1 //===-- TypeStreamMerger.cpp ------------------------------------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 #include "llvm/DebugInfo/CodeView/TypeStreamMerger.h"
11 #include "llvm/ADT/SmallString.h"
12 #include "llvm/ADT/StringExtras.h"
13 #include "llvm/DebugInfo/CodeView/GlobalTypeTableBuilder.h"
14 #include "llvm/DebugInfo/CodeView/MergingTypeTableBuilder.h"
15 #include "llvm/DebugInfo/CodeView/TypeDeserializer.h"
16 #include "llvm/DebugInfo/CodeView/TypeIndex.h"
17 #include "llvm/DebugInfo/CodeView/TypeIndexDiscovery.h"
18 #include "llvm/DebugInfo/CodeView/TypeRecord.h"
19 #include "llvm/Support/Error.h"
22 using namespace llvm::codeview;
24 static inline size_t slotForIndex(TypeIndex Idx) {
25 assert(!Idx.isSimple() && "simple type indices have no slots");
26 return Idx.getIndex() - TypeIndex::FirstNonSimpleIndex;
31 /// Implementation of CodeView type stream merging.
33 /// A CodeView type stream is a series of records that reference each other
34 /// through type indices. A type index is either "simple", meaning it is less
35 /// than 0x1000 and refers to a builtin type, or it is complex, meaning it
36 /// refers to a prior type record in the current stream. The type index of a
37 /// record is equal to the number of records before it in the stream plus
40 /// Type records are only allowed to use type indices smaller than their own, so
41 /// a type stream is effectively a topologically sorted DAG. Cycles occuring in
42 /// the type graph of the source program are resolved with forward declarations
43 /// of composite types. This class implements the following type stream merging
44 /// algorithm, which relies on this DAG structure:
46 /// - Begin with a new empty stream, and a new empty hash table that maps from
47 /// type record contents to new type index.
48 /// - For each new type stream, maintain a map from source type index to
49 /// destination type index.
50 /// - For each record, copy it and rewrite its type indices to be valid in the
51 /// destination type stream.
52 /// - If the new type record is not already present in the destination stream
53 /// hash table, append it to the destination type stream, assign it the next
54 /// type index, and update the two hash tables.
55 /// - If the type record already exists in the destination stream, discard it
56 /// and update the type index map to forward the source type index to the
57 /// existing destination type index.
59 /// As an additional complication, type stream merging actually produces two
60 /// streams: an item (or IPI) stream and a type stream, as this is what is
61 /// actually stored in the final PDB. We choose which records go where by
62 /// looking at the record kind.
63 class TypeStreamMerger {
65 explicit TypeStreamMerger(SmallVectorImpl<TypeIndex> &SourceToDest)
66 : IndexMap(SourceToDest) {
67 // When dealing with precompiled headers objects, all data in SourceToDest
68 // belongs to the precompiled headers object, and is assumed to be already
69 // remapped to the target PDB. Any forthcoming type that will be merged in
70 // might potentially back-reference this data. We also don't want to resolve
71 // twice the types in the precompiled object.
