1 //===- PDB.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 "PDB.h" 10 #include "COFFLinkerContext.h" 11 #include "Chunks.h" 12 #include "Config.h" 13 #include "DebugTypes.h" 14 #include "Driver.h" 15 #include "SymbolTable.h" 16 #include "Symbols.h" 17 #include "TypeMerger.h" 18 #include "Writer.h" 19 #include "lld/Common/Timer.h" 20 #include "llvm/DebugInfo/CodeView/DebugFrameDataSubsection.h" 21 #include "llvm/DebugInfo/CodeView/DebugInlineeLinesSubsection.h" 22 #include "llvm/DebugInfo/CodeView/DebugLinesSubsection.h" 23 #include "llvm/DebugInfo/CodeView/DebugSubsectionRecord.h" 24 #include "llvm/DebugInfo/CodeView/GlobalTypeTableBuilder.h" 25 #include "llvm/DebugInfo/CodeView/LazyRandomTypeCollection.h" 26 #include "llvm/DebugInfo/CodeView/MergingTypeTableBuilder.h" 27 #include "llvm/DebugInfo/CodeView/RecordName.h" 28 #include "llvm/DebugInfo/CodeView/SymbolDeserializer.h" 29 #include "llvm/DebugInfo/CodeView/SymbolRecordHelpers.h" 30 #include "llvm/DebugInfo/CodeView/SymbolSerializer.h" 31 #include "llvm/DebugInfo/CodeView/TypeIndexDiscovery.h" 32 #include "llvm/DebugInfo/MSF/MSFBuilder.h" 33 #include "llvm/DebugInfo/MSF/MSFCommon.h" 34 #include "llvm/DebugInfo/MSF/MSFError.h" 35 #include "llvm/DebugInfo/PDB/GenericError.h" 36 #include "llvm/DebugInfo/PDB/Native/DbiModuleDescriptorBuilder.h" 37 #include "llvm/DebugInfo/PDB/Native/DbiStream.h" 38 #include "llvm/DebugInfo/PDB/Native/DbiStreamBuilder.h" 39 #include "llvm/DebugInfo/PDB/Native/GSIStreamBuilder.h" 40 #include "llvm/DebugInfo/PDB/Native/InfoStream.h" 41 #include "llvm/DebugInfo/PDB/Native/InfoStreamBuilder.h" 42 #include "llvm/DebugInfo/PDB/Native/NativeSession.h" 43 #include "llvm/DebugInfo/PDB/Native/PDBFile.h" 44 #include "llvm/DebugInfo/PDB/Native/PDBFileBuilder.h" 45 #include "llvm/DebugInfo/PDB/Native/PDBStringTableBuilder.h" 46 #include "llvm/DebugInfo/PDB/Native/TpiHashing.h" 47 #include "llvm/DebugInfo/PDB/Native/TpiStream.h" 48 #include "llvm/DebugInfo/PDB/Native/TpiStreamBuilder.h" 49 #include "llvm/DebugInfo/PDB/PDB.h" 50 #include "llvm/Object/COFF.h" 51 #include "llvm/Object/CVDebugRecord.h" 52 #include "llvm/Support/BinaryByteStream.h" 53 #include "llvm/Support/CRC.h" 54 #include "llvm/Support/Endian.h" 55 #include "llvm/Support/Errc.h" 56 #include "llvm/Support/FormatAdapters.h" 57 #include "llvm/Support/FormatVariadic.h" 58 #include "llvm/Support/Path.h" 59 #include "llvm/Support/ScopedPrinter.h" 60 #include "llvm/Support/TimeProfiler.h" 61 #include <memory> 62 #include <optional> 63 64 using namespace llvm; 65 using namespace llvm::codeview; 66 using namespace lld; 67 using namespace lld::coff; 68 69 using llvm::object::coff_section; 70 using llvm::pdb::StringTableFixup; 71 72 namespace { 73 class DebugSHandler; 74 75 class PDBLinker { 76 friend DebugSHandler; 77 78 public: 79 PDBLinker(COFFLinkerContext &ctx) 80 : builder(bAlloc()), tMerger(ctx, bAlloc()), ctx(ctx) { 81 // This isn't strictly necessary, but link.exe usually puts an empty string 82 // as the first "valid" string in the string table, so we do the same in 83 // order to maintain as much byte-for-byte compatibility as possible. 84 pdbStrTab.insert(""); 85 } 86 87 /// Emit the basic PDB structure: initial streams, headers, etc. 88 void initialize(llvm::codeview::DebugInfo *buildId); 89 90 /// Add natvis files specified on the command line. 91 void addNatvisFiles(); 92 93 /// Add named streams specified on the command line. 94 void addNamedStreams(); 95 96 /// Link CodeView from each object file in the symbol table into the PDB. 97 void addObjectsToPDB(); 98 99 /// Add every live, defined public symbol to the PDB. 100 void addPublicsToPDB(); 101 102 /// Link info for each import file in the symbol table into the PDB. 103 void addImportFilesToPDB(); 104 105 void createModuleDBI(ObjFile *file); 106 107 /// Link CodeView from a single object file into the target (output) PDB. 108 /// When a precompiled headers object is linked, its TPI map might be provided 109 /// externally. 110 void addDebug(TpiSource *source); 111 112 void addDebugSymbols(TpiSource *source); 113 114 // Analyze the symbol records to separate module symbols from global symbols, 115 // find string references, and calculate how large the symbol stream will be 116 // in the PDB. 117 void analyzeSymbolSubsection(SectionChunk *debugChunk, 118 uint32_t &moduleSymOffset, 119 uint32_t &nextRelocIndex, 120 std::vector<StringTableFixup> &stringTableFixups, 121 BinaryStreamRef symData); 122 123 // Write all module symbols from all live debug symbol subsections of the 124 // given object file into the given stream writer. 125 Error writeAllModuleSymbolRecords(ObjFile *file, BinaryStreamWriter &writer); 126 127 // Callback to copy and relocate debug symbols during PDB file writing. 128 static Error commitSymbolsForObject(void *ctx, void *obj, 129 BinaryStreamWriter &writer); 130 131 // Copy the symbol record, relocate it, and fix the alignment if necessary. 132 // Rewrite type indices in the record. Replace unrecognized symbol records 133 // with S_SKIP records. 134 void writeSymbolRecord(SectionChunk *debugChunk, 135 ArrayRef<uint8_t> sectionContents, CVSymbol sym, 136 size_t alignedSize, uint32_t &nextRelocIndex, 137 std::vector<uint8_t> &storage); 138 139 /// Add the section map and section contributions to the PDB. 140 void addSections(ArrayRef<uint8_t> sectionTable); 141 142 /// Write the PDB to disk and store the Guid generated for it in *Guid. 143 void commit(codeview::GUID *guid); 144 145 // Print statistics regarding the final PDB 146 void printStats(); 147 148 private: 149 void pdbMakeAbsolute(SmallVectorImpl<char> &fileName); 150 void translateIdSymbols(MutableArrayRef<uint8_t> &recordData, 151 TpiSource *source); 152 void addCommonLinkerModuleSymbols(StringRef path, 153 pdb::DbiModuleDescriptorBuilder &mod); 154 155 pdb::PDBFileBuilder builder; 156 157 TypeMerger tMerger; 158 159 COFFLinkerContext &ctx; 160 161 /// PDBs use a single global string table for filenames in the file checksum 162 /// table. 163 DebugStringTableSubsection pdbStrTab; 164 165 llvm::SmallString<128> nativePath; 166 167 // For statistics 168 uint64_t globalSymbols = 0; 169 uint64_t moduleSymbols = 0; 170 uint64_t publicSymbols = 0; 171 uint64_t nbTypeRecords = 0; 172 uint64_t nbTypeRecordsBytes = 0; 173 }; 174 175 /// Represents an unrelocated DEBUG_S_FRAMEDATA subsection. 176 struct UnrelocatedFpoData { 177 SectionChunk *debugChunk = nullptr; 178 ArrayRef<uint8_t> subsecData; 179 uint32_t relocIndex = 0; 180 }; 181 182 /// The size of the magic bytes at the beginning of a symbol section or stream. 183 enum : uint32_t { kSymbolStreamMagicSize = 4 }; 184 185 class DebugSHandler { 186 COFFLinkerContext &ctx; 187 PDBLinker &linker; 188 189 /// The object file whose .debug$S sections we're processing. 190 ObjFile &file; 191 192 /// The DEBUG_S_STRINGTABLE subsection. These strings are referred to by 193 /// index from other records in the .debug$S section. All of these strings 194 /// need to be added to the global PDB string table, and all references to 195 /// these strings need to have their indices re-written to refer to the 196 /// global PDB string table. 197 DebugStringTableSubsectionRef cvStrTab; 198 199 /// The DEBUG_S_FILECHKSMS subsection. As above, these are referred to 200 /// by other records in the .debug$S section and need to be merged into the 201 /// PDB. 202 DebugChecksumsSubsectionRef checksums; 203 204 /// The DEBUG_S_FRAMEDATA subsection(s). There can be more than one of 205 /// these and they need not appear in any specific order. However, they 206 /// contain string table references which need to be re-written, so we 207 /// collect them all here and re-write them after all subsections have been 208 /// discovered and processed. 209 std::vector<UnrelocatedFpoData> frameDataSubsecs; 210 211 /// List of string table references in symbol records. Later they will be 212 /// applied to the symbols during PDB writing. 213 std::vector<StringTableFixup> stringTableFixups; 214 215 /// Sum of the size of all module symbol records across all .debug$S sections. 216 /// Includes record realignment and the size of the symbol stream magic 217 /// prefix. 