1 //===- DWARFDebugLine.cpp -------------------------------------------------===// 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 #include "llvm/DebugInfo/DWARF/DWARFDebugLine.h" 10 #include "llvm/ADT/SmallString.h" 11 #include "llvm/ADT/SmallVector.h" 12 #include "llvm/ADT/StringRef.h" 13 #include "llvm/BinaryFormat/Dwarf.h" 14 #include "llvm/DebugInfo/DWARF/DWARFDataExtractor.h" 15 #include "llvm/DebugInfo/DWARF/DWARFDie.h" 16 #include "llvm/DebugInfo/DWARF/DWARFFormValue.h" 17 #include "llvm/Support/Errc.h" 18 #include "llvm/Support/Format.h" 19 #include "llvm/Support/FormatVariadic.h" 20 #include "llvm/Support/raw_ostream.h" 21 #include <algorithm> 22 #include <cassert> 23 #include <cinttypes> 24 #include <cstdint> 25 #include <cstdio> 26 #include <utility> 27 28 using namespace llvm; 29 using namespace dwarf; 30 31 using FileLineInfoKind = DILineInfoSpecifier::FileLineInfoKind; 32 33 namespace { 34 35 struct ContentDescriptor { 36 dwarf::LineNumberEntryFormat Type; 37 dwarf::Form Form; 38 }; 39 40 using ContentDescriptors = SmallVector<ContentDescriptor, 4>; 41 42 } // end anonymous namespace 43 44 static bool versionIsSupported(uint16_t Version) { 45 return Version >= 2 && Version <= 5; 46 } 47 48 void DWARFDebugLine::ContentTypeTracker::trackContentType( 49 dwarf::LineNumberEntryFormat ContentType) { 50 switch (ContentType) { 51 case dwarf::DW_LNCT_timestamp: 52 HasModTime = true; 53 break; 54 case dwarf::DW_LNCT_size: 55 HasLength = true; 56 break; 57 case dwarf::DW_LNCT_MD5: 58 HasMD5 = true; 59 break; 60 case dwarf::DW_LNCT_LLVM_source: 61 HasSource = true; 62 break; 63 default: 64 // We only care about values we consider optional, and new values may be 65 // added in the vendor extension range, so we do not match exhaustively. 66 break; 67 } 68 } 69 70 DWARFDebugLine::Prologue::Prologue() { clear(); } 71 72 bool DWARFDebugLine::Prologue::hasFileAtIndex(uint64_t FileIndex) const { 73 uint16_t DwarfVersion = getVersion(); 74 assert(DwarfVersion != 0 && 75 "line table prologue has no dwarf version information"); 76 if (DwarfVersion >= 5) 77 return FileIndex < FileNames.size(); 78 return FileIndex != 0 && FileIndex <= FileNames.size(); 79 } 80 81 std::optional<uint64_t> 82 DWARFDebugLine::Prologue::getLastValidFileIndex() const { 83 if (FileNames.empty()) 84 return std::nullopt; 85 uint16_t DwarfVersion = getVersion(); 86 assert(DwarfVersion != 0 && 87 "line table prologue has no dwarf version information"); 88 // In DWARF v5 the file names are 0-indexed. 89 if (DwarfVersion >= 5) 90 return FileNames.size() - 1; 91 return FileNames.size(); 92 } 93 94 const llvm::DWARFDebugLine::FileNameEntry & 95 DWARFDebugLine::Prologue::getFileNameEntry(uint64_t Index) const { 96 uint16_t DwarfVersion = getVersion(); 97 assert(DwarfVersion != 0 && 98 "line table prologue has no dwarf version information"); 99 // In DWARF v5 the file names are 0-indexed. 100 if (DwarfVersion >= 5) 101 return FileNames[Index]; 102 return FileNames[Index - 1]; 103 } 104 105 void DWARFDebugLine::Prologue::clear() { 106 TotalLength = PrologueLength = 0; 107 SegSelectorSize = 0; 108 MinInstLength = MaxOpsPerInst = DefaultIsStmt = LineBase = LineRange = 0; 109 OpcodeBase = 0; 110 FormParams = dwarf::FormParams({0, 0, DWARF32}); 111 ContentTypes = ContentTypeTracker(); 112 StandardOpcodeLengths.clear(); 113 IncludeDirectories.clear(); 114 FileNames.clear(); 115 } 116 117 void DWARFDebugLine::Prologue::dump(raw_ostream &OS, 118 DIDumpOptions DumpOptions) const { 119 if (!totalLengthIsValid()) 120 return; 121 int OffsetDumpWidth = 2 * dwarf::getDwarfOffsetByteSize(FormParams.Format); 122 OS << "Line table prologue:\n" 123 << format(" total_length: 0x%0*" PRIx64 "\n", OffsetDumpWidth, 124 TotalLength) 125 << " format: " << dwarf::FormatString(FormParams.Format) << "\n" 126 << format(" version: %u\n", getVersion()); 127 if (!versionIsSupported(getVersion())) 128 return; 129 if (getVersion() >= 5) 130 OS << format(" address_size: %u\n", getAddressSize()) 131 << format(" seg_select_size: %u\n", SegSelectorSize); 132 OS << format(" prologue_length: 0x%0*" PRIx64 "\n", OffsetDumpWidth, 133 PrologueLength) 134 << format(" min_inst_length: %u\n", MinInstLength) 135 << format(getVersion() >= 4 ? "max_ops_per_inst: %u\n" : "", MaxOpsPerInst) 136 << format(" default_is_stmt: %u\n", DefaultIsStmt) 137 << format(" line_base: %i\n", LineBase) 138 << format(" line_range: %u\n", LineRange) 139 << format(" opcode_base: %u\n", OpcodeBase); 140 141 for (uint32_t I = 0; I != StandardOpcodeLengths.size(); ++I) 142 OS << formatv("standard_opcode_lengths[{0}] = {1}\n", 143 static_cast<dwarf::LineNumberOps>(I + 1), 144 StandardOpcodeLengths[I]); 145 146 if (!IncludeDirectories.empty()) { 147 // DWARF v5 starts directory indexes at 0. 148 uint32_t DirBase = getVersion() >= 5 ? 0 : 1; 149 for (uint32_t I = 0; I != IncludeDirectories.size(); ++I) { 150 OS << format("include_directories[%3u] = ", I + DirBase); 151 IncludeDirectories[I].dump(OS, DumpOptions); 152 OS << '\n'; 153 } 154 } 155 156 if (!FileNames.empty()) { 157 // DWARF v5 starts file indexes at 0. 158 uint32_t FileBase = getVersion() >= 5 ? 0 : 1; 159 for (uint32_t I = 0; I != FileNames.size(); ++I) { 160 const FileNameEntry &FileEntry = FileNames[I]; 161 OS << format("file_names[%3u]:\n", I + FileBase); 162 OS << " name: "; 163 FileEntry.Name.dump(OS, DumpOptions); 164 OS << '\n' 165 << format(" dir_index: %" PRIu64 "\n", FileEntry.DirIdx); 166 if (ContentTypes.HasMD5) 167 OS << " md5_checksum: " << FileEntry.Checksum.digest() << '\n'; 168 if (ContentTypes.HasModTime) 169 OS << format(" mod_time: 0x%8.8" PRIx64 "\n", FileEntry.ModTime); 170 if (ContentTypes.HasLength) 171 OS << format(" length: 0x%8.8" PRIx64 "\n", FileEntry.Length); 172 if (ContentTypes.HasSource) { 173 OS << " source: "; 174 FileEntry.Source.dump(OS, DumpOptions); 175 OS << '\n'; 176 } 177 } 178 } 179 } 180 181 // Parse v2-v4 directory and file tables. 182 static Error 183 parseV2DirFileTables(const DWARFDataExtractor &DebugLineData, 184 uint64_t *OffsetPtr, 185 DWARFDebugLine::ContentTypeTracker &ContentTypes, 186 std::vector<DWARFFormValue> &IncludeDirectories, 187 std::vector<DWARFDebugLine::FileNameEntry> &FileNames) { 188 while (true) { 189 Error Err = Error::success(); 190 StringRef S = DebugLineData.getCStrRef(OffsetPtr, &Err); 191 if (Err) { 192 consumeError(std::move(Err)); 193 return createStringError(errc::invalid_argument, 194 "include directories table was not null " 195 "terminated before the end of the prologue"); 196 } 197 if (S.empty()) 198 break; 199 DWARFFormValue Dir = 200 DWARFFormValue::createFromPValue(dwarf::DW_FORM_string, S.data()); 201 IncludeDirectories.push_back(Dir); 202 } 203 204 ContentTypes.HasModTime = true; 205 ContentTypes.HasLength = true; 206 207 while (true) { 208 Error Err = Error::success(); 209 StringRef Name = DebugLineData.getCStrRef(OffsetPtr, &Err); 210 if (!Err && Name.empty()) 211 break; 212 213 DWARFDebugLine::FileNameEntry FileEntry; 214 FileEntry.Name = 215 DWARFFormValue::createFromPValue(dwarf::DW_FORM_string, Name.data()); 216 FileEntry.DirIdx = DebugLineData.getULEB128(OffsetPtr, &Err); 217 FileEntry.ModTime = DebugLineData.getULEB128(OffsetPtr, &Err); 218 FileEntry.Length = DebugLineData.getULEB128(OffsetPtr, &Err); 219 220 if (Err) { 221 consumeError(std::move(Err)); 222 return createStringError( 223 errc::invalid_argument, 224 "file names table was not null terminated before " 225 "the end of the prologue"); 226 } 227 FileNames.push_back(FileEntry); 228 } 229 230 return Error::success(); 231 } 232 233 // Parse v5 directory/file entry content descriptions. 