1 //===- CoverageMappingReader.cpp - Code coverage mapping reader -----------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file contains support for reading coverage mapping data for
10 // instrumentation based coverage.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "llvm/ProfileData/Coverage/CoverageMappingReader.h"
15 #include "llvm/ADT/ArrayRef.h"
16 #include "llvm/ADT/DenseMap.h"
17 #include "llvm/ADT/STLExtras.h"
18 #include "llvm/ADT/SmallVector.h"
19 #include "llvm/ADT/Statistic.h"
20 #include "llvm/ADT/StringRef.h"
21 #include "llvm/Object/Archive.h"
22 #include "llvm/Object/Binary.h"
23 #include "llvm/Object/COFF.h"
24 #include "llvm/Object/Error.h"
25 #include "llvm/Object/MachOUniversal.h"
26 #include "llvm/Object/ObjectFile.h"
27 #include "llvm/ProfileData/InstrProf.h"
28 #include "llvm/Support/Casting.h"
29 #include "llvm/Support/Compression.h"
30 #include "llvm/Support/Debug.h"
31 #include "llvm/Support/Endian.h"
32 #include "llvm/Support/Error.h"
33 #include "llvm/Support/ErrorHandling.h"
34 #include "llvm/Support/LEB128.h"
35 #include "llvm/Support/MathExtras.h"
36 #include "llvm/Support/Path.h"
37 #include "llvm/Support/raw_ostream.h"
38 #include "llvm/TargetParser/Triple.h"
39 #include <vector>
40 
41 using namespace llvm;
42 using namespace coverage;
43 using namespace object;
44 
45 #define DEBUG_TYPE "coverage-mapping"
46 
47 STATISTIC(CovMapNumRecords, "The # of coverage function records");
48 STATISTIC(CovMapNumUsedRecords, "The # of used coverage function records");
49 
increment()50 void CoverageMappingIterator::increment() {
51   if (ReadErr != coveragemap_error::success)
52     return;
53 
54   // Check if all the records were read or if an error occurred while reading
55   // the next record.
56   if (auto E = Reader->readNextRecord(Record))
57     handleAllErrors(std::move(E), [&](const CoverageMapError &CME) {
58       if (CME.get() == coveragemap_error::eof)
59         *this = CoverageMappingIterator();
60       else
61         ReadErr = CME.get();
62     });
63 }
64 
readULEB128(uint64_t & Result)65 Error RawCoverageReader::readULEB128(uint64_t &Result) {
66   if (Data.empty())
67     return make_error<CoverageMapError>(coveragemap_error::truncated);
68   unsigned N = 0;
69   Result = decodeULEB128(Data.bytes_begin(), &N);
70   if (N > Data.size())
71     return make_error<CoverageMapError>(coveragemap_error::malformed,
72                                         "the size of ULEB128 is too big");
73   Data = Data.substr(N);
74   return Error::success();
75 }
76 
readIntMax(uint64_t & Result,uint64_t MaxPlus1)77 Error RawCoverageReader::readIntMax(uint64_t &Result, uint64_t MaxPlus1) {
78   if (auto Err = readULEB128(Result))
79     return Err;
80   if (Result >= MaxPlus1)
81     return make_error<CoverageMapError>(
82         coveragemap_error::malformed,
83         "the value of ULEB128 is greater than or equal to MaxPlus1");
84   return Error::success();
85 }
86 
readSize(uint64_t & Result)87 Error RawCoverageReader::readSize(uint64_t &Result) {
88   if (auto Err = readULEB128(Result))
89     return Err;
90   if (Result > Data.size())
91     return make_error<CoverageMapError>(coveragemap_error::malformed,
92                                         "the value of ULEB128 is too big");
93   return Error::success();
94 }
95 
readString(StringRef & Result)96 Error RawCoverageReader::readString(StringRef &Result) {
97   uint64_t Length;
98   if (auto Err = readSize(Length))
99     return Err;
100   Result = Data.substr(0, Length);
101   Data = Data.substr(Length);
102   return Error::success();
103 }
104 
read(CovMapVersion Version)105 Error RawCoverageFilenamesReader::read(CovMapVersion Version) {
106   uint64_t NumFilenames;
107   if (auto Err = readSize(NumFilenames))
108     return Err;
109   if (!NumFilenames)
110     return make_error<CoverageMapError>(coveragemap_error::malformed,
111                                         "number of filenames is zero");
112 
113   if (Version < CovMapVersion::Version4)
114     return readUncompressed(Version, NumFilenames);
115 
116   // The uncompressed length may exceed the size of the encoded filenames.
117   // Skip size validation.
118   uint64_t UncompressedLen;
119   if (auto Err = readULEB128(UncompressedLen))
120     return Err;
121 
122   uint64_t CompressedLen;
123   if (auto Err = readSize(CompressedLen))
124     return Err;
125 
126   if (CompressedLen > 0) {
127     if (!compression::zlib::isAvailable())
128       return make_error<CoverageMapError>(
129           coveragemap_error::decompression_failed);
130 
131     // Allocate memory for the decompressed filenames.
132     SmallVector<uint8_t, 0> StorageBuf;
133 
134     // Read compressed filenames.
135     StringRef CompressedFilenames = Data.substr(0, CompressedLen);
136     Data = Data.substr(CompressedLen);
137     auto Err = compression::zlib::decompress(
138         arrayRefFromStringRef(CompressedFilenames), StorageBuf,
139         UncompressedLen);
140     if (Err) {
141       consumeError(std::move(Err));
142       return make_error<CoverageMapError>(
143           coveragemap_error::decompression_failed);
144     }
145 
146     RawCoverageFilenamesReader Delegate(toStringRef(StorageBuf), Filenames,
147                                         CompilationDir);
148     return Delegate.readUncompressed(Version, NumFilenames);
149   }
150 
151   return readUncompressed(Version, NumFilenames);
152 }
153 
readUncompressed(CovMapVersion Version,uint64_t NumFilenames)154 Error RawCoverageFilenamesReader::readUncompressed(CovMapVersion Version,
155                                                    uint64_t NumFilenames) {
156   // Read uncompressed filenames.
157   if (Version < CovMapVersion::Version6) {
158     for (size_t I = 0; I < NumFilenames; ++I) {
159       StringRef Filename;
160       if (auto Err = readString(Filename))
161         return Err;
162       Filenames.push_back(Filename.str());
163     }
164   } else {
165     StringRef CWD;
166     if (auto Err = readString(CWD))
167       return Err;
168     Filenames.push_back(CWD.str());
169 
170     for (size_t I = 1; I < NumFilenames; ++I) {
171       StringRef Filename;
172       if (auto Err = readString(Filename))
173         return Err;
174       if (sys::path::is_absolute(Filename)) {
175         Filenames.push_back(Filename.str());
176       } else {
177         SmallString<256> P;
178         if (!CompilationDir.empty())
179           P.assign(CompilationDir);
180         else
181           P.assign(CWD);
182         llvm::sys::path::append(P, Filename);
183         sys::path::remove_dots(P, /*remove_dot_dot=*/true);
184         Filenames.push_back(static_cast<std::string>(P.str()));
185       }
186     }
187   }
188   return Error::success();
189 }
190 
decodeCounter(unsigned Value,Counter & C)191 Error RawCoverageMappingReader::decodeCounter(unsigned Value, Counter &C) {
192   auto Tag = Value & Counter::EncodingTagMask;
193   switch (Tag) {
194   case Counter::Zero:
195     C = Counter::getZero();
196     return Error::success();
197   case Counter::CounterValueReference:
198     C = Counter::getCounter(Value >> Counter::EncodingTagBits);
199     return Error::success();
200   default:
201     break;
202   }
203   Tag -= Counter::Expression;
204   switch (Tag) {
205   case CounterExpression::Subtract:
206   case CounterExpression::Add: {
207     auto ID = Value >> Counter::EncodingTagBits;
208     if (ID >= Expressions.size())
209       return make_error<CoverageMapError>(coveragemap_error::malformed,
210                                           "counter expression is invalid");
211     Expressions[ID].Kind = CounterExpression::ExprKind(Tag);
212     C = Counter::getExpression(ID);
213     break;
214   }
215   default:
216     return make_error<CoverageMapError>(coveragemap_error::malformed,
217                                         "counter expression kind is invalid");
218   }
219   return Error::success();
220 }
221 
readCounter(Counter & C)222 Error RawCoverageMappingReader::readCounter(Counter &C) {
223   uint64_t EncodedCounter;
224   if (auto Err =
225           readIntMax(EncodedCounter, std::numeric_limits<unsigned>::max()))
226     return Err;
227   if (auto Err = decodeCounter(EncodedCounter, C))
228     return Err;
229   return Error::success();
230 }
231 
232 static const unsigned EncodingExpansionRegionBit = 1
233                                                    << Counter::EncodingTagBits;
234 
235 /// Read the sub-array of regions for the given inferred file id.
