1 //=-- InstrProfReader.cpp - Instrumented profiling reader -------------------=//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file contains support for reading profiling data for clang's
11 // instrumentation based PGO and coverage.
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #include "llvm/ProfileData/InstrProfReader.h"
16 #include "llvm/ProfileData/InstrProf.h"
17 
18 #include "InstrProfIndexed.h"
19 
20 #include <cassert>
21 
22 using namespace llvm;
23 
24 static error_code setupMemoryBuffer(std::string Path,
25                                     std::unique_ptr<MemoryBuffer> &Buffer) {
26   if (error_code EC = MemoryBuffer::getFileOrSTDIN(Path, Buffer))
27     return EC;
28 
29   // Sanity check the file.
30   if (Buffer->getBufferSize() > std::numeric_limits<unsigned>::max())
31     return instrprof_error::too_large;
32   return instrprof_error::success;
33 }
34 
35 static error_code initializeReader(InstrProfReader &Reader) {
36   return Reader.readHeader();
37 }
38 
39 error_code InstrProfReader::create(std::string Path,
40                                    std::unique_ptr<InstrProfReader> &Result) {
41   // Set up the buffer to read.
42   std::unique_ptr<MemoryBuffer> Buffer;
43   if (error_code EC = setupMemoryBuffer(Path, Buffer))
44     return EC;
45 
46   // Create the reader.
47   if (IndexedInstrProfReader::hasFormat(*Buffer))
48     Result.reset(new IndexedInstrProfReader(std::move(Buffer)));
49   else if (RawInstrProfReader64::hasFormat(*Buffer))
50     Result.reset(new RawInstrProfReader64(std::move(Buffer)));
51   else if (RawInstrProfReader32::hasFormat(*Buffer))
52     Result.reset(new RawInstrProfReader32(std::move(Buffer)));
53   else
54     Result.reset(new TextInstrProfReader(std::move(Buffer)));
55 
56   // Initialize the reader and return the result.
57   return initializeReader(*Result);
58 }
59 
60 error_code IndexedInstrProfReader::create(
61     std::string Path, std::unique_ptr<IndexedInstrProfReader> &Result) {
62   // Set up the buffer to read.
63   std::unique_ptr<MemoryBuffer> Buffer;
64   if (error_code EC = setupMemoryBuffer(Path, Buffer))
65     return EC;
66 
67   // Create the reader.
68   if (!IndexedInstrProfReader::hasFormat(*Buffer))
69     return instrprof_error::bad_magic;
70   Result.reset(new IndexedInstrProfReader(std::move(Buffer)));
71 
72   // Initialize the reader and return the result.
73   return initializeReader(*Result);
74 }
75 
76 void InstrProfIterator::Increment() {
77   if (Reader->readNextRecord(Record))
78     *this = InstrProfIterator();
79 }
80 
81 error_code TextInstrProfReader::readNextRecord(InstrProfRecord &Record) {
82   // Skip empty lines.
83   while (!Line.is_at_end() && Line->empty())
84     ++Line;
85   // If we hit EOF while looking for a name, we're done.
86   if (Line.is_at_end())
87     return error(instrprof_error::eof);
88 
89   // Read the function name.
90   Record.Name = *Line++;
91 
92   // Read the function hash.
93   if (Line.is_at_end())
94     return error(instrprof_error::truncated);
95   if ((Line++)->getAsInteger(10, Record.Hash))
96     return error(instrprof_error::malformed);
97 
98   // Read the number of counters.
99   uint64_t NumCounters;
100   if (Line.is_at_end())
101     return error(instrprof_error::truncated);
102   if ((Line++)->getAsInteger(10, NumCounters))
103     return error(instrprof_error::malformed);
104 
105   // Read each counter and fill our internal storage with the values.
106   Counts.clear();
107   Counts.reserve(NumCounters);
108   for (uint64_t I = 0; I < NumCounters; ++I) {
109     if (Line.is_at_end())
110       return error(instrprof_error::truncated);
111     uint64_t Count;
112     if ((Line++)->getAsInteger(10, Count))
113       return error(instrprof_error::malformed);
114     Counts.push_back(Count);
115   }
116   // Give the record a reference to our internal counter storage.