234 // Returns the descriptors, or an error if we did not find a path or ran off 235 // the end of the prologue. 236 static llvm::Expected<ContentDescriptors> 237 parseV5EntryFormat(const DWARFDataExtractor &DebugLineData, uint64_t *OffsetPtr, 238 DWARFDebugLine::ContentTypeTracker *ContentTypes) { 239 Error Err = Error::success(); 240 ContentDescriptors Descriptors; 241 int FormatCount = DebugLineData.getU8(OffsetPtr, &Err); 242 bool HasPath = false; 243 for (int I = 0; I != FormatCount && !Err; ++I) { 244 ContentDescriptor Descriptor; 245 Descriptor.Type = 246 dwarf::LineNumberEntryFormat(DebugLineData.getULEB128(OffsetPtr, &Err)); 247 Descriptor.Form = dwarf::Form(DebugLineData.getULEB128(OffsetPtr, &Err)); 248 if (Descriptor.Type == dwarf::DW_LNCT_path) 249 HasPath = true; 250 if (ContentTypes) 251 ContentTypes->trackContentType(Descriptor.Type); 252 Descriptors.push_back(Descriptor); 253 } 254 255 if (Err) 256 return createStringError(errc::invalid_argument, 257 "failed to parse entry content descriptors: %s", 258 toString(std::move(Err)).c_str()); 259 260 if (!HasPath) 261 return createStringError(errc::invalid_argument, 262 "failed to parse entry content descriptions" 263 " because no path was found"); 264 return Descriptors; 265 } 266 267 static Error 268 parseV5DirFileTables(const DWARFDataExtractor &DebugLineData, 269 uint64_t *OffsetPtr, const dwarf::FormParams &FormParams, 270 const DWARFContext &Ctx, const DWARFUnit *U, 271 DWARFDebugLine::ContentTypeTracker &ContentTypes, 272 std::vector<DWARFFormValue> &IncludeDirectories, 273 std::vector<DWARFDebugLine::FileNameEntry> &FileNames) { 274 // Get the directory entry description. 275 llvm::Expected<ContentDescriptors> DirDescriptors = 276 parseV5EntryFormat(DebugLineData, OffsetPtr, nullptr); 277 if (!DirDescriptors) 278 return DirDescriptors.takeError(); 279 280 // Get the directory entries, according to the format described above. 281 uint64_t DirEntryCount = DebugLineData.getULEB128(OffsetPtr); 282 for (uint64_t I = 0; I != DirEntryCount; ++I) { 283 for (auto Descriptor : *DirDescriptors) { 284 DWARFFormValue Value(Descriptor.Form); 285 switch (Descriptor.Type) { 286 case DW_LNCT_path: 287 if (!Value.extractValue(DebugLineData, OffsetPtr, FormParams, &Ctx, U)) 288 return createStringError(errc::invalid_argument, 289 "failed to parse directory entry because " 290 "extracting the form value failed"); 291 IncludeDirectories.push_back(Value); 292 break; 293 default: 294 if (!Value.skipValue(DebugLineData, OffsetPtr, FormParams)) 295 return createStringError(errc::invalid_argument, 296 "failed to parse directory entry because " 297 "skipping the form value failed"); 298 } 299 } 300 } 301 302 // Get the file entry description. 303 llvm::Expected<ContentDescriptors> FileDescriptors = 304 parseV5EntryFormat(DebugLineData, OffsetPtr, &ContentTypes); 305 if (!FileDescriptors) 306 return FileDescriptors.takeError(); 307 308 // Get the file entries, according to the format described above. 309 uint64_t FileEntryCount = DebugLineData.getULEB128(OffsetPtr); 310 for (uint64_t I = 0; I != FileEntryCount; ++I) { 311 DWARFDebugLine::FileNameEntry FileEntry; 312 for (auto Descriptor : *FileDescriptors) { 313 DWARFFormValue Value(Descriptor.Form); 314 if (!Value.extractValue(DebugLineData, OffsetPtr, FormParams, &Ctx, U)) 315 return createStringError(errc::invalid_argument, 316 "failed to parse file entry because " 317 "extracting the form value failed"); 318 switch (Descriptor.Type) { 319 case DW_LNCT_path: 320 FileEntry.Name = Value; 321 break; 322 case DW_LNCT_LLVM_source: 323 FileEntry.Source = Value; 324 break; 325 case DW_LNCT_directory_index: 326 FileEntry.DirIdx = *Value.getAsUnsignedConstant(); 327 break; 328 case DW_LNCT_timestamp: 329 FileEntry.ModTime = *Value.getAsUnsignedConstant(); 330 break; 331 case DW_LNCT_size: 332 FileEntry.Length = *Value.getAsUnsignedConstant(); 333 break; 334 case DW_LNCT_MD5: 335 if (!Value.getAsBlock() || Value.getAsBlock()->size() != 16) 336 return createStringError( 337 errc::invalid_argument, 338 "failed to parse file entry because the MD5 hash is invalid"); 339 std::uninitialized_copy_n(Value.getAsBlock()->begin(), 16, 340 FileEntry.Checksum.begin()); 341 break; 342 default: 343 break; 344 } 345 } 346 FileNames.push_back(FileEntry); 347 } 348 return Error::success(); 349 } 350 351 uint64_t DWARFDebugLine::Prologue::getLength() const { 352 uint64_t Length = PrologueLength + sizeofTotalLength() + 353 sizeof(getVersion()) + sizeofPrologueLength(); 354 if (getVersion() >= 5) 355 Length += 2; // Address + Segment selector sizes. 356 return Length; 357 } 358 359 Error DWARFDebugLine::Prologue::parse( 360 DWARFDataExtractor DebugLineData, uint64_t *OffsetPtr, 361 function_ref<void(Error)> RecoverableErrorHandler, const DWARFContext &Ctx, 362 const DWARFUnit *U) { 363 const uint64_t PrologueOffset = *OffsetPtr; 364 365 clear(); 366 DataExtractor::Cursor Cursor(*OffsetPtr); 367 std::tie(TotalLength, FormParams.Format) = 368 DebugLineData.getInitialLength(Cursor); 369 370 DebugLineData = 371 DWARFDataExtractor(DebugLineData, Cursor.tell() + TotalLength); 372 FormParams.Version = DebugLineData.getU16(Cursor); 373 if (Cursor && !versionIsSupported(getVersion())) { 374 // Treat this error as unrecoverable - we cannot be sure what any of 375 // the data represents including the length field, so cannot skip it or make 376 // any reasonable assumptions. 377 *OffsetPtr = Cursor.tell(); 378 return createStringError( 379 errc::not_supported, 380 "parsing line table prologue at offset 0x%8.8" PRIx64 381 ": unsupported version %" PRIu16, 382 PrologueOffset, getVersion()); 383 } 384 385 if (getVersion() >= 5) { 386 FormParams.AddrSize = DebugLineData.getU8(Cursor); 387 assert((!Cursor || DebugLineData.getAddressSize() == 0 || 388 DebugLineData.getAddressSize() == getAddressSize()) && 389 "Line table header and data extractor disagree"); 390 SegSelectorSize = DebugLineData.getU8(Cursor); 391 } 392 393 PrologueLength = 394 DebugLineData.getRelocatedValue(Cursor, sizeofPrologueLength()); 395 const uint64_t EndPrologueOffset = PrologueLength + Cursor.tell(); 396 DebugLineData = DWARFDataExtractor(DebugLineData, EndPrologueOffset); 397 MinInstLength = DebugLineData.getU8(Cursor); 398 if (getVersion() >= 4) 399 MaxOpsPerInst = DebugLineData.getU8(Cursor); 400 DefaultIsStmt = DebugLineData.getU8(Cursor); 401 LineBase = DebugLineData.getU8(Cursor); 402 LineRange = DebugLineData.getU8(Cursor); 403 OpcodeBase = DebugLineData.getU8(Cursor); 404 405 if (Cursor && OpcodeBase == 0) { 406 // If the opcode base is 0, we cannot read the standard opcode lengths (of 407 // which there are supposed to be one fewer than the opcode base). Assume 408 // there are no standard opcodes and continue parsing. 