236 /// \param NumFileIDs the number of file ids that are defined for this
237 /// function.
readMappingRegionsSubArray(std::vector<CounterMappingRegion> & MappingRegions,unsigned InferredFileID,size_t NumFileIDs)238 Error RawCoverageMappingReader::readMappingRegionsSubArray(
239     std::vector<CounterMappingRegion> &MappingRegions, unsigned InferredFileID,
240     size_t NumFileIDs) {
241   uint64_t NumRegions;
242   if (auto Err = readSize(NumRegions))
243     return Err;
244   unsigned LineStart = 0;
245   for (size_t I = 0; I < NumRegions; ++I) {
246     Counter C, C2;
247     uint64_t BIDX = 0, NC = 0, ID = 0, TID = 0, FID = 0;
248     CounterMappingRegion::RegionKind Kind = CounterMappingRegion::CodeRegion;
249 
250     // Read the combined counter + region kind.
251     uint64_t EncodedCounterAndRegion;
252     if (auto Err = readIntMax(EncodedCounterAndRegion,
253                               std::numeric_limits<unsigned>::max()))
254       return Err;
255     unsigned Tag = EncodedCounterAndRegion & Counter::EncodingTagMask;
256     uint64_t ExpandedFileID = 0;
257 
258     // If Tag does not represent a ZeroCounter, then it is understood to refer
259     // to a counter or counter expression with region kind assumed to be
260     // "CodeRegion". In that case, EncodedCounterAndRegion actually encodes the
261     // referenced counter or counter expression (and nothing else).
262     //
263     // If Tag represents a ZeroCounter and EncodingExpansionRegionBit is set,
264     // then EncodedCounterAndRegion is interpreted to represent an
265     // ExpansionRegion. In all other cases, EncodedCounterAndRegion is
266     // interpreted to refer to a specific region kind, after which additional
267     // fields may be read (e.g. BranchRegions have two encoded counters that
268     // follow an encoded region kind value).
269     if (Tag != Counter::Zero) {
270       if (auto Err = decodeCounter(EncodedCounterAndRegion, C))
271         return Err;
272     } else {
273       // Is it an expansion region?
274       if (EncodedCounterAndRegion & EncodingExpansionRegionBit) {
275         Kind = CounterMappingRegion::ExpansionRegion;
276         ExpandedFileID = EncodedCounterAndRegion >>
277                          Counter::EncodingCounterTagAndExpansionRegionTagBits;
278         if (ExpandedFileID >= NumFileIDs)
279           return make_error<CoverageMapError>(coveragemap_error::malformed,
280                                               "ExpandedFileID is invalid");
281       } else {
282         switch (EncodedCounterAndRegion >>
283                 Counter::EncodingCounterTagAndExpansionRegionTagBits) {
284         case CounterMappingRegion::CodeRegion:
285           // Don't do anything when we have a code region with a zero counter.
286           break;
287         case CounterMappingRegion::SkippedRegion:
288           Kind = CounterMappingRegion::SkippedRegion;
289           break;
290         case CounterMappingRegion::BranchRegion:
291           // For a Branch Region, read two successive counters.
292           Kind = CounterMappingRegion::BranchRegion;
293           if (auto Err = readCounter(C))
294             return Err;
295           if (auto Err = readCounter(C2))
296             return Err;
297           break;
298         case CounterMappingRegion::MCDCBranchRegion:
299           // For a MCDC Branch Region, read two successive counters and 3 IDs.
300           Kind = CounterMappingRegion::MCDCBranchRegion;
301           if (auto Err = readCounter(C))
302             return Err;
303           if (auto Err = readCounter(C2))
304             return Err;
305           if (auto Err = readIntMax(ID, std::numeric_limits<unsigned>::max()))
306             return Err;
307           if (auto Err = readIntMax(TID, std::numeric_limits<unsigned>::max()))
308             return Err;
309           if (auto Err = readIntMax(FID, std::numeric_limits<unsigned>::max()))
310             return Err;
311           break;
312         case CounterMappingRegion::MCDCDecisionRegion:
313           Kind = CounterMappingRegion::MCDCDecisionRegion;
314           if (auto Err = readIntMax(BIDX, std::numeric_limits<unsigned>::max()))
315             return Err;
316           if (auto Err = readIntMax(NC, std::numeric_limits<unsigned>::max()))
317             return Err;
318           break;
319         default:
320           return make_error<CoverageMapError>(coveragemap_error::malformed,
321                                               "region kind is incorrect");
322         }
323       }
324     }
325 
326     // Read the source range.
327     uint64_t LineStartDelta, ColumnStart, NumLines, ColumnEnd;
328     if (auto Err =
329             readIntMax(LineStartDelta, std::numeric_limits<unsigned>::max()))
330       return Err;
331     if (auto Err = readULEB128(ColumnStart))
332       return Err;
333     if (ColumnStart > std::numeric_limits<unsigned>::max())
334       return make_error<CoverageMapError>(coveragemap_error::malformed,
335                                           "start column is too big");
336     if (auto Err = readIntMax(NumLines, std::numeric_limits<unsigned>::max()))
337       return Err;
338     if (auto Err = readIntMax(ColumnEnd, std::numeric_limits<unsigned>::max()))
339       return Err;
340     LineStart += LineStartDelta;
341 
342     // If the high bit of ColumnEnd is set, this is a gap region.
343     if (ColumnEnd & (1U << 31)) {
344       Kind = CounterMappingRegion::GapRegion;
345       ColumnEnd &= ~(1U << 31);
346     }
347 
348     // Adjust the column locations for the empty regions that are supposed to
349     // cover whole lines. Those regions should be encoded with the
350     // column range (1 -> std::numeric_limits<unsigned>::max()), but because
351     // the encoded std::numeric_limits<unsigned>::max() is several bytes long,
352     // we set the column range to (0 -> 0) to ensure that the column start and
353     // column end take up one byte each.
354     // The std::numeric_limits<unsigned>::max() is used to represent a column
355     // position at the end of the line without knowing the length of that line.
356     if (ColumnStart == 0 && ColumnEnd == 0) {
357       ColumnStart = 1;
358       ColumnEnd = std::numeric_limits<unsigned>::max();
359     }
360 
361     LLVM_DEBUG({
362       dbgs() << "Counter in file " << InferredFileID << " " << LineStart << ":"
363              << ColumnStart << " -> " << (LineStart + NumLines) << ":"
364              << ColumnEnd << ", ";
365       if (Kind == CounterMappingRegion::ExpansionRegion)
366         dbgs() << "Expands to file " << ExpandedFileID;
367       else
368         CounterMappingContext(Expressions).dump(C, dbgs());
369       dbgs() << "\n";
370     });
371 
372     auto CMR = CounterMappingRegion(
373         C, C2,
374         CounterMappingRegion::MCDCParameters{
375             static_cast<unsigned>(BIDX), static_cast<unsigned>(NC),
376             static_cast<unsigned>(ID), static_cast<unsigned>(TID),
377             static_cast<unsigned>(FID)},
378         InferredFileID, ExpandedFileID, LineStart, ColumnStart,
379         LineStart + NumLines, ColumnEnd, Kind);
380     if (CMR.startLoc() > CMR.endLoc())
381       return make_error<CoverageMapError>(
382           coveragemap_error::malformed,
383           "counter mapping region locations are incorrect");
384     MappingRegions.push_back(CMR);
385   }
386   return Error::success();
387 }
388 
read()389 Error RawCoverageMappingReader::read() {
390   // Read the virtual file mapping.
