if (!StrTabSec) 435 return errorToErrorCode(StrTabSec.takeError()); 436 auto StrTabOrErr = EF.getStringTable(*StrTabSec); 437 if (!StrTabOrErr) 438 return errorToErrorCode(StrTabOrErr.takeError()); 439 StringRef StrTab = *StrTabOrErr; 440 int64_t addend = 0; 441 // If there is no Symbol associated with the relocation, we set the undef 442 // boolean value to 'true'. This will prevent us from calling functions that 443 // requires the relocation to be associated with a symbol. 444 bool undef = false; 445 switch (Sec->sh_type) { 446 default: 447 return object_error::parse_failed; 448 case ELF::SHT_REL: { 449 // TODO: Read implicit addend from section data. 450 break; 451 } 452 case ELF::SHT_RELA: { 453 const Elf_Rela *ERela = Obj->getRela(Rel); 454 addend = ERela->r_addend; 455 undef = ERela->getSymbol(false) == 0; 456 break; 457 } 458 } 459 StringRef Target; 460 if (!undef) { 461 symbol_iterator SI = RelRef.getSymbol(); 462 const Elf_Sym *symb = Obj->getSymbol(SI->getRawDataRefImpl()); 463 if (symb->getType() == ELF::STT_SECTION) { 464 Expected<section_iterator> SymSI = SI->getSection(); 465 if (!SymSI) 466 return errorToErrorCode(SymSI.takeError()); 467 const Elf_Shdr *SymSec = Obj->getSection((*SymSI)->getRawDataRefImpl()); 468 auto SecName = EF.getSectionName(SymSec); 469 if (!SecName) 470 return errorToErrorCode(SecName.takeError()); 471 Target = *SecName; 472 } else { 473 Expected<StringRef> SymName = symb->getName(StrTab); 474 if (!SymName) 475 return errorToErrorCode(SymName.takeError()); 476 Target = *SymName; 477 } 478 } else 479 Target = "*ABS*"; 480 481 // Default scheme is to print Target, as well as "+ <addend>" for nonzero 482 // addend. Should be acceptable for all normal purposes. 483 std::string fmtbuf; 484 raw_string_ostream fmt(fmtbuf); 485 fmt << Target; 486 if (addend != 0) 487 fmt << (addend < 0 ? "" : "+") << addend; 488 fmt.flush(); 489 Result.append(fmtbuf.begin(), fmtbuf.end()); 490 return std::error_code(); 491 } 492 493 static std::error_code getRelocationValueString(const ELFObjectFileBase *Obj, 494 const RelocationRef &Rel, 495 SmallVectorImpl<char> &Result) { 496 if (auto *ELF32LE = dyn_cast<ELF32LEObjectFile>(Obj)) 497 return getRelocationValueString(ELF32LE, Rel, Result); 498 if (auto *ELF64LE = dyn_cast<ELF64LEObjectFile>(Obj)) 499 return getRelocationValueString(ELF64LE, Rel, Result); 500 if (auto *ELF32BE = dyn_cast<ELF32BEObjectFile>(Obj)) 501 return getRelocationValueString(ELF32BE, Rel, Result); 502 auto *ELF64BE = cast<ELF64BEObjectFile>(Obj); 503 return getRelocationValueString(ELF64BE, Rel, Result); 504 } 505 506 static std::error_code getRelocationValueString(const COFFObjectFile *Obj, 507 const RelocationRef &Rel, 508 SmallVectorImpl<char> &Result) { 509 symbol_iterator SymI = Rel.getSymbol(); 510 Expected<StringRef> SymNameOrErr = SymI->getName(); 511 if (!SymNameOrErr) 512 return errorToErrorCode(SymNameOrErr.takeError()); 513 StringRef SymName = *SymNameOrErr; 514 Result.append(SymName.begin(), SymName.end()); 515 return std::error_code(); 516 } 517 518 static void printRelocationTargetName(const MachOObjectFile *O, 519 const MachO::any_relocation_info &RE, 520 raw_string_ostream &fmt) { 521 bool IsScattered = O->isRelocationScattered(RE); 522 523 // Target of a scattered relocation is an address. In the interest of 524 // generating pretty output, scan through the symbol table looking for a 525 // symbol that aligns with that address. If we find one, print it. 526 // Otherwise, we just print the hex address of the target. 527 if (IsScattered) { 528 uint32_t Val = O->getPlainRelocationSymbolNum(RE); 529 530 for (const SymbolRef &Symbol : O->symbols()) { 531 std::error_code ec; 532 Expected<uint64_t> Addr = Symbol.getAddress(); 533 if (!Addr) 534 report_error(O->getFileName(), Addr.takeError()); 535 if (*Addr != Val) 536 continue; 537 Expected<StringRef> Name = Symbol.getName(); 538 if (!Name) 539 report_error(O->getFileName(), Name.takeError()); 540 fmt << *Name; 541 return; 542 } 543 544 // If we couldn't find a symbol that this relocation refers to, try 545 // to find a section beginning instead. 546 for (const SectionRef &Section : ToolSectionFilter(*O)) { 547 std::error_code ec; 548 549 StringRef Name; 550 uint64_t Addr = Section.getAddress(); 551 if (Addr != Val) 552 continue; 553 if ((ec = Section.getName(Name))) 554 report_error(O->getFileName(), ec); 555 fmt << Name; 556 return; 557 } 558 559 fmt << format("0x%x", Val); 560 return; 561 } 562 563 StringRef S; 564 bool isExtern = O->getPlainRelocationExternal(RE); 565 uint64_t Val = O->getPlainRelocationSymbolNum(RE); 566 567 if (O->getAnyRelocationType(RE) == MachO::ARM64_RELOC_ADDEND) { 568 fmt << format("0x%0" PRIx64, Val); 569 return; 570 } else if (isExtern) { 571 symbol_iterator SI = O->symbol_begin(); 572 advance(SI, Val); 573 Expected<StringRef> SOrErr = SI->getName(); 574 if (!SOrErr) 575 report_error(O->getFileName(), SOrErr.takeError()); 576 S = *SOrErr; 577 } else { 578 section_iterator SI = O->section_begin(); 579 // Adjust for the fact that sections are 1-indexed. 580 if (Val == 0) { 581 fmt << "0 (?,?)"; 582 return; 583 } 584 uint32_t i = Val - 1; 585 while (i != 0 && SI != O->section_end()) { 586 i--; 587 advance(SI, 1); 588 } 589 if (SI == O->section_end()) 590 fmt << Val << " (?,?)"; 591 else 592 SI->getName(S); 593 } 594 595 fmt << S; 596 } 597 598 static std::error_code getRelocationValueString(const WasmObjectFile *Obj, 599 const RelocationRef &RelRef, 600 SmallVectorImpl<char> &Result) { 601 const wasm::WasmRelocation& Rel = Obj->getWasmRelocation(RelRef); 602 symbol_iterator SI = RelRef.getSymbol(); 603 std::string fmtbuf; 604 raw_string_ostream fmt(fmtbuf); 605 if (SI == Obj->symbol_end()) { 606 // Not all wasm relocations have symbols associated with them. 607 // In particular R_WEBASSEMBLY_TYPE_INDEX_LEB. 608 fmt << Rel.Index; 609 } else { 610 Expected<StringRef> SymNameOrErr = SI->getName(); 611 if (!SymNameOrErr) 612 return errorToErrorCode(SymNameOrErr.takeError()); 613 StringRef SymName = *SymNameOrErr; 614 Result.append(SymName.begin(), SymName.end()); 615 } 616 fmt << (Rel.Addend < 0 ? "" : "+") << Rel.Addend; 617 fmt.flush(); 618 Result.append(fmtbuf.begin(), fmtbuf.end()); 619 return std::error_code(); 620 } 621 622 static std::error_code getRelocationValueString(const MachOObjectFile *Obj, 623 const RelocationRef &RelRef, 624 SmallVectorImpl<char> &Result) { 625 DataRefImpl Rel = RelRef.getRawDataRefImpl(); 626 MachO::any_relocation_info RE = Obj->getRelocation(Rel); 627 628 unsigned Arch = Obj->getArch(); 629 630 std::string fmtbuf; 631 raw_string_ostream fmt(fmtbuf); 632 unsigned Type = Obj->getAnyRelocationType(RE); 633 bool IsPCRel = Obj->getAnyRelocationPCRel(RE); 634 635 // Determine any addends that should be displayed with the relocation. 636 // These require decoding the relocation type, which is triple-specific. 637 638 // X86_64 has entirely custom relocation types. 639 if (Arch == Triple::x86_64) { 640 bool isPCRel = Obj->getAnyRelocationPCRel(RE); 641 642 switch (Type) { 643 case MachO::X86_64_RELOC_GOT_LOAD: 644 case MachO::X86_64_RELOC_GOT: { 645 printRelocationTargetName(Obj, RE, fmt); 646 fmt << "@GOT"; 647 if (isPCRel) 648 fmt << "PCREL"; 649 break; 650 } 651 case MachO::X86_64_RELOC_SUBTRACTOR: { 652 DataRefImpl RelNext = Rel; 653 Obj->moveRelocationNext(RelNext); 654 MachO::any_relocation_info RENext = Obj->getRelocation(RelNext); 655 656 // X86_64_RELOC_SUBTRACTOR must be followed by a relocation of type 657 // X86_64_RELOC_UNSIGNED. 658 // NOTE: Scattered relocations don't exist on x86_64. 659 unsigned RType = Obj->getAnyRelocationType(RENext); 660 if (RType != MachO::X86_64_RELOC_UNSIGNED) 661 report_error(Obj->getFileName(), "Expected X86_64_RELOC_UNSIGNED after " 662 "X86_64_RELOC_SUBTRACTOR."); 663 664 // The X86_64_RELOC_UNSIGNED contains the minuend symbol; 665 // X86_64_RELOC_SUBTRACTOR contains the subtrahend. 