409 RecoverableErrorHandler(createStringError( 410 errc::invalid_argument, 411 "parsing line table prologue at offset 0x%8.8" PRIx64 412 " found opcode base of 0. Assuming no standard opcodes", 413 PrologueOffset)); 414 } else if (Cursor) { 415 StandardOpcodeLengths.reserve(OpcodeBase - 1); 416 for (uint32_t I = 1; I < OpcodeBase; ++I) { 417 uint8_t OpLen = DebugLineData.getU8(Cursor); 418 StandardOpcodeLengths.push_back(OpLen); 419 } 420 } 421 422 *OffsetPtr = Cursor.tell(); 423 // A corrupt file name or directory table does not prevent interpretation of 424 // the main line program, so check the cursor state now so that its errors can 425 // be handled separately. 426 if (!Cursor) 427 return createStringError( 428 errc::invalid_argument, 429 "parsing line table prologue at offset 0x%8.8" PRIx64 ": %s", 430 PrologueOffset, toString(Cursor.takeError()).c_str()); 431 432 Error E = 433 getVersion() >= 5 434 ? parseV5DirFileTables(DebugLineData, OffsetPtr, FormParams, Ctx, U, 435 ContentTypes, IncludeDirectories, FileNames) 436 : parseV2DirFileTables(DebugLineData, OffsetPtr, ContentTypes, 437 IncludeDirectories, FileNames); 438 if (E) { 439 RecoverableErrorHandler(joinErrors( 440 createStringError( 441 errc::invalid_argument, 442 "parsing line table prologue at 0x%8.8" PRIx64 443 " found an invalid directory or file table description at" 444 " 0x%8.8" PRIx64, 445 PrologueOffset, *OffsetPtr), 446 std::move(E))); 447 return Error::success(); 448 } 449 450 assert(*OffsetPtr <= EndPrologueOffset); 451 if (*OffsetPtr != EndPrologueOffset) { 452 RecoverableErrorHandler(createStringError( 453 errc::invalid_argument, 454 "unknown data in line table prologue at offset 0x%8.8" PRIx64 455 ": parsing ended (at offset 0x%8.8" PRIx64 456 ") before reaching the prologue end at offset 0x%8.8" PRIx64, 457 PrologueOffset, *OffsetPtr, EndPrologueOffset)); 458 } 459 return Error::success(); 460 } 461 462 DWARFDebugLine::Row::Row(bool DefaultIsStmt) { reset(DefaultIsStmt); } 463 464 void DWARFDebugLine::Row::postAppend() { 465 Discriminator = 0; 466 BasicBlock = false; 467 PrologueEnd = false; 468 EpilogueBegin = false; 469 } 470 471 void DWARFDebugLine::Row::reset(bool DefaultIsStmt) { 472 Address.Address = 0; 473 Address.SectionIndex = object::SectionedAddress::UndefSection; 474 Line = 1; 475 Column = 0; 476 File = 1; 477 Isa = 0; 478 Discriminator = 0; 479 IsStmt = DefaultIsStmt; 480 OpIndex = 0; 481 BasicBlock = false; 482 EndSequence = false; 483 PrologueEnd = false; 484 EpilogueBegin = false; 485 } 486 487 void DWARFDebugLine::Row::dumpTableHeader(raw_ostream &OS, unsigned Indent) { 488 OS.indent(Indent) 489 << "Address Line Column File ISA Discriminator OpIndex " 490 "Flags\n"; 491 OS.indent(Indent) 492 << "------------------ ------ ------ ------ --- ------------- ------- " 493 "-------------\n"; 494 } 495 496 void DWARFDebugLine::Row::dump(raw_ostream &OS) const { 497 OS << format("0x%16.16" PRIx64 " %6u %6u", Address.Address, Line, Column) 498 << format(" %6u %3u %13u %7u ", File, Isa, Discriminator, OpIndex) 499 << (IsStmt ? " is_stmt" : "") << (BasicBlock ? " basic_block" : "") 500 << (PrologueEnd ? " prologue_end" : "") 501 << (EpilogueBegin ? " epilogue_begin" : "") 502 << (EndSequence ? " end_sequence" : "") << '\n'; 503 } 504 505 DWARFDebugLine::Sequence::Sequence() { reset(); } 506 507 void DWARFDebugLine::Sequence::reset() { 508 LowPC = 0; 509 HighPC = 0; 510 SectionIndex = object::SectionedAddress::UndefSection; 511 FirstRowIndex = 0; 512 LastRowIndex = 0; 513 Empty = true; 514 } 515 516 DWARFDebugLine::LineTable::LineTable() { clear(); } 517 518 void DWARFDebugLine::LineTable::dump(raw_ostream &OS, 519 DIDumpOptions DumpOptions) const { 520 Prologue.dump(OS, DumpOptions); 521 522 if (!Rows.empty()) { 523 OS << '\n'; 524 Row::dumpTableHeader(OS, 0); 525 for (const Row &R : Rows) { 526 R.dump(OS); 527 } 528 } 529 530 // Terminate the table with a final blank line to clearly delineate it from 531 // later dumps. 532 OS << '\n'; 533 } 534 535 void DWARFDebugLine::LineTable::clear() { 536 Prologue.clear(); 537 Rows.clear(); 538 Sequences.clear(); 539 } 540 541 DWARFDebugLine::ParsingState::ParsingState( 542 struct LineTable *LT, uint64_t TableOffset, 543 function_ref<void(Error)> ErrorHandler) 544 : LineTable(LT), LineTableOffset(TableOffset), ErrorHandler(ErrorHandler) { 545 resetRowAndSequence(); 546 } 547 548 void DWARFDebugLine::ParsingState::resetRowAndSequence() { 549 Row.reset(LineTable->Prologue.DefaultIsStmt); 550 Sequence.reset(); 551 } 552 553 void DWARFDebugLine::ParsingState::appendRowToMatrix() { 554 unsigned RowNumber = LineTable->Rows.size(); 555 if (Sequence.Empty) { 556 // Record the beginning of instruction sequence. 557 Sequence.Empty = false; 558 Sequence.LowPC = Row.Address.Address; 559 Sequence.FirstRowIndex = RowNumber; 560 } 561 LineTable->appendRow(Row); 562 if (Row.EndSequence) { 563 // Record the end of instruction sequence. 564 Sequence.HighPC = Row.Address.Address; 565 Sequence.LastRowIndex = RowNumber + 1; 566 Sequence.SectionIndex = Row.Address.SectionIndex; 567 if (Sequence.isValid()) 568 LineTable->appendSequence(Sequence); 569 Sequence.reset(); 570 } 571 Row.postAppend(); 572 } 573 574 const DWARFDebugLine::LineTable * 575 DWARFDebugLine::getLineTable(uint64_t Offset) const { 576 LineTableConstIter Pos = LineTableMap.find(Offset); 577 if (Pos != LineTableMap.end()) 578 return &Pos->second; 579 return nullptr; 580 } 581 582 Expected<const DWARFDebugLine::LineTable *> DWARFDebugLine::getOrParseLineTable( 583 DWARFDataExtractor &DebugLineData, uint64_t Offset, const DWARFContext &Ctx, 584 const DWARFUnit *U, function_ref<void(Error)> RecoverableErrorHandler) { 585 if (!DebugLineData.isValidOffset(Offset)) 586 return createStringError(errc::invalid_argument, "offset 0x%8.8" PRIx64 587 " is not a valid debug line section offset", 588 Offset); 589 590 std::pair<LineTableIter, bool> Pos = 591 LineTableMap.insert(LineTableMapTy::value_type(Offset, LineTable())); 592 LineTable *LT = &Pos.first->second; 593 if (Pos.second) { 594 if (Error Err = 595 LT->parse(DebugLineData, &Offset, Ctx, U, RecoverableErrorHandler)) 596 return std::move(Err); 597 return LT; 598 } 599 return LT; 600 } 601 602 void DWARFDebugLine::clearLineTable(uint64_t Offset) { 603 LineTableMap.erase(Offset); 604 } 605 606 static StringRef getOpcodeName(uint8_t Opcode, uint8_t OpcodeBase) { 607 assert(Opcode != 0); 608 if (Opcode < OpcodeBase) 609 return LNStandardString(Opcode); 610 return "special"; 611 } 612 613 uint64_t DWARFDebugLine::ParsingState::advanceAddr(uint64_t OperationAdvance, 614 uint8_t Opcode, 615 uint64_t OpcodeOffset) { 616 StringRef OpcodeName = getOpcodeName(Opcode, LineTable->Prologue.OpcodeBase); 617 // For versions less than 4, the MaxOpsPerInst member is set to 0, as the 618 // maximum_operations_per_instruction field wasn't introduced until DWARFv4. 619 // Don't warn about bad values in this situation. 