391   SmallVector<unsigned, 8> VirtualFileMapping;
392   uint64_t NumFileMappings;
393   if (auto Err = readSize(NumFileMappings))
394     return Err;
395   for (size_t I = 0; I < NumFileMappings; ++I) {
396     uint64_t FilenameIndex;
397     if (auto Err = readIntMax(FilenameIndex, TranslationUnitFilenames.size()))
398       return Err;
399     VirtualFileMapping.push_back(FilenameIndex);
400   }
401 
402   // Construct the files using unique filenames and virtual file mapping.
403   for (auto I : VirtualFileMapping) {
404     Filenames.push_back(TranslationUnitFilenames[I]);
405   }
406 
407   // Read the expressions.
408   uint64_t NumExpressions;
409   if (auto Err = readSize(NumExpressions))
410     return Err;
411   // Create an array of dummy expressions that get the proper counters
412   // when the expressions are read, and the proper kinds when the counters
413   // are decoded.
414   Expressions.resize(
415       NumExpressions,
416       CounterExpression(CounterExpression::Subtract, Counter(), Counter()));
417   for (size_t I = 0; I < NumExpressions; ++I) {
418     if (auto Err = readCounter(Expressions[I].LHS))
419       return Err;
420     if (auto Err = readCounter(Expressions[I].RHS))
421       return Err;
422   }
423 
424   // Read the mapping regions sub-arrays.
425   for (unsigned InferredFileID = 0, S = VirtualFileMapping.size();
426        InferredFileID < S; ++InferredFileID) {
427     if (auto Err = readMappingRegionsSubArray(MappingRegions, InferredFileID,
428                                               VirtualFileMapping.size()))
429       return Err;
430   }
431 
432   // Set the counters for the expansion regions.
433   // i.e. Counter of expansion region = counter of the first region
434   // from the expanded file.
435   // Perform multiple passes to correctly propagate the counters through
436   // all the nested expansion regions.
437   SmallVector<CounterMappingRegion *, 8> FileIDExpansionRegionMapping;
438   FileIDExpansionRegionMapping.resize(VirtualFileMapping.size(), nullptr);
439   for (unsigned Pass = 1, S = VirtualFileMapping.size(); Pass < S; ++Pass) {
440     for (auto &R : MappingRegions) {
441       if (R.Kind != CounterMappingRegion::ExpansionRegion)
442         continue;
443       assert(!FileIDExpansionRegionMapping[R.ExpandedFileID]);
444       FileIDExpansionRegionMapping[R.ExpandedFileID] = &R;
445     }
446     for (auto &R : MappingRegions) {
447       if (FileIDExpansionRegionMapping[R.FileID]) {
448         FileIDExpansionRegionMapping[R.FileID]->Count = R.Count;
449         FileIDExpansionRegionMapping[R.FileID] = nullptr;
450       }
451     }
452   }
453 
454   return Error::success();
455 }
456 
isDummy()457 Expected<bool> RawCoverageMappingDummyChecker::isDummy() {
458   // A dummy coverage mapping data consists of just one region with zero count.
459   uint64_t NumFileMappings;
460   if (Error Err = readSize(NumFileMappings))
461     return std::move(Err);
462   if (NumFileMappings != 1)
463     return false;
464   // We don't expect any specific value for the filename index, just skip it.
465   uint64_t FilenameIndex;
466   if (Error Err =
467           readIntMax(FilenameIndex, std::numeric_limits<unsigned>::max()))
468     return std::move(Err);
469   uint64_t NumExpressions;
470   if (Error Err = readSize(NumExpressions))
471     return std::move(Err);
472   if (NumExpressions != 0)
473     return false;
474   uint64_t NumRegions;
475   if (Error Err = readSize(NumRegions))
476     return std::move(Err);
477   if (NumRegions != 1)
478     return false;
479   uint64_t EncodedCounterAndRegion;
480   if (Error Err = readIntMax(EncodedCounterAndRegion,
481                              std::numeric_limits<unsigned>::max()))
482     return std::move(Err);
483   unsigned Tag = EncodedCounterAndRegion & Counter::EncodingTagMask;
484   return Tag == Counter::Zero;
485 }
486 
create(SectionRef & Section)487 Error InstrProfSymtab::create(SectionRef &Section) {
488   Expected<StringRef> DataOrErr = Section.getContents();
489   if (!DataOrErr)
490     return DataOrErr.takeError();
491   Data = *DataOrErr;
492   Address = Section.getAddress();
493 
494   // If this is a linked PE/COFF file, then we have to skip over the null byte
495   // that is allocated in the .lprfn$A section in the LLVM profiling runtime.
496   // If the name section is .lprfcovnames, it doesn't have the null byte at the
497   // beginning.
498   const ObjectFile *Obj = Section.getObject();
499   if (isa<COFFObjectFile>(Obj) && !Obj->isRelocatableObject())
500     if (Expected<StringRef> NameOrErr = Section.getName())
501       if (*NameOrErr != getInstrProfSectionName(IPSK_covname, Triple::COFF))
502         Data = Data.drop_front(1);
503 
504   return Error::success();
505 }
506 
getFuncName(uint64_t Pointer,size_t Size)507 StringRef InstrProfSymtab::getFuncName(uint64_t Pointer, size_t Size) {
508   if (Pointer < Address)
509     return StringRef();
510   auto Offset = Pointer - Address;
511   if (Offset + Size > Data.size())
512     return StringRef();
513   return Data.substr(Pointer - Address, Size);
514 }
515 
516 // Check if the mapping data is a dummy, i.e. is emitted for an unused function.
isCoverageMappingDummy(uint64_t Hash,StringRef Mapping)517 static Expected<bool> isCoverageMappingDummy(uint64_t Hash, StringRef Mapping) {
518   // The hash value of dummy mapping records is always zero.
519   if (Hash)
520     return false;
521   return RawCoverageMappingDummyChecker(Mapping).isDummy();
522 }
523 
524 /// A range of filename indices. Used to specify the location of a batch of
525 /// filenames in a vector-like container.
526 struct FilenameRange {
527   unsigned StartingIndex;
528   unsigned Length;
529 
FilenameRangeFilenameRange530   FilenameRange(unsigned StartingIndex, unsigned Length)
531       : StartingIndex(StartingIndex), Length(Length) {}
532 
markInvalidFilenameRange533   void markInvalid() { Length = 0; }
isInvalidFilenameRange534   bool isInvalid() const { return Length == 0; }
535 };
536 
537 namespace {
538 
539 /// The interface to read coverage mapping function records for a module.
540 struct CovMapFuncRecordReader {
541   virtual ~CovMapFuncRecordReader() = default;
542 
543   // Read a coverage header.
544   //
545   // \p CovBuf points to the buffer containing the \c CovHeader of the coverage
546   // mapping data associated with the module.
547   //
548   // Returns a pointer to the next \c CovHeader if it exists, or to an address
549   // greater than \p CovEnd if not.
550   virtual Expected<const char *> readCoverageHeader(const char *CovBuf,
551                                                     const char *CovBufEnd) = 0;
552 
553   // Read function records.
554   //
555   // \p FuncRecBuf points to the buffer containing a batch of function records.
556   // \p FuncRecBufEnd points past the end of the batch of records.