72 CurIndex += SourceToDest.size();
75 static const TypeIndex Untranslated;
77 // Local hashing entry points
78 Error mergeTypesAndIds(MergingTypeTableBuilder &DestIds,
79 MergingTypeTableBuilder &DestTypes,
80 const CVTypeArray &IdsAndTypes, Optional<uint32_t> &S);
81 Error mergeIdRecords(MergingTypeTableBuilder &Dest,
82 ArrayRef<TypeIndex> TypeSourceToDest,
83 const CVTypeArray &Ids);
84 Error mergeTypeRecords(MergingTypeTableBuilder &Dest,
85 const CVTypeArray &Types);
87 // Global hashing entry points
88 Error mergeTypesAndIds(GlobalTypeTableBuilder &DestIds,
89 GlobalTypeTableBuilder &DestTypes,
90 const CVTypeArray &IdsAndTypes,
91 ArrayRef<GloballyHashedType> Hashes,
92 Optional<uint32_t> &S);
93 Error mergeIdRecords(GlobalTypeTableBuilder &Dest,
94 ArrayRef<TypeIndex> TypeSourceToDest,
95 const CVTypeArray &Ids,
96 ArrayRef<GloballyHashedType> Hashes);
97 Error mergeTypeRecords(GlobalTypeTableBuilder &Dest, const CVTypeArray &Types,
98 ArrayRef<GloballyHashedType> Hashes,
99 Optional<uint32_t> &S);
102 Error doit(const CVTypeArray &Types);
104 Error remapAllTypes(const CVTypeArray &Types);
106 Error remapType(const CVType &Type);
108 void addMapping(TypeIndex Idx);
110 inline bool remapTypeIndex(TypeIndex &Idx) {
111 // If we're mapping a pure index stream, then IndexMap only contains
112 // mappings from OldIdStream -> NewIdStream, in which case we will need to
113 // use the special mapping from OldTypeStream -> NewTypeStream which was
114 // computed externally. Regardless, we use this special map if and only if
115 // we are doing an id-only mapping.
116 if (!hasTypeStream())
117 return remapIndex(Idx, TypeLookup);
119 assert(TypeLookup.empty());
120 return remapIndex(Idx, IndexMap);
122 inline bool remapItemIndex(TypeIndex &Idx) {
123 assert(hasIdStream());
124 return remapIndex(Idx, IndexMap);
127 bool hasTypeStream() const {
128 return (UseGlobalHashes) ? (!!DestGlobalTypeStream) : (!!DestTypeStream);
131 bool hasIdStream() const {
132 return (UseGlobalHashes) ? (!!DestGlobalIdStream) : (!!DestIdStream);
135 ArrayRef<uint8_t> remapIndices(const CVType &OriginalType,
136 MutableArrayRef<uint8_t> Storage);
138 inline bool remapIndex(TypeIndex &Idx, ArrayRef<TypeIndex> Map) {
139 if (LLVM_LIKELY(remapIndexSimple(Idx, Map)))
142 return remapIndexFallback(Idx, Map);
145 inline bool remapIndexSimple(TypeIndex &Idx, ArrayRef<TypeIndex> Map) const {
146 // Simple types are unchanged.
150 // Check if this type index refers to a record we've already translated
151 // successfully. If it refers to a type later in the stream or a record we
152 // had to defer, defer it until later pass.
153 unsigned MapPos = slotForIndex(Idx);
154 if (LLVM_UNLIKELY(MapPos >= Map.size() || Map[MapPos] == Untranslated))
161 bool remapIndexFallback(TypeIndex &Idx, ArrayRef<TypeIndex> Map);
163 Error errorCorruptRecord() const {
164 return llvm::make_error<CodeViewError>(cv_error_code::corrupt_record);
167 Expected<bool> shouldRemapType(const CVType &Type);
169 Optional<Error> LastError;
171 bool UseGlobalHashes = false;
173 bool IsSecondPass = false;
175 unsigned NumBadIndices = 0;
177 TypeIndex CurIndex{TypeIndex::FirstNonSimpleIndex};
179 MergingTypeTableBuilder *DestIdStream = nullptr;
180 MergingTypeTableBuilder *DestTypeStream = nullptr;
182 GlobalTypeTableBuilder *DestGlobalIdStream = nullptr;
183 GlobalTypeTableBuilder *DestGlobalTypeStream = nullptr;
185 ArrayRef<GloballyHashedType> GlobalHashes;
187 // If we're only mapping id records, this array contains the mapping for
189 ArrayRef<TypeIndex> TypeLookup;
191 /// Map from source type index to destination type index. Indexed by source
192 /// type index minus 0x1000.