620 if (ReportAdvanceAddrProblem && LineTable->Prologue.getVersion() >= 4 && 621 LineTable->Prologue.MaxOpsPerInst != 1) 622 ErrorHandler(createStringError( 623 errc::not_supported, 624 "line table program at offset 0x%8.8" PRIx64 625 " contains a %s opcode at offset 0x%8.8" PRIx64 626 ", but the prologue maximum_operations_per_instruction value is %" PRId8 627 ", which is unsupported. Assuming a value of 1 instead", 628 LineTableOffset, OpcodeName.data(), OpcodeOffset, 629 LineTable->Prologue.MaxOpsPerInst)); 630 if (ReportAdvanceAddrProblem && LineTable->Prologue.MinInstLength == 0) 631 ErrorHandler( 632 createStringError(errc::invalid_argument, 633 "line table program at offset 0x%8.8" PRIx64 634 " contains a %s opcode at offset 0x%8.8" PRIx64 635 ", but the prologue minimum_instruction_length value " 636 "is 0, which prevents any address advancing", 637 LineTableOffset, OpcodeName.data(), OpcodeOffset)); 638 ReportAdvanceAddrProblem = false; 639 uint64_t AddrOffset = OperationAdvance * LineTable->Prologue.MinInstLength; 640 Row.Address.Address += AddrOffset; 641 return AddrOffset; 642 } 643 644 DWARFDebugLine::ParsingState::AddrAndAdjustedOpcode 645 DWARFDebugLine::ParsingState::advanceAddrForOpcode(uint8_t Opcode, 646 uint64_t OpcodeOffset) { 647 assert(Opcode == DW_LNS_const_add_pc || 648 Opcode >= LineTable->Prologue.OpcodeBase); 649 if (ReportBadLineRange && LineTable->Prologue.LineRange == 0) { 650 StringRef OpcodeName = 651 getOpcodeName(Opcode, LineTable->Prologue.OpcodeBase); 652 ErrorHandler( 653 createStringError(errc::not_supported, 654 "line table program at offset 0x%8.8" PRIx64 655 " contains a %s opcode at offset 0x%8.8" PRIx64 656 ", but the prologue line_range value is 0. The " 657 "address and line will not be adjusted", 658 LineTableOffset, OpcodeName.data(), OpcodeOffset)); 659 ReportBadLineRange = false; 660 } 661 662 uint8_t OpcodeValue = Opcode; 663 if (Opcode == DW_LNS_const_add_pc) 664 OpcodeValue = 255; 665 uint8_t AdjustedOpcode = OpcodeValue - LineTable->Prologue.OpcodeBase; 666 uint64_t OperationAdvance = 667 LineTable->Prologue.LineRange != 0 668 ? AdjustedOpcode / LineTable->Prologue.LineRange 669 : 0; 670 uint64_t AddrOffset = advanceAddr(OperationAdvance, Opcode, OpcodeOffset); 671 return {AddrOffset, AdjustedOpcode}; 672 } 673 674 DWARFDebugLine::ParsingState::AddrAndLineDelta 675 DWARFDebugLine::ParsingState::handleSpecialOpcode(uint8_t Opcode, 676 uint64_t OpcodeOffset) { 677 // A special opcode value is chosen based on the amount that needs 678 // to be added to the line and address registers. The maximum line 679 // increment for a special opcode is the value of the line_base 680 // field in the header, plus the value of the line_range field, 681 // minus 1 (line base + line range - 1). If the desired line 682 // increment is greater than the maximum line increment, a standard 683 // opcode must be used instead of a special opcode. The "address 684 // advance" is calculated by dividing the desired address increment 685 // by the minimum_instruction_length field from the header. The 686 // special opcode is then calculated using the following formula: 687 // 688 // opcode = (desired line increment - line_base) + 689 // (line_range * address advance) + opcode_base 690 // 691 // If the resulting opcode is greater than 255, a standard opcode 692 // must be used instead. 693 // 694 // To decode a special opcode, subtract the opcode_base from the 695 // opcode itself to give the adjusted opcode. The amount to 696 // increment the address register is the result of the adjusted 697 // opcode divided by the line_range multiplied by the 698 // minimum_instruction_length field from the header. That is: 699 // 700 // address increment = (adjusted opcode / line_range) * 701 // minimum_instruction_length 702 // 703 // The amount to increment the line register is the line_base plus 704 // the result of the adjusted opcode modulo the line_range. That is: 705 // 706 // line increment = line_base + (adjusted opcode % line_range) 707 708 DWARFDebugLine::ParsingState::AddrAndAdjustedOpcode AddrAdvanceResult = 709 advanceAddrForOpcode(Opcode, OpcodeOffset); 710 int32_t LineOffset = 0; 711 if (LineTable->Prologue.LineRange != 0) 712 LineOffset = 713 LineTable->Prologue.LineBase + 714 (AddrAdvanceResult.AdjustedOpcode % LineTable->Prologue.LineRange); 715 Row.Line += LineOffset; 716 return {AddrAdvanceResult.AddrDelta, LineOffset}; 717 } 718 719 /// Parse a ULEB128 using the specified \p Cursor. \returns the parsed value on 720 /// success, or std::nullopt if \p Cursor is in a failing state. 721 template <typename T> 722 static std::optional<T> parseULEB128(DWARFDataExtractor &Data, 723 DataExtractor::Cursor &Cursor) { 724 T Value = Data.getULEB128(Cursor); 725 if (Cursor) 726 return Value; 727 return std::nullopt; 728 } 729 730 Error DWARFDebugLine::LineTable::parse( 731 DWARFDataExtractor &DebugLineData, uint64_t *OffsetPtr, 732 const DWARFContext &Ctx, const DWARFUnit *U, 733 function_ref<void(Error)> RecoverableErrorHandler, raw_ostream *OS, 734 bool Verbose) { 735 assert((OS || !Verbose) && "cannot have verbose output without stream"); 736 const uint64_t DebugLineOffset = *OffsetPtr; 737 738 clear(); 739 740 Error PrologueErr = 741 Prologue.parse(DebugLineData, OffsetPtr, RecoverableErrorHandler, Ctx, U); 742 743 if (OS) { 744 DIDumpOptions DumpOptions; 745 DumpOptions.Verbose = Verbose; 746 Prologue.dump(*OS, DumpOptions); 747 } 748 749 if (PrologueErr) { 750 // Ensure there is a blank line after the prologue to clearly delineate it 751 // from later dumps. 752 if (OS) 753 *OS << "\n"; 754 return PrologueErr; 755 } 756 757 uint64_t ProgramLength = Prologue.TotalLength + Prologue.sizeofTotalLength(); 758 if (!DebugLineData.isValidOffsetForDataOfSize(DebugLineOffset, 759 ProgramLength)) { 760 assert(DebugLineData.size() > DebugLineOffset && 761 "prologue parsing should handle invalid offset"); 762 uint64_t BytesRemaining = DebugLineData.size() - DebugLineOffset; 763 RecoverableErrorHandler( 764 createStringError(errc::invalid_argument, 765 "line table program with offset 0x%8.8" PRIx64 766 " has length 0x%8.8" PRIx64 " but only 0x%8.8" PRIx64 767 " bytes are available", 768 DebugLineOffset, ProgramLength, BytesRemaining)); 769 // Continue by capping the length at the number of remaining bytes. 770 ProgramLength = BytesRemaining; 771 } 772 773 // Create a DataExtractor which can only see the data up to the end of the 774 // table, to prevent reading past the end. 775 const uint64_t EndOffset = DebugLineOffset + ProgramLength; 776 DWARFDataExtractor TableData(DebugLineData, EndOffset); 777 778 // See if we should tell the data extractor the address size. 779 if (TableData.getAddressSize() == 0) 780 TableData.setAddressSize(Prologue.getAddressSize()); 781 else 782 assert(Prologue.getAddressSize() == 0 || 783 Prologue.getAddressSize() == TableData.getAddressSize()); 784 785 ParsingState State(this, DebugLineOffset, RecoverableErrorHandler); 786 787 *OffsetPtr = DebugLineOffset + Prologue.getLength(); 788 if (OS && *OffsetPtr < EndOffset) { 789 *OS << '\n'; 790 Row::dumpTableHeader(*OS, /*Indent=*/Verbose ? 