557   //
558   // Prior to Version4, \p OutOfLineFileRange points to a sequence of filenames
559   // associated with the function records. It is unused in Version4.
560   //
561   // Prior to Version4, \p OutOfLineMappingBuf points to a sequence of coverage
562   // mappings associated with the function records. It is unused in Version4.
563   virtual Error
564   readFunctionRecords(const char *FuncRecBuf, const char *FuncRecBufEnd,
565                       std::optional<FilenameRange> OutOfLineFileRange,
566                       const char *OutOfLineMappingBuf,
567                       const char *OutOfLineMappingBufEnd) = 0;
568 
569   template <class IntPtrT, llvm::endianness Endian>
570   static Expected<std::unique_ptr<CovMapFuncRecordReader>>
571   get(CovMapVersion Version, InstrProfSymtab &P,
572       std::vector<BinaryCoverageReader::ProfileMappingRecord> &R, StringRef D,
573       std::vector<std::string> &F);
574 };
575 
576 // A class for reading coverage mapping function records for a module.
577 template <CovMapVersion Version, class IntPtrT, llvm::endianness Endian>
578 class VersionedCovMapFuncRecordReader : public CovMapFuncRecordReader {
579   using FuncRecordType =
580       typename CovMapTraits<Version, IntPtrT>::CovMapFuncRecordType;
581   using NameRefType = typename CovMapTraits<Version, IntPtrT>::NameRefType;
582 
583   // Maps function's name references to the indexes of their records
584   // in \c Records.
585   DenseMap<NameRefType, size_t> FunctionRecords;
586   InstrProfSymtab &ProfileNames;
587   StringRef CompilationDir;
588   std::vector<std::string> &Filenames;
589   std::vector<BinaryCoverageReader::ProfileMappingRecord> &Records;
590 
591   // Maps a hash of the filenames in a TU to a \c FileRange. The range
592   // specifies the location of the hashed filenames in \c Filenames.
593   DenseMap<uint64_t, FilenameRange> FileRangeMap;
594 
595   // Add the record to the collection if we don't already have a record that
596   // points to the same function name. This is useful to ignore the redundant
597   // records for the functions with ODR linkage.
598   // In addition, prefer records with real coverage mapping data to dummy
599   // records, which were emitted for inline functions which were seen but
600   // not used in the corresponding translation unit.
insertFunctionRecordIfNeeded(const FuncRecordType * CFR,StringRef Mapping,FilenameRange FileRange)601   Error insertFunctionRecordIfNeeded(const FuncRecordType *CFR,
602                                      StringRef Mapping,
603                                      FilenameRange FileRange) {
604     ++CovMapNumRecords;
605     uint64_t FuncHash = CFR->template getFuncHash<Endian>();
606     NameRefType NameRef = CFR->template getFuncNameRef<Endian>();
607     auto InsertResult =
608         FunctionRecords.insert(std::make_pair(NameRef, Records.size()));
609     if (InsertResult.second) {
610       StringRef FuncName;
611       if (Error Err = CFR->template getFuncName<Endian>(ProfileNames, FuncName))
612         return Err;
613       if (FuncName.empty())
614         return make_error<InstrProfError>(instrprof_error::malformed,
615                                           "function name is empty");
616       ++CovMapNumUsedRecords;
617       Records.emplace_back(Version, FuncName, FuncHash, Mapping,
618                            FileRange.StartingIndex, FileRange.Length);
619       return Error::success();
620     }
621     // Update the existing record if it's a dummy and the new record is real.
622     size_t OldRecordIndex = InsertResult.first->second;
623     BinaryCoverageReader::ProfileMappingRecord &OldRecord =
624         Records[OldRecordIndex];
625     Expected<bool> OldIsDummyExpected = isCoverageMappingDummy(
626         OldRecord.FunctionHash, OldRecord.CoverageMapping);
627     if (Error Err = OldIsDummyExpected.takeError())
628       return Err;
629     if (!*OldIsDummyExpected)
630       return Error::success();
631     Expected<bool> NewIsDummyExpected =
632         isCoverageMappingDummy(FuncHash, Mapping);
633     if (Error Err = NewIsDummyExpected.takeError())
634       return Err;
635     if (*NewIsDummyExpected)
636       return Error::success();
637     ++CovMapNumUsedRecords;
638     OldRecord.FunctionHash = FuncHash;
639     OldRecord.CoverageMapping = Mapping;
640     OldRecord.FilenamesBegin = FileRange.StartingIndex;
641     OldRecord.FilenamesSize = FileRange.Length;
642     return Error::success();
643   }
644 
645 public:
VersionedCovMapFuncRecordReader(InstrProfSymtab & P,std::vector<BinaryCoverageReader::ProfileMappingRecord> & R,StringRef D,std::vector<std::string> & F)646   VersionedCovMapFuncRecordReader(
647       InstrProfSymtab &P,
648       std::vector<BinaryCoverageReader::ProfileMappingRecord> &R, StringRef D,
649       std::vector<std::string> &F)
650       : ProfileNames(P), CompilationDir(D), Filenames(F), Records(R) {}
651 
652   ~VersionedCovMapFuncRecordReader() override = default;
653 
readCoverageHeader(const char * CovBuf,const char * CovBufEnd)654   Expected<const char *> readCoverageHeader(const char *CovBuf,
655                                             const char *CovBufEnd) override {
656     using namespace support;
657 
658     if (CovBuf + sizeof(CovMapHeader) > CovBufEnd)
659       return make_error<CoverageMapError>(
660           coveragemap_error::malformed,
661           "coverage mapping header section is larger than buffer size");
662     auto CovHeader = reinterpret_cast<const CovMapHeader *>(CovBuf);
663     uint32_t NRecords = CovHeader->getNRecords<Endian>();
664     uint32_t FilenamesSize = CovHeader->getFilenamesSize<Endian>();
665     uint32_t CoverageSize = CovHeader->getCoverageSize<Endian>();
666     assert((CovMapVersion)CovHeader->getVersion<Endian>() == Version);
667     CovBuf = reinterpret_cast<const char *>(CovHeader + 1);
668 
669     // Skip past the function records, saving the start and end for later.
670     // This is a no-op in Version4 (function records are read after all headers
671     // are read).
672     const char *FuncRecBuf = nullptr;
673     const char *FuncRecBufEnd = nullptr;
674     if (Version < CovMapVersion::Version4)
675       FuncRecBuf = CovBuf;
676     CovBuf += NRecords * sizeof(FuncRecordType);
677     if (Version < CovMapVersion::Version4)
678       FuncRecBufEnd = CovBuf;
679 
680     // Get the filenames.
681     if (CovBuf + FilenamesSize > CovBufEnd)
682       return make_error<CoverageMapError>(
683           coveragemap_error::malformed,
684           "filenames section is larger than buffer size");
685     size_t FilenamesBegin = Filenames.size();
686     StringRef FilenameRegion(CovBuf, FilenamesSize);
687     RawCoverageFilenamesReader Reader(FilenameRegion, Filenames,
688                                       CompilationDir);
689     if (auto Err = Reader.read(Version))
690       return std::move(Err);
691     CovBuf += FilenamesSize;
692     FilenameRange FileRange(FilenamesBegin, Filenames.size() - FilenamesBegin);
693 
694     if (Version >= CovMapVersion::Version4) {
695       // Map a hash of the filenames region to the filename range associated
696       // with this coverage header.
697       int64_t FilenamesRef =
698           llvm::IndexedInstrProf::ComputeHash(FilenameRegion);
699       auto Insert =
700           FileRangeMap.insert(std::make_pair(FilenamesRef, FileRange));
701       if (!Insert.second) {
702         // The same filenames ref was encountered twice. It's possible that
703         // the associated filenames are the same.