193 SmallVectorImpl<TypeIndex> &IndexMap;
195 /// Temporary storage that we use to copy a record's data while re-writing
196 /// its type indices.
197 SmallVector<uint8_t, 256> RemapStorage;
199 Optional<uint32_t> PCHSignature;
202 } // end anonymous namespace
204 const TypeIndex TypeStreamMerger::Untranslated(SimpleTypeKind::NotTranslated);
206 static bool isIdRecord(TypeLeafKind K) {
208 case TypeLeafKind::LF_FUNC_ID:
209 case TypeLeafKind::LF_MFUNC_ID:
210 case TypeLeafKind::LF_STRING_ID:
211 case TypeLeafKind::LF_SUBSTR_LIST:
212 case TypeLeafKind::LF_BUILDINFO:
213 case TypeLeafKind::LF_UDT_SRC_LINE:
214 case TypeLeafKind::LF_UDT_MOD_SRC_LINE:
221 void TypeStreamMerger::addMapping(TypeIndex Idx) {
223 assert(IndexMap.size() == slotForIndex(CurIndex) &&
224 "visitKnownRecord should add one index map entry");
225 IndexMap.push_back(Idx);
227 assert(slotForIndex(CurIndex) < IndexMap.size());
228 IndexMap[slotForIndex(CurIndex)] = Idx;
232 bool TypeStreamMerger::remapIndexFallback(TypeIndex &Idx,
233 ArrayRef<TypeIndex> Map) {
234 size_t MapPos = slotForIndex(Idx);
236 // If this is the second pass and this index isn't in the map, then it points
237 // outside the current type stream, and this is a corrupt record.
238 if (IsSecondPass && MapPos >= Map.size()) {
239 // FIXME: Print a more useful error. We can give the current record and the
240 // index that we think its pointing to.
242 LastError = joinErrors(std::move(*LastError), errorCorruptRecord());
244 LastError = errorCorruptRecord();
249 // This type index is invalid. Remap this to "not translated by cvpack",
250 // and return failure.
255 // Local hashing entry points
256 Error TypeStreamMerger::mergeTypeRecords(MergingTypeTableBuilder &Dest,
257 const CVTypeArray &Types) {
258 DestTypeStream = &Dest;
259 UseGlobalHashes = false;
264 Error TypeStreamMerger::mergeIdRecords(MergingTypeTableBuilder &Dest,
265 ArrayRef<TypeIndex> TypeSourceToDest,
266 const CVTypeArray &Ids) {
267 DestIdStream = &Dest;
268 TypeLookup = TypeSourceToDest;
269 UseGlobalHashes = false;
274 Error TypeStreamMerger::mergeTypesAndIds(MergingTypeTableBuilder &DestIds,
275 MergingTypeTableBuilder &DestTypes,
276 const CVTypeArray &IdsAndTypes,
277 Optional<uint32_t> &S) {
278 DestIdStream = &DestIds;
279 DestTypeStream = &DestTypes;
280 UseGlobalHashes = false;
281 auto Err = doit(IdsAndTypes);
286 // Global hashing entry points
287 Error TypeStreamMerger::mergeTypeRecords(GlobalTypeTableBuilder &Dest,
288 const CVTypeArray &Types,
289 ArrayRef<GloballyHashedType> Hashes,
290 Optional<uint32_t> &S) {
291 DestGlobalTypeStream = &Dest;
292 UseGlobalHashes = true;
293 GlobalHashes = Hashes;
294 auto Err = doit(Types);
299 Error TypeStreamMerger::mergeIdRecords(GlobalTypeTableBuilder &Dest,
300 ArrayRef<TypeIndex> TypeSourceToDest,
301 const CVTypeArray &Ids,
302 ArrayRef<GloballyHashedType> Hashes) {
303 DestGlobalIdStream = &Dest;
304 TypeLookup = TypeSourceToDest;
305 UseGlobalHashes = true;
306 GlobalHashes = Hashes;
311 Error TypeStreamMerger::mergeTypesAndIds(GlobalTypeTableBuilder &DestIds,
312 GlobalTypeTableBuilder &DestTypes,
313 const CVTypeArray &IdsAndTypes,
314 ArrayRef<GloballyHashedType> Hashes,
315 Optional<uint32_t> &S) {
316 DestGlobalIdStream = &DestIds;
317 DestGlobalTypeStream = &DestTypes;
318 UseGlobalHashes = true;
319 GlobalHashes = Hashes;
320 auto Err = doit(IdsAndTypes);
325 Error TypeStreamMerger::doit(const CVTypeArray &Types) {
326 if (auto EC = remapAllTypes(Types))
329 // If we found bad indices but no other errors, try doing another pass and see
330 // if we can resolve the indices that weren't in the map on the first pass.