218 uint32_t moduleStreamSize = kSymbolStreamMagicSize; 219 220 /// Next relocation index in the current .debug$S section. Resets every 221 /// handleDebugS call. 222 uint32_t nextRelocIndex = 0; 223 224 void advanceRelocIndex(SectionChunk *debugChunk, ArrayRef<uint8_t> subsec); 225 226 void addUnrelocatedSubsection(SectionChunk *debugChunk, 227 const DebugSubsectionRecord &ss); 228 229 void addFrameDataSubsection(SectionChunk *debugChunk, 230 const DebugSubsectionRecord &ss); 231 232 public: 233 DebugSHandler(COFFLinkerContext &ctx, PDBLinker &linker, ObjFile &file) 234 : ctx(ctx), linker(linker), file(file) {} 235 236 void handleDebugS(SectionChunk *debugChunk); 237 238 void finish(); 239 }; 240 } 241 242 // Visual Studio's debugger requires absolute paths in various places in the 243 // PDB to work without additional configuration: 244 // https://docs.microsoft.com/en-us/visualstudio/debugger/debug-source-files-common-properties-solution-property-pages-dialog-box 245 void PDBLinker::pdbMakeAbsolute(SmallVectorImpl<char> &fileName) { 246 // The default behavior is to produce paths that are valid within the context 247 // of the machine that you perform the link on. If the linker is running on 248 // a POSIX system, we will output absolute POSIX paths. If the linker is 249 // running on a Windows system, we will output absolute Windows paths. If the 250 // user desires any other kind of behavior, they should explicitly pass 251 // /pdbsourcepath, in which case we will treat the exact string the user 252 // passed in as the gospel and not normalize, canonicalize it. 253 if (sys::path::is_absolute(fileName, sys::path::Style::windows) || 254 sys::path::is_absolute(fileName, sys::path::Style::posix)) 255 return; 256 257 // It's not absolute in any path syntax. Relative paths necessarily refer to 258 // the local file system, so we can make it native without ending up with a 259 // nonsensical path. 260 if (ctx.config.pdbSourcePath.empty()) { 261 sys::path::native(fileName); 262 sys::fs::make_absolute(fileName); 263 sys::path::remove_dots(fileName, true); 264 return; 265 } 266 267 // Try to guess whether /PDBSOURCEPATH is a unix path or a windows path. 268 // Since PDB's are more of a Windows thing, we make this conservative and only 269 // decide that it's a unix path if we're fairly certain. Specifically, if 270 // it starts with a forward slash. 271 SmallString<128> absoluteFileName = ctx.config.pdbSourcePath; 272 sys::path::Style guessedStyle = absoluteFileName.starts_with("/") 273 ? sys::path::Style::posix 274 : sys::path::Style::windows; 275 sys::path::append(absoluteFileName, guessedStyle, fileName); 276 sys::path::native(absoluteFileName, guessedStyle); 277 sys::path::remove_dots(absoluteFileName, true, guessedStyle); 278 279 fileName = std::move(absoluteFileName); 280 } 281 282 static void addTypeInfo(pdb::TpiStreamBuilder &tpiBuilder, 283 TypeCollection &typeTable) { 284 // Start the TPI or IPI stream header. 285 tpiBuilder.setVersionHeader(pdb::PdbTpiV80); 286 287 // Flatten the in memory type table and hash each type. 288 typeTable.ForEachRecord([&](TypeIndex ti, const CVType &type) { 289 auto hash = pdb::hashTypeRecord(type); 290 if (auto e = hash.takeError()) 291 fatal("type hashing error"); 292 tpiBuilder.addTypeRecord(type.RecordData, *hash); 293 }); 294 } 295 296 static void addGHashTypeInfo(COFFLinkerContext &ctx, 297 pdb::PDBFileBuilder &builder) { 298 // Start the TPI or IPI stream header. 299 builder.getTpiBuilder().setVersionHeader(pdb::PdbTpiV80); 300 builder.getIpiBuilder().setVersionHeader(pdb::PdbTpiV80); 301 for (TpiSource *source : ctx.tpiSourceList) { 302 builder.getTpiBuilder().addTypeRecords(source->mergedTpi.recs, 303 source->mergedTpi.recSizes, 304 source->mergedTpi.recHashes); 305 builder.getIpiBuilder().addTypeRecords(source->mergedIpi.recs, 306 source->mergedIpi.recSizes, 307 source->mergedIpi.recHashes); 308 } 309 } 310 311 static void 312 recordStringTableReferences(CVSymbol sym, uint32_t symOffset, 313 std::vector<StringTableFixup> &stringTableFixups) { 314 // For now we only handle S_FILESTATIC, but we may need the same logic for 315 // S_DEFRANGE and S_DEFRANGE_SUBFIELD. However, I cannot seem to generate any 316 // PDBs that contain these types of records, so because of the uncertainty 317 // they are omitted here until we can prove that it's necessary. 318 switch (sym.kind()) { 319 case SymbolKind::S_FILESTATIC: { 320 // FileStaticSym::ModFileOffset 321 uint32_t ref = *reinterpret_cast<const ulittle32_t *>(&sym.data()[8]); 322 stringTableFixups.push_back({ref, symOffset + 8}); 323 break; 324 } 325 case SymbolKind::S_DEFRANGE: 326 case SymbolKind::S_DEFRANGE_SUBFIELD: 327 log("Not fixing up string table reference in S_DEFRANGE / " 328 "S_DEFRANGE_SUBFIELD record"); 329 break; 330 default: 331 break; 332 } 333 } 334 335 static SymbolKind symbolKind(ArrayRef<uint8_t> recordData) { 336 const RecordPrefix *prefix = 337 reinterpret_cast<const RecordPrefix *>(recordData.data()); 338 return static_cast<SymbolKind>(uint16_t(prefix->RecordKind)); 339 } 340 341 /// MSVC translates S_PROC_ID_END to S_END, and S_[LG]PROC32_ID to S_[LG]PROC32 342 void PDBLinker::translateIdSymbols(MutableArrayRef<uint8_t> &recordData, 343 TpiSource *source) { 344 RecordPrefix *prefix = reinterpret_cast<RecordPrefix *>(recordData.data()); 345 346 SymbolKind kind = symbolKind(recordData); 347 348 if (kind == SymbolKind::S_PROC_ID_END) { 349 prefix->RecordKind = SymbolKind::S_END; 350 return; 351 } 352 353 // In an object file, GPROC32_ID has an embedded reference which refers to the 354 // single object file type index namespace. This has already been translated 355 // to the PDB file's ID stream index space, but we need to convert this to a 356 // symbol that refers to the type stream index space. So we remap again from 357 // ID index space to type index space. 358 if (kind == SymbolKind::S_GPROC32_ID || kind == SymbolKind::S_LPROC32_ID) { 359 SmallVector<TiReference, 1> refs; 360 auto content = recordData.drop_front(sizeof(RecordPrefix)); 361 CVSymbol sym(recordData); 362 discoverTypeIndicesInSymbol(sym, refs); 363 assert(refs.size() == 1); 364 assert(refs.front().Count == 1); 365 366 TypeIndex *ti = 367 reinterpret_cast<TypeIndex *>(content.data() + refs[0].Offset); 368 // `ti` is the index of a FuncIdRecord or MemberFuncIdRecord which lives in 369 // the IPI stream, whose `FunctionType` member refers to the TPI stream. 370 // Note that LF_FUNC_ID and LF_MFUNC_ID have the same record layout, and 371 // in both cases we just need the second type index. 372 if (!ti->isSimple() && !ti->isNoneType()) { 373 TypeIndex newType = TypeIndex(SimpleTypeKind::NotTranslated); 374 if (ctx.config.debugGHashes) { 375 auto idToType = tMerger.funcIdToType.find(*ti); 376 if (idToType != tMerger.funcIdToType.end()) 377 newType = idToType->second; 378 } else { 379 if (tMerger.getIDTable().contains(*ti)) { 380 CVType funcIdData = tMerger.getIDTable().getType(*ti); 381 if (funcIdData.length() >= 8 && (funcIdData.kind() == LF_FUNC_ID || 382 funcIdData.kind() == LF_MFUNC_ID)) { 383 newType = *reinterpret_cast<const TypeIndex *>(&funcIdData.data()[8]); 384 } 385 } 386 } 387 if (newType == TypeIndex(SimpleTypeKind::NotTranslated)) { 388 Warn(ctx) << formatv( 389 "procedure symbol record for `{0}` in {1} refers to PDB " 390 "item index {2:X} which is not a valid function ID record", 391 getSymbolName(CVSymbol(recordData)), source->file->getName(), 392 ti->getIndex()); 393 } 394 *ti = newType; 395 } 396 397 kind = (kind == SymbolKind::S_GPROC32_ID) ? SymbolKind::S_GPROC32 398 : SymbolKind::S_LPROC32; 399 prefix->RecordKind = uint16_t(kind); 400 } 401 } 402 403 namespace { 404 struct ScopeRecord { 405 ulittle32_t ptrParent; 406 ulittle32_t ptrEnd; 407 }; 408 } // namespace 409 410 /// Given a pointer to a symbol record that opens a scope, return a pointer to 411 /// the scope fields. 412 static ScopeRecord *getSymbolScopeFields(void *sym) { 413 return reinterpret_cast<ScopeRecord *>(reinterpret_cast<char *>(sym) + 414 sizeof(RecordPrefix)); 415 } 416 417 // To open a scope, push the offset of the current symbol record onto the 418 // stack. 419 static void scopeStackOpen(SmallVectorImpl<uint32_t> &stack, 420 std::vector<uint8_t> &storage) { 421 stack.push_back(storage.size()); 422 } 423 424 // To close a scope, update the record that opened the scope. 425 static void scopeStackClose(COFFLinkerContext &ctx, 426 SmallVectorImpl<uint32_t> &stack, 427 std::vector<uint8_t> &storage, 428 uint32_t storageBaseOffset, ObjFile *file) { 429 if (stack.empty()) { 430 Warn(ctx) << "symbol scopes are not balanced in " << file->getName(); 431 return; 432 } 433 434 // Update ptrEnd of the record that opened the scope to point to the 435 // current record, if we are writing into the module symbol stream. 436 uint32_t offOpen = stack.pop_back_val(); 437 uint32_t offEnd = storageBaseOffset + storage.size(); 438 uint32_t offParent = stack.empty() ? 0 : (stack.back() + storageBaseOffset); 439 ScopeRecord *scopeRec = getSymbolScopeFields(&(storage)[offOpen]); 440 scopeRec->ptrParent = offParent; 441 scopeRec->ptrEnd = offEnd; 442 } 443 444 static bool symbolGoesInModuleStream(const CVSymbol &sym, 445 unsigned symbolScopeDepth) { 446 switch (sym.kind()) { 447 case SymbolKind::S_GDATA32: 448 case SymbolKind::S_GTHREAD32: 449 // We really should not be seeing S_PROCREF and S_LPROCREF in the first place 450 // since they are synthesized by the linker in response to S_GPROC32 and 451 // S_LPROC32, but if we do see them, don't put them in the module stream I 452 // guess. 453 case SymbolKind::S_PROCREF: 454 case SymbolKind::S_LPROCREF: 455 return false; 456 // S_UDT and S_CONSTANT records go in the module stream if it is not a global record. 