704         auto It = Filenames.begin();
705         FilenameRange &OrigRange = Insert.first->getSecond();
706         if (std::equal(It + OrigRange.StartingIndex,
707                        It + OrigRange.StartingIndex + OrigRange.Length,
708                        It + FileRange.StartingIndex,
709                        It + FileRange.StartingIndex + FileRange.Length))
710           // Map the new range to the original one.
711           FileRange = OrigRange;
712         else
713           // This is a hash collision. Mark the filenames ref invalid.
714           OrigRange.markInvalid();
715       }
716     }
717 
718     // We'll read the coverage mapping records in the loop below.
719     // This is a no-op in Version4 (coverage mappings are not affixed to the
720     // coverage header).
721     const char *MappingBuf = CovBuf;
722     if (Version >= CovMapVersion::Version4 && CoverageSize != 0)
723       return make_error<CoverageMapError>(coveragemap_error::malformed,
724                                           "coverage mapping size is not zero");
725     CovBuf += CoverageSize;
726     const char *MappingEnd = CovBuf;
727 
728     if (CovBuf > CovBufEnd)
729       return make_error<CoverageMapError>(
730           coveragemap_error::malformed,
731           "function records section is larger than buffer size");
732 
733     if (Version < CovMapVersion::Version4) {
734       // Read each function record.
735       if (Error E = readFunctionRecords(FuncRecBuf, FuncRecBufEnd, FileRange,
736                                         MappingBuf, MappingEnd))
737         return std::move(E);
738     }
739 
740     // Each coverage map has an alignment of 8, so we need to adjust alignment
741     // before reading the next map.
742     CovBuf += offsetToAlignedAddr(CovBuf, Align(8));
743 
744     return CovBuf;
745   }
746 
readFunctionRecords(const char * FuncRecBuf,const char * FuncRecBufEnd,std::optional<FilenameRange> OutOfLineFileRange,const char * OutOfLineMappingBuf,const char * OutOfLineMappingBufEnd)747   Error readFunctionRecords(const char *FuncRecBuf, const char *FuncRecBufEnd,
748                             std::optional<FilenameRange> OutOfLineFileRange,
749                             const char *OutOfLineMappingBuf,
750                             const char *OutOfLineMappingBufEnd) override {
751     auto CFR = reinterpret_cast<const FuncRecordType *>(FuncRecBuf);
752     while ((const char *)CFR < FuncRecBufEnd) {
753       // Validate the length of the coverage mapping for this function.
754       const char *NextMappingBuf;
755       const FuncRecordType *NextCFR;
756       std::tie(NextMappingBuf, NextCFR) =
757           CFR->template advanceByOne<Endian>(OutOfLineMappingBuf);
758       if (Version < CovMapVersion::Version4)
759         if (NextMappingBuf > OutOfLineMappingBufEnd)
760           return make_error<CoverageMapError>(
761               coveragemap_error::malformed,
762               "next mapping buffer is larger than buffer size");
763 
764       // Look up the set of filenames associated with this function record.
765       std::optional<FilenameRange> FileRange;
766       if (Version < CovMapVersion::Version4) {
767         FileRange = OutOfLineFileRange;
768       } else {
769         uint64_t FilenamesRef = CFR->template getFilenamesRef<Endian>();
770         auto It = FileRangeMap.find(FilenamesRef);
771         if (It == FileRangeMap.end())
772           return make_error<CoverageMapError>(
773               coveragemap_error::malformed,
774               "no filename found for function with hash=0x" +
775                   Twine::utohexstr(FilenamesRef));
776         else
777           FileRange = It->getSecond();
778       }
779 
780       // Now, read the coverage data.
781       if (FileRange && !FileRange->isInvalid()) {
782         StringRef Mapping =
783             CFR->template getCoverageMapping<Endian>(OutOfLineMappingBuf);
784         if (Version >= CovMapVersion::Version4 &&
785             Mapping.data() + Mapping.size() > FuncRecBufEnd)
786           return make_error<CoverageMapError>(
787               coveragemap_error::malformed,
788               "coverage mapping data is larger than buffer size");
789         if (Error Err = insertFunctionRecordIfNeeded(CFR, Mapping, *FileRange))
790           return Err;
791       }
792 
793       std::tie(OutOfLineMappingBuf, CFR) = std::tie(NextMappingBuf, NextCFR);
794     }
795     return Error::success();
796   }
797 };
798 
799 } // end anonymous namespace
800 
801 template <class IntPtrT, llvm::endianness Endian>
get(CovMapVersion Version,InstrProfSymtab & P,std::vector<BinaryCoverageReader::ProfileMappingRecord> & R,StringRef D,std::vector<std::string> & F)802 Expected<std::unique_ptr<CovMapFuncRecordReader>> CovMapFuncRecordReader::get(
803     CovMapVersion Version, InstrProfSymtab &P,
804     std::vector<BinaryCoverageReader::ProfileMappingRecord> &R, StringRef D,
805     std::vector<std::string> &F) {
806   using namespace coverage;
807 
808   switch (Version) {
809   case CovMapVersion::Version1:
810     return std::make_unique<VersionedCovMapFuncRecordReader<
811         CovMapVersion::Version1, IntPtrT, Endian>>(P, R, D, F);
812   case CovMapVersion::Version2:
813   case CovMapVersion::Version3:
814   case CovMapVersion::Version4:
815   case CovMapVersion::Version5:
816   case CovMapVersion::Version6:
817   case CovMapVersion::Version7:
818     // Decompress the name data.
819     if (Error E = P.create(P.getNameData()))
820       return std::move(E);
821     if (Version == CovMapVersion::Version2)
822       return std::make_unique<VersionedCovMapFuncRecordReader<
823           CovMapVersion::Version2, IntPtrT, Endian>>(P, R, D, F);
824     else if (Version == CovMapVersion::Version3)
825       return std::make_unique<VersionedCovMapFuncRecordReader<
826           CovMapVersion::Version3, IntPtrT, Endian>>(P, R, D, F);
827     else if (Version == CovMapVersion::Version4)
828       return std::make_unique<VersionedCovMapFuncRecordReader<
829           CovMapVersion::Version4, IntPtrT, Endian>>(P, R, D, F);
830     else if (Version == CovMapVersion::Version5)
831       return std::make_unique<VersionedCovMapFuncRecordReader<
832           CovMapVersion::Version5, IntPtrT, Endian>>(P, R, D, F);
833     else if (Version == CovMapVersion::Version6)
834       return std::make_unique<VersionedCovMapFuncRecordReader<
835           CovMapVersion::Version6, IntPtrT, Endian>>(P, R, D, F);
836     else if (Version == CovMapVersion::Version7)
837       return std::make_unique<VersionedCovMapFuncRecordReader<
838           CovMapVersion::Version7, IntPtrT, Endian>>(P, R, D, F);
839   }
840   llvm_unreachable("Unsupported version");
841 }
842 
843 template <typename T, llvm::endianness Endian>
readCoverageMappingData(InstrProfSymtab & ProfileNames,StringRef CovMap,StringRef FuncRecords,std::vector<BinaryCoverageReader::ProfileMappingRecord> & Records,StringRef CompilationDir,std::vector<std::string> & Filenames)844 static Error readCoverageMappingData(
845     InstrProfSymtab &ProfileNames, StringRef CovMap, StringRef FuncRecords,
846     std::vector<BinaryCoverageReader::ProfileMappingRecord> &Records,
847     StringRef CompilationDir, std::vector<std::string> &Filenames) {
848   using namespace coverage;
849 
850   // Read the records in the coverage data section.