666 printRelocationTargetName(Obj, RENext, fmt); 667 fmt << "-"; 668 printRelocationTargetName(Obj, RE, fmt); 669 break; 670 } 671 case MachO::X86_64_RELOC_TLV: 672 printRelocationTargetName(Obj, RE, fmt); 673 fmt << "@TLV"; 674 if (isPCRel) 675 fmt << "P"; 676 break; 677 case MachO::X86_64_RELOC_SIGNED_1: 678 printRelocationTargetName(Obj, RE, fmt); 679 fmt << "-1"; 680 break; 681 case MachO::X86_64_RELOC_SIGNED_2: 682 printRelocationTargetName(Obj, RE, fmt); 683 fmt << "-2"; 684 break; 685 case MachO::X86_64_RELOC_SIGNED_4: 686 printRelocationTargetName(Obj, RE, fmt); 687 fmt << "-4"; 688 break; 689 default: 690 printRelocationTargetName(Obj, RE, fmt); 691 break; 692 } 693 // X86 and ARM share some relocation types in common. 694 } else if (Arch == Triple::x86 || Arch == Triple::arm || 695 Arch == Triple::ppc) { 696 // Generic relocation types... 697 switch (Type) { 698 case MachO::GENERIC_RELOC_PAIR: // prints no info 699 return std::error_code(); 700 case MachO::GENERIC_RELOC_SECTDIFF: { 701 DataRefImpl RelNext = Rel; 702 Obj->moveRelocationNext(RelNext); 703 MachO::any_relocation_info RENext = Obj->getRelocation(RelNext); 704 705 // X86 sect diff's must be followed by a relocation of type 706 // GENERIC_RELOC_PAIR. 707 unsigned RType = Obj->getAnyRelocationType(RENext); 708 709 if (RType != MachO::GENERIC_RELOC_PAIR) 710 report_error(Obj->getFileName(), "Expected GENERIC_RELOC_PAIR after " 711 "GENERIC_RELOC_SECTDIFF."); 712 713 printRelocationTargetName(Obj, RE, fmt); 714 fmt << "-"; 715 printRelocationTargetName(Obj, RENext, fmt); 716 break; 717 } 718 } 719 720 if (Arch == Triple::x86 || Arch == Triple::ppc) { 721 switch (Type) { 722 case MachO::GENERIC_RELOC_LOCAL_SECTDIFF: { 723 DataRefImpl RelNext = Rel; 724 Obj->moveRelocationNext(RelNext); 725 MachO::any_relocation_info RENext = Obj->getRelocation(RelNext); 726 727 // X86 sect diff's must be followed by a relocation of type 728 // GENERIC_RELOC_PAIR. 729 unsigned RType = Obj->getAnyRelocationType(RENext); 730 if (RType != MachO::GENERIC_RELOC_PAIR) 731 report_error(Obj->getFileName(), "Expected GENERIC_RELOC_PAIR after " 732 "GENERIC_RELOC_LOCAL_SECTDIFF."); 733 734 printRelocationTargetName(Obj, RE, fmt); 735 fmt << "-"; 736 printRelocationTargetName(Obj, RENext, fmt); 737 break; 738 } 739 case MachO::GENERIC_RELOC_TLV: { 740 printRelocationTargetName(Obj, RE, fmt); 741 fmt << "@TLV"; 742 if (IsPCRel) 743 fmt << "P"; 744 break; 745 } 746 default: 747 printRelocationTargetName(Obj, RE, fmt); 748 } 749 } else { // ARM-specific relocations 750 switch (Type) { 751 case MachO::ARM_RELOC_HALF: 752 case MachO::ARM_RELOC_HALF_SECTDIFF: { 753 // Half relocations steal a bit from the length field to encode 754 // whether this is an upper16 or a lower16 relocation. 755 bool isUpper = (Obj->getAnyRelocationLength(RE) & 0x1) == 1; 756 757 if (isUpper) 758 fmt << ":upper16:("; 759 else 760 fmt << ":lower16:("; 761 printRelocationTargetName(Obj, RE, fmt); 762 763 DataRefImpl RelNext = Rel; 764 Obj->moveRelocationNext(RelNext); 765 MachO::any_relocation_info RENext = Obj->getRelocation(RelNext); 766 767 // ARM half relocs must be followed by a relocation of type 768 // ARM_RELOC_PAIR. 769 unsigned RType = Obj->getAnyRelocationType(RENext); 770 if (RType != MachO::ARM_RELOC_PAIR) 771 report_error(Obj->getFileName(), "Expected ARM_RELOC_PAIR after " 772 "ARM_RELOC_HALF"); 773 774 // NOTE: The half of the target virtual address is stashed in the 775 // address field of the secondary relocation, but we can't reverse 776 // engineer the constant offset from it without decoding the movw/movt 777 // instruction to find the other half in its immediate field. 778 779 // ARM_RELOC_HALF_SECTDIFF encodes the second section in the 780 // symbol/section pointer of the follow-on relocation. 781 if (Type == MachO::ARM_RELOC_HALF_SECTDIFF) { 782 fmt << "-"; 783 printRelocationTargetName(Obj, RENext, fmt); 784 } 785 786 fmt << ")"; 787 break; 788 } 789 default: { printRelocationTargetName(Obj, RE, fmt); } 790 } 791 } 792 } else 793 printRelocationTargetName(Obj, RE, fmt); 794 795 fmt.flush(); 796 Result.append(fmtbuf.begin(), fmtbuf.end()); 797 return std::error_code(); 798 } 799 800 static std::error_code getRelocationValueString(const RelocationRef &Rel, 801 SmallVectorImpl<char> &Result) { 802 const ObjectFile *Obj = Rel.getObject(); 803 if (auto *ELF = dyn_cast<ELFObjectFileBase>(Obj)) 804 return getRelocationValueString(ELF, Rel, Result); 805 if (auto *COFF = dyn_cast<COFFObjectFile>(Obj)) 806 return getRelocationValueString(COFF, Rel, Result); 807 if (auto *Wasm = dyn_cast<WasmObjectFile>(Obj)) 808 return getRelocationValueString(Wasm, Rel, Result); 809 if (auto *MachO = dyn_cast<MachOObjectFile>(Obj)) 810 return getRelocationValueString(MachO, Rel, Result); 811 llvm_unreachable("unknown object file format"); 812 } 813 814 /// Indicates whether this relocation should hidden when listing 815 /// relocations, usually because it is the trailing part of a multipart 816 /// relocation that will be printed as part of the leading relocation. 817 static bool getHidden(RelocationRef RelRef) { 818 const ObjectFile *Obj = RelRef.getObject(); 819 auto *MachO = dyn_cast<MachOObjectFile>(Obj); 820 if (!MachO) 821 return false; 822 823 unsigned Arch = MachO->getArch(); 824 DataRefImpl Rel = RelRef.getRawDataRefImpl(); 825 uint64_t Type = MachO->getRelocationType(Rel); 826 827 // On arches that use the generic relocations, GENERIC_RELOC_PAIR 828 // is always hidden. 829 if (Arch == Triple::x86 || Arch == Triple::arm || Arch == Triple::ppc) { 830 if (Type == MachO::GENERIC_RELOC_PAIR) 831 return true; 832 } else if (Arch == Triple::x86_64) { 833 // On x86_64, X86_64_RELOC_UNSIGNED is hidden only when it follows 834 // an X86_64_RELOC_SUBTRACTOR. 835 if (Type == MachO::X86_64_RELOC_UNSIGNED && Rel.d.a > 0) { 836 DataRefImpl RelPrev = Rel; 837 RelPrev.d.a--; 838 uint64_t PrevType = MachO->getRelocationType(RelPrev); 839 if (PrevType == MachO::X86_64_RELOC_SUBTRACTOR) 840 return true; 841 } 842 } 843 844 return false; 845 } 846 847 namespace { 848 class SourcePrinter { 849 protected: 850 DILineInfo OldLineInfo; 851 const ObjectFile *Obj = nullptr; 852 std::unique_ptr<symbolize::LLVMSymbolizer> Symbolizer; 853 // File name to file contents of source 854 std::unordered_map<std::string, std::unique_ptr<MemoryBuffer>> SourceCache; 855 // Mark the line endings of the cached source 856 std::unordered_map<std::string, std::vector<StringRef>> LineCache; 857 858 private: 859 bool cacheSource(const DILineInfo& LineInfoFile); 860 861 public: 862 SourcePrinter() = default; 863 SourcePrinter(const ObjectFile *Obj, StringRef DefaultArch) : Obj(Obj) { 864 symbolize::LLVMSymbolizer::Options SymbolizerOpts( 865 DILineInfoSpecifier::FunctionNameKind::None, true, false, false, 866 DefaultArch); 867 Symbolizer.reset(new symbolize::LLVMSymbolizer(SymbolizerOpts)); 868 } 869 virtual ~SourcePrinter() = default; 870 virtual void printSourceLine(raw_ostream &OS, uint64_t Address, 871 StringRef Delimiter = "; "); 872 }; 873 874 bool SourcePrinter::cacheSource(const DILineInfo &LineInfo) { 875 std::unique_ptr<MemoryBuffer> Buffer; 876 if (LineInfo.Source) { 877 Buffer = MemoryBuffer::getMemBuffer(*LineInfo.Source); 878 } else { 879 auto BufferOrError = MemoryBuffer::getFile(LineInfo.FileName); 880 if (!BufferOrError) 881 return false; 882 Buffer = std::move(*BufferOrError); 883 } 884 // Chomp the file to get lines 885 size_t BufferSize = Buffer->getBufferSize(); 886 const char *BufferStart = Buffer->getBufferStart(); 887 for (const char *Start = BufferStart, *End = BufferStart; 