117   Record.Counts = Counts;
118 
119   return success();
120 }
121 
122 template <class IntPtrT>
123 static uint64_t getRawMagic();
124 
125 template <>
126 uint64_t getRawMagic<uint64_t>() {
127   return
128     uint64_t(255) << 56 |
129     uint64_t('l') << 48 |
130     uint64_t('p') << 40 |
131     uint64_t('r') << 32 |
132     uint64_t('o') << 24 |
133     uint64_t('f') << 16 |
134     uint64_t('r') <<  8 |
135     uint64_t(129);
136 }
137 
138 template <>
139 uint64_t getRawMagic<uint32_t>() {
140   return
141     uint64_t(255) << 56 |
142     uint64_t('l') << 48 |
143     uint64_t('p') << 40 |
144     uint64_t('r') << 32 |
145     uint64_t('o') << 24 |
146     uint64_t('f') << 16 |
147     uint64_t('R') <<  8 |
148     uint64_t(129);
149 }
150 
151 template <class IntPtrT>
152 bool RawInstrProfReader<IntPtrT>::hasFormat(const MemoryBuffer &DataBuffer) {
153   if (DataBuffer.getBufferSize() < sizeof(uint64_t))
154     return false;
155   uint64_t Magic =
156     *reinterpret_cast<const uint64_t *>(DataBuffer.getBufferStart());
157   return getRawMagic<IntPtrT>() == Magic ||
158     sys::SwapByteOrder(getRawMagic<IntPtrT>()) == Magic;
159 }
160 
161 template <class IntPtrT>
162 error_code RawInstrProfReader<IntPtrT>::readHeader() {
163   if (!hasFormat(*DataBuffer))
164     return error(instrprof_error::bad_magic);
165   if (DataBuffer->getBufferSize() < sizeof(RawHeader))
166     return error(instrprof_error::bad_header);
167   auto *Header =
168     reinterpret_cast<const RawHeader *>(DataBuffer->getBufferStart());
169   ShouldSwapBytes = Header->Magic != getRawMagic<IntPtrT>();
170   return readHeader(*Header);
171 }
172 
173 static uint64_t getRawVersion() {
174   return 1;
175 }
176 
177 template <class IntPtrT>
178 error_code RawInstrProfReader<IntPtrT>::readHeader(const RawHeader &Header) {
179   if (swap(Header.Version) != getRawVersion())
180     return error(instrprof_error::unsupported_version);
181 
182   CountersDelta = swap(Header.CountersDelta);
183   NamesDelta = swap(Header.NamesDelta);
184   auto DataSize = swap(Header.DataSize);
185   auto CountersSize = swap(Header.CountersSize);
186   auto NamesSize = swap(Header.NamesSize);
187 
188   ptrdiff_t DataOffset = sizeof(RawHeader);
189   ptrdiff_t CountersOffset = DataOffset + sizeof(ProfileData) * DataSize;
190   ptrdiff_t NamesOffset = CountersOffset + sizeof(uint64_t) * CountersSize;
191   size_t FileSize = NamesOffset + sizeof(char) * NamesSize;
192 
193   if (FileSize != DataBuffer->getBufferSize())
194     return error(instrprof_error::bad_header);
195 
196   const char *Start = DataBuffer->getBufferStart();
197   Data = reinterpret_cast<const ProfileData *>(Start + DataOffset);
198   DataEnd = Data + DataSize;
199   CountersStart = reinterpret_cast<const uint64_t *>(Start + CountersOffset);
200   NamesStart = Start + NamesOffset;
201 
202   return success();
203 }
204 
205 template <class IntPtrT>
206 error_code
207 RawInstrProfReader<IntPtrT>::readNextRecord(InstrProfRecord &Record) {
208   if (Data == DataEnd)
209     return error(instrprof_error::eof);
210 
211   // Get the raw data.
212   StringRef RawName(getName(Data->NamePtr), swap(Data->NameSize));
213   auto RawCounts = makeArrayRef(getCounter(Data->CounterPtr),
214                                 swap(Data->NumCounters));
215 
216   // Check bounds.
217   auto *NamesStartAsCounter = reinterpret_cast<const uint64_t *>(NamesStart);
218   if (RawName.data() < NamesStart ||
219       RawName.data() + RawName.size() > DataBuffer->getBufferEnd() ||
220       RawCounts.data() < CountersStart ||
221       RawCounts.data() + RawCounts.size() > NamesStartAsCounter)
222     return error(instrprof_error::malformed);
223 
224   // Store the data in Record, byte-swapping as necessary.
225   Record.Hash = swap(Data->FuncHash);
226   Record.Name = RawName;
227   if (ShouldSwapBytes) {
228     Counts.clear();
229     