851   auto CovHeader =
852       reinterpret_cast<const CovMapHeader *>(CovMap.data());
853   CovMapVersion Version = (CovMapVersion)CovHeader->getVersion<Endian>();
854   if (Version > CovMapVersion::CurrentVersion)
855     return make_error<CoverageMapError>(coveragemap_error::unsupported_version);
856   Expected<std::unique_ptr<CovMapFuncRecordReader>> ReaderExpected =
857       CovMapFuncRecordReader::get<T, Endian>(Version, ProfileNames, Records,
858                                              CompilationDir, Filenames);
859   if (Error E = ReaderExpected.takeError())
860     return E;
861   auto Reader = std::move(ReaderExpected.get());
862   const char *CovBuf = CovMap.data();
863   const char *CovBufEnd = CovBuf + CovMap.size();
864   const char *FuncRecBuf = FuncRecords.data();
865   const char *FuncRecBufEnd = FuncRecords.data() + FuncRecords.size();
866   while (CovBuf < CovBufEnd) {
867     // Read the current coverage header & filename data.
868     //
869     // Prior to Version4, this also reads all function records affixed to the
870     // header.
871     //
872     // Return a pointer to the next coverage header.
873     auto NextOrErr = Reader->readCoverageHeader(CovBuf, CovBufEnd);
874     if (auto E = NextOrErr.takeError())
875       return E;
876     CovBuf = NextOrErr.get();
877   }
878   // In Version4, function records are not affixed to coverage headers. Read
879   // the records from their dedicated section.
880   if (Version >= CovMapVersion::Version4)
881     return Reader->readFunctionRecords(FuncRecBuf, FuncRecBufEnd, std::nullopt,
882                                        nullptr, nullptr);
883   return Error::success();
884 }
885 
886 Expected<std::unique_ptr<BinaryCoverageReader>>
createCoverageReaderFromBuffer(StringRef Coverage,FuncRecordsStorage && FuncRecords,InstrProfSymtab && ProfileNames,uint8_t BytesInAddress,llvm::endianness Endian,StringRef CompilationDir)887 BinaryCoverageReader::createCoverageReaderFromBuffer(
888     StringRef Coverage, FuncRecordsStorage &&FuncRecords,
889     InstrProfSymtab &&ProfileNames, uint8_t BytesInAddress,
890     llvm::endianness Endian, StringRef CompilationDir) {
891   std::unique_ptr<BinaryCoverageReader> Reader(
892       new BinaryCoverageReader(std::move(FuncRecords)));
893   Reader->ProfileNames = std::move(ProfileNames);
894   StringRef FuncRecordsRef = Reader->FuncRecords->getBuffer();
895   if (BytesInAddress == 4 && Endian == llvm::endianness::little) {
896     if (Error E = readCoverageMappingData<uint32_t, llvm::endianness::little>(
897             Reader->ProfileNames, Coverage, FuncRecordsRef,
898             Reader->MappingRecords, CompilationDir, Reader->Filenames))
899       return std::move(E);
900   } else if (BytesInAddress == 4 && Endian == llvm::endianness::big) {
901     if (Error E = readCoverageMappingData<uint32_t, llvm::endianness::big>(
902             Reader->ProfileNames, Coverage, FuncRecordsRef,
903             Reader->MappingRecords, CompilationDir, Reader->Filenames))
904       return std::move(E);
905   } else if (BytesInAddress == 8 && Endian == llvm::endianness::little) {
906     if (Error E = readCoverageMappingData<uint64_t, llvm::endianness::little>(
907             Reader->ProfileNames, Coverage, FuncRecordsRef,
908             Reader->MappingRecords, CompilationDir, Reader->Filenames))
909       return std::move(E);
910   } else if (BytesInAddress == 8 && Endian == llvm::endianness::big) {
911     if (Error E = readCoverageMappingData<uint64_t, llvm::endianness::big>(
912             Reader->ProfileNames, Coverage, FuncRecordsRef,
913             Reader->MappingRecords, CompilationDir, Reader->Filenames))
914       return std::move(E);
915   } else
916     return make_error<CoverageMapError>(
917         coveragemap_error::malformed,
918         "not supported endianness or bytes in address");
919   return std::move(Reader);
920 }
921 
922 static Expected<std::unique_ptr<BinaryCoverageReader>>
loadTestingFormat(StringRef Data,StringRef CompilationDir)923 loadTestingFormat(StringRef Data, StringRef CompilationDir) {
924   uint8_t BytesInAddress = 8;
925   llvm::endianness Endian = llvm::endianness::little;
926 
927   // Read the magic and version.
928   Data = Data.substr(sizeof(TestingFormatMagic));
929   if (Data.size() < sizeof(uint64_t))
930     return make_error<CoverageMapError>(coveragemap_error::malformed,
931                                         "the size of data is too small");
932   auto TestingVersion =
933       support::endian::byte_swap<uint64_t, llvm::endianness::little>(
934           *reinterpret_cast<const uint64_t *>(Data.data()));
935   Data = Data.substr(sizeof(uint64_t));
936 
937   // Read the ProfileNames data.
938   if (Data.empty())
939     return make_error<CoverageMapError>(coveragemap_error::truncated);
940   unsigned N = 0;
941   uint64_t ProfileNamesSize = decodeULEB128(Data.bytes_begin(), &N);
942   if (N > Data.size())
943     return make_error<CoverageMapError>(
944         coveragemap_error::malformed,
945         "the size of TestingFormatMagic is too big");
946   Data = Data.substr(N);
947   if (Data.empty())
948     return make_error<CoverageMapError>(coveragemap_error::truncated);
949   N = 0;
950   uint64_t Address = decodeULEB128(Data.bytes_begin(), &N);
951   if (N > Data.size())
952     return make_error<CoverageMapError>(coveragemap_error::malformed,
953                                         "the size of ULEB128 is too big");
954   Data = Data.substr(N);
955   if (Data.size() < ProfileNamesSize)
956     return make_error<CoverageMapError>(coveragemap_error::malformed,
957                                         "the size of ProfileNames is too big");
958   InstrProfSymtab ProfileNames;
959   if (Error E = ProfileNames.create(Data.substr(0, ProfileNamesSize), Address))
960     return std::move(E);
961   Data = Data.substr(ProfileNamesSize);
962 
963   // In Version2, the size of CoverageMapping is stored directly.
964   uint64_t CoverageMappingSize;
965   if (TestingVersion == uint64_t(TestingFormatVersion::Version2)) {
966     N = 0;
967     CoverageMappingSize = decodeULEB128(Data.bytes_begin(), &N);
968     if (N > Data.size())
969       return make_error<CoverageMapError>(coveragemap_error::malformed,
970                                           "the size of ULEB128 is too big");
971     Data = Data.substr(N);
972     if (CoverageMappingSize < sizeof(CovMapHeader))
973       return make_error<CoverageMapError>(
974           coveragemap_error::malformed,
975           "the size of CoverageMapping is teoo small");
976   } else if (TestingVersion != uint64_t(TestingFormatVersion::Version1)) {
977     return make_error<CoverageMapError>(coveragemap_error::unsupported_version);
978   }
979 
980   // Skip the padding bytes because coverage map data has an alignment of 8.
981   auto Pad = offsetToAlignedAddr(Data.data(), Align(8));
982   if (Data.size() < Pad)
983     return make_error<CoverageMapError>(coveragemap_error::malformed,
984                                         "insufficient padding");
985   Data = Data.substr(Pad);
986   if (Data.size() < sizeof(CovMapHeader))
987     return make_error<CoverageMapError>(
988         coveragemap_error::malformed,
989         "coverage mapping header section is larger than data size");
990   auto const *CovHeader = reinterpret_cast<const CovMapHeader *>(
991       Data.substr(0, sizeof(CovMapHeader)).data());
992   auto Version =
993       CovMapVersion(CovHeader->getVersion<llvm::endianness::little>());
994 
995   // In Version1, the size of CoverageMapping is calculated.
996   if (TestingVersion == uint64_t(TestingFormatVersion::Version1)) {
997     if (Version < CovMapVersion::Version4) {
998       CoverageMappingSize = Data.size();
999     } else {
1000       auto FilenamesSize =
1001           CovHeader->getFilenamesSize<llvm::endianness::little>();
1002       CoverageMappingSize = sizeof(CovMapHeader) + FilenamesSize;
1003     }
1004   }
1005 
1006   auto CoverageMapping = Data.substr(0, CoverageMappingSize);
1007   Data = Data.substr(CoverageMappingSize);
1008 
1009   // Read the CoverageRecords data.