888 End < BufferStart + BufferSize; End++) 889 if (*End == '\n' || End == BufferStart + BufferSize - 1 || 890 (*End == '\r' && *(End + 1) == '\n')) { 891 LineCache[LineInfo.FileName].push_back(StringRef(Start, End - Start)); 892 if (*End == '\r') 893 End++; 894 Start = End + 1; 895 } 896 SourceCache[LineInfo.FileName] = std::move(Buffer); 897 return true; 898 } 899 900 void SourcePrinter::printSourceLine(raw_ostream &OS, uint64_t Address, 901 StringRef Delimiter) { 902 if (!Symbolizer) 903 return; 904 DILineInfo LineInfo = DILineInfo(); 905 auto ExpectecLineInfo = 906 Symbolizer->symbolizeCode(Obj->getFileName(), Address); 907 if (!ExpectecLineInfo) 908 consumeError(ExpectecLineInfo.takeError()); 909 else 910 LineInfo = *ExpectecLineInfo; 911 912 if ((LineInfo.FileName == "<invalid>") || OldLineInfo.Line == LineInfo.Line || 913 LineInfo.Line == 0) 914 return; 915 916 if (PrintLines) 917 OS << Delimiter << LineInfo.FileName << ":" << LineInfo.Line << "\n"; 918 if (PrintSource) { 919 if (SourceCache.find(LineInfo.FileName) == SourceCache.end()) 920 if (!cacheSource(LineInfo)) 921 return; 922 auto FileBuffer = SourceCache.find(LineInfo.FileName); 923 if (FileBuffer != SourceCache.end()) { 924 auto LineBuffer = LineCache.find(LineInfo.FileName); 925 if (LineBuffer != LineCache.end()) { 926 if (LineInfo.Line > LineBuffer->second.size()) 927 return; 928 // Vector begins at 0, line numbers are non-zero 929 OS << Delimiter << LineBuffer->second[LineInfo.Line - 1].ltrim() 930 << "\n"; 931 } 932 } 933 } 934 OldLineInfo = LineInfo; 935 } 936 937 static bool isArmElf(const ObjectFile *Obj) { 938 return (Obj->isELF() && 939 (Obj->getArch() == Triple::aarch64 || 940 Obj->getArch() == Triple::aarch64_be || 941 Obj->getArch() == Triple::arm || Obj->getArch() == Triple::armeb || 942 Obj->getArch() == Triple::thumb || 943 Obj->getArch() == Triple::thumbeb)); 944 } 945 946 class PrettyPrinter { 947 public: 948 virtual ~PrettyPrinter() = default; 949 virtual void printInst(MCInstPrinter &IP, const MCInst *MI, 950 ArrayRef<uint8_t> Bytes, uint64_t Address, 951 raw_ostream &OS, StringRef Annot, 952 MCSubtargetInfo const &STI, SourcePrinter *SP, 953 std::vector<RelocationRef> *Rels = nullptr) { 954 if (SP && (PrintSource || PrintLines)) 955 SP->printSourceLine(OS, Address); 956 if (!NoLeadingAddr) 957 OS << format("%8" PRIx64 ":", Address); 958 if (!NoShowRawInsn) { 959 OS << "\t"; 960 dumpBytes(Bytes, OS); 961 } 962 if (MI) 963 IP.printInst(MI, OS, "", STI); 964 else 965 OS << " <unknown>"; 966 } 967 }; 968 PrettyPrinter PrettyPrinterInst; 969 class HexagonPrettyPrinter : public PrettyPrinter { 970 public: 971 void printLead(ArrayRef<uint8_t> Bytes, uint64_t Address, 972 raw_ostream &OS) { 973 uint32_t opcode = 974 (Bytes[3] << 24) | (Bytes[2] << 16) | (Bytes[1] << 8) | Bytes[0]; 975 if (!NoLeadingAddr) 976 OS << format("%8" PRIx64 ":", Address); 977 if (!NoShowRawInsn) { 978 OS << "\t"; 979 dumpBytes(Bytes.slice(0, 4), OS); 980 OS << format("%08" PRIx32, opcode); 981 } 982 } 983 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 984 uint64_t Address, raw_ostream &OS, StringRef Annot, 985 MCSubtargetInfo const &STI, SourcePrinter *SP, 986 std::vector<RelocationRef> *Rels) override { 987 if (SP && (PrintSource || PrintLines)) 988 SP->printSourceLine(OS, Address, ""); 989 if (!MI) { 990 printLead(Bytes, Address, OS); 991 OS << " <unknown>"; 992 return; 993 } 994 std::string Buffer; 995 { 996 raw_string_ostream TempStream(Buffer); 997 IP.printInst(MI, TempStream, "", STI); 998 } 999 StringRef Contents(Buffer); 1000 // Split off bundle attributes 1001 auto PacketBundle = Contents.rsplit('\n'); 1002 // Split off first instruction from the rest 1003 auto HeadTail = PacketBundle.first.split('\n'); 1004 auto Preamble = " { "; 1005 auto Separator = ""; 1006 StringRef Fmt = "\t\t\t%08" PRIx64 ": "; 1007 std::vector<RelocationRef>::const_iterator rel_cur = Rels->begin(); 1008 std::vector<RelocationRef>::const_iterator rel_end = Rels->end(); 1009 1010 // Hexagon's packets require relocations to be inline rather than 1011 // clustered at the end of the packet. 1012 auto PrintReloc = [&]() -> void { 1013 while ((rel_cur != rel_end) && (rel_cur->getOffset() <= Address)) { 1014 if (rel_cur->getOffset() == Address) { 1015 SmallString<16> name; 1016 SmallString<32> val; 1017 rel_cur->getTypeName(name); 1018 error(getRelocationValueString(*rel_cur, val)); 1019 OS << Separator << format(Fmt.data(), Address) << name << "\t" << val 1020 << "\n"; 1021 return; 1022 } 1023 rel_cur++; 1024 } 1025 }; 1026 1027 while(!HeadTail.first.empty()) { 1028 OS << Separator; 1029 Separator = "\n"; 1030 if (SP && (PrintSource || PrintLines)) 1031 SP->printSourceLine(OS, Address, ""); 1032 printLead(Bytes, Address, OS); 1033 OS << Preamble; 1034 Preamble = " "; 1035 StringRef Inst; 1036 auto Duplex = HeadTail.first.split('\v'); 1037 if(!Duplex.second.empty()){ 1038 OS << Duplex.first; 1039 OS << "; "; 1040 Inst = Duplex.second; 1041 } 1042 else 1043 Inst = HeadTail.first; 1044 OS << Inst; 1045 HeadTail = HeadTail.second.split('\n'); 1046 if (HeadTail.first.empty()) 1047 OS << " } " << PacketBundle.second; 1048 PrintReloc(); 1049 Bytes = Bytes.slice(4); 1050 Address += 4; 1051 } 1052 } 1053 }; 1054 HexagonPrettyPrinter HexagonPrettyPrinterInst; 1055 1056 class AMDGCNPrettyPrinter : public PrettyPrinter { 1057 public: 1058 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 1059 uint64_t Address, raw_ostream &OS, StringRef Annot, 1060 MCSubtargetInfo const &STI, SourcePrinter *SP, 1061 std::vector<RelocationRef> *Rels) override { 1062 if (SP && (PrintSource || PrintLines)) 1063 SP->printSourceLine(OS, Address); 1064 1065 typedef support::ulittle32_t U32; 1066 1067 if (MI) { 1068 SmallString<40> InstStr; 1069 raw_svector_ostream IS(InstStr); 1070 1071 IP.printInst(MI, IS, "", STI); 1072 1073 OS << left_justify(IS.str(), 60); 1074 } else { 1075 // an unrecognized encoding - this is probably data so represent it 1076 // using the .long directive, or .byte directive if fewer than 4 bytes 1077 // remaining 1078 if (Bytes.size() >= 4) { 1079 OS << format("\t.long 0x%08" PRIx32 " ", 1080 static_cast<uint32_t>(*reinterpret_cast<const U32*>(Bytes.data()))); 1081 OS.indent(42); 1082 } else { 1083 OS << format("\t.byte 0x%02" PRIx8, Bytes[0]); 1084 for (unsigned int i = 1; i < Bytes.size(); i++) 1085 OS << format(", 0x%02" PRIx8, Bytes[i]); 1086 OS.indent(55 - (6 * Bytes.size())); 1087 } 1088 } 1089 1090 OS << format("// %012" PRIX64 ": ", Address); 1091 if (Bytes.size() >=4) { 1092 for (auto D : makeArrayRef(reinterpret_cast<const U32*>(Bytes.data()), 1093 Bytes.size() / sizeof(U32))) 1094 // D should be explicitly casted to uint32_t here as it is passed 1095 // by format to snprintf as vararg. 