457 case SymbolKind::S_UDT: 458 case SymbolKind::S_CONSTANT: 459 return symbolScopeDepth > 0; 460 // S_GDATA32 does not go in the module stream, but S_LDATA32 does. 461 case SymbolKind::S_LDATA32: 462 case SymbolKind::S_LTHREAD32: 463 default: 464 return true; 465 } 466 } 467 468 static bool symbolGoesInGlobalsStream(const CVSymbol &sym, 469 unsigned symbolScopeDepth) { 470 switch (sym.kind()) { 471 case SymbolKind::S_GDATA32: 472 case SymbolKind::S_GTHREAD32: 473 case SymbolKind::S_GPROC32: 474 case SymbolKind::S_LPROC32: 475 case SymbolKind::S_GPROC32_ID: 476 case SymbolKind::S_LPROC32_ID: 477 // We really should not be seeing S_PROCREF and S_LPROCREF in the first place 478 // since they are synthesized by the linker in response to S_GPROC32 and 479 // S_LPROC32, but if we do see them, copy them straight through. 480 case SymbolKind::S_PROCREF: 481 case SymbolKind::S_LPROCREF: 482 return true; 483 // Records that go in the globals stream, unless they are function-local. 484 case SymbolKind::S_UDT: 485 case SymbolKind::S_LDATA32: 486 case SymbolKind::S_LTHREAD32: 487 case SymbolKind::S_CONSTANT: 488 return symbolScopeDepth == 0; 489 default: 490 return false; 491 } 492 } 493 494 static void addGlobalSymbol(pdb::GSIStreamBuilder &builder, uint16_t modIndex, 495 unsigned symOffset, 496 std::vector<uint8_t> &symStorage) { 497 CVSymbol sym{ArrayRef(symStorage)}; 498 switch (sym.kind()) { 499 case SymbolKind::S_CONSTANT: 500 case SymbolKind::S_UDT: 501 case SymbolKind::S_GDATA32: 502 case SymbolKind::S_GTHREAD32: 503 case SymbolKind::S_LTHREAD32: 504 case SymbolKind::S_LDATA32: 505 case SymbolKind::S_PROCREF: 506 case SymbolKind::S_LPROCREF: { 507 // sym is a temporary object, so we have to copy and reallocate the record 508 // to stabilize it. 509 uint8_t *mem = bAlloc().Allocate<uint8_t>(sym.length()); 510 memcpy(mem, sym.data().data(), sym.length()); 511 builder.addGlobalSymbol(CVSymbol(ArrayRef(mem, sym.length()))); 512 break; 513 } 514 case SymbolKind::S_GPROC32: 515 case SymbolKind::S_LPROC32: { 516 SymbolRecordKind k = SymbolRecordKind::ProcRefSym; 517 if (sym.kind() == SymbolKind::S_LPROC32) 518 k = SymbolRecordKind::LocalProcRef; 519 ProcRefSym ps(k); 520 ps.Module = modIndex; 521 // For some reason, MSVC seems to add one to this value. 522 ++ps.Module; 523 ps.Name = getSymbolName(sym); 524 ps.SumName = 0; 525 ps.SymOffset = symOffset; 526 builder.addGlobalSymbol(ps); 527 break; 528 } 529 default: 530 llvm_unreachable("Invalid symbol kind!"); 531 } 532 } 533 534 // Check if the given symbol record was padded for alignment. If so, zero out 535 // the padding bytes and update the record prefix with the new size. 536 static void fixRecordAlignment(MutableArrayRef<uint8_t> recordBytes, 537 size_t oldSize) { 538 size_t alignedSize = recordBytes.size(); 539 if (oldSize == alignedSize) 540 return; 541 reinterpret_cast<RecordPrefix *>(recordBytes.data())->RecordLen = 542 alignedSize - 2; 543 memset(recordBytes.data() + oldSize, 0, alignedSize - oldSize); 544 } 545 546 // Replace any record with a skip record of the same size. This is useful when 547 // we have reserved size for a symbol record, but type index remapping fails. 548 static void replaceWithSkipRecord(MutableArrayRef<uint8_t> recordBytes) { 549 memset(recordBytes.data(), 0, recordBytes.size()); 550 auto *prefix = reinterpret_cast<RecordPrefix *>(recordBytes.data()); 551 prefix->RecordKind = SymbolKind::S_SKIP; 552 prefix->RecordLen = recordBytes.size() - 2; 553 } 554 555 // Copy the symbol record, relocate it, and fix the alignment if necessary. 556 // Rewrite type indices in the record. Replace unrecognized symbol records with 557 // S_SKIP records. 558 void PDBLinker::writeSymbolRecord(SectionChunk *debugChunk, 559 ArrayRef<uint8_t> sectionContents, 560 CVSymbol sym, size_t alignedSize, 561 uint32_t &nextRelocIndex, 562 std::vector<uint8_t> &storage) { 563 // Allocate space for the new record at the end of the storage. 564 storage.resize(storage.size() + alignedSize); 565 auto recordBytes = MutableArrayRef<uint8_t>(storage).take_back(alignedSize); 566 567 // Copy the symbol record and relocate it. 568 debugChunk->writeAndRelocateSubsection(sectionContents, sym.data(), 569 nextRelocIndex, recordBytes.data()); 570 fixRecordAlignment(recordBytes, sym.length()); 571 572 // Re-map all the type index references. 573 TpiSource *source = debugChunk->file->debugTypesObj; 574 if (!source->remapTypesInSymbolRecord(recordBytes)) { 575 Log(ctx) << "ignoring unknown symbol record with kind 0x" 576 << utohexstr(sym.kind()); 577 replaceWithSkipRecord(recordBytes); 578 } 579 580 // An object file may have S_xxx_ID symbols, but these get converted to 581 // "real" symbols in a PDB. 582 translateIdSymbols(recordBytes, source); 583 } 584 585 void PDBLinker::analyzeSymbolSubsection( 586 SectionChunk *debugChunk, uint32_t &moduleSymOffset, 587 uint32_t &nextRelocIndex, std::vector<StringTableFixup> &stringTableFixups, 588 BinaryStreamRef symData) { 589 ObjFile *file = debugChunk->file; 590 uint32_t moduleSymStart = moduleSymOffset; 591 592 uint32_t scopeLevel = 0; 593 std::vector<uint8_t> storage; 594 ArrayRef<uint8_t> sectionContents = debugChunk->getContents(); 595 596 ArrayRef<uint8_t> symsBuffer; 597 cantFail(symData.readBytes(0, symData.getLength(), symsBuffer)); 598 599 if (symsBuffer.empty()) 600 Warn(ctx) << "empty symbols subsection in " << file->getName(); 601 602 Error ec = forEachCodeViewRecord<CVSymbol>( 603 symsBuffer, [&](CVSymbol sym) -> llvm::Error { 604 // Track the current scope. 605 if (symbolOpensScope(sym.kind())) 606 ++scopeLevel; 607 else if (symbolEndsScope(sym.kind())) 608 --scopeLevel; 609 610 uint32_t alignedSize = 611 alignTo(sym.length(), alignOf(CodeViewContainer::Pdb)); 612 613 // Copy global records. Some global records (mainly procedures) 614 // reference the current offset into the module stream. 615 if (symbolGoesInGlobalsStream(sym, scopeLevel)) { 616 storage.clear(); 617 writeSymbolRecord(debugChunk, sectionContents, sym, alignedSize, 618 nextRelocIndex, storage); 619 addGlobalSymbol(builder.getGsiBuilder(), 620 file->moduleDBI->getModuleIndex(), moduleSymOffset, 621 storage); 622 ++globalSymbols; 623 } 624 625 // Update the module stream offset and record any string table index 626 // references. There are very few of these and they will be rewritten 627 // later during PDB writing. 628 if (symbolGoesInModuleStream(sym, scopeLevel)) { 629 recordStringTableReferences(sym, moduleSymOffset, stringTableFixups); 630 moduleSymOffset += alignedSize; 631 ++moduleSymbols; 632 } 633 634 return Error::success(); 635 }); 636 637 // If we encountered corrupt records, ignore the whole subsection. If we wrote 638 // any partial records, undo that. For globals, we just keep what we have and 639 // continue. 640 if (ec) { 641 Warn(ctx) << "corrupt symbol records in " << file->getName(); 642 moduleSymOffset = moduleSymStart; 643 consumeError(std::move(ec)); 644 } 645 } 646 647 Error PDBLinker::writeAllModuleSymbolRecords(ObjFile *file, 648 BinaryStreamWriter &writer) { 649 ExitOnError exitOnErr; 650 std::vector<uint8_t> storage; 651 SmallVector<uint32_t, 4> scopes; 652 653 // Visit all live .debug$S sections a second time, and write them to the PDB. 654 for (SectionChunk *debugChunk : file->getDebugChunks()) { 655 if (!debugChunk->live || debugChunk->getSize() == 0 || 656 debugChunk->getSectionName() != ".debug$S") 657 continue; 658 659 ArrayRef<uint8_t> sectionContents = debugChunk->getContents(); 660 auto contents = 661 SectionChunk::consumeDebugMagic(sectionContents, ".debug$S"); 662 DebugSubsectionArray subsections; 663 BinaryStreamReader reader(contents, llvm::endianness::little); 664 exitOnErr(reader.readArray(subsections, contents.size())); 665 666 uint32_t nextRelocIndex = 0; 667 for (const DebugSubsectionRecord &ss : subsections) { 668 if (ss.kind() != DebugSubsectionKind::Symbols) 669 continue; 670 671 uint32_t moduleSymStart = writer.getOffset(); 672 scopes.clear(); 673 storage.clear(); 674 ArrayRef<uint8_t> symsBuffer; 675 BinaryStreamRef sr = ss.getRecordData(); 676 cantFail(sr.readBytes(0, sr.getLength(), symsBuffer)); 677 auto ec = forEachCodeViewRecord<CVSymbol>( 678 symsBuffer, [&](CVSymbol sym) -> llvm::Error { 679 // Track the current scope. Only update records in the postmerge 680 // pass. 681 if (symbolOpensScope(sym.kind())) 682 scopeStackOpen(scopes, storage); 683 else if (symbolEndsScope(sym.kind())) 684 scopeStackClose(ctx, scopes, storage, moduleSymStart, file); 685 686 // Copy, relocate, and rewrite each module symbol. 687 if (symbolGoesInModuleStream(sym, scopes.size())) { 688 uint32_t alignedSize = 689 alignTo(sym.length(), alignOf(CodeViewContainer::Pdb)); 690 writeSymbolRecord(debugChunk, sectionContents, sym, alignedSize, 691 nextRelocIndex, storage); 692 } 693 return Error::success(); 694 }); 695 696 // If we encounter corrupt records in the second pass, ignore them. We 697 // already warned about them in the first analysis pass. 698 if (ec) { 699 consumeError(std::move(ec)); 700 storage.clear(); 701 } 702 703 // Writing bytes has a very high overhead, so write the entire subsection 704 // at once. 705 // TODO: Consider buffering symbols for the entire object file to reduce 706 // overhead even further. 