1010   if (Version < CovMapVersion::Version4) {
1011     if (!Data.empty())
1012       return make_error<CoverageMapError>(coveragemap_error::malformed,
1013                                           "data is not empty");
1014   } else {
1015     // Skip the padding bytes because coverage records data has an alignment
1016     // of 8.
1017     Pad = offsetToAlignedAddr(Data.data(), Align(8));
1018     if (Data.size() < Pad)
1019       return make_error<CoverageMapError>(coveragemap_error::malformed,
1020                                           "insufficient padding");
1021     Data = Data.substr(Pad);
1022   }
1023   BinaryCoverageReader::FuncRecordsStorage CoverageRecords =
1024       MemoryBuffer::getMemBuffer(Data);
1025 
1026   return BinaryCoverageReader::createCoverageReaderFromBuffer(
1027       CoverageMapping, std::move(CoverageRecords), std::move(ProfileNames),
1028       BytesInAddress, Endian, CompilationDir);
1029 }
1030 
1031 /// Find all sections that match \p IPSK name. There may be more than one if
1032 /// comdats are in use, e.g. for the __llvm_covfun section on ELF.
1033 static Expected<std::vector<SectionRef>>
lookupSections(ObjectFile & OF,InstrProfSectKind IPSK)1034 lookupSections(ObjectFile &OF, InstrProfSectKind IPSK) {
1035   auto ObjFormat = OF.getTripleObjectFormat();
1036   auto Name =
1037       getInstrProfSectionName(IPSK, ObjFormat, /*AddSegmentInfo=*/false);
1038   // On COFF, the object file section name may end in "$M". This tells the
1039   // linker to sort these sections between "$A" and "$Z". The linker removes the
1040   // dollar and everything after it in the final binary. Do the same to match.
1041   bool IsCOFF = isa<COFFObjectFile>(OF);
1042   auto stripSuffix = [IsCOFF](StringRef N) {
1043     return IsCOFF ? N.split('$').first : N;
1044   };
1045   Name = stripSuffix(Name);
1046 
1047   std::vector<SectionRef> Sections;
1048   for (const auto &Section : OF.sections()) {
1049     Expected<StringRef> NameOrErr = Section.getName();
1050     if (!NameOrErr)
1051       return NameOrErr.takeError();
1052     if (stripSuffix(*NameOrErr) == Name) {
1053       // COFF profile name section contains two null bytes indicating the
1054       // start/end of the section. If its size is 2 bytes, it's empty.
1055       if (IsCOFF && IPSK == IPSK_name && Section.getSize() == 2)
1056         continue;
1057       Sections.push_back(Section);
1058     }
1059   }
1060   if (Sections.empty())
1061     return make_error<CoverageMapError>(coveragemap_error::no_data_found);
1062   return Sections;
1063 }
1064 
1065 static Expected<std::unique_ptr<BinaryCoverageReader>>
loadBinaryFormat(std::unique_ptr<Binary> Bin,StringRef Arch,StringRef CompilationDir="",object::BuildIDRef * BinaryID=nullptr)1066 loadBinaryFormat(std::unique_ptr<Binary> Bin, StringRef Arch,
1067                  StringRef CompilationDir = "",
1068                  object::BuildIDRef *BinaryID = nullptr) {
1069   std::unique_ptr<ObjectFile> OF;
1070   if (auto *Universal = dyn_cast<MachOUniversalBinary>(Bin.get())) {
1071     // If we have a universal binary, try to look up the object for the
1072     // appropriate architecture.
1073     auto ObjectFileOrErr = Universal->getMachOObjectForArch(Arch);
1074     if (!ObjectFileOrErr)
1075       return ObjectFileOrErr.takeError();
1076     OF = std::move(ObjectFileOrErr.get());
1077   } else if (isa<ObjectFile>(Bin.get())) {
1078     // For any other object file, upcast and take ownership.
1079     OF.reset(cast<ObjectFile>(Bin.release()));
1080     // If we've asked for a particular arch, make sure they match.
1081     if (!Arch.empty() && OF->getArch() != Triple(Arch).getArch())
1082       return errorCodeToError(object_error::arch_not_found);
1083   } else
1084     // We can only handle object files.
1085     return make_error<CoverageMapError>(coveragemap_error::malformed,
1086                                         "binary is not an object file");
1087 
1088   // The coverage uses native pointer sizes for the object it's written in.
1089   uint8_t BytesInAddress = OF->getBytesInAddress();
1090   llvm::endianness Endian =
1091       OF->isLittleEndian() ? llvm::endianness::little : llvm::endianness::big;
1092 
1093   // Look for the sections that we are interested in.
1094   InstrProfSymtab ProfileNames;
1095   std::vector<SectionRef> NamesSectionRefs;
1096   // If IPSK_name is not found, fallback to search for IPK_covname, which is
1097   // used when binary correlation is enabled.
1098   auto NamesSection = lookupSections(*OF, IPSK_name);
1099   if (auto E = NamesSection.takeError()) {
1100     consumeError(std::move(E));
1101     NamesSection = lookupSections(*OF, IPSK_covname);
1102     if (auto E = NamesSection.takeError())
1103       return std::move(E);
1104   }
1105   NamesSectionRefs = *NamesSection;
1106 
1107   if (NamesSectionRefs.size() != 1)
1108     return make_error<CoverageMapError>(
1109         coveragemap_error::malformed,
1110         "the size of coverage mapping section is not one");
1111   if (Error E = ProfileNames.create(NamesSectionRefs.back()))
1112     return std::move(E);
1113 
1114   auto CoverageSection = lookupSections(*OF, IPSK_covmap);
1115   if (auto E = CoverageSection.takeError())
1116     return std::move(E);
1117   std::vector<SectionRef> CoverageSectionRefs = *CoverageSection;
1118   if (CoverageSectionRefs.size() != 1)
1119     return make_error<CoverageMapError>(coveragemap_error::malformed,
1120                                         "the size of name section is not one");
1121   auto CoverageMappingOrErr = CoverageSectionRefs.back().getContents();
1122   if (!CoverageMappingOrErr)
1123     return CoverageMappingOrErr.takeError();
1124   StringRef CoverageMapping = CoverageMappingOrErr.get();
1125 
1126   // Look for the coverage records section (Version4 only).
1127   auto CoverageRecordsSections = lookupSections(*OF, IPSK_covfun);
1128 
1129   BinaryCoverageReader::FuncRecordsStorage FuncRecords;
1130   if (auto E = CoverageRecordsSections.takeError()) {
1131     consumeError(std::move(E));
1132     FuncRecords = MemoryBuffer::getMemBuffer("");
1133   } else {
1134     // Compute the FuncRecordsBuffer of the buffer, taking into account the
1135     // padding between each record, and making sure the first block is aligned
1136     // in memory to maintain consistency between buffer address and size
1137     // alignment.