1096 OS << format("%08" PRIX32 " ", static_cast<uint32_t>(D)); 1097 } else { 1098 for (unsigned int i = 0; i < Bytes.size(); i++) 1099 OS << format("%02" PRIX8 " ", Bytes[i]); 1100 } 1101 1102 if (!Annot.empty()) 1103 OS << "// " << Annot; 1104 } 1105 }; 1106 AMDGCNPrettyPrinter AMDGCNPrettyPrinterInst; 1107 1108 class BPFPrettyPrinter : public PrettyPrinter { 1109 public: 1110 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 1111 uint64_t Address, raw_ostream &OS, StringRef Annot, 1112 MCSubtargetInfo const &STI, SourcePrinter *SP, 1113 std::vector<RelocationRef> *Rels) override { 1114 if (SP && (PrintSource || PrintLines)) 1115 SP->printSourceLine(OS, Address); 1116 if (!NoLeadingAddr) 1117 OS << format("%8" PRId64 ":", Address / 8); 1118 if (!NoShowRawInsn) { 1119 OS << "\t"; 1120 dumpBytes(Bytes, OS); 1121 } 1122 if (MI) 1123 IP.printInst(MI, OS, "", STI); 1124 else 1125 OS << " <unknown>"; 1126 } 1127 }; 1128 BPFPrettyPrinter BPFPrettyPrinterInst; 1129 1130 PrettyPrinter &selectPrettyPrinter(Triple const &Triple) { 1131 switch(Triple.getArch()) { 1132 default: 1133 return PrettyPrinterInst; 1134 case Triple::hexagon: 1135 return HexagonPrettyPrinterInst; 1136 case Triple::amdgcn: 1137 return AMDGCNPrettyPrinterInst; 1138 case Triple::bpfel: 1139 case Triple::bpfeb: 1140 return BPFPrettyPrinterInst; 1141 } 1142 } 1143 } 1144 1145 static uint8_t getElfSymbolType(const ObjectFile *Obj, const SymbolRef &Sym) { 1146 assert(Obj->isELF()); 1147 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(Obj)) 1148 return Elf32LEObj->getSymbol(Sym.getRawDataRefImpl())->getType(); 1149 if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(Obj)) 1150 return Elf64LEObj->getSymbol(Sym.getRawDataRefImpl())->getType(); 1151 if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(Obj)) 1152 return Elf32BEObj->getSymbol(Sym.getRawDataRefImpl())->getType(); 1153 if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(Obj)) 1154 return Elf64BEObj->getSymbol(Sym.getRawDataRefImpl())->getType(); 1155 llvm_unreachable("Unsupported binary format"); 1156 } 1157 1158 template <class ELFT> static void 1159 addDynamicElfSymbols(const ELFObjectFile<ELFT> *Obj, 1160 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 1161 for (auto Symbol : Obj->getDynamicSymbolIterators()) { 1162 uint8_t SymbolType = Symbol.getELFType(); 1163 if (SymbolType != ELF::STT_FUNC || Symbol.getSize() == 0) 1164 continue; 1165 1166 Expected<uint64_t> AddressOrErr = Symbol.getAddress(); 1167 if (!AddressOrErr) 1168 report_error(Obj->getFileName(), AddressOrErr.takeError()); 1169 uint64_t Address = *AddressOrErr; 1170 1171 Expected<StringRef> Name = Symbol.getName(); 1172 if (!Name) 1173 report_error(Obj->getFileName(), Name.takeError()); 1174 if (Name->empty()) 1175 continue; 1176 1177 Expected<section_iterator> SectionOrErr = Symbol.getSection(); 1178 if (!SectionOrErr) 1179 report_error(Obj->getFileName(), SectionOrErr.takeError()); 1180 section_iterator SecI = *SectionOrErr; 1181 if (SecI == Obj->section_end()) 1182 continue; 1183 1184 AllSymbols[*SecI].emplace_back(Address, *Name, SymbolType); 1185 } 1186 } 1187 1188 static void 1189 addDynamicElfSymbols(const ObjectFile *Obj, 1190 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 1191 assert(Obj->isELF()); 1192 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(Obj)) 1193 addDynamicElfSymbols(Elf32LEObj, AllSymbols); 1194 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(Obj)) 1195 addDynamicElfSymbols(Elf64LEObj, AllSymbols); 1196 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(Obj)) 1197 addDynamicElfSymbols(Elf32BEObj, AllSymbols); 1198 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(Obj)) 1199 addDynamicElfSymbols(Elf64BEObj, AllSymbols); 1200 else 1201 llvm_unreachable("Unsupported binary format"); 1202 } 1203 1204 static void DisassembleObject(const ObjectFile *Obj, bool InlineRelocs) { 1205 if (StartAddress > StopAddress) 1206 error("Start address should be less than stop address"); 1207 1208 const Target *TheTarget = getTarget(Obj); 1209 1210 // Package up features to be passed to target/subtarget 1211 SubtargetFeatures Features = Obj->getFeatures(); 1212 if (MAttrs.size()) { 1213 for (unsigned i = 0; i != MAttrs.size(); ++i) 1214 Features.AddFeature(MAttrs[i]); 1215 } 1216 1217 std::unique_ptr<const MCRegisterInfo> MRI( 1218 TheTarget->createMCRegInfo(TripleName)); 1219 if (!MRI) 1220 report_error(Obj->getFileName(), "no register info for target " + 1221 TripleName); 1222 1223 // Set up disassembler. 1224 std::unique_ptr<const MCAsmInfo> AsmInfo( 1225 TheTarget->createMCAsmInfo(*MRI, TripleName)); 1226 if (!AsmInfo) 1227 report_error(Obj->getFileName(), "no assembly info for target " + 1228 TripleName); 1229 std::unique_ptr<const MCSubtargetInfo> STI( 1230 TheTarget->createMCSubtargetInfo(TripleName, MCPU, Features.getString())); 1231 if (!STI) 1232 report_error(Obj->getFileName(), "no subtarget info for target " + 1233 TripleName); 1234 std::unique_ptr<const MCInstrInfo> MII(TheTarget->createMCInstrInfo()); 1235 if (!MII) 1236 report_error(Obj->getFileName(), "no instruction info for target " + 1237 TripleName); 1238 MCObjectFileInfo MOFI; 1239 MCContext Ctx(AsmInfo.get(), MRI.get(), &MOFI); 1240 // FIXME: for now initialize MCObjectFileInfo with default values 1241 MOFI.InitMCObjectFileInfo(Triple(TripleName), false, Ctx); 1242 1243 std::unique_ptr<MCDisassembler> DisAsm( 1244 TheTarget->createMCDisassembler(*STI, Ctx)); 1245 if (!DisAsm) 1246 report_error(Obj->getFileName(), "no disassembler for target " + 1247 TripleName); 1248 1249 std::unique_ptr<const MCInstrAnalysis> MIA( 1250 TheTarget->createMCInstrAnalysis(MII.get())); 1251 1252 int AsmPrinterVariant = AsmInfo->getAssemblerDialect(); 1253 std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter( 1254 Triple(TripleName), AsmPrinterVariant, *AsmInfo, *MII, *MRI)); 1255 if (!IP) 1256 report_error(Obj->getFileName(), "no instruction printer for target " + 1257 TripleName); 1258 IP->setPrintImmHex(PrintImmHex); 1259 PrettyPrinter &PIP = selectPrettyPrinter(Triple(TripleName)); 1260 1261 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "\t\t%016" PRIx64 ": " : 1262 "\t\t\t%08" PRIx64 ": "; 1263 1264 SourcePrinter SP(Obj, TheTarget->getName()); 1265 1266 // Create a mapping, RelocSecs = SectionRelocMap[S], where sections 1267 // in RelocSecs contain the relocations for section S. 1268 std::error_code EC; 1269 std::map<SectionRef, SmallVector<SectionRef, 1>> SectionRelocMap; 1270 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1271 section_iterator Sec2 = Section.getRelocatedSection(); 1272 if (Sec2 != Obj->section_end()) 1273 SectionRelocMap[*Sec2].push_back(Section); 1274 } 1275 1276 // Create a mapping from virtual address to symbol name. This is used to 1277 // pretty print the symbols while disassembling. 1278 std::map<SectionRef, SectionSymbolsTy> AllSymbols; 1279 for (const SymbolRef &Symbol : Obj->symbols()) { 1280 Expected<uint64_t> AddressOrErr = Symbol.getAddress(); 1281 if (!AddressOrErr) 1282 report_error(Obj->getFileName(), AddressOrErr.takeError()); 1283 uint64_t Address = *AddressOrErr; 1284 1285 Expected<StringRef> Name = Symbol.getName(); 1286 if (!Name) 1287 report_error(Obj->getFileName(), Name.takeError()); 1288 if (Name->empty()) 1289 continue; 1290 1291 Expected<section_iterator> SectionOrErr = Symbol.getSection(); 1292 if (!SectionOrErr) 1293 report_error(Obj->getFileName(), SectionOrErr.takeError()); 1294 section_iterator SecI = *SectionOrErr; 1295 if (SecI == Obj->section_end()) 1296 continue; 1297 1298 uint8_t SymbolType = ELF::STT_NOTYPE; 1299 if (Obj->isELF()) 1300 SymbolType = getElfSymbolType(Obj, Symbol); 1301 1302 AllSymbols[*SecI].emplace_back(Address, *Name, SymbolType); 1303 1304 } 1305 if (AllSymbols.empty() && Obj->isELF()) 1306 addDynamicElfSymbols(Obj, AllSymbols); 1307 1308 // Create a mapping from virtual address to section. 1309 std::vector<std::pair<uint64_t, SectionRef>> SectionAddresses; 1310 for (SectionRef Sec : Obj->sections()) 1311 SectionAddresses.emplace_back(Sec.getAddress(), Sec); 1312 array_pod_sort(SectionAddresses.begin(), SectionAddresses.end()); 1313 1314 // Linked executables (.exe and .dll files) typically don't include a real 1315 // symbol table but they might contain an export table. 1316 if (const auto *COFFObj = dyn_cast<COFFObjectFile>(Obj)) { 1317 for (const auto &ExportEntry : COFFObj->export_directories()) { 1318 StringRef Name; 1319 error(ExportEntry.getSymbolName(Name)); 1320 if (Name.empty()) 1321 continue; 1322 uint32_t RVA; 1323 error(ExportEntry.getExportRVA(RVA)); 1324 1325 uint64_t VA = COFFObj->getImageBase() + RVA; 1326 auto Sec = std::upper_bound( 1327 SectionAddresses.begin(), SectionAddresses.end(), VA, 1328 [](uint64_t LHS, const std::pair<uint64_t, SectionRef> &RHS) { 1329 return LHS < RHS.first; 1330 }); 1331 if (Sec != SectionAddresses.begin()) 1332 --Sec; 1333 else 1334 Sec = SectionAddresses.end(); 1335 1336 if (Sec != SectionAddresses.end()) 1337 AllSymbols[Sec->second].emplace_back(VA, Name, ELF::STT_NOTYPE); 1338 } 1339 } 1340 1341 // Sort all the symbols, this allows us to use a simple binary search to find 1342 // a symbol near an address. 