707 if (Error e = writer.writeBytes(storage)) 708 return e; 709 } 710 } 711 712 return Error::success(); 713 } 714 715 Error PDBLinker::commitSymbolsForObject(void *ctx, void *obj, 716 BinaryStreamWriter &writer) { 717 return static_cast<PDBLinker *>(ctx)->writeAllModuleSymbolRecords( 718 static_cast<ObjFile *>(obj), writer); 719 } 720 721 static pdb::SectionContrib createSectionContrib(COFFLinkerContext &ctx, 722 const Chunk *c, uint32_t modi) { 723 OutputSection *os = c ? ctx.getOutputSection(c) : nullptr; 724 pdb::SectionContrib sc; 725 memset(&sc, 0, sizeof(sc)); 726 sc.ISect = os ? os->sectionIndex : llvm::pdb::kInvalidStreamIndex; 727 sc.Off = c && os ? c->getRVA() - os->getRVA() : 0; 728 sc.Size = c ? c->getSize() : -1; 729 if (auto *secChunk = dyn_cast_or_null<SectionChunk>(c)) { 730 sc.Characteristics = secChunk->header->Characteristics; 731 sc.Imod = secChunk->file->moduleDBI->getModuleIndex(); 732 ArrayRef<uint8_t> contents = secChunk->getContents(); 733 JamCRC crc(0); 734 crc.update(contents); 735 sc.DataCrc = crc.getCRC(); 736 } else { 737 sc.Characteristics = os ? os->header.Characteristics : 0; 738 sc.Imod = modi; 739 } 740 sc.RelocCrc = 0; // FIXME 741 742 return sc; 743 } 744 745 static uint32_t 746 translateStringTableIndex(COFFLinkerContext &ctx, uint32_t objIndex, 747 const DebugStringTableSubsectionRef &objStrTable, 748 DebugStringTableSubsection &pdbStrTable) { 749 auto expectedString = objStrTable.getString(objIndex); 750 if (!expectedString) { 751 Warn(ctx) << "Invalid string table reference"; 752 consumeError(expectedString.takeError()); 753 return 0; 754 } 755 756 return pdbStrTable.insert(*expectedString); 757 } 758 759 void DebugSHandler::handleDebugS(SectionChunk *debugChunk) { 760 // Note that we are processing the *unrelocated* section contents. They will 761 // be relocated later during PDB writing. 762 ArrayRef<uint8_t> contents = debugChunk->getContents(); 763 contents = SectionChunk::consumeDebugMagic(contents, ".debug$S"); 764 DebugSubsectionArray subsections; 765 BinaryStreamReader reader(contents, llvm::endianness::little); 766 ExitOnError exitOnErr; 767 exitOnErr(reader.readArray(subsections, contents.size())); 768 debugChunk->sortRelocations(); 769 770 // Reset the relocation index, since this is a new section. 771 nextRelocIndex = 0; 772 773 for (const DebugSubsectionRecord &ss : subsections) { 774 // Ignore subsections with the 'ignore' bit. Some versions of the Visual C++ 775 // runtime have subsections with this bit set. 776 if (uint32_t(ss.kind()) & codeview::SubsectionIgnoreFlag) 777 continue; 778 779 switch (ss.kind()) { 780 case DebugSubsectionKind::StringTable: { 781 assert(!cvStrTab.valid() && 782 "Encountered multiple string table subsections!"); 783 exitOnErr(cvStrTab.initialize(ss.getRecordData())); 784 break; 785 } 786 case DebugSubsectionKind::FileChecksums: 787 assert(!checksums.valid() && 788 "Encountered multiple checksum subsections!"); 789 exitOnErr(checksums.initialize(ss.getRecordData())); 790 break; 791 case DebugSubsectionKind::Lines: 792 case DebugSubsectionKind::InlineeLines: 793 addUnrelocatedSubsection(debugChunk, ss); 794 break; 795 case DebugSubsectionKind::FrameData: 796 addFrameDataSubsection(debugChunk, ss); 797 break; 798 case DebugSubsectionKind::Symbols: 799 linker.analyzeSymbolSubsection(debugChunk, moduleStreamSize, 800 nextRelocIndex, stringTableFixups, 801 ss.getRecordData()); 802 break; 803 804 case DebugSubsectionKind::CrossScopeImports: 805 case DebugSubsectionKind::CrossScopeExports: 806 // These appear to relate to cross-module optimization, so we might use 807 // these for ThinLTO. 808 break; 809 810 case DebugSubsectionKind::ILLines: 811 case DebugSubsectionKind::FuncMDTokenMap: 812 case DebugSubsectionKind::TypeMDTokenMap: 813 case DebugSubsectionKind::MergedAssemblyInput: 814 // These appear to relate to .Net assembly info. 815 break; 816 817 case DebugSubsectionKind::CoffSymbolRVA: 818 // Unclear what this is for. 819 break; 820 821 case DebugSubsectionKind::XfgHashType: 822 case DebugSubsectionKind::XfgHashVirtual: 823 break; 824 825 default: 826 Warn(ctx) << "ignoring unknown debug$S subsection kind 0x" 827 << utohexstr(uint32_t(ss.kind())) << " in file " 828 << toString(&file); 829 break; 830 } 831 } 832 } 833 834 void DebugSHandler::advanceRelocIndex(SectionChunk *sc, 835 ArrayRef<uint8_t> subsec) { 836 ptrdiff_t vaBegin = subsec.data() - sc->getContents().data(); 837 assert(vaBegin > 0); 838 auto relocs = sc->getRelocs(); 839 for (; nextRelocIndex < relocs.size(); ++nextRelocIndex) { 840 if (relocs[nextRelocIndex].VirtualAddress >= (uint32_t)vaBegin) 841 break; 842 } 843 } 844 845 namespace { 846 /// Wrapper class for unrelocated line and inlinee line subsections, which 847 /// require only relocation and type index remapping to add to the PDB. 848 class UnrelocatedDebugSubsection : public DebugSubsection { 849 public: 850 UnrelocatedDebugSubsection(DebugSubsectionKind k, SectionChunk *debugChunk, 851 ArrayRef<uint8_t> subsec, uint32_t relocIndex) 852 : DebugSubsection(k), debugChunk(debugChunk), subsec(subsec), 853 relocIndex(relocIndex) {} 854 855 Error commit(BinaryStreamWriter &writer) const override; 856 uint32_t calculateSerializedSize() const override { return subsec.size(); } 857 858 SectionChunk *debugChunk; 859 ArrayRef<uint8_t> subsec; 860 uint32_t relocIndex; 861 }; 862 } // namespace 863 864 Error UnrelocatedDebugSubsection::commit(BinaryStreamWriter &writer) const { 865 std::vector<uint8_t> relocatedBytes(subsec.size()); 866 uint32_t tmpRelocIndex = relocIndex; 867 debugChunk->writeAndRelocateSubsection(debugChunk->getContents(), subsec, 868 tmpRelocIndex, relocatedBytes.data()); 869 870 // Remap type indices in inlinee line records in place. Skip the remapping if 871 // there is no type source info. 872 if (kind() == DebugSubsectionKind::InlineeLines && 873 debugChunk->file->debugTypesObj) { 874 TpiSource *source = debugChunk->file->debugTypesObj; 875 DebugInlineeLinesSubsectionRef inlineeLines; 876 BinaryStreamReader storageReader(relocatedBytes, llvm::endianness::little); 877 ExitOnError exitOnErr; 878 exitOnErr(inlineeLines.initialize(storageReader)); 879 for (const InlineeSourceLine &line : inlineeLines) { 880 TypeIndex &inlinee = *const_cast<TypeIndex *>(&line.Header->Inlinee); 881 if (!source->remapTypeIndex(inlinee, TiRefKind::IndexRef)) { 882 log("bad inlinee line record in " + debugChunk->file->getName() + 883 " with bad inlinee index 0x" + utohexstr(inlinee.getIndex())); 884 } 885 } 886 } 887 888 return writer.writeBytes(relocatedBytes); 889 } 890 891 void DebugSHandler::addUnrelocatedSubsection(SectionChunk *debugChunk, 892 const DebugSubsectionRecord &ss) { 893 ArrayRef<uint8_t> subsec; 894 BinaryStreamRef sr = ss.getRecordData(); 895 cantFail(sr.readBytes(0, sr.getLength(), subsec)); 896 advanceRelocIndex(debugChunk, subsec); 897 file.moduleDBI->addDebugSubsection( 898 std::make_shared<UnrelocatedDebugSubsection>(ss.kind(), debugChunk, 899 subsec, nextRelocIndex)); 900 } 901 902 void DebugSHandler::addFrameDataSubsection(SectionChunk *debugChunk, 903 const DebugSubsectionRecord &ss) { 904 // We need to re-write string table indices here, so save off all 905 // frame data subsections until we've processed the entire list of 906 // subsections so that we can be sure we have the string table. 907 ArrayRef<uint8_t> subsec; 908 BinaryStreamRef sr = ss.getRecordData(); 909 cantFail(sr.readBytes(0, sr.getLength(), subsec)); 910 advanceRelocIndex(debugChunk, subsec); 911 frameDataSubsecs.push_back({debugChunk, subsec, nextRelocIndex}); 912 } 913 914 static Expected<StringRef> 915 getFileName(const DebugStringTableSubsectionRef &strings, 916 const DebugChecksumsSubsectionRef &checksums, uint32_t fileID) { 917 auto iter = checksums.getArray().at(fileID); 918 if (iter == checksums.getArray().end()) 919 return make_error<CodeViewError>(cv_error_code::no_records); 920 uint32_t offset = iter->FileNameOffset; 921 return strings.getString(offset); 922 } 923 924 void DebugSHandler::finish() { 925 pdb::DbiStreamBuilder &dbiBuilder = linker.builder.getDbiBuilder(); 926 927 // If we found any symbol records for the module symbol stream, defer them. 928 if (moduleStreamSize > kSymbolStreamMagicSize) 929 file.moduleDBI->addUnmergedSymbols(&file, moduleStreamSize - 930 kSymbolStreamMagicSize); 931 932 // We should have seen all debug subsections across the entire object file now 933 // which means that if a StringTable subsection and Checksums subsection were 934 // present, now is the time to handle them. 935 if (!cvStrTab.valid()) { 936 if (checksums.valid()) 937 fatal(".debug$S sections with a checksums subsection must also contain a " 938 "string table subsection"); 939 940 if (!stringTableFixups.empty()) 941 Warn(ctx) 942 << "No StringTable subsection was encountered, but there are string " 943 "table references"; 944 return; 945 } 946 947 ExitOnError exitOnErr; 948 949 // Handle FPO data. Each subsection begins with a single image base 950 // relocation, which is then added to the RvaStart of each frame data record 951 // when it is added to the PDB. The string table indices for the FPO program 952 // must also be rewritten to use the PDB string table. 