331 // This may require multiple passes, but we should always make progress. MASM
332 // is the only known CodeView producer that makes type streams that aren't
333 // topologically sorted. The standard library contains MASM-produced objects,
334 // so this is important to handle correctly, but we don't have to be too
335 // efficient. MASM type streams are usually very small.
336 while (!LastError && NumBadIndices > 0) {
337 unsigned BadIndicesRemaining = NumBadIndices;
340 CurIndex = TypeIndex(TypeIndex::FirstNonSimpleIndex);
342 if (auto EC = remapAllTypes(Types))
345 assert(NumBadIndices <= BadIndicesRemaining &&
346 "second pass found more bad indices");
347 if (!LastError && NumBadIndices == BadIndicesRemaining) {
348 return llvm::make_error<CodeViewError>(
349 cv_error_code::corrupt_record, "Input type graph contains cycles");
354 return std::move(*LastError);
355 return Error::success();
358 Error TypeStreamMerger::remapAllTypes(const CVTypeArray &Types) {
359 BinaryStreamRef Stream = Types.getUnderlyingStream();
360 ArrayRef<uint8_t> Buffer;
361 cantFail(Stream.readBytes(0, Stream.getLength(), Buffer));
363 return forEachCodeViewRecord<CVType>(
364 Buffer, [this](const CVType &T) { return remapType(T); });
367 Error TypeStreamMerger::remapType(const CVType &Type) {
368 auto R = shouldRemapType(Type);
370 return R.takeError();
372 TypeIndex DestIdx = Untranslated;
375 [this, Type](MutableArrayRef<uint8_t> Storage) -> ArrayRef<uint8_t> {
376 return remapIndices(Type, Storage);
378 if (LLVM_LIKELY(UseGlobalHashes)) {
379 GlobalTypeTableBuilder &Dest =
380 isIdRecord(Type.kind()) ? *DestGlobalIdStream : *DestGlobalTypeStream;
381 GloballyHashedType H = GlobalHashes[CurIndex.toArrayIndex()];
382 DestIdx = Dest.insertRecordAs(H, Type.RecordData.size(), DoSerialize);
384 MergingTypeTableBuilder &Dest =
385 isIdRecord(Type.kind()) ? *DestIdStream : *DestTypeStream;
387 RemapStorage.resize(Type.RecordData.size());
388 ArrayRef<uint8_t> Result = DoSerialize(RemapStorage);
390 DestIdx = Dest.insertRecordBytes(Result);
396 assert((IsSecondPass || IndexMap.size() == slotForIndex(CurIndex)) &&
397 "visitKnownRecord should add one index map entry");
398 return Error::success();
402 TypeStreamMerger::remapIndices(const CVType &OriginalType,
403 MutableArrayRef<uint8_t> Storage) {
404 SmallVector<TiReference, 4> Refs;
405 discoverTypeIndices(OriginalType.RecordData, Refs);
407 return OriginalType.RecordData;
409 ::memcpy(Storage.data(), OriginalType.RecordData.data(),
410 OriginalType.RecordData.size());
412 uint8_t *DestContent = Storage.data() + sizeof(RecordPrefix);
414 for (auto &Ref : Refs) {
416 reinterpret_cast<TypeIndex *>(DestContent + Ref.Offset);