1343 for (std::pair<const SectionRef, SectionSymbolsTy> &SecSyms : AllSymbols) 1344 array_pod_sort(SecSyms.second.begin(), SecSyms.second.end()); 1345 1346 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1347 if (!DisassembleAll && (!Section.isText() || Section.isVirtual())) 1348 continue; 1349 1350 uint64_t SectionAddr = Section.getAddress(); 1351 uint64_t SectSize = Section.getSize(); 1352 if (!SectSize) 1353 continue; 1354 1355 // Get the list of all the symbols in this section. 1356 SectionSymbolsTy &Symbols = AllSymbols[Section]; 1357 std::vector<uint64_t> DataMappingSymsAddr; 1358 std::vector<uint64_t> TextMappingSymsAddr; 1359 if (isArmElf(Obj)) { 1360 for (const auto &Symb : Symbols) { 1361 uint64_t Address = std::get<0>(Symb); 1362 StringRef Name = std::get<1>(Symb); 1363 if (Name.startswith("$d")) 1364 DataMappingSymsAddr.push_back(Address - SectionAddr); 1365 if (Name.startswith("$x")) 1366 TextMappingSymsAddr.push_back(Address - SectionAddr); 1367 if (Name.startswith("$a")) 1368 TextMappingSymsAddr.push_back(Address - SectionAddr); 1369 if (Name.startswith("$t")) 1370 TextMappingSymsAddr.push_back(Address - SectionAddr); 1371 } 1372 } 1373 1374 llvm::sort(DataMappingSymsAddr.begin(), DataMappingSymsAddr.end()); 1375 llvm::sort(TextMappingSymsAddr.begin(), TextMappingSymsAddr.end()); 1376 1377 if (Obj->isELF() && Obj->getArch() == Triple::amdgcn) { 1378 // AMDGPU disassembler uses symbolizer for printing labels 1379 std::unique_ptr<MCRelocationInfo> RelInfo( 1380 TheTarget->createMCRelocationInfo(TripleName, Ctx)); 1381 if (RelInfo) { 1382 std::unique_ptr<MCSymbolizer> Symbolizer( 1383 TheTarget->createMCSymbolizer( 1384 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1385 DisAsm->setSymbolizer(std::move(Symbolizer)); 1386 } 1387 } 1388 1389 // Make a list of all the relocations for this section. 1390 std::vector<RelocationRef> Rels; 1391 if (InlineRelocs) { 1392 for (const SectionRef &RelocSec : SectionRelocMap[Section]) { 1393 for (const RelocationRef &Reloc : RelocSec.relocations()) { 1394 Rels.push_back(Reloc); 1395 } 1396 } 1397 } 1398 1399 // Sort relocations by address. 1400 llvm::sort(Rels.begin(), Rels.end(), RelocAddressLess); 1401 1402 StringRef SegmentName = ""; 1403 if (const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj)) { 1404 DataRefImpl DR = Section.getRawDataRefImpl(); 1405 SegmentName = MachO->getSectionFinalSegmentName(DR); 1406 } 1407 StringRef SectionName; 1408 error(Section.getName(SectionName)); 1409 1410 // If the section has no symbol at the start, just insert a dummy one. 1411 if (Symbols.empty() || std::get<0>(Symbols[0]) != 0) { 1412 Symbols.insert( 1413 Symbols.begin(), 1414 std::make_tuple(SectionAddr, SectionName, 1415 Section.isText() ? ELF::STT_FUNC : ELF::STT_OBJECT)); 1416 } 1417 1418 SmallString<40> Comments; 1419 raw_svector_ostream CommentStream(Comments); 1420 1421 StringRef BytesStr; 1422 error(Section.getContents(BytesStr)); 1423 ArrayRef<uint8_t> Bytes(reinterpret_cast<const uint8_t *>(BytesStr.data()), 1424 BytesStr.size()); 1425 1426 uint64_t Size; 1427 uint64_t Index; 1428 bool PrintedSection = false; 1429 1430 std::vector<RelocationRef>::const_iterator rel_cur = Rels.begin(); 1431 std::vector<RelocationRef>::const_iterator rel_end = Rels.end(); 1432 // Disassemble symbol by symbol. 1433 for (unsigned si = 0, se = Symbols.size(); si != se; ++si) { 1434 uint64_t Start = std::get<0>(Symbols[si]) - SectionAddr; 1435 // The end is either the section end or the beginning of the next 1436 // symbol. 1437 uint64_t End = 1438 (si == se - 1) ? SectSize : std::get<0>(Symbols[si + 1]) - SectionAddr; 1439 // Don't try to disassemble beyond the end of section contents. 1440 if (End > SectSize) 1441 End = SectSize; 1442 // If this symbol has the same address as the next symbol, then skip it. 1443 if (Start >= End) 1444 continue; 1445 1446 // Check if we need to skip symbol 1447 // Skip if the symbol's data is not between StartAddress and StopAddress 1448 if (End + SectionAddr < StartAddress || 1449 Start + SectionAddr > StopAddress) { 1450 continue; 1451 } 1452 1453 /// Skip if user requested specific symbols and this is not in the list 1454 if (!DisasmFuncsSet.empty() && 1455 !DisasmFuncsSet.count(std::get<1>(Symbols[si]))) 1456 continue; 1457 1458 if (!PrintedSection) { 1459 PrintedSection = true; 1460 outs() << "Disassembly of section "; 1461 if (!SegmentName.empty()) 1462 outs() << SegmentName << ","; 1463 outs() << SectionName << ':'; 1464 } 1465 1466 // Stop disassembly at the stop address specified 1467 if (End + SectionAddr > StopAddress) 1468 End = StopAddress - SectionAddr; 1469 1470 if (Obj->isELF() && Obj->getArch() == Triple::amdgcn) { 1471 if (std::get<2>(Symbols[si]) == ELF::STT_AMDGPU_HSA_KERNEL) { 1472 // skip amd_kernel_code_t at the begining of kernel symbol (256 bytes) 1473 Start += 256; 1474 } 1475 if (si == se - 1 || 1476 std::get<2>(Symbols[si + 1]) == ELF::STT_AMDGPU_HSA_KERNEL) { 1477 // cut trailing zeroes at the end of kernel 1478 // cut up to 256 bytes 1479 const uint64_t EndAlign = 256; 1480 const auto Limit = End - (std::min)(EndAlign, End - Start); 1481 while (End > Limit && 1482 *reinterpret_cast<const support::ulittle32_t*>(&Bytes[End - 4]) == 0) 1483 End -= 4; 1484 } 1485 } 1486 1487 outs() << '\n' << std::get<1>(Symbols[si]) << ":\n"; 1488 1489 // Don't print raw contents of a virtual section. A virtual section 1490 // doesn't have any contents in the file. 1491 if (Section.isVirtual()) { 1492 outs() << "...\n"; 1493 continue; 1494 } 1495 1496 #ifndef NDEBUG 1497 raw_ostream &DebugOut = DebugFlag ? dbgs() : nulls(); 1498 #else 1499 raw_ostream &DebugOut = nulls(); 1500 #endif 1501 1502 for (Index = Start; Index < End; Index += Size) { 1503 MCInst Inst; 1504 1505 if (Index + SectionAddr < StartAddress || 1506 Index + SectionAddr > StopAddress) { 1507 // skip byte by byte till StartAddress is reached 1508 Size = 1; 1509 continue; 1510 } 1511 // AArch64 ELF binaries can interleave data and text in the 1512 // same section. We rely on the markers introduced to 1513 // understand what we need to dump. If the data marker is within a 1514 // function, it is denoted as a word/short etc 1515 if (isArmElf(Obj) && std::get<2>(Symbols[si]) != ELF::STT_OBJECT && 1516 !DisassembleAll) { 1517 uint64_t Stride = 0; 1518 1519 auto DAI = std::lower_bound(DataMappingSymsAddr.begin(), 1520 DataMappingSymsAddr.end(), Index); 1521 if (DAI != DataMappingSymsAddr.end() && *DAI == Index) { 1522 // Switch to data. 1523 while (Index < End) { 1524 outs() << format("%8" PRIx64 ":", SectionAddr + Index); 1525 outs() << "\t"; 1526 if (Index + 4 <= End) { 1527 Stride = 4; 1528 dumpBytes(Bytes.slice(Index, 4), outs()); 1529 outs() << "\t.word\t"; 1530 uint32_t Data = 0; 1531 if (Obj->isLittleEndian()) { 1532 const auto Word = 1533 reinterpret_cast<const support::ulittle32_t *>( 1534 Bytes.data() + Index); 1535 Data = *Word; 1536 } else { 1537 const auto Word = reinterpret_cast<const support::ubig32_t *>( 1538 Bytes.data() + Index); 1539 Data = *Word; 1540 } 1541 outs() << "0x" << format("%08" PRIx32, Data); 1542 } else if (Index + 2 <= End) { 1543 Stride = 2; 1544 dumpBytes(Bytes.slice(Index, 2), outs()); 1545 outs() << "\t\t.short\t"; 1546 uint16_t Data = 0; 1547 if (Obj->isLittleEndian()) { 1548 const auto Short = 1549 