953 for (const UnrelocatedFpoData &subsec : frameDataSubsecs) { 954 // Relocate the first four bytes of the subection and reinterpret them as a 955 // 32 bit little-endian integer. 956 SectionChunk *debugChunk = subsec.debugChunk; 957 ArrayRef<uint8_t> subsecData = subsec.subsecData; 958 uint32_t relocIndex = subsec.relocIndex; 959 auto unrelocatedRvaStart = subsecData.take_front(sizeof(uint32_t)); 960 uint8_t relocatedRvaStart[sizeof(uint32_t)]; 961 debugChunk->writeAndRelocateSubsection(debugChunk->getContents(), 962 unrelocatedRvaStart, relocIndex, 963 &relocatedRvaStart[0]); 964 // Use of memcpy here avoids violating type-based aliasing rules. 965 support::ulittle32_t rvaStart; 966 memcpy(&rvaStart, &relocatedRvaStart[0], sizeof(support::ulittle32_t)); 967 968 // Copy each frame data record, add in rvaStart, translate string table 969 // indices, and add the record to the PDB. 970 DebugFrameDataSubsectionRef fds; 971 BinaryStreamReader reader(subsecData, llvm::endianness::little); 972 exitOnErr(fds.initialize(reader)); 973 for (codeview::FrameData fd : fds) { 974 fd.RvaStart += rvaStart; 975 fd.FrameFunc = translateStringTableIndex(ctx, fd.FrameFunc, cvStrTab, 976 linker.pdbStrTab); 977 dbiBuilder.addNewFpoData(fd); 978 } 979 } 980 981 // Translate the fixups and pass them off to the module builder so they will 982 // be applied during writing. 983 for (StringTableFixup &ref : stringTableFixups) { 984 ref.StrTabOffset = translateStringTableIndex(ctx, ref.StrTabOffset, 985 cvStrTab, linker.pdbStrTab); 986 } 987 file.moduleDBI->setStringTableFixups(std::move(stringTableFixups)); 988 989 // Make a new file checksum table that refers to offsets in the PDB-wide 990 // string table. Generally the string table subsection appears after the 991 // checksum table, so we have to do this after looping over all the 992 // subsections. The new checksum table must have the exact same layout and 993 // size as the original. Otherwise, the file references in the line and 994 // inlinee line tables will be incorrect. 995 auto newChecksums = std::make_unique<DebugChecksumsSubsection>(linker.pdbStrTab); 996 for (const FileChecksumEntry &fc : checksums) { 997 SmallString<128> filename = 998 exitOnErr(cvStrTab.getString(fc.FileNameOffset)); 999 linker.pdbMakeAbsolute(filename); 1000 exitOnErr(dbiBuilder.addModuleSourceFile(*file.moduleDBI, filename)); 1001 newChecksums->addChecksum(filename, fc.Kind, fc.Checksum); 1002 } 1003 assert(checksums.getArray().getUnderlyingStream().getLength() == 1004 newChecksums->calculateSerializedSize() && 1005 "file checksum table must have same layout"); 1006 1007 file.moduleDBI->addDebugSubsection(std::move(newChecksums)); 1008 } 1009 1010 static void warnUnusable(InputFile *f, Error e, bool shouldWarn) { 1011 if (!shouldWarn) { 1012 consumeError(std::move(e)); 1013 return; 1014 } 1015 auto diag = Warn(f->symtab.ctx); 1016 diag << "Cannot use debug info for '" << f << "' [LNK4099]"; 1017 if (e) 1018 diag << "\n>>> failed to load reference " << std::move(e); 1019 } 1020 1021 // Allocate memory for a .debug$S / .debug$F section and relocate it. 1022 static ArrayRef<uint8_t> relocateDebugChunk(SectionChunk &debugChunk) { 1023 uint8_t *buffer = bAlloc().Allocate<uint8_t>(debugChunk.getSize()); 1024 assert(debugChunk.getOutputSectionIdx() == 0 && 1025 "debug sections should not be in output sections"); 1026 debugChunk.writeTo(buffer); 1027 return ArrayRef(buffer, debugChunk.getSize()); 1028 } 1029 1030 void PDBLinker::addDebugSymbols(TpiSource *source) { 1031 // If this TpiSource doesn't have an object file, it must be from a type 1032 // server PDB. Type server PDBs do not contain symbols, so stop here. 1033 if (!source->file) 1034 return; 1035 1036 llvm::TimeTraceScope timeScope("Merge symbols"); 1037 ScopedTimer t(ctx.symbolMergingTimer); 1038 ExitOnError exitOnErr; 1039 pdb::DbiStreamBuilder &dbiBuilder = builder.getDbiBuilder(); 1040 DebugSHandler dsh(ctx, *this, *source->file); 1041 // Now do all live .debug$S and .debug$F sections. 1042 for (SectionChunk *debugChunk : source->file->getDebugChunks()) { 1043 if (!debugChunk->live || debugChunk->getSize() == 0) 1044 continue; 1045 1046 bool isDebugS = debugChunk->getSectionName() == ".debug$S"; 1047 bool isDebugF = debugChunk->getSectionName() == ".debug$F"; 1048 if (!isDebugS && !isDebugF) 1049 continue; 1050 1051 if (isDebugS) { 1052 dsh.handleDebugS(debugChunk); 1053 } else if (isDebugF) { 1054 // Handle old FPO data .debug$F sections. These are relatively rare. 1055 ArrayRef<uint8_t> relocatedDebugContents = 1056 relocateDebugChunk(*debugChunk); 1057 FixedStreamArray<object::FpoData> fpoRecords; 1058 BinaryStreamReader reader(relocatedDebugContents, 1059 llvm::endianness::little); 1060 uint32_t count = relocatedDebugContents.size() / sizeof(object::FpoData); 1061 exitOnErr(reader.readArray(fpoRecords, count)); 1062 1063 // These are already relocated and don't refer to the string table, so we 1064 // can just copy it. 1065 for (const object::FpoData &fd : fpoRecords) 1066 dbiBuilder.addOldFpoData(fd); 1067 } 1068 } 1069 1070 // Do any post-processing now that all .debug$S sections have been processed. 1071 dsh.finish(); 1072 } 1073 1074 // Add a module descriptor for every object file. We need to put an absolute 1075 // path to the object into the PDB. If this is a plain object, we make its 1076 // path absolute. If it's an object in an archive, we make the archive path 1077 // absolute. 1078 void PDBLinker::createModuleDBI(ObjFile *file) { 1079 pdb::DbiStreamBuilder &dbiBuilder = builder.getDbiBuilder(); 1080 SmallString<128> objName; 1081 ExitOnError exitOnErr; 1082 1083 bool inArchive = !file->parentName.empty(); 1084 objName = inArchive ? file->parentName : file->getName(); 1085 pdbMakeAbsolute(objName); 1086 StringRef modName = inArchive ? file->getName() : objName.str(); 1087 1088 file->moduleDBI = &exitOnErr(dbiBuilder.addModuleInfo(modName)); 1089 file->moduleDBI->setObjFileName(objName); 1090 file->moduleDBI->setMergeSymbolsCallback(this, &commitSymbolsForObject); 1091 1092 ArrayRef<Chunk *> chunks = file->getChunks(); 1093 uint32_t modi = file->moduleDBI->getModuleIndex(); 1094 1095 for (Chunk *c : chunks) { 1096 auto *secChunk = dyn_cast<SectionChunk>(c); 1097 if (!secChunk || !secChunk->live) 1098 continue; 1099 pdb::SectionContrib sc = createSectionContrib(ctx, secChunk, modi); 1100 file->moduleDBI->setFirstSectionContrib(sc); 1101 break; 1102 } 1103 } 1104 1105 void PDBLinker::addDebug(TpiSource *source) { 1106 // Before we can process symbol substreams from .debug$S, we need to process 1107 // type information, file checksums, and the string table. Add type info to 1108 // the PDB first, so that we can get the map from object file type and item 1109 // indices to PDB type and item indices. If we are using ghashes, types have 1110 // already been merged. 1111 if (!ctx.config.debugGHashes) { 1112 llvm::TimeTraceScope timeScope("Merge types (Non-GHASH)"); 1113 ScopedTimer t(ctx.typeMergingTimer); 1114 if (Error e = source->mergeDebugT(&tMerger)) { 1115 // If type merging failed, ignore the symbols. 1116 warnUnusable(source->file, std::move(e), 1117 ctx.config.warnDebugInfoUnusable); 1118 return; 1119 } 1120 } 1121 1122 // If type merging failed, ignore the symbols. 1123 Error typeError = std::move(source->typeMergingError); 1124 if (typeError) { 1125 warnUnusable(source->file, std::move(typeError), 1126 ctx.config.warnDebugInfoUnusable); 1127 return; 1128 } 1129 1130 addDebugSymbols(source); 1131 } 1132 1133 static pdb::BulkPublic createPublic(COFFLinkerContext &ctx, Defined *def) { 1134 pdb::BulkPublic pub; 1135 pub.Name = def->getName().data(); 1136 pub.NameLen = def->getName().size(); 1137 1138 PublicSymFlags flags = PublicSymFlags::None; 1139 if (auto *d = dyn_cast<DefinedCOFF>(def)) { 1140 if (d->getCOFFSymbol().isFunctionDefinition()) 1141 flags = PublicSymFlags::Function; 1142 } else if (isa<DefinedImportThunk>(def)) { 1143 flags = PublicSymFlags::Function; 1144 } 1145 pub.setFlags(flags); 1146 1147 OutputSection *os = ctx.getOutputSection(def->getChunk()); 1148 assert(os && "all publics should be in final image"); 1149 pub.Offset = def->getRVA() - os->getRVA(); 1150 pub.Segment = os->sectionIndex; 1151 return pub; 1152 } 1153 1154 // Add all object files to the PDB. Merge .debug$T sections into IpiData and 1155 // TpiData. 