1138     const Align RecordAlignment(8);
1139     uint64_t FuncRecordsSize = 0;
1140     for (SectionRef Section : *CoverageRecordsSections) {
1141       auto CoverageRecordsOrErr = Section.getContents();
1142       if (!CoverageRecordsOrErr)
1143         return CoverageRecordsOrErr.takeError();
1144       FuncRecordsSize += alignTo(CoverageRecordsOrErr->size(), RecordAlignment);
1145     }
1146     auto WritableBuffer =
1147         WritableMemoryBuffer::getNewUninitMemBuffer(FuncRecordsSize);
1148     char *FuncRecordsBuffer = WritableBuffer->getBufferStart();
1149     assert(isAddrAligned(RecordAlignment, FuncRecordsBuffer) &&
1150            "Allocated memory is correctly aligned");
1151 
1152     for (SectionRef Section : *CoverageRecordsSections) {
1153       auto CoverageRecordsOrErr = Section.getContents();
1154       if (!CoverageRecordsOrErr)
1155         return CoverageRecordsOrErr.takeError();
1156       const auto &CoverageRecords = CoverageRecordsOrErr.get();
1157       FuncRecordsBuffer = std::copy(CoverageRecords.begin(),
1158                                     CoverageRecords.end(), FuncRecordsBuffer);
1159       FuncRecordsBuffer =
1160           std::fill_n(FuncRecordsBuffer,
1161                       alignAddr(FuncRecordsBuffer, RecordAlignment) -
1162                           (uintptr_t)FuncRecordsBuffer,
1163                       '\0');
1164     }
1165     assert(FuncRecordsBuffer == WritableBuffer->getBufferEnd() &&
1166            "consistent init");
1167     FuncRecords = std::move(WritableBuffer);
1168   }
1169 
1170   if (BinaryID)
1171     *BinaryID = getBuildID(OF.get());
1172 
1173   return BinaryCoverageReader::createCoverageReaderFromBuffer(
1174       CoverageMapping, std::move(FuncRecords), std::move(ProfileNames),
1175       BytesInAddress, Endian, CompilationDir);
1176 }
1177 
1178 /// Determine whether \p Arch is invalid or empty, given \p Bin.
isArchSpecifierInvalidOrMissing(Binary * Bin,StringRef Arch)1179 static bool isArchSpecifierInvalidOrMissing(Binary *Bin, StringRef Arch) {
1180   // If we have a universal binary and Arch doesn't identify any of its slices,
1181   // it's user error.
1182   if (auto *Universal = dyn_cast<MachOUniversalBinary>(Bin)) {
1183     for (auto &ObjForArch : Universal->objects())
1184       if (Arch == ObjForArch.getArchFlagName())
1185         return false;
1186     return true;
1187   }
1188   return false;
1189 }
1190 
1191 Expected<std::vector<std::unique_ptr<BinaryCoverageReader>>>
create(MemoryBufferRef ObjectBuffer,StringRef Arch,SmallVectorImpl<std::unique_ptr<MemoryBuffer>> & ObjectFileBuffers,StringRef CompilationDir,SmallVectorImpl<object::BuildIDRef> * BinaryIDs)1192 BinaryCoverageReader::create(
1193     MemoryBufferRef ObjectBuffer, StringRef Arch,
1194     SmallVectorImpl<std::unique_ptr<MemoryBuffer>> &ObjectFileBuffers,
1195     StringRef CompilationDir, SmallVectorImpl<object::BuildIDRef> *BinaryIDs) {
1196   std::vector<std::unique_ptr<BinaryCoverageReader>> Readers;
1197 
1198   if (ObjectBuffer.getBuffer().size() > sizeof(TestingFormatMagic)) {
1199     uint64_t Magic =
1200         support::endian::byte_swap<uint64_t, llvm::endianness::little>(
1201             *reinterpret_cast<const uint64_t *>(ObjectBuffer.getBufferStart()));
1202     if (Magic == TestingFormatMagic) {
1203       // This is a special format used for testing.
1204       auto ReaderOrErr =
1205           loadTestingFormat(ObjectBuffer.getBuffer(), CompilationDir);
1206       if (!ReaderOrErr)
1207         return ReaderOrErr.takeError();
1208       Readers.push_back(std::move(ReaderOrErr.get()));
1209       return std::move(Readers);
1210     }
1211   }
1212 
1213   auto BinOrErr = createBinary(ObjectBuffer);
1214   if (!BinOrErr)
1215     return BinOrErr.takeError();
1216   std::unique_ptr<Binary> Bin = std::move(BinOrErr.get());
1217 
1218   if (isArchSpecifierInvalidOrMissing(Bin.get(), Arch))
1219     return make_error<CoverageMapError>(
1220         coveragemap_error::invalid_or_missing_arch_specifier);
1221 
1222   // MachO universal binaries which contain archives need to be treated as
1223   // archives, not as regular binaries.
1224   if (auto *Universal = dyn_cast<MachOUniversalBinary>(Bin.get())) {
1225     for (auto &ObjForArch : Universal->objects()) {
1226       // Skip slices within the universal binary which target the wrong arch.
1227       std::string ObjArch = ObjForArch.getArchFlagName();
1228       if (Arch != ObjArch)
1229         continue;
1230 
1231       auto ArchiveOrErr = ObjForArch.getAsArchive();
1232       if (!ArchiveOrErr) {
1233         // If this is not an archive, try treating it as a regular object.
1234         consumeError(ArchiveOrErr.takeError());
1235         break;
1236       }
1237 
1238       return BinaryCoverageReader::create(
1239           ArchiveOrErr.get()->getMemoryBufferRef(), Arch, ObjectFileBuffers,
1240           CompilationDir, BinaryIDs);
1241     }
1242   }
1243 
1244   // Load coverage out of archive members.
1245   if (auto *Ar = dyn_cast<Archive>(Bin.get())) {
1246     Error Err = Error::success();
1247     for (auto &Child : Ar->children(Err)) {
1248       Expected<MemoryBufferRef> ChildBufOrErr = Child.getMemoryBufferRef();
1249       if (!ChildBufOrErr)
1250         return ChildBufOrErr.takeError();
1251 
1252       auto ChildReadersOrErr = BinaryCoverageReader::create(
1253           ChildBufOrErr.get(), Arch, ObjectFileBuffers, CompilationDir,
1254           BinaryIDs);
1255       if (!ChildReadersOrErr)
1256         return ChildReadersOrErr.takeError();
1257       for (auto &Reader : ChildReadersOrErr.get())
1258         Readers.push_back(std::move(Reader));
1259     }
1260     if (Err)
1261       return std::move(Err);
1262 
1263     // Thin archives reference object files outside of the archive file, i.e.
1264     // files which reside in memory not owned by the caller. Transfer ownership
1265     // to the caller.
1266     if (Ar->isThin())
1267       for (auto &Buffer : Ar->takeThinBuffers())
1268         ObjectFileBuffers.push_back(std::move(Buffer));
1269 
1270     return std::move(Readers);
1271   }
1272 
1273   object::BuildIDRef BinaryID;
1274   auto ReaderOrErr = loadBinaryFormat(std::move(Bin), Arch, CompilationDir,
1275                                       BinaryIDs ? &BinaryID : nullptr);
1276   if (!ReaderOrErr)
1277     return ReaderOrErr.takeError();
1278   Readers.push_back(std::move(ReaderOrErr.get()));
1279   if (!BinaryID.empty())
1280     BinaryIDs->push_back(BinaryID);
1281   return std::move(Readers);
1282 }
1283 
readNextRecord(CoverageMappingRecord & Record)1284 Error BinaryCoverageReader::readNextRecord(CoverageMappingRecord &Record) {
1285   if (CurrentRecord >= MappingRecords.size())
1286     return make_error<CoverageMapError>(coveragemap_error::eof);
1287 
1288   FunctionsFilenames.clear();
1289   Expressions.clear();
1290   MappingRegions.clear();
1291   auto &R = MappingRecords[CurrentRecord];
1292   auto F = ArrayRef(Filenames).slice(R.FilenamesBegin, R.FilenamesSize);
1293   RawCoverageMappingReader Reader(R.CoverageMapping, F, FunctionsFilenames,
1294                                   Expressions, MappingRegions);
1295   if (auto Err = Reader.read())
1296     return Err;
1297 
1298   Record.FunctionName = R.FunctionName;
1299   Record.FunctionHash = R.FunctionHash;
1300   Record.Filenames = FunctionsFilenames;
1301   Record.Expressions = Expressions;
1302   Record.MappingRegions = MappingRegions;
1303 
1304   ++CurrentRecord;
1305   return Error::success();
1306 }
1307