reinterpret_cast<const support::ulittle16_t *>( 1550 Bytes.data() + Index); 1551 Data = *Short; 1552 } else { 1553 const auto Short = 1554 reinterpret_cast<const support::ubig16_t *>(Bytes.data() + 1555 Index); 1556 Data = *Short; 1557 } 1558 outs() << "0x" << format("%04" PRIx16, Data); 1559 } else { 1560 Stride = 1; 1561 dumpBytes(Bytes.slice(Index, 1), outs()); 1562 outs() << "\t\t.byte\t"; 1563 outs() << "0x" << format("%02" PRIx8, Bytes.slice(Index, 1)[0]); 1564 } 1565 Index += Stride; 1566 outs() << "\n"; 1567 auto TAI = std::lower_bound(TextMappingSymsAddr.begin(), 1568 TextMappingSymsAddr.end(), Index); 1569 if (TAI != TextMappingSymsAddr.end() && *TAI == Index) 1570 break; 1571 } 1572 } 1573 } 1574 1575 // If there is a data symbol inside an ELF text section and we are only 1576 // disassembling text (applicable all architectures), 1577 // we are in a situation where we must print the data and not 1578 // disassemble it. 1579 if (Obj->isELF() && std::get<2>(Symbols[si]) == ELF::STT_OBJECT && 1580 !DisassembleAll && Section.isText()) { 1581 // print out data up to 8 bytes at a time in hex and ascii 1582 uint8_t AsciiData[9] = {'\0'}; 1583 uint8_t Byte; 1584 int NumBytes = 0; 1585 1586 for (Index = Start; Index < End; Index += 1) { 1587 if (((SectionAddr + Index) < StartAddress) || 1588 ((SectionAddr + Index) > StopAddress)) 1589 continue; 1590 if (NumBytes == 0) { 1591 outs() << format("%8" PRIx64 ":", SectionAddr + Index); 1592 outs() << "\t"; 1593 } 1594 Byte = Bytes.slice(Index)[0]; 1595 outs() << format(" %02x", Byte); 1596 AsciiData[NumBytes] = isprint(Byte) ? Byte : '.'; 1597 1598 uint8_t IndentOffset = 0; 1599 NumBytes++; 1600 if (Index == End - 1 || NumBytes > 8) { 1601 // Indent the space for less than 8 bytes data. 1602 // 2 spaces for byte and one for space between bytes 1603 IndentOffset = 3 * (8 - NumBytes); 1604 for (int Excess = 8 - NumBytes; Excess < 8; Excess++) 1605 AsciiData[Excess] = '\0'; 1606 NumBytes = 8; 1607 } 1608 if (NumBytes == 8) { 1609 AsciiData[8] = '\0'; 1610 outs() << std::string(IndentOffset, ' ') << " "; 1611 outs() << reinterpret_cast<char *>(AsciiData); 1612 outs() << '\n'; 1613 NumBytes = 0; 1614 } 1615 } 1616 } 1617 if (Index >= End) 1618 break; 1619 1620 // Disassemble a real instruction or a data when disassemble all is 1621 // provided 1622 bool Disassembled = DisAsm->getInstruction(Inst, Size, Bytes.slice(Index), 1623 SectionAddr + Index, DebugOut, 1624 CommentStream); 1625 if (Size == 0) 1626 Size = 1; 1627 1628 PIP.printInst(*IP, Disassembled ? &Inst : nullptr, 1629 Bytes.slice(Index, Size), SectionAddr + Index, outs(), "", 1630 *STI, &SP, &Rels); 1631 outs() << CommentStream.str(); 1632 Comments.clear(); 1633 1634 // Try to resolve the target of a call, tail call, etc. to a specific 1635 // symbol. 1636 if (MIA && (MIA->isCall(Inst) || MIA->isUnconditionalBranch(Inst) || 1637 MIA->isConditionalBranch(Inst))) { 1638 uint64_t Target; 1639 if (MIA->evaluateBranch(Inst, SectionAddr + Index, Size, Target)) { 1640 // In a relocatable object, the target's section must reside in 1641 // the same section as the call instruction or it is accessed 1642 // through a relocation. 1643 // 1644 // In a non-relocatable object, the target may be in any section. 1645 // 1646 // N.B. We don't walk the relocations in the relocatable case yet. 1647 auto *TargetSectionSymbols = &Symbols; 1648 if (!Obj->isRelocatableObject()) { 1649 auto SectionAddress = std::upper_bound( 1650 SectionAddresses.begin(), SectionAddresses.end(), Target, 1651 [](uint64_t LHS, 1652 const std::pair<uint64_t, SectionRef> &RHS) { 1653 return LHS < RHS.first; 1654 }); 1655 if (SectionAddress != SectionAddresses.begin()) { 1656 --SectionAddress; 1657 TargetSectionSymbols = &AllSymbols[SectionAddress->second]; 1658 } else { 1659 TargetSectionSymbols = nullptr; 1660 } 1661 } 1662 1663 // Find the first symbol in the section whose offset is less than 1664 // or equal to the target. 1665 if (TargetSectionSymbols) { 1666 auto TargetSym = std::upper_bound( 1667 TargetSectionSymbols->begin(), TargetSectionSymbols->end(), 1668 Target, [](uint64_t LHS, 1669 const std::tuple<uint64_t, StringRef, uint8_t> &RHS) { 1670 return LHS < std::get<0>(RHS); 1671 }); 1672 if (TargetSym != TargetSectionSymbols->begin()) { 1673 --TargetSym; 1674 uint64_t TargetAddress = std::get<0>(*TargetSym); 1675 StringRef TargetName = std::get<1>(*TargetSym); 1676 outs() << " <" << TargetName; 1677 uint64_t Disp = Target - TargetAddress; 1678 if (Disp) 1679 outs() << "+0x" << Twine::utohexstr(Disp); 1680 outs() << '>'; 1681 } 1682 } 1683 } 1684 } 1685 outs() << "\n"; 1686 1687 // Hexagon does this in pretty printer 1688 if (Obj->getArch() != Triple::hexagon) 1689 // Print relocation for instruction. 1690 while (rel_cur != rel_end) { 1691 bool hidden = getHidden(*rel_cur); 1692 uint64_t addr = rel_cur->getOffset(); 1693 SmallString<16> name; 1694 SmallString<32> val; 1695 1696 // If this relocation is hidden, skip it. 1697 if (hidden || ((SectionAddr + addr) < StartAddress)) { 1698 ++rel_cur; 1699 continue; 1700 } 1701 1702 // Stop when rel_cur's address is past the current instruction. 1703 if (addr >= Index + Size) break; 1704 rel_cur->getTypeName(name); 1705 error(getRelocationValueString(*rel_cur, val)); 1706 outs() << format(Fmt.data(), SectionAddr + addr) << name 1707 << "\t" << val << "\n"; 1708 ++rel_cur; 1709 } 1710 } 1711 } 1712 } 1713 } 1714 1715 void llvm::PrintRelocations(const ObjectFile *Obj) { 1716 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : 1717 "%08" PRIx64; 1718 // Regular objdump doesn't print relocations in non-relocatable object 1719 // files. 1720 if (!Obj->isRelocatableObject()) 1721 return; 1722 1723 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1724 if (Section.relocation_begin() == Section.relocation_end()) 1725 continue; 1726 StringRef secname; 1727 error(Section.getName(secname)); 1728 outs() << "RELOCATION RECORDS FOR [" << secname << "]:\n"; 1729 for (const RelocationRef &Reloc : Section.relocations()) { 1730 bool hidden = getHidden(Reloc); 1731 uint64_t address = Reloc.getOffset(); 1732 SmallString<32> relocname; 1733 SmallString<32> valuestr; 1734 if (address < StartAddress || address > StopAddress || hidden) 1735 continue; 1736 Reloc.getTypeName(relocname); 1737 error(getRelocationValueString(Reloc, valuestr)); 1738 outs() << format(Fmt.data(), address) << " " << relocname << " " 1739 << valuestr << "\n"; 1740 } 1741 outs() << "\n"; 1742 } 1743 } 1744 1745 void llvm::PrintDynamicRelocations(const ObjectFile *Obj) { 1746 1747 // For the moment, this option is for ELF only 1748 if (!Obj->isELF()) 1749 return; 1750 1751 const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj); 1752 1753 if (!Elf || Elf->getEType() != ELF::ET_DYN) { 1754 error("not a dynamic object"); 1755 return; 1756 } 1757 1758 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 1759 1760 std::vector<SectionRef> DynRelSec = Obj->dynamic_relocation_sections(); 1761 if (DynRelSec.empty()) 1762 return; 1763 1764 outs() << "DYNAMIC RELOCATION RECORDS\n"; 1765 for (const SectionRef &Section : DynRelSec) { 1766 if (Section.relocation_begin() == Section.relocation_end()) 1767 continue; 1768 for (const RelocationRef &Reloc : Section.relocations()) { 1769 uint64_t address = Reloc.getOffset(); 1770 SmallString<32> relocname; 1771 SmallString<32> valuestr; 1772 Reloc.getTypeName(relocname); 1773 error(getRelocationValueString(Reloc, valuestr)); 1774 outs() << format(Fmt.data(), address) << " " << relocname << " " 1775 << valuestr << "\n"; 1776 } 1777 } 1778 } 1779 1780 void llvm::PrintSectionHeaders(const ObjectFile *Obj) { 1781 outs() << "Sections:\n" 1782 "Idx Name Size Address Type\n"; 1783 unsigned i = 0; 1784 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1785 