1156 void PDBLinker::addObjectsToPDB() { 1157 { 1158 llvm::TimeTraceScope timeScope("Add objects to PDB"); 1159 ScopedTimer t1(ctx.addObjectsTimer); 1160 1161 // Create module descriptors 1162 for (ObjFile *obj : ctx.objFileInstances) 1163 createModuleDBI(obj); 1164 1165 // Reorder dependency type sources to come first. 1166 tMerger.sortDependencies(); 1167 1168 // Merge type information from input files using global type hashing. 1169 if (ctx.config.debugGHashes) 1170 tMerger.mergeTypesWithGHash(); 1171 1172 // Merge dependencies and then regular objects. 1173 { 1174 llvm::TimeTraceScope timeScope("Merge debug info (dependencies)"); 1175 for (TpiSource *source : tMerger.dependencySources) 1176 addDebug(source); 1177 } 1178 { 1179 llvm::TimeTraceScope timeScope("Merge debug info (objects)"); 1180 for (TpiSource *source : tMerger.objectSources) 1181 addDebug(source); 1182 } 1183 1184 builder.getStringTableBuilder().setStrings(pdbStrTab); 1185 } 1186 1187 // Construct TPI and IPI stream contents. 1188 { 1189 llvm::TimeTraceScope timeScope("TPI/IPI stream layout"); 1190 ScopedTimer t2(ctx.tpiStreamLayoutTimer); 1191 1192 // Collect all the merged types. 1193 if (ctx.config.debugGHashes) { 1194 addGHashTypeInfo(ctx, builder); 1195 } else { 1196 addTypeInfo(builder.getTpiBuilder(), tMerger.getTypeTable()); 1197 addTypeInfo(builder.getIpiBuilder(), tMerger.getIDTable()); 1198 } 1199 } 1200 1201 if (ctx.config.showSummary) { 1202 for (TpiSource *source : ctx.tpiSourceList) { 1203 nbTypeRecords += source->nbTypeRecords; 1204 nbTypeRecordsBytes += source->nbTypeRecordsBytes; 1205 } 1206 } 1207 } 1208 1209 void PDBLinker::addPublicsToPDB() { 1210 llvm::TimeTraceScope timeScope("Publics layout"); 1211 ScopedTimer t3(ctx.publicsLayoutTimer); 1212 // Compute the public symbols. 1213 auto &gsiBuilder = builder.getGsiBuilder(); 1214 std::vector<pdb::BulkPublic> publics; 1215 ctx.symtab.forEachSymbol([&publics, this](Symbol *s) { 1216 // Only emit external, defined, live symbols that have a chunk. Static, 1217 // non-external symbols do not appear in the symbol table. 1218 auto *def = dyn_cast<Defined>(s); 1219 if (def && def->isLive() && def->getChunk()) { 1220 // Don't emit a public symbol for coverage data symbols. LLVM code 1221 // coverage (and PGO) create a __profd_ and __profc_ symbol for every 1222 // function. C++ mangled names are long, and tend to dominate symbol size. 1223 // Including these names triples the size of the public stream, which 1224 // results in bloated PDB files. These symbols generally are not helpful 1225 // for debugging, so suppress them. 1226 StringRef name = def->getName(); 1227 if (name.data()[0] == '_' && name.data()[1] == '_') { 1228 // Drop the '_' prefix for x86. 1229 if (ctx.config.machine == I386) 1230 name = name.drop_front(1); 1231 if (name.starts_with("__profd_") || name.starts_with("__profc_") || 1232 name.starts_with("__covrec_")) { 1233 return; 1234 } 1235 } 1236 publics.push_back(createPublic(ctx, def)); 1237 } 1238 }); 1239 1240 if (!publics.empty()) { 1241 publicSymbols = publics.size(); 1242 gsiBuilder.addPublicSymbols(std::move(publics)); 1243 } 1244 } 1245 1246 void PDBLinker::printStats() { 1247 if (!ctx.config.showSummary) 1248 return; 1249 1250 SmallString<256> buffer; 1251 raw_svector_ostream stream(buffer); 1252 1253 stream << center_justify("Summary", 80) << '\n' 1254 << std::string(80, '-') << '\n'; 1255 1256 auto print = [&](uint64_t v, StringRef s) { 1257 stream << format_decimal(v, 15) << " " << s << '\n'; 1258 }; 1259 1260 print(ctx.objFileInstances.size(), 1261 "Input OBJ files (expanded from all cmd-line inputs)"); 1262 print(ctx.typeServerSourceMappings.size(), "PDB type server dependencies"); 1263 print(ctx.precompSourceMappings.size(), "Precomp OBJ dependencies"); 1264 print(nbTypeRecords, "Input type records"); 1265 print(nbTypeRecordsBytes, "Input type records bytes"); 1266 print(builder.getTpiBuilder().getRecordCount(), "Merged TPI records"); 1267 print(builder.getIpiBuilder().getRecordCount(), "Merged IPI records"); 1268 print(pdbStrTab.size(), "Output PDB strings"); 1269 print(globalSymbols, "Global symbol records"); 1270 print(moduleSymbols, "Module symbol records"); 1271 print(publicSymbols, "Public symbol records"); 1272 1273 auto printLargeInputTypeRecs = [&](StringRef name, 1274 ArrayRef<uint32_t> recCounts, 1275 TypeCollection &records) { 1276 // Figure out which type indices were responsible for the most duplicate 1277 // bytes in the input files. These should be frequently emitted LF_CLASS and 1278 // LF_FIELDLIST records. 1279 struct TypeSizeInfo { 1280 uint32_t typeSize; 1281 uint32_t dupCount; 1282 TypeIndex typeIndex; 1283 uint64_t totalInputSize() const { return uint64_t(dupCount) * typeSize; } 1284 bool operator<(const TypeSizeInfo &rhs) const { 1285 if (totalInputSize() == rhs.totalInputSize()) 1286 return typeIndex < rhs.typeIndex; 1287 return totalInputSize() < rhs.totalInputSize(); 1288 } 1289 }; 1290 SmallVector<TypeSizeInfo, 0> tsis; 1291 for (auto e : enumerate(recCounts)) { 1292 TypeIndex typeIndex = TypeIndex::fromArrayIndex(e.index()); 1293 uint32_t typeSize = records.getType(typeIndex).length(); 1294 uint32_t dupCount = e.value(); 1295 tsis.push_back({typeSize, dupCount, typeIndex}); 1296 } 1297 1298 if (!tsis.empty()) { 1299 stream << "\nTop 10 types responsible for the most " << name 1300 << " input:\n"; 1301 stream << " index total bytes count size\n"; 1302 llvm::sort(tsis); 1303 unsigned i = 0; 1304 for (const auto &tsi : reverse(tsis)) { 1305 stream << formatv(" {0,10:X}: {1,14:N} = {2,5:N} * {3,6:N}\n", 1306 tsi.typeIndex.getIndex(), tsi.totalInputSize(), 1307 tsi.dupCount, tsi.typeSize); 1308 if (++i >= 10) 1309 break; 1310 } 1311 stream 1312 << "Run llvm-pdbutil to print details about a particular record:\n"; 1313 stream << formatv("llvm-pdbutil dump -{0}s -{0}-index {1:X} {2}\n", 1314 (name == "TPI" ? "type" : "id"), 1315 tsis.back().typeIndex.getIndex(), ctx.config.pdbPath); 1316 } 1317 }; 1318 1319 if (!ctx.config.debugGHashes) { 1320 // FIXME: Reimplement for ghash. 1321 printLargeInputTypeRecs("TPI", tMerger.tpiCounts, tMerger.getTypeTable()); 1322 printLargeInputTypeRecs("IPI", tMerger.ipiCounts, tMerger.getIDTable()); 1323 } 1324 1325 Msg(ctx) << buffer; 1326 } 1327 1328 void PDBLinker::addNatvisFiles() { 1329 llvm::TimeTraceScope timeScope("Natvis files"); 1330 for (StringRef file : ctx.config.natvisFiles) { 1331 ErrorOr<std::unique_ptr<MemoryBuffer>> dataOrErr = 1332 MemoryBuffer::getFile(file); 1333 if (!dataOrErr) { 1334 Warn(ctx) << "Cannot open input file: " << file; 1335 continue; 1336 } 1337 std::unique_ptr<MemoryBuffer> data = std::move(*dataOrErr); 1338 1339 // Can't use takeBuffer() here since addInjectedSource() takes ownership. 1340 if (ctx.driver.tar) 1341 ctx.driver.tar->append(relativeToRoot(data->getBufferIdentifier()), 1342 data->getBuffer()); 1343 1344 builder.addInjectedSource(file, std::move(data)); 1345 } 1346 } 1347 1348 void PDBLinker::addNamedStreams() { 1349 llvm::TimeTraceScope timeScope("Named streams"); 1350 ExitOnError exitOnErr; 1351 for (const auto &streamFile : ctx.config.namedStreams) { 1352 const StringRef stream = streamFile.getKey(), file = streamFile.getValue(); 1353 ErrorOr<std::unique_ptr<MemoryBuffer>> dataOrErr = 1354 MemoryBuffer::getFile(file); 1355 if (!dataOrErr) { 1356 Warn(ctx) << "Cannot open input file: " << file; 1357 continue; 1358 } 1359 std::unique_ptr<MemoryBuffer> data = std::move(*dataOrErr); 1360 exitOnErr(builder.addNamedStream(stream, data->getBuffer())); 1361 ctx.driver.takeBuffer(std::move(data)); 1362 } 1363 } 1364 1365 static codeview::CPUType toCodeViewMachine(COFF::MachineTypes machine) { 1366 switch (machine) { 1367 case COFF::IMAGE_FILE_MACHINE_AMD64: 1368 return codeview::CPUType::X64; 1369 case COFF::IMAGE_FILE_MACHINE_ARM: 1370 return codeview::CPUType::ARM7; 1371 case COFF::IMAGE_FILE_MACHINE_ARM64: 1372 return codeview::CPUType::ARM64; 1373 case COFF::IMAGE_FILE_MACHINE_ARM64EC: 1374 return codeview::CPUType::ARM64EC; 1375 case COFF::IMAGE_FILE_MACHINE_ARM64X: 1376 return codeview::CPUType::ARM64X; 1377 case COFF::IMAGE_FILE_MACHINE_ARMNT: 1378 return codeview::CPUType::ARMNT; 1379 case COFF::IMAGE_FILE_MACHINE_I386: 1380 return codeview::CPUType::Intel80386; 1381 default: 1382 llvm_unreachable("Unsupported CPU Type"); 1383 } 1384 } 1385 1386 // Mimic MSVC which surrounds arguments containing whitespace with quotes. 1387 // Double double-quotes are handled, so that the resulting string can be 1388 // executed again on the cmd-line. 1389 static std::string quote(ArrayRef<StringRef> args) { 1390 std::string r; 1391 r.reserve(256); 1392 for (StringRef a : args) { 1393 if (!r.empty()) 1394 r.push_back(' '); 1395 bool hasWS = a.contains(' '); 1396 bool hasQ = a.contains('"'); 1397 if (hasWS || hasQ) 1398 r.push_back('"'); 1399 if (hasQ) { 1400 SmallVector<StringRef, 4> s; 1401 a.split(s, '"'); 1402 r.append(join(s, "\"\"")); 1403 } else { 1404 r.append(std::string(a)); 1405 } 1406 if (hasWS || hasQ) 1407 r.push_back('"'); 1408 } 1409 return r; 1410 } 1411 1412 static void fillLinkerVerRecord(Compile3Sym &cs, MachineTypes machine) { 1413 cs.Machine = toCodeViewMachine(machine); 1414 // Interestingly, if we set the string to 0.0.0.0, then when trying to view 1415 // local variables WinDbg emits an error that private symbols are not present. 