418 for (size_t I = 0; I < Ref.Count; ++I) {
419 TypeIndex &TI = DestTIs[I];
420 bool Success = (Ref.Kind == TiRefKind::IndexRef) ? remapItemIndex(TI)
421 : remapTypeIndex(TI);
422 if (LLVM_UNLIKELY(!Success))
429 Error llvm::codeview::mergeTypeRecords(MergingTypeTableBuilder &Dest,
430 SmallVectorImpl<TypeIndex> &SourceToDest,
431 const CVTypeArray &Types) {
432 TypeStreamMerger M(SourceToDest);
433 return M.mergeTypeRecords(Dest, Types);
436 Error llvm::codeview::mergeIdRecords(MergingTypeTableBuilder &Dest,
437 ArrayRef<TypeIndex> TypeSourceToDest,
438 SmallVectorImpl<TypeIndex> &SourceToDest,
439 const CVTypeArray &Ids) {
440 TypeStreamMerger M(SourceToDest);
441 return M.mergeIdRecords(Dest, TypeSourceToDest, Ids);
444 Error llvm::codeview::mergeTypeAndIdRecords(
445 MergingTypeTableBuilder &DestIds, MergingTypeTableBuilder &DestTypes,
446 SmallVectorImpl<TypeIndex> &SourceToDest, const CVTypeArray &IdsAndTypes,
447 Optional<uint32_t> &PCHSignature) {
448 TypeStreamMerger M(SourceToDest);
449 return M.mergeTypesAndIds(DestIds, DestTypes, IdsAndTypes, PCHSignature);
452 Error llvm::codeview::mergeTypeAndIdRecords(
453 GlobalTypeTableBuilder &DestIds, GlobalTypeTableBuilder &DestTypes,
454 SmallVectorImpl<TypeIndex> &SourceToDest, const CVTypeArray &IdsAndTypes,
455 ArrayRef<GloballyHashedType> Hashes, Optional<uint32_t> &PCHSignature) {
456 TypeStreamMerger M(SourceToDest);
457 return M.mergeTypesAndIds(DestIds, DestTypes, IdsAndTypes, Hashes,
461 Error llvm::codeview::mergeTypeRecords(GlobalTypeTableBuilder &Dest,
462 SmallVectorImpl<TypeIndex> &SourceToDest,
463 const CVTypeArray &Types,
464 ArrayRef<GloballyHashedType> Hashes,
465 Optional<uint32_t> &PCHSignature) {
466 TypeStreamMerger M(SourceToDest);
467 return M.mergeTypeRecords(Dest, Types, Hashes, PCHSignature);
470 Error llvm::codeview::mergeIdRecords(GlobalTypeTableBuilder &Dest,
471 ArrayRef<TypeIndex> Types,
472 SmallVectorImpl<TypeIndex> &SourceToDest,
473 const CVTypeArray &Ids,
474 ArrayRef<GloballyHashedType> Hashes) {
475 TypeStreamMerger M(SourceToDest);
476 return M.mergeIdRecords(Dest, Types, Ids, Hashes);
479 Expected<bool> TypeStreamMerger::shouldRemapType(const CVType &Type) {
480 // For object files containing precompiled types, we need to extract the
481 // signature, through EndPrecompRecord. This is done here for performance
482 // reasons, to avoid re-parsing the Types stream.
483 if (Type.kind() == LF_ENDPRECOMP) {
485 if (auto EC = TypeDeserializer::deserializeAs(const_cast<CVType &>(Type),
487 return joinErrors(std::move(EC), errorCorruptRecord());
488 if (PCHSignature.hasValue())
489 return errorCorruptRecord();
490 PCHSignature.emplace(EP.getSignature());