StringRef Name; 1786 error(Section.getName(Name)); 1787 uint64_t Address = Section.getAddress(); 1788 uint64_t Size = Section.getSize(); 1789 bool Text = Section.isText(); 1790 bool Data = Section.isData(); 1791 bool BSS = Section.isBSS(); 1792 std::string Type = (std::string(Text ? "TEXT " : "") + 1793 (Data ? "DATA " : "") + (BSS ? "BSS" : "")); 1794 outs() << format("%3d %-13s %08" PRIx64 " %016" PRIx64 " %s\n", i, 1795 Name.str().c_str(), Size, Address, Type.c_str()); 1796 ++i; 1797 } 1798 } 1799 1800 void llvm::PrintSectionContents(const ObjectFile *Obj) { 1801 std::error_code EC; 1802 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1803 StringRef Name; 1804 StringRef Contents; 1805 error(Section.getName(Name)); 1806 uint64_t BaseAddr = Section.getAddress(); 1807 uint64_t Size = Section.getSize(); 1808 if (!Size) 1809 continue; 1810 1811 outs() << "Contents of section " << Name << ":\n"; 1812 if (Section.isBSS()) { 1813 outs() << format("<skipping contents of bss section at [%04" PRIx64 1814 ", %04" PRIx64 ")>\n", 1815 BaseAddr, BaseAddr + Size); 1816 continue; 1817 } 1818 1819 error(Section.getContents(Contents)); 1820 1821 // Dump out the content as hex and printable ascii characters. 1822 for (std::size_t addr = 0, end = Contents.size(); addr < end; addr += 16) { 1823 outs() << format(" %04" PRIx64 " ", BaseAddr + addr); 1824 // Dump line of hex. 1825 for (std::size_t i = 0; i < 16; ++i) { 1826 if (i != 0 && i % 4 == 0) 1827 outs() << ' '; 1828 if (addr + i < end) 1829 outs() << hexdigit((Contents[addr + i] >> 4) & 0xF, true) 1830 << hexdigit(Contents[addr + i] & 0xF, true); 1831 else 1832 outs() << " "; 1833 } 1834 // Print ascii. 1835 outs() << " "; 1836 for (std::size_t i = 0; i < 16 && addr + i < end; ++i) { 1837 if (std::isprint(static_cast<unsigned char>(Contents[addr + i]) & 0xFF)) 1838 outs() << Contents[addr + i]; 1839 else 1840 outs() << "."; 1841 } 1842 outs() << "\n"; 1843 } 1844 } 1845 } 1846 1847 void llvm::PrintSymbolTable(const ObjectFile *o, StringRef ArchiveName, 1848 StringRef ArchitectureName) { 1849 outs() << "SYMBOL TABLE:\n"; 1850 1851 if (const COFFObjectFile *coff = dyn_cast<const COFFObjectFile>(o)) { 1852 printCOFFSymbolTable(coff); 1853 return; 1854 } 1855 for (const SymbolRef &Symbol : o->symbols()) { 1856 Expected<uint64_t> AddressOrError = Symbol.getAddress(); 1857 if (!AddressOrError) 1858 report_error(ArchiveName, o->getFileName(), AddressOrError.takeError(), 1859 ArchitectureName); 1860 uint64_t Address = *AddressOrError; 1861 if ((Address < StartAddress) || (Address > StopAddress)) 1862 continue; 1863 Expected<SymbolRef::Type> TypeOrError = Symbol.getType(); 1864 if (!TypeOrError) 1865 report_error(ArchiveName, o->getFileName(), TypeOrError.takeError(), 1866 ArchitectureName); 1867 SymbolRef::Type Type = *TypeOrError; 1868 uint32_t Flags = Symbol.getFlags(); 1869 Expected<section_iterator> SectionOrErr = Symbol.getSection(); 1870 if (!SectionOrErr) 1871 report_error(ArchiveName, o->getFileName(), SectionOrErr.takeError(), 1872 ArchitectureName); 1873 section_iterator Section = *SectionOrErr; 1874 StringRef Name; 1875 if (Type == SymbolRef::ST_Debug && Section != o->section_end()) { 1876 Section->getName(Name); 1877 } else { 1878 Expected<StringRef> NameOrErr = Symbol.getName(); 1879 if (!NameOrErr) 1880 report_error(ArchiveName, o->getFileName(), NameOrErr.takeError(), 1881 ArchitectureName); 1882 Name = *NameOrErr; 1883 } 1884 1885 bool Global = Flags & SymbolRef::SF_Global; 1886 bool Weak = Flags & SymbolRef::SF_Weak; 1887 bool Absolute = Flags & SymbolRef::SF_Absolute; 1888 bool Common = Flags & SymbolRef::SF_Common; 1889 bool Hidden = Flags & SymbolRef::SF_Hidden; 1890 1891 char GlobLoc = ' '; 1892 if (Type != SymbolRef::ST_Unknown) 1893 GlobLoc = Global ? 'g' : 'l'; 1894 char Debug = (Type == SymbolRef::ST_Debug || Type == SymbolRef::ST_File) 1895 ? 'd' : ' '; 1896 char FileFunc = ' '; 1897 if (Type == SymbolRef::ST_File) 1898 FileFunc = 'f'; 1899 else if (Type == SymbolRef::ST_Function) 1900 FileFunc = 'F'; 1901 1902 const char *Fmt = o->getBytesInAddress() > 4 ? "%016" PRIx64 : 1903 "%08" PRIx64; 1904 1905 outs() << format(Fmt, Address) << " " 1906 << GlobLoc // Local -> 'l', Global -> 'g', Neither -> ' ' 1907 << (Weak ? 'w' : ' ') // Weak? 1908 << ' ' // Constructor. Not supported yet. 1909 << ' ' // Warning. Not supported yet. 1910 << ' ' // Indirect reference to another symbol. 1911 << Debug // Debugging (d) or dynamic (D) symbol. 1912 << FileFunc // Name of function (F), file (f) or object (O). 1913 << ' '; 1914 if (Absolute) { 1915 outs() << "*ABS*"; 1916 } else if (Common) { 1917 outs() << "*COM*"; 1918 } else if (Section == o->section_end()) { 1919 outs() << "*UND*"; 1920 } else { 1921 if (const MachOObjectFile *MachO = 1922 dyn_cast<const MachOObjectFile>(o)) { 1923 DataRefImpl DR = Section->getRawDataRefImpl(); 1924 StringRef SegmentName = MachO->getSectionFinalSegmentName(DR); 1925 outs() << SegmentName << ","; 1926 } 1927 StringRef SectionName; 1928 error(Section->getName(SectionName)); 1929 outs() << SectionName; 1930 } 1931 1932 outs() << '\t'; 1933 if (Common || isa<ELFObjectFileBase>(o)) { 1934 uint64_t Val = 1935 Common ? Symbol.getAlignment() : ELFSymbolRef(Symbol).getSize(); 1936 outs() << format("\t %08" PRIx64 " ", Val); 1937 } 1938 1939 if (Hidden) { 1940 outs() << ".hidden "; 1941 } 1942 outs() << Name 1943 << '\n'; 1944 } 1945 } 1946 1947 static void PrintUnwindInfo(const ObjectFile *o) { 1948 outs() << "Unwind info:\n\n"; 1949 1950 if (const COFFObjectFile *coff = dyn_cast<COFFObjectFile>(o)) { 1951 printCOFFUnwindInfo(coff); 1952 } else if (const MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(o)) 1953 printMachOUnwindInfo(MachO); 1954 else { 1955 // TODO: Extract DWARF dump tool to objdump. 1956 errs() << "This operation is only currently supported " 1957 "for COFF and MachO object files.\n"; 1958 return; 1959 } 1960 } 1961 1962 void llvm::printExportsTrie(const ObjectFile *o) { 1963 outs() << "Exports trie:\n"; 1964 if (const MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(o)) 1965 printMachOExportsTrie(MachO); 1966 else { 1967 errs() << "This operation is only currently supported " 1968 "for Mach-O executable files.\n"; 1969 return; 1970 } 1971 } 1972 1973 void llvm::printRebaseTable(ObjectFile *o) { 1974 outs() << "Rebase table:\n"; 1975 if (MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(o)) 1976 printMachORebaseTable(MachO); 1977 else { 1978 errs() << "This operation is only currently supported " 1979 "for Mach-O executable files.\n"; 1980 return; 1981 } 1982 } 1983 1984 void llvm::printBindTable(ObjectFile *o) { 1985 outs() << "Bind table:\n"; 1986 if (MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(o)) 1987 printMachOBindTable(MachO); 1988 else { 1989 errs() << "This operation is only currently supported " 1990 "for Mach-O executable files.\n"; 1991 return; 1992 } 1993 } 1994 1995 void llvm::printLazyBindTable(ObjectFile *o) { 1996 outs() << "Lazy bind table:\n"; 1997 if (MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(o)) 1998 printMachOLazyBindTable(MachO); 1999 else { 2000 errs() << "This operation is only currently supported " 2001 "for Mach-O executable files.\n"; 2002 return; 2003 } 2004 } 2005 2006 void llvm::printWeakBindTable(ObjectFile *o) { 2007 outs() << "Weak bind table:\n"; 2008 if (MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(o)) 2009 printMachOWeakBindTable(MachO); 2010 else { 2011 errs() << "This operation is only currently supported " 2012 "for Mach-O executable files.\n"; 2013 return; 2014 } 2015 } 2016 2017 /// Dump the raw contents of the __clangast section so the output can be piped 2018 /// into llvm-bcanalyzer. 