1416 // By setting this to a valid MSVC linker version string, local variables are 1417 // displayed properly. As such, even though it is not representative of 1418 // LLVM's version information, we need this for compatibility. 1419 cs.Flags = CompileSym3Flags::None; 1420 cs.VersionBackendBuild = 25019; 1421 cs.VersionBackendMajor = 14; 1422 cs.VersionBackendMinor = 10; 1423 cs.VersionBackendQFE = 0; 1424 1425 // MSVC also sets the frontend to 0.0.0.0 since this is specifically for the 1426 // linker module (which is by definition a backend), so we don't need to do 1427 // anything here. Also, it seems we can use "LLVM Linker" for the linker name 1428 // without any problems. Only the backend version has to be hardcoded to a 1429 // magic number. 1430 cs.VersionFrontendBuild = 0; 1431 cs.VersionFrontendMajor = 0; 1432 cs.VersionFrontendMinor = 0; 1433 cs.VersionFrontendQFE = 0; 1434 cs.Version = "LLVM Linker"; 1435 cs.setLanguage(SourceLanguage::Link); 1436 } 1437 1438 void PDBLinker::addCommonLinkerModuleSymbols( 1439 StringRef path, pdb::DbiModuleDescriptorBuilder &mod) { 1440 ObjNameSym ons(SymbolRecordKind::ObjNameSym); 1441 EnvBlockSym ebs(SymbolRecordKind::EnvBlockSym); 1442 Compile3Sym cs(SymbolRecordKind::Compile3Sym); 1443 1444 MachineTypes machine = ctx.config.machine; 1445 // MSVC uses the ARM64X machine type for ARM64EC targets in the common linker 1446 // module record. 1447 if (isArm64EC(machine)) 1448 machine = ARM64X; 1449 fillLinkerVerRecord(cs, machine); 1450 1451 ons.Name = "* Linker *"; 1452 ons.Signature = 0; 1453 1454 ArrayRef<StringRef> args = ArrayRef(ctx.config.argv).drop_front(); 1455 std::string argStr = quote(args); 1456 ebs.Fields.push_back("cwd"); 1457 SmallString<64> cwd; 1458 if (ctx.config.pdbSourcePath.empty()) 1459 sys::fs::current_path(cwd); 1460 else 1461 cwd = ctx.config.pdbSourcePath; 1462 ebs.Fields.push_back(cwd); 1463 ebs.Fields.push_back("exe"); 1464 SmallString<64> exe = ctx.config.argv[0]; 1465 pdbMakeAbsolute(exe); 1466 ebs.Fields.push_back(exe); 1467 ebs.Fields.push_back("pdb"); 1468 ebs.Fields.push_back(path); 1469 ebs.Fields.push_back("cmd"); 1470 ebs.Fields.push_back(argStr); 1471 llvm::BumpPtrAllocator &bAlloc = lld::bAlloc(); 1472 mod.addSymbol(codeview::SymbolSerializer::writeOneSymbol( 1473 ons, bAlloc, CodeViewContainer::Pdb)); 1474 mod.addSymbol(codeview::SymbolSerializer::writeOneSymbol( 1475 cs, bAlloc, CodeViewContainer::Pdb)); 1476 mod.addSymbol(codeview::SymbolSerializer::writeOneSymbol( 1477 ebs, bAlloc, CodeViewContainer::Pdb)); 1478 } 1479 1480 static void addLinkerModuleCoffGroup(PartialSection *sec, 1481 pdb::DbiModuleDescriptorBuilder &mod, 1482 OutputSection &os) { 1483 // If there's a section, there's at least one chunk 1484 assert(!sec->chunks.empty()); 1485 const Chunk *firstChunk = *sec->chunks.begin(); 1486 const Chunk *lastChunk = *sec->chunks.rbegin(); 1487 1488 // Emit COFF group 1489 CoffGroupSym cgs(SymbolRecordKind::CoffGroupSym); 1490 cgs.Name = sec->name; 1491 cgs.Segment = os.sectionIndex; 1492 cgs.Offset = firstChunk->getRVA() - os.getRVA(); 1493 cgs.Size = lastChunk->getRVA() + lastChunk->getSize() - firstChunk->getRVA(); 1494 cgs.Characteristics = sec->characteristics; 1495 1496 // Somehow .idata sections & sections groups in the debug symbol stream have 1497 // the "write" flag set. However the section header for the corresponding 1498 // .idata section doesn't have it. 1499 if (cgs.Name.starts_with(".idata")) 1500 cgs.Characteristics |= llvm::COFF::IMAGE_SCN_MEM_WRITE; 1501 1502 mod.addSymbol(codeview::SymbolSerializer::writeOneSymbol( 1503 cgs, bAlloc(), CodeViewContainer::Pdb)); 1504 } 1505 1506 static void addLinkerModuleSectionSymbol(pdb::DbiModuleDescriptorBuilder &mod, 1507 OutputSection &os, bool isMinGW) { 1508 SectionSym sym(SymbolRecordKind::SectionSym); 1509 sym.Alignment = 12; // 2^12 = 4KB 1510 sym.Characteristics = os.header.Characteristics; 1511 sym.Length = os.getVirtualSize(); 1512 sym.Name = os.name; 1513 sym.Rva = os.getRVA(); 1514 sym.SectionNumber = os.sectionIndex; 1515 mod.addSymbol(codeview::SymbolSerializer::writeOneSymbol( 1516 sym, bAlloc(), CodeViewContainer::Pdb)); 1517 1518 // Skip COFF groups in MinGW because it adds a significant footprint to the 1519 // PDB, due to each function being in its own section 1520 if (isMinGW) 1521 return; 1522 1523 // Output COFF groups for individual chunks of this section. 1524 for (PartialSection *sec : os.contribSections) { 1525 addLinkerModuleCoffGroup(sec, mod, os); 1526 } 1527 } 1528 1529 // Add all import files as modules to the PDB. 1530 void PDBLinker::addImportFilesToPDB() { 1531 if (ctx.importFileInstances.empty()) 1532 return; 1533 1534 llvm::TimeTraceScope timeScope("Import files"); 1535 ExitOnError exitOnErr; 1536 std::map<std::string, llvm::pdb::DbiModuleDescriptorBuilder *> dllToModuleDbi; 1537 1538 for (ImportFile *file : ctx.importFileInstances) { 1539 if (!file->live) 1540 continue; 1541 1542 if (!file->thunkSym) 1543 continue; 1544 1545 if (!file->thunkSym->isLive()) 1546 continue; 1547 1548 std::string dll = StringRef(file->dllName).lower(); 1549 llvm::pdb::DbiModuleDescriptorBuilder *&mod = dllToModuleDbi[dll]; 1550 if (!mod) { 1551 pdb::DbiStreamBuilder &dbiBuilder = builder.getDbiBuilder(); 1552 SmallString<128> libPath = file->parentName; 1553 pdbMakeAbsolute(libPath); 1554 sys::path::native(libPath); 1555 1556 // Name modules similar to MSVC's link.exe. 1557 // The first module is the simple dll filename 1558 llvm::pdb::DbiModuleDescriptorBuilder &firstMod = 1559 exitOnErr(dbiBuilder.addModuleInfo(file->dllName)); 1560 firstMod.setObjFileName(libPath); 1561 pdb::SectionContrib sc = 1562 createSectionContrib(ctx, nullptr, llvm::pdb::kInvalidStreamIndex); 1563 firstMod.setFirstSectionContrib(sc); 1564 1565 // The second module is where the import stream goes. 1566 mod = &exitOnErr(dbiBuilder.addModuleInfo("Import:" + file->dllName)); 1567 mod->setObjFileName(libPath); 1568 } 1569 1570 DefinedImportThunk *thunk = cast<DefinedImportThunk>(file->thunkSym); 1571 Chunk *thunkChunk = thunk->getChunk(); 1572 OutputSection *thunkOS = ctx.getOutputSection(thunkChunk); 1573 1574 ObjNameSym ons(SymbolRecordKind::ObjNameSym); 1575 Compile3Sym cs(SymbolRecordKind::Compile3Sym); 1576 Thunk32Sym ts(SymbolRecordKind::Thunk32Sym); 1577 ScopeEndSym es(SymbolRecordKind::ScopeEndSym); 1578 1579 ons.Name = file->dllName; 1580 ons.Signature = 0; 1581 1582 fillLinkerVerRecord(cs, ctx.config.machine); 1583 1584 ts.Name = thunk->getName(); 1585 ts.Parent = 0; 1586 ts.End = 0; 1587 ts.Next = 0; 1588 ts.Thunk = ThunkOrdinal::Standard; 1589 ts.Length = thunkChunk->getSize(); 1590 ts.Segment = thunkOS->sectionIndex; 1591 ts.Offset = thunkChunk->getRVA() - thunkOS->getRVA(); 1592 1593 llvm::BumpPtrAllocator &bAlloc = lld::bAlloc(); 1594 mod->addSymbol(codeview::SymbolSerializer::writeOneSymbol( 1595 ons, bAlloc, CodeViewContainer::Pdb)); 1596 mod->addSymbol(codeview::SymbolSerializer::writeOneSymbol( 1597 cs, bAlloc, CodeViewContainer::Pdb)); 1598 1599 CVSymbol newSym = codeview::SymbolSerializer::writeOneSymbol( 1600 ts, bAlloc, CodeViewContainer::Pdb); 1601 1602 // Write ptrEnd for the S_THUNK32. 1603 ScopeRecord *thunkSymScope = 1604 getSymbolScopeFields(const_cast<uint8_t *>(newSym.data().data())); 1605 1606 mod->addSymbol(newSym); 1607 1608 newSym = codeview::SymbolSerializer::writeOneSymbol(es, bAlloc, 1609 CodeViewContainer::Pdb); 1610 thunkSymScope->ptrEnd = mod->getNextSymbolOffset(); 1611 1612 mod->addSymbol(newSym); 1613 1614 pdb::SectionContrib sc = 1615 createSectionContrib(ctx, thunk->getChunk(), mod->getModuleIndex()); 1616 mod->setFirstSectionContrib(sc); 1617 } 1618 } 1619 1620 // Creates a PDB file. 1621 void lld::coff::createPDB(COFFLinkerContext &ctx, 1622 ArrayRef<uint8_t> sectionTable, 1623 llvm::codeview::DebugInfo *buildId) { 1624 llvm::TimeTraceScope timeScope("PDB file"); 1625 ScopedTimer t1(ctx.totalPdbLinkTimer); 1626 { 1627 PDBLinker pdb(ctx); 1628 1629 pdb.initialize(buildId); 1630 pdb.addObjectsToPDB(); 1631 pdb.addImportFilesToPDB(); 1632 pdb.addSections(sectionTable); 1633 pdb.addNatvisFiles(); 1634 pdb.addNamedStreams(); 1635 pdb.addPublicsToPDB(); 1636 1637 { 1638 llvm::TimeTraceScope timeScope("Commit PDB file to disk"); 1639 ScopedTimer t2(ctx.diskCommitTimer); 1640 codeview::GUID guid; 1641 pdb.commit(&guid); 1642 memcpy(&buildId->PDB70.Signature, &guid, 16); 1643 } 1644 1645 t1.stop(); 1646 pdb.printStats(); 1647 1648 // Manually start this profile point to measure ~PDBLinker(). 