12 : 0); 791 } 792 bool TombstonedAddress = false; 793 auto EmitRow = [&] { 794 if (!TombstonedAddress) { 795 if (Verbose) { 796 *OS << "\n"; 797 OS->indent(12); 798 } 799 if (OS) 800 State.Row.dump(*OS); 801 State.appendRowToMatrix(); 802 } 803 }; 804 while (*OffsetPtr < EndOffset) { 805 DataExtractor::Cursor Cursor(*OffsetPtr); 806 807 if (Verbose) 808 *OS << format("0x%08.08" PRIx64 ": ", *OffsetPtr); 809 810 uint64_t OpcodeOffset = *OffsetPtr; 811 uint8_t Opcode = TableData.getU8(Cursor); 812 size_t RowCount = Rows.size(); 813 814 if (Cursor && Verbose) 815 *OS << format("%02.02" PRIx8 " ", Opcode); 816 817 if (Opcode == 0) { 818 // Extended Opcodes always start with a zero opcode followed by 819 // a uleb128 length so you can skip ones you don't know about 820 uint64_t Len = TableData.getULEB128(Cursor); 821 uint64_t ExtOffset = Cursor.tell(); 822 823 // Tolerate zero-length; assume length is correct and soldier on. 824 if (Len == 0) { 825 if (Cursor && Verbose) 826 *OS << "Badly formed extended line op (length 0)\n"; 827 if (!Cursor) { 828 if (Verbose) 829 *OS << "\n"; 830 RecoverableErrorHandler(Cursor.takeError()); 831 } 832 *OffsetPtr = Cursor.tell(); 833 continue; 834 } 835 836 uint8_t SubOpcode = TableData.getU8(Cursor); 837 // OperandOffset will be the same as ExtOffset, if it was not possible to 838 // read the SubOpcode. 839 uint64_t OperandOffset = Cursor.tell(); 840 if (Verbose) 841 *OS << LNExtendedString(SubOpcode); 842 switch (SubOpcode) { 843 case DW_LNE_end_sequence: 844 // Set the end_sequence register of the state machine to true and 845 // append a row to the matrix using the current values of the 846 // state-machine registers. Then reset the registers to the initial 847 // values specified above. Every statement program sequence must end 848 // with a DW_LNE_end_sequence instruction which creates a row whose 849 // address is that of the byte after the last target machine instruction 850 // of the sequence. 851 State.Row.EndSequence = true; 852 // No need to test the Cursor is valid here, since it must be to get 853 // into this code path - if it were invalid, the default case would be 854 // followed. 855 EmitRow(); 856 State.resetRowAndSequence(); 857 break; 858 859 case DW_LNE_set_address: 860 // Takes a single relocatable address as an operand. The size of the 861 // operand is the size appropriate to hold an address on the target 862 // machine. Set the address register to the value given by the 863 // relocatable address. All of the other statement program opcodes 864 // that affect the address register add a delta to it. This instruction 865 // stores a relocatable value into it instead. 866 // 867 // Make sure the extractor knows the address size. If not, infer it 868 // from the size of the operand. 869 { 870 uint8_t ExtractorAddressSize = TableData.getAddressSize(); 871 uint64_t OpcodeAddressSize = Len - 1; 872 if (ExtractorAddressSize != OpcodeAddressSize && 873 ExtractorAddressSize != 0) 874 RecoverableErrorHandler(createStringError( 875 errc::invalid_argument, 876 "mismatching address size at offset 0x%8.8" PRIx64 877 " expected 0x%2.2" PRIx8 " found 0x%2.2" PRIx64, 878 ExtOffset, ExtractorAddressSize, Len - 1)); 879 880 // Assume that the line table is correct and temporarily override the 881 // address size. If the size is unsupported, give up trying to read 882 // the address and continue to the next opcode. 883 if (OpcodeAddressSize != 1 && OpcodeAddressSize != 2 && 884 OpcodeAddressSize != 4 && OpcodeAddressSize != 8) { 885 RecoverableErrorHandler(createStringError( 886 errc::invalid_argument, 887 "address size 0x%2.2" PRIx64 888 " of DW_LNE_set_address opcode at offset 0x%8.8" PRIx64 889 " is unsupported", 890 OpcodeAddressSize, ExtOffset)); 891 TableData.skip(Cursor, OpcodeAddressSize); 892 } else { 893 TableData.setAddressSize(OpcodeAddressSize); 894 State.Row.Address.Address = TableData.getRelocatedAddress( 895 Cursor, &State.Row.Address.SectionIndex); 896 897 uint64_t Tombstone = 898 dwarf::computeTombstoneAddress(OpcodeAddressSize); 899 TombstonedAddress = State.Row.Address.Address == Tombstone; 900 901 // Restore the address size if the extractor already had it. 902 if (ExtractorAddressSize != 0) 903 TableData.setAddressSize(ExtractorAddressSize); 904 } 905 906 if (Cursor && Verbose) { 907 *OS << " ("; 908 DWARFFormValue::dumpAddress(*OS, OpcodeAddressSize, State.Row.Address.Address); 909 *OS << ')'; 910 } 911 } 912 break; 913 914 case DW_LNE_define_file: 915 // Takes 4 arguments. The first is a null terminated string containing 916 // a source file name. The second is an unsigned LEB128 number 917 // representing the directory index of the directory in which the file 918 // was found. The third is an unsigned LEB128 number representing the 919 // time of last modification of the file. The fourth is an unsigned 920 // LEB128 number representing the length in bytes of the file. The time 921 // and length fields may contain LEB128(0) if the information is not 922 // available. 923 // 924 // The directory index represents an entry in the include_directories 925 // section of the statement program prologue. The index is LEB128(0) 926 // if the file was found in the current directory of the compilation, 927 // LEB128(1) if it was found in the first directory in the 928 // include_directories section, and so on. The directory index is 929 // ignored for file names that represent full path names. 930 // 931 // The files are numbered, starting at 1, in the order in which they 932 // appear; the names in the prologue come before names defined by 933 // the DW_LNE_define_file instruction. These numbers are used in the 934 // the file register of the state machine. 935 { 936 FileNameEntry FileEntry; 937 const char *Name = TableData.getCStr(Cursor); 938 FileEntry.Name = 939 DWARFFormValue::createFromPValue(dwarf::DW_FORM_string, Name); 940 FileEntry.DirIdx = TableData.getULEB128(Cursor); 941 FileEntry.ModTime = TableData.getULEB128(Cursor); 942 FileEntry.Length = TableData.getULEB128(Cursor); 943 Prologue.FileNames.push_back(FileEntry); 944 if (Cursor && Verbose) 945 *OS << " (" << Name << ", dir=" << FileEntry.DirIdx << ", mod_time=" 946 << format("(0x%16.16" PRIx64 ")", FileEntry.ModTime) 947 << ", length=" << FileEntry.Length << ")"; 948 } 949 break; 950 951 case DW_LNE_set_discriminator: 952 State.Row.Discriminator = TableData.getULEB128(Cursor); 953 if (Cursor && Verbose) 954 *OS << " (" << State.Row.Discriminator << ")"; 955 break; 956 957 default: 958 if (Cursor && Verbose) 959 *OS << format("Unrecognized extended op 0x%02.02" PRIx8, SubOpcode) 960 << format(" length %" PRIx64, Len); 961 // Len doesn't include the zero opcode byte or the length itself, but 962 // it does include the sub_opcode, so we have to adjust for that. 963 TableData.skip(Cursor, Len - 1); 964 break; 965 } 966 // Make sure the length as recorded in the table and the standard length 967 // for the opcode match. If they don't, continue from the end as claimed 968 // by the table. Similarly, continue from the claimed end in the event of 969 // a parsing error. 970 uint64_t End = ExtOffset + Len; 971 if (Cursor && Cursor.tell() != End) 972 RecoverableErrorHandler(createStringError( 973 errc::illegal_byte_sequence, 974 "unexpected line op length at offset 0x%8.8" PRIx64 975 " expected 0x%2.2" PRIx64 " found 0x%2.2" PRIx64, 976 ExtOffset, Len, Cursor.tell() - ExtOffset)); 977 if (!Cursor && Verbose) { 978 DWARFDataExtractor::Cursor ByteCursor(OperandOffset); 979 uint8_t Byte = TableData.getU8(ByteCursor); 980 if (ByteCursor) { 981 *OS << " (<parsing error>"; 982 do { 983 *OS << format(" %2.2" PRIx8, Byte); 984 Byte = TableData.getU8(ByteCursor); 985 } while (ByteCursor); 986 *OS << ")"; 987 } 988 989 // The only parse failure in this case should be if the end was reached. 