2019 void llvm::printRawClangAST(const ObjectFile *Obj) { 2020 if (outs().is_displayed()) { 2021 errs() << "The -raw-clang-ast option will dump the raw binary contents of " 2022 "the clang ast section.\n" 2023 "Please redirect the output to a file or another program such as " 2024 "llvm-bcanalyzer.\n"; 2025 return; 2026 } 2027 2028 StringRef ClangASTSectionName("__clangast"); 2029 if (isa<COFFObjectFile>(Obj)) { 2030 ClangASTSectionName = "clangast"; 2031 } 2032 2033 Optional<object::SectionRef> ClangASTSection; 2034 for (auto Sec : ToolSectionFilter(*Obj)) { 2035 StringRef Name; 2036 Sec.getName(Name); 2037 if (Name == ClangASTSectionName) { 2038 ClangASTSection = Sec; 2039 break; 2040 } 2041 } 2042 if (!ClangASTSection) 2043 return; 2044 2045 StringRef ClangASTContents; 2046 error(ClangASTSection.getValue().getContents(ClangASTContents)); 2047 outs().write(ClangASTContents.data(), ClangASTContents.size()); 2048 } 2049 2050 static void printFaultMaps(const ObjectFile *Obj) { 2051 const char *FaultMapSectionName = nullptr; 2052 2053 if (isa<ELFObjectFileBase>(Obj)) { 2054 FaultMapSectionName = ".llvm_faultmaps"; 2055 } else if (isa<MachOObjectFile>(Obj)) { 2056 FaultMapSectionName = "__llvm_faultmaps"; 2057 } else { 2058 errs() << "This operation is only currently supported " 2059 "for ELF and Mach-O executable files.\n"; 2060 return; 2061 } 2062 2063 Optional<object::SectionRef> FaultMapSection; 2064 2065 for (auto Sec : ToolSectionFilter(*Obj)) { 2066 StringRef Name; 2067 Sec.getName(Name); 2068 if (Name == FaultMapSectionName) { 2069 FaultMapSection = Sec; 2070 break; 2071 } 2072 } 2073 2074 outs() << "FaultMap table:\n"; 2075 2076 if (!FaultMapSection.hasValue()) { 2077 outs() << "<not found>\n"; 2078 return; 2079 } 2080 2081 StringRef FaultMapContents; 2082 error(FaultMapSection.getValue().getContents(FaultMapContents)); 2083 2084 FaultMapParser FMP(FaultMapContents.bytes_begin(), 2085 FaultMapContents.bytes_end()); 2086 2087 outs() << FMP; 2088 } 2089 2090 static void printPrivateFileHeaders(const ObjectFile *o, bool onlyFirst) { 2091 if (o->isELF()) 2092 return printELFFileHeader(o); 2093 if (o->isCOFF()) 2094 return printCOFFFileHeader(o); 2095 if (o->isWasm()) 2096 return printWasmFileHeader(o); 2097 if (o->isMachO()) { 2098 printMachOFileHeader(o); 2099 if (!onlyFirst) 2100 printMachOLoadCommands(o); 2101 return; 2102 } 2103 report_error(o->getFileName(), "Invalid/Unsupported object file format"); 2104 } 2105 2106 static void DumpObject(ObjectFile *o, const Archive *a = nullptr) { 2107 StringRef ArchiveName = a != nullptr ? a->getFileName() : ""; 2108 // Avoid other output when using a raw option. 2109 if (!RawClangAST) { 2110 outs() << '\n'; 2111 if (a) 2112 outs() << a->getFileName() << "(" << o->getFileName() << ")"; 2113 else 2114 outs() << o->getFileName(); 2115 outs() << ":\tfile format " << o->getFileFormatName() << "\n\n"; 2116 } 2117 2118 if (Disassemble) 2119 DisassembleObject(o, Relocations); 2120 if (Relocations && !Disassemble) 2121 PrintRelocations(o); 2122 if (DynamicRelocations) 2123 PrintDynamicRelocations(o); 2124 if (SectionHeaders) 2125 PrintSectionHeaders(o); 2126 if (SectionContents) 2127 PrintSectionContents(o); 2128 if (SymbolTable) 2129 PrintSymbolTable(o, ArchiveName); 2130 if (UnwindInfo) 2131 PrintUnwindInfo(o); 2132 if (PrivateHeaders || FirstPrivateHeader) 2133 printPrivateFileHeaders(o, FirstPrivateHeader); 2134 if (ExportsTrie) 2135 printExportsTrie(o); 2136 if (Rebase) 2137 printRebaseTable(o); 2138 if (Bind) 2139 printBindTable(o); 2140 if (LazyBind) 2141 printLazyBindTable(o); 2142 if (WeakBind) 2143 printWeakBindTable(o); 2144 if (RawClangAST) 2145 printRawClangAST(o); 2146 if (PrintFaultMaps) 2147 printFaultMaps(o); 2148 if (DwarfDumpType != DIDT_Null) { 2149 std::unique_ptr<DIContext> DICtx = DWARFContext::create(*o); 2150 // Dump the complete DWARF structure. 2151 DIDumpOptions DumpOpts; 2152 DumpOpts.DumpType = DwarfDumpType; 2153 DICtx->dump(outs(), DumpOpts); 2154 } 2155 } 2156 2157 static void DumpObject(const COFFImportFile *I, const Archive *A) { 2158 StringRef ArchiveName = A ? A->getFileName() : ""; 2159 2160 // Avoid other output when using a raw option. 2161 if (!RawClangAST) 2162 outs() << '\n' 2163 << ArchiveName << "(" << I->getFileName() << ")" 2164 << ":\tfile format COFF-import-file" 2165 << "\n\n"; 2166 2167 if (SymbolTable) 2168 printCOFFSymbolTable(I); 2169 } 2170 2171 /// Dump each object file in \a a; 2172 static void DumpArchive(const Archive *a) { 2173 Error Err = Error::success(); 2174 for (auto &C : a->children(Err)) { 2175 Expected<std::unique_ptr<Binary>> ChildOrErr = C.getAsBinary(); 2176 if (!ChildOrErr) { 2177 if (auto E = isNotObjectErrorInvalidFileType(ChildOrErr.takeError())) 2178 report_error(a->getFileName(), C, std::move(E)); 2179 continue; 2180 } 2181 if (ObjectFile *o = dyn_cast<ObjectFile>(&*ChildOrErr.get())) 2182 DumpObject(o, a); 2183 else if (COFFImportFile *I = dyn_cast<COFFImportFile>(&*ChildOrErr.get())) 2184 DumpObject(I, a); 2185 else 2186 report_error(a->getFileName(), object_error::invalid_file_type); 2187 } 2188 if (Err) 2189 report_error(a->getFileName(), std::move(Err)); 2190 } 2191 2192 /// Open file and figure out how to dump it. 2193 static void DumpInput(StringRef file) { 2194 2195 // If we are using the Mach-O specific object file parser, then let it parse 2196 // the file and process the command line options. So the -arch flags can 2197 // be used to select specific slices, etc. 2198 if (MachOOpt) { 2199 ParseInputMachO(file); 2200 return; 2201 } 2202 2203 // Attempt to open the binary. 2204 Expected<OwningBinary<Binary>> BinaryOrErr = createBinary(file); 2205 if (!BinaryOrErr) 2206 report_error(file, BinaryOrErr.takeError()); 2207 Binary &Binary = *BinaryOrErr.get().getBinary(); 2208 2209 if (Archive *a = dyn_cast<Archive>(&Binary)) 2210 DumpArchive(a); 2211 else if (ObjectFile *o = dyn_cast<ObjectFile>(&Binary)) 2212 DumpObject(o); 2213 else 2214 report_error(file, object_error::invalid_file_type); 2215 } 2216 2217 int main(int argc, char **argv) { 2218 InitLLVM X(argc, argv); 2219 2220 // Initialize targets and assembly printers/parsers. 2221 llvm::InitializeAllTargetInfos(); 2222 llvm::InitializeAllTargetMCs(); 2223 llvm::InitializeAllDisassemblers(); 2224 2225 // Register the target printer for --version. 2226 cl::AddExtraVersionPrinter(TargetRegistry::printRegisteredTargetsForVersion); 2227 2228 cl::ParseCommandLineOptions(argc, argv, "llvm object file dumper\n"); 2229 TripleName = Triple::normalize(TripleName); 2230 2231 ToolName = argv[0]; 2232 2233 // Defaults to a.out if no filenames specified. 2234 if (InputFilenames.size() == 0) 2235 InputFilenames.push_back("a.out"); 2236 2237 if (DisassembleAll || PrintSource || PrintLines) 2238 Disassemble = true; 2239 if (!Disassemble 2240 && !Relocations 2241 && !DynamicRelocations 2242 && !SectionHeaders 2243 && !SectionContents 2244 && !SymbolTable 2245 && !UnwindInfo 2246 && !PrivateHeaders 2247 && !FirstPrivateHeader 2248 && !ExportsTrie 2249 && !Rebase 2250 && !Bind 2251 && !LazyBind 2252 && !WeakBind 2253 && !RawClangAST 2254 && !(UniversalHeaders && MachOOpt) 2255 && !(ArchiveHeaders && MachOOpt) 2256 && !(IndirectSymbols && MachOOpt) 2257 && !(DataInCode && MachOOpt) 2258 && !(LinkOptHints && MachOOpt) 2259 && !(InfoPlist && MachOOpt) 2260 && !(DylibsUsed && MachOOpt) 2261 && !(DylibId && MachOOpt) 2262 && !(ObjcMetaData && MachOOpt) 2263 && !(FilterSections.size() != 0 && MachOOpt) 2264 && !PrintFaultMaps 2265 && DwarfDumpType == DIDT_Null) { 2266 cl::PrintHelpMessage(); 2267 return 2; 2268 } 2269 2270 DisasmFuncsSet.insert(DisassembleFunctions.begin(), 2271 DisassembleFunctions.end()); 2272 2273 llvm::for_each(InputFilenames, DumpInput); 2274 2275 return EXIT_SUCCESS; 2276 } 2277