1649 if (getTimeTraceProfilerInstance() != nullptr) 1650 timeTraceProfilerBegin("PDBLinker destructor", StringRef("")); 1651 } 1652 // Manually end this profile point to measure ~PDBLinker(). 1653 if (getTimeTraceProfilerInstance() != nullptr) 1654 timeTraceProfilerEnd(); 1655 } 1656 1657 void PDBLinker::initialize(llvm::codeview::DebugInfo *buildId) { 1658 ExitOnError exitOnErr; 1659 exitOnErr(builder.initialize(ctx.config.pdbPageSize)); 1660 1661 buildId->Signature.CVSignature = OMF::Signature::PDB70; 1662 // Signature is set to a hash of the PDB contents when the PDB is done. 1663 memset(buildId->PDB70.Signature, 0, 16); 1664 buildId->PDB70.Age = 1; 1665 1666 // Create streams in MSF for predefined streams, namely 1667 // PDB, TPI, DBI and IPI. 1668 for (int i = 0; i < (int)pdb::kSpecialStreamCount; ++i) 1669 exitOnErr(builder.getMsfBuilder().addStream(0)); 1670 1671 // Add an Info stream. 1672 auto &infoBuilder = builder.getInfoBuilder(); 1673 infoBuilder.setVersion(pdb::PdbRaw_ImplVer::PdbImplVC70); 1674 infoBuilder.setHashPDBContentsToGUID(true); 1675 1676 // Add an empty DBI stream. 1677 pdb::DbiStreamBuilder &dbiBuilder = builder.getDbiBuilder(); 1678 dbiBuilder.setAge(buildId->PDB70.Age); 1679 dbiBuilder.setVersionHeader(pdb::PdbDbiV70); 1680 dbiBuilder.setMachineType(ctx.config.machine); 1681 // Technically we are not link.exe 14.11, but there are known cases where 1682 // debugging tools on Windows expect Microsoft-specific version numbers or 1683 // they fail to work at all. Since we know we produce PDBs that are 1684 // compatible with LINK 14.11, we set that version number here. 1685 dbiBuilder.setBuildNumber(14, 11); 1686 } 1687 1688 void PDBLinker::addSections(ArrayRef<uint8_t> sectionTable) { 1689 llvm::TimeTraceScope timeScope("PDB output sections"); 1690 ExitOnError exitOnErr; 1691 // It's not entirely clear what this is, but the * Linker * module uses it. 1692 pdb::DbiStreamBuilder &dbiBuilder = builder.getDbiBuilder(); 1693 nativePath = ctx.config.pdbPath; 1694 pdbMakeAbsolute(nativePath); 1695 uint32_t pdbFilePathNI = dbiBuilder.addECName(nativePath); 1696 auto &linkerModule = exitOnErr(dbiBuilder.addModuleInfo("* Linker *")); 1697 linkerModule.setPdbFilePathNI(pdbFilePathNI); 1698 addCommonLinkerModuleSymbols(nativePath, linkerModule); 1699 1700 // Add section contributions. They must be ordered by ascending RVA. 1701 for (OutputSection *os : ctx.outputSections) { 1702 addLinkerModuleSectionSymbol(linkerModule, *os, ctx.config.mingw); 1703 for (Chunk *c : os->chunks) { 1704 pdb::SectionContrib sc = 1705 createSectionContrib(ctx, c, linkerModule.getModuleIndex()); 1706 builder.getDbiBuilder().addSectionContrib(sc); 1707 } 1708 } 1709 1710 // The * Linker * first section contrib is only used along with /INCREMENTAL, 1711 // to provide trampolines thunks for incremental function patching. Set this 1712 // as "unused" because LLD doesn't support /INCREMENTAL link. 1713 pdb::SectionContrib sc = 1714 createSectionContrib(ctx, nullptr, llvm::pdb::kInvalidStreamIndex); 1715 linkerModule.setFirstSectionContrib(sc); 1716 1717 // Add Section Map stream. 1718 ArrayRef<object::coff_section> sections = { 1719 (const object::coff_section *)sectionTable.data(), 1720 sectionTable.size() / sizeof(object::coff_section)}; 1721 dbiBuilder.createSectionMap(sections); 1722 1723 // Add COFF section header stream. 1724 exitOnErr( 1725 dbiBuilder.addDbgStream(pdb::DbgHeaderType::SectionHdr, sectionTable)); 1726 } 1727 1728 void PDBLinker::commit(codeview::GUID *guid) { 1729 // Print an error and continue if PDB writing fails. This is done mainly so 1730 // the user can see the output of /time and /summary, which is very helpful 1731 // when trying to figure out why a PDB file is too large. 1732 if (Error e = builder.commit(ctx.config.pdbPath, guid)) { 1733 e = handleErrors(std::move(e), [&](const llvm::msf::MSFError &me) { 1734 Err(ctx) << me.message(); 1735 if (me.isPageOverflow()) 1736 Err(ctx) << "try setting a larger /pdbpagesize"; 1737 }); 1738 checkError(std::move(e)); 1739 Err(ctx) << "failed to write PDB file " << Twine(ctx.config.pdbPath); 1740 } 1741 } 1742 1743 static uint32_t getSecrelReloc(Triple::ArchType arch) { 1744 switch (arch) { 1745 case Triple::x86_64: 1746 return COFF::IMAGE_REL_AMD64_SECREL; 1747 case Triple::x86: 1748 return COFF::IMAGE_REL_I386_SECREL; 1749 case Triple::thumb: 1750 return COFF::IMAGE_REL_ARM_SECREL; 1751 case Triple::aarch64: 1752 return COFF::IMAGE_REL_ARM64_SECREL; 1753 default: 1754 llvm_unreachable("unknown machine type"); 1755 } 1756 } 1757 1758 // Try to find a line table for the given offset Addr into the given chunk C. 1759 // If a line table was found, the line table, the string and checksum tables 1760 // that are used to interpret the line table, and the offset of Addr in the line 1761 // table are stored in the output arguments. Returns whether a line table was 1762 // found. 1763 static bool findLineTable(const SectionChunk *c, uint32_t addr, 1764 DebugStringTableSubsectionRef &cvStrTab, 1765 DebugChecksumsSubsectionRef &checksums, 1766 DebugLinesSubsectionRef &lines, 1767 uint32_t &offsetInLinetable) { 1768 ExitOnError exitOnErr; 1769 const uint32_t secrelReloc = getSecrelReloc(c->getArch()); 1770 1771 for (SectionChunk *dbgC : c->file->getDebugChunks()) { 1772 if (dbgC->getSectionName() != ".debug$S") 1773 continue; 1774 1775 // Build a mapping of SECREL relocations in dbgC that refer to `c`. 1776 DenseMap<uint32_t, uint32_t> secrels; 1777 for (const coff_relocation &r : dbgC->getRelocs()) { 1778 if (r.Type != secrelReloc) 1779 continue; 1780 1781 if (auto *s = dyn_cast_or_null<DefinedRegular>( 1782 c->file->getSymbols()[r.SymbolTableIndex])) 1783 if (s->getChunk() == c) 1784 secrels[r.VirtualAddress] = s->getValue(); 1785 } 1786 1787 ArrayRef<uint8_t> contents = 1788 SectionChunk::consumeDebugMagic(dbgC->getContents(), ".debug$S"); 1789 DebugSubsectionArray subsections; 1790 BinaryStreamReader reader(contents, llvm::endianness::little); 1791 exitOnErr(reader.readArray(subsections, contents.size())); 1792 1793 for (const DebugSubsectionRecord &ss : subsections) { 1794 switch (ss.kind()) { 1795 case DebugSubsectionKind::StringTable: { 1796 assert(!cvStrTab.valid() && 1797 "Encountered multiple string table subsections!"); 1798 exitOnErr(cvStrTab.initialize(ss.getRecordData())); 1799 break; 1800 } 1801 case DebugSubsectionKind::FileChecksums: 1802 assert(!checksums.valid() && 1803 "Encountered multiple checksum subsections!"); 1804 exitOnErr(checksums.initialize(ss.getRecordData())); 1805 break; 1806 case DebugSubsectionKind::Lines: { 1807 ArrayRef<uint8_t> bytes; 1808 auto ref = ss.getRecordData(); 1809 exitOnErr(ref.readLongestContiguousChunk(0, bytes)); 1810 size_t offsetInDbgC = bytes.data() - dbgC->getContents().data(); 1811 1812 // Check whether this line table refers to C. 1813 auto i = secrels.find(offsetInDbgC); 1814 if (i == secrels.end()) 1815 break; 1816 1817 // Check whether this line table covers Addr in C. 1818 DebugLinesSubsectionRef linesTmp; 1819 exitOnErr(linesTmp.initialize(BinaryStreamReader(ref))); 1820 uint32_t offsetInC = i->second + linesTmp.header()->RelocOffset; 1821 if (addr < offsetInC || addr >= offsetInC + linesTmp.header()->CodeSize) 1822 break; 1823 1824 assert(!lines.header() && 1825 "Encountered multiple line tables for function!"); 1826 exitOnErr(lines.initialize(BinaryStreamReader(ref))); 1827 offsetInLinetable = addr - offsetInC; 1828 break; 1829 } 1830 default: 1831 break; 1832 } 1833 1834 if (cvStrTab.valid() && checksums.valid() && lines.header()) 1835 return true; 1836 } 1837 } 1838 1839 return false; 1840 } 1841 1842 // Use CodeView line tables to resolve a file and line number for the given 1843 // offset into the given chunk and return them, or std::nullopt if a line table 1844 // was not found. 1845 std::optional<std::pair<StringRef, uint32_t>> 1846 lld::coff::getFileLineCodeView(const SectionChunk *c, uint32_t addr) { 1847 ExitOnError exitOnErr; 1848 1849 DebugStringTableSubsectionRef cvStrTab; 1850 DebugChecksumsSubsectionRef checksums; 1851 DebugLinesSubsectionRef lines; 1852 uint32_t offsetInLinetable; 1853 1854 if (!findLineTable(c, addr, cvStrTab, checksums, lines, offsetInLinetable)) 1855 return std::nullopt; 1856 1857 std::optional<uint32_t> nameIndex; 1858 std::optional<uint32_t> lineNumber; 1859 for (const LineColumnEntry &entry : lines) { 1860 for (const LineNumberEntry &ln : entry.LineNumbers) { 1861 LineInfo li(ln.Flags); 1862 if (ln.Offset > offsetInLinetable) { 1863 if (!nameIndex) { 1864 nameIndex = entry.NameIndex; 1865 lineNumber = li.getStartLine(); 1866 } 1867 StringRef filename = 1868 exitOnErr(getFileName(cvStrTab, checksums, *nameIndex)); 1869 return std::make_pair(filename, *lineNumber); 1870 } 1871 nameIndex = entry.NameIndex; 1872 lineNumber = li.getStartLine(); 1873 } 1874 } 1875 if (!nameIndex) 1876 return std::nullopt; 1877 StringRef filename = exitOnErr(getFileName(cvStrTab, checksums, *nameIndex)); 1878 return std::make_pair(filename, *lineNumber); 1879 } 1880