990 // In that case, throw away the error, as the main Cursor's error will 991 // be sufficient. 992 consumeError(ByteCursor.takeError()); 993 } 994 *OffsetPtr = End; 995 } else if (Opcode < Prologue.OpcodeBase) { 996 if (Verbose) 997 *OS << LNStandardString(Opcode); 998 switch (Opcode) { 999 // Standard Opcodes 1000 case DW_LNS_copy: 1001 // Takes no arguments. Append a row to the matrix using the 1002 // current values of the state-machine registers. 1003 EmitRow(); 1004 break; 1005 1006 case DW_LNS_advance_pc: 1007 // Takes a single unsigned LEB128 operand, multiplies it by the 1008 // min_inst_length field of the prologue, and adds the 1009 // result to the address register of the state machine. 1010 if (std::optional<uint64_t> Operand = 1011 parseULEB128<uint64_t>(TableData, Cursor)) { 1012 uint64_t AddrOffset = 1013 State.advanceAddr(*Operand, Opcode, OpcodeOffset); 1014 if (Verbose) 1015 *OS << " (" << AddrOffset << ")"; 1016 } 1017 break; 1018 1019 case DW_LNS_advance_line: 1020 // Takes a single signed LEB128 operand and adds that value to 1021 // the line register of the state machine. 1022 { 1023 int64_t LineDelta = TableData.getSLEB128(Cursor); 1024 if (Cursor) { 1025 State.Row.Line += LineDelta; 1026 if (Verbose) 1027 *OS << " (" << State.Row.Line << ")"; 1028 } 1029 } 1030 break; 1031 1032 case DW_LNS_set_file: 1033 // Takes a single unsigned LEB128 operand and stores it in the file 1034 // register of the state machine. 1035 if (std::optional<uint16_t> File = 1036 parseULEB128<uint16_t>(TableData, Cursor)) { 1037 State.Row.File = *File; 1038 if (Verbose) 1039 *OS << " (" << State.Row.File << ")"; 1040 } 1041 break; 1042 1043 case DW_LNS_set_column: 1044 // Takes a single unsigned LEB128 operand and stores it in the 1045 // column register of the state machine. 1046 if (std::optional<uint16_t> Column = 1047 parseULEB128<uint16_t>(TableData, Cursor)) { 1048 State.Row.Column = *Column; 1049 if (Verbose) 1050 *OS << " (" << State.Row.Column << ")"; 1051 } 1052 break; 1053 1054 case DW_LNS_negate_stmt: 1055 // Takes no arguments. Set the is_stmt register of the state 1056 // machine to the logical negation of its current value. 1057 State.Row.IsStmt = !State.Row.IsStmt; 1058 break; 1059 1060 case DW_LNS_set_basic_block: 1061 // Takes no arguments. Set the basic_block register of the 1062 // state machine to true 1063 State.Row.BasicBlock = true; 1064 break; 1065 1066 case DW_LNS_const_add_pc: 1067 // Takes no arguments. Add to the address register of the state 1068 // machine the address increment value corresponding to special 1069 // opcode 255. The motivation for DW_LNS_const_add_pc is this: 1070 // when the statement program needs to advance the address by a 1071 // small amount, it can use a single special opcode, which occupies 1072 // a single byte. When it needs to advance the address by up to 1073 // twice the range of the last special opcode, it can use 1074 // DW_LNS_const_add_pc followed by a special opcode, for a total 1075 // of two bytes. Only if it needs to advance the address by more 1076 // than twice that range will it need to use both DW_LNS_advance_pc 1077 // and a special opcode, requiring three or more bytes. 1078 { 1079 uint64_t AddrOffset = 1080 State.advanceAddrForOpcode(Opcode, OpcodeOffset).AddrDelta; 1081 if (Verbose) 1082 *OS << format(" (0x%16.16" PRIx64 ")", AddrOffset); 1083 } 1084 break; 1085 1086 case DW_LNS_fixed_advance_pc: 1087 // Takes a single uhalf operand. Add to the address register of 1088 // the state machine the value of the (unencoded) operand. This 1089 // is the only extended opcode that takes an argument that is not 1090 // a variable length number. The motivation for DW_LNS_fixed_advance_pc 1091 // is this: existing assemblers cannot emit DW_LNS_advance_pc or 1092 // special opcodes because they cannot encode LEB128 numbers or 1093 // judge when the computation of a special opcode overflows and 1094 // requires the use of DW_LNS_advance_pc. Such assemblers, however, 1095 // can use DW_LNS_fixed_advance_pc instead, sacrificing compression. 1096 { 1097 uint16_t PCOffset = 1098 TableData.getRelocatedValue(Cursor, 2); 1099 if (Cursor) { 1100 State.Row.Address.Address += PCOffset; 1101 if (Verbose) 1102 *OS << format(" (0x%4.4" PRIx16 ")", PCOffset); 1103 } 1104 } 1105 break; 1106 1107 case DW_LNS_set_prologue_end: 1108 // Takes no arguments. Set the prologue_end register of the 1109 // state machine to true 1110 State.Row.PrologueEnd = true; 1111 break; 1112 1113 case DW_LNS_set_epilogue_begin: 1114 // Takes no arguments. Set the basic_block register of the 1115 // state machine to true 1116 State.Row.EpilogueBegin = true; 1117 break; 1118 1119 case DW_LNS_set_isa: 1120 // Takes a single unsigned LEB128 operand and stores it in the 1121 // ISA register of the state machine. 1122 if (std::optional<uint8_t> Isa = 1123 parseULEB128<uint8_t>(TableData, Cursor)) { 1124 State.Row.Isa = *Isa; 1125 if (Verbose) 1126 *OS << " (" << (uint64_t)State.Row.Isa << ")"; 1127 } 1128 break; 1129 1130 default: 1131 // Handle any unknown standard opcodes here. We know the lengths 1132 // of such opcodes because they are specified in the prologue 1133 // as a multiple of LEB128 operands for each opcode. 1134 { 1135 assert(Opcode - 1U < Prologue.StandardOpcodeLengths.size()); 1136 if (Verbose) 1137 *OS << "Unrecognized standard opcode"; 1138 uint8_t OpcodeLength = Prologue.StandardOpcodeLengths[Opcode - 1]; 1139 std::vector<uint64_t> Operands; 1140 for (uint8_t I = 0; I < OpcodeLength; ++I) { 1141 if (std::optional<uint64_t> Value = 1142 parseULEB128<uint64_t>(TableData, Cursor)) 1143 Operands.push_back(*Value); 1144 else 1145 break; 1146 } 1147 if (Verbose && !Operands.empty()) { 1148 *OS << " (operands: "; 1149 bool First = true; 1150 for (uint64_t Value : Operands) { 1151 if (!First) 1152 *OS << ", "; 1153 First = false; 1154 *OS << format("0x%16.16" PRIx64, Value); 1155 } 1156 if (Verbose) 1157 *OS << ')'; 1158 } 1159 } 1160 break; 1161 } 1162 1163 *OffsetPtr = Cursor.tell(); 1164 } else { 1165 // Special Opcodes. 1166 ParsingState::AddrAndLineDelta Delta = 1167 State.handleSpecialOpcode(Opcode, OpcodeOffset); 1168 1169 if (Verbose) 1170 *OS << "address += " << Delta.Address << ", line += " << Delta.Line; 1171 EmitRow(); 1172 *OffsetPtr = Cursor.tell(); 1173 } 1174 1175 // When a row is added to the matrix, it is also dumped, which includes a 1176 // new line already, so don't add an extra one. 1177 if (Verbose && Rows.size() == RowCount) 1178 *OS << "\n"; 1179 1180 // Most parse failures other than when parsing extended opcodes are due to 1181 // failures to read ULEBs. Bail out of parsing, since we don't know where to 1182 // continue reading from as there is no stated length for such byte 1183 // sequences. Print the final trailing new line if needed before doing so. 1184 if (!Cursor && Opcode != 0) { 1185 if (Verbose) 1186 *OS << "\n"; 1187 return Cursor.takeError(); 1188 } 1189 1190 if (!Cursor) 1191 RecoverableErrorHandler(Cursor.takeError()); 1192 } 1193 1194 if (!State.Sequence.Empty) 1195 RecoverableErrorHandler(createStringError( 1196 errc::illegal_byte_sequence, 1197 "last sequence in debug line table at offset 0x%8.8" PRIx64 1198 " is not terminated", 1199 DebugLineOffset)); 1200 1201 // Sort all sequences so that address lookup will work faster. 1202 if (!Sequences.empty()) { 1203 llvm::sort(Sequences, Sequence::orderByHighPC); 1204 // Note: actually, instruction address ranges of sequences should not 1205 // overlap (in shared objects and executables). If they do, the address 1206 // lookup would still work, though, but result would be ambiguous. 1207 // We don't report warning in this case. For example, 1208 // sometimes .so compiled from multiple object files contains a few 1209 // rudimentary sequences for address ranges [0x0, 0xsomething). 1210 } 1211 1212 // Terminate the table with a final blank line to clearly delineate it from 1213 // later dumps. 1214 if (OS) 1215 *OS << "\n"; 1216 1217 return Error::success(); 1218 } 1219 1220 uint32_t DWARFDebugLine::LineTable::findRowInSeq( 1221 const DWARFDebugLine::Sequence &Seq, 1222 object::SectionedAddress Address) const { 1223 if (!Seq.containsPC(Address)) 1224 return UnknownRowIndex; 1225 assert(Seq.SectionIndex == Address.SectionIndex); 1226 // In some cases, e.g. first instruction in a function, the compiler generates 1227 // two entries, both with the same address. We want the last one. 1228 // 1229 // In general we want a non-empty range: the last row whose address is less 1230 // than or equal to Address. This can be computed as upper_bound - 1. 1231 DWARFDebugLine::Row Row; 1232 Row.Address = Address; 1233 RowIter FirstRow = Rows.begin() + Seq.FirstRowIndex; 1234 RowIter LastRow = Rows.begin() + Seq.LastRowIndex; 1235 assert(FirstRow->Address.Address <= Row.Address.Address && 1236 Row.Address.Address < LastRow[-1].Address.Address); 1237 RowIter RowPos = std::upper_bound(FirstRow + 1, LastRow - 1, Row, 1238 DWARFDebugLine::Row::orderByAddress) - 1239 1; 1240 assert(Seq.SectionIndex == RowPos->Address.SectionIndex); 1241 return RowPos - Rows.begin(); 1242 } 1243 1244 uint32_t DWARFDebugLine::LineTable::lookupAddress( 1245 object::SectionedAddress Address) const { 1246 1247 // Search for relocatable addresses 1248 uint32_t Result = lookupAddressImpl(Address); 1249 1250 if (Result != UnknownRowIndex || 1251 Address.SectionIndex == object::SectionedAddress::UndefSection) 1252 return Result; 1253 1254 // Search for absolute addresses 1255 Address.SectionIndex = object::SectionedAddress::UndefSection; 1256 return lookupAddressImpl(Address); 1257 } 1258 1259 uint32_t DWARFDebugLine::LineTable::lookupAddressImpl( 1260 object::SectionedAddress Address) const { 1261 // First, find an instruction sequence containing the given address. 1262 DWARFDebugLine::Sequence Sequence; 1263 Sequence.SectionIndex = Address.SectionIndex; 1264 Sequence.HighPC = Address.Address; 1265 SequenceIter It = llvm::upper_bound(Sequences, Sequence, 1266 DWARFDebugLine::Sequence::orderByHighPC); 1267 if (It == Sequences.end() || It->SectionIndex != Address.SectionIndex) 1268 return UnknownRowIndex; 1269 return findRowInSeq(*It, Address); 1270 } 1271 1272 bool DWARFDebugLine::LineTable::lookupAddressRange( 1273 object::SectionedAddress Address, uint64_t Size, 1274 std::vector<uint32_t> &Result) const { 1275 1276 // Search for relocatable addresses 1277 if (lookupAddressRangeImpl(Address, Size, Result)) 1278 return true; 1279 1280 if (Address.SectionIndex == object::SectionedAddress::UndefSection) 1281 return false; 1282 1283 // Search for absolute addresses 1284 Address.SectionIndex = object::SectionedAddress::UndefSection; 1285 return lookupAddressRangeImpl(Address, Size, Result); 1286 } 1287 1288 bool DWARFDebugLine::LineTable::lookupAddressRangeImpl( 1289 object::SectionedAddress Address, uint64_t Size, 1290 std::vector<uint32_t> &Result) const { 1291 if (Sequences.empty()) 1292 return false; 1293 uint64_t EndAddr = Address.Address + Size; 1294 // First, find an instruction sequence containing the given address. 1295 DWARFDebugLine::Sequence Sequence; 1296 Sequence.SectionIndex = Address.SectionIndex; 1297 Sequence.HighPC = Address.Address; 1298 SequenceIter LastSeq = Sequences.end(); 1299 SequenceIter SeqPos = llvm::upper_bound( 1300 Sequences, Sequence, DWARFDebugLine::Sequence::orderByHighPC); 1301 if (SeqPos == LastSeq || !SeqPos->containsPC(Address)) 1302 return false; 1303 1304 SequenceIter StartPos = SeqPos; 1305 1306 // Add the rows from the first sequence to the vector, starting with the 1307 // index we just calculated 1308 1309 while (SeqPos != LastSeq && SeqPos->LowPC < EndAddr) { 1310 const DWARFDebugLine::Sequence &CurSeq = *SeqPos; 1311 // For the first sequence, we need to find which row in the sequence is the 1312 // first in our range. 1313 uint32_t FirstRowIndex = CurSeq.FirstRowIndex; 1314 if (SeqPos == StartPos) 1315 FirstRowIndex = findRowInSeq(CurSeq, Address); 1316 1317 // Figure out the last row in the range. 1318 uint32_t LastRowIndex = 1319 findRowInSeq(CurSeq, {EndAddr - 1, Address.SectionIndex}); 1320 if (LastRowIndex == UnknownRowIndex) 1321 LastRowIndex = CurSeq.LastRowIndex - 1; 1322 1323 assert(FirstRowIndex != UnknownRowIndex); 1324 assert(LastRowIndex != UnknownRowIndex); 1325 1326 for (uint32_t I = FirstRowIndex; I <= LastRowIndex; ++I) { 1327 Result.push_back(I); 1328 } 1329 1330 ++SeqPos; 1331 } 1332 1333 return true; 1334 } 1335 1336 std::optional<StringRef> 1337 DWARFDebugLine::LineTable::getSourceByIndex(uint64_t FileIndex, 1338 FileLineInfoKind Kind) const { 1339 if (Kind == FileLineInfoKind::None || !Prologue.hasFileAtIndex(FileIndex)) 1340 return std::nullopt; 1341 const FileNameEntry &Entry = Prologue.getFileNameEntry(FileIndex); 1342 if (auto E = dwarf::toString(Entry.Source)) 1343 return StringRef(*E); 1344 return std::nullopt; 1345 } 1346 1347 static bool isPathAbsoluteOnWindowsOrPosix(const Twine &Path) { 1348 // Debug info can contain paths from any OS, not necessarily 1349 // an OS we're currently running on. Moreover different compilation units can 1350 // be compiled on different operating systems and linked together later. 1351 return sys::path::is_absolute(Path, sys::path::Style::posix) || 1352 sys::path::is_absolute(Path, sys::path::Style::windows); 1353 } 1354 1355 bool DWARFDebugLine::Prologue::getFileNameByIndex( 1356 uint64_t FileIndex, StringRef CompDir, FileLineInfoKind Kind, 1357 std::string &Result, sys::path::Style Style) const { 1358 if (Kind == FileLineInfoKind::None || !hasFileAtIndex(FileIndex)) 1359 return false; 1360 const FileNameEntry &Entry = getFileNameEntry(FileIndex); 1361 auto E = dwarf::toString(Entry.Name); 1362 if (!E) 1363 return false; 1364 StringRef FileName = *E; 1365 if (Kind == FileLineInfoKind::RawValue || 1366 isPathAbsoluteOnWindowsOrPosix(FileName)) { 1367 Result = std::string(FileName); 1368 return true; 1369 } 1370 if (Kind == FileLineInfoKind::BaseNameOnly) { 1371 Result = std::string(llvm::sys::path::filename(FileName)); 1372 return true; 1373 } 1374 1375 SmallString<16> FilePath; 1376 StringRef IncludeDir; 1377 // Be defensive about the contents of Entry. 1378 if (getVersion() >= 5) { 1379 // DirIdx 0 is the compilation directory, so don't include it for 1380 // relative names. 1381 if ((Entry.DirIdx != 0 || Kind != FileLineInfoKind::RelativeFilePath) && 1382 Entry.DirIdx < IncludeDirectories.size()) 1383 IncludeDir = dwarf::toStringRef(IncludeDirectories[Entry.DirIdx]); 1384 } else { 1385 if (0 < Entry.DirIdx && Entry.DirIdx <= IncludeDirectories.size()) 1386 IncludeDir = dwarf::toStringRef(IncludeDirectories[Entry.DirIdx - 1]); 1387 } 1388 1389 // For absolute paths only, include the compilation directory of compile unit, 1390 // unless v5 DirIdx == 0 (IncludeDir indicates the compilation directory). We 1391 // know that FileName is not absolute, the only way to have an absolute path 1392 // at this point would be if IncludeDir is absolute. 1393 if (Kind == FileLineInfoKind::AbsoluteFilePath && 1394 (getVersion() < 5 || Entry.DirIdx != 0) && !CompDir.empty() && 1395 !isPathAbsoluteOnWindowsOrPosix(IncludeDir)) 1396 sys::path::append(FilePath, Style, CompDir); 1397 1398 assert((Kind == FileLineInfoKind::AbsoluteFilePath || 1399 Kind == FileLineInfoKind::RelativeFilePath) && 1400 "invalid FileLineInfo Kind"); 1401 1402 // sys::path::append skips empty strings. 1403 sys::path::append(FilePath, Style, IncludeDir, FileName); 1404 Result = std::string(FilePath.str()); 1405 return true; 1406 } 1407 1408 bool DWARFDebugLine::LineTable::getFileLineInfoForAddress( 1409 object::SectionedAddress Address, const char *CompDir, 1410 FileLineInfoKind Kind, DILineInfo &Result) const { 1411 // Get the index of row we're looking for in the line table. 1412 uint32_t RowIndex = lookupAddress(Address); 1413 if (RowIndex == -1U) 1414 return false; 1415 // Take file number and line/column from the row. 1416 const auto &Row = Rows[RowIndex]; 1417 if (!getFileNameByIndex(Row.File, CompDir, Kind, Result.FileName)) 1418 return false; 1419 Result.Line = Row.Line; 1420 Result.Column = Row.Column; 1421 Result.Discriminator = Row.Discriminator; 1422 Result.Source = getSourceByIndex(Row.File, Kind); 1423 return true; 1424 } 1425 1426 bool DWARFDebugLine::LineTable::getDirectoryForEntry( 1427 const FileNameEntry &Entry, std::string &Directory) const { 1428 if (Prologue.getVersion() >= 5) { 1429 if (Entry.DirIdx < Prologue.IncludeDirectories.size()) { 1430 Directory = 1431 dwarf::toString(Prologue.IncludeDirectories[Entry.DirIdx], ""); 1432 return true; 1433 } 1434 return false; 1435 } 1436 if (0 < Entry.DirIdx && Entry.DirIdx <= Prologue.IncludeDirectories.size()) { 1437 Directory = 1438 dwarf::toString(Prologue.IncludeDirectories[Entry.DirIdx - 1], ""); 1439 return true; 1440 } 1441 return false; 1442 } 1443 1444 // We want to supply the Unit associated with a .debug_line[.dwo] table when 1445 // we dump it, if possible, but still dump the table even if there isn't a Unit. 1446 // Therefore, collect up handles on all the Units that point into the 1447 // line-table section. 1448 static DWARFDebugLine::SectionParser::LineToUnitMap 1449 buildLineToUnitMap(DWARFUnitVector::iterator_range Units) { 1450 DWARFDebugLine::SectionParser::LineToUnitMap LineToUnit; 1451 for (const auto &U : Units) 1452 if (auto CUDIE = U->getUnitDIE()) 1453 if (auto StmtOffset = toSectionOffset(CUDIE.find(DW_AT_stmt_list))) 1454 LineToUnit.insert(std::make_pair(*StmtOffset, &*U)); 1455 return LineToUnit; 1456 } 1457 1458 DWARFDebugLine::SectionParser::SectionParser( 1459 DWARFDataExtractor &Data, const DWARFContext &C, 1460 DWARFUnitVector::iterator_range Units) 1461 : DebugLineData(Data), Context(C) { 1462 LineToUnit = buildLineToUnitMap(Units); 1463 if (!DebugLineData.isValidOffset(Offset)) 1464 Done = true; 1465 } 1466 1467 bool DWARFDebugLine::Prologue::totalLengthIsValid() const { 1468 return TotalLength != 0u; 1469 } 1470 1471 DWARFDebugLine::LineTable DWARFDebugLine::SectionParser::parseNext( 1472 function_ref<void(Error)> RecoverableErrorHandler, 1473 function_ref<void(Error)> UnrecoverableErrorHandler, raw_ostream *OS, 1474 bool Verbose) { 1475 assert(DebugLineData.isValidOffset(Offset) && 1476 "parsing should have terminated"); 1477 DWARFUnit *U = prepareToParse(Offset); 1478 uint64_t OldOffset = Offset; 1479 LineTable LT; 1480 if (Error Err = LT.parse(DebugLineData, &Offset, Context, U, 1481 RecoverableErrorHandler, OS, Verbose)) 1482 UnrecoverableErrorHandler(std::move(Err)); 1483 moveToNextTable(OldOffset, LT.Prologue); 1484 return LT; 1485 } 1486 1487 void DWARFDebugLine::SectionParser::skip( 1488 function_ref<void(Error)> RecoverableErrorHandler, 1489 function_ref<void(Error)> UnrecoverableErrorHandler) { 1490 assert(DebugLineData.isValidOffset(Offset) && 1491 "parsing should have terminated"); 1492 DWARFUnit *U = prepareToParse(Offset); 1493 uint64_t OldOffset = Offset; 1494 LineTable LT; 1495 if (Error Err = LT.Prologue.parse(DebugLineData, &Offset, 1496 RecoverableErrorHandler, Context, U)) 1497 UnrecoverableErrorHandler(std::move(Err)); 1498 moveToNextTable(OldOffset, LT.Prologue); 1499 } 1500 1501 DWARFUnit *DWARFDebugLine::SectionParser::prepareToParse(uint64_t Offset) { 1502 DWARFUnit *U = nullptr; 1503 auto It = LineToUnit.find(Offset); 1504 if (It != LineToUnit.end()) 1505 U = It->second; 1506 DebugLineData.setAddressSize(U ? U->getAddressByteSize() : 0); 1507 return U; 1508 } 1509 1510 bool DWARFDebugLine::SectionParser::hasValidVersion(uint64_t Offset) { 1511 DataExtractor::Cursor Cursor(Offset); 1512 auto [TotalLength, _] = DebugLineData.getInitialLength(Cursor); 1513 DWARFDataExtractor HeaderData(DebugLineData, Cursor.tell() + TotalLength); 1514 uint16_t Version = HeaderData.getU16(Cursor); 1515 if (!Cursor) { 1516 // Ignore any error here. 1517 // If this is not the end of the section parseNext() will still be 1518 // attempted, where this error will occur again (and can be handled). 1519 consumeError(Cursor.takeError()); 1520 return false; 1521 } 1522 return versionIsSupported(Version); 1523 } 1524 1525 void DWARFDebugLine::SectionParser::moveToNextTable(uint64_t OldOffset, 1526 const Prologue &P) { 1527 // If the length field is not valid, we don't know where the next table is, so 1528 // cannot continue to parse. Mark the parser as done, and leave the Offset 1529 // value as it currently is. This will be the end of the bad length field. 1530 if (!P.totalLengthIsValid()) { 1531 Done = true; 1532 return; 1533 } 1534 1535 Offset = OldOffset + P.TotalLength + P.sizeofTotalLength(); 1536 if (!DebugLineData.isValidOffset(Offset)) { 1537 Done = true; 1538 return; 1539 } 1540 1541 // Heuristic: If the version is valid, then this is probably a line table. 1542 // Otherwise, the offset might need alignment (to a 4 or 8 byte boundary). 1543 if (hasValidVersion(Offset)) 1544 return; 1545 1546 // ARM C/C++ Compiler aligns each line table to word boundaries and pads out 1547 // the .debug_line section to a word multiple. Note that in the specification 1548 // this does not seem forbidden since each unit has a DW_AT_stmt_list. 1549 for (unsigned Align : {4, 8}) { 1550 uint64_t AlignedOffset = alignTo(Offset, Align); 1551 if (!DebugLineData.isValidOffset(AlignedOffset)) { 1552 // This is almost certainly not another line table but some alignment 1553 // padding. This assumes the alignments tested are ordered, and are 1554 // smaller than the header size (which is true for 4 and 8). 1555 Done = true; 1556 return; 1557 } 1558 if (hasValidVersion(AlignedOffset)) { 1559 Offset = AlignedOffset; 1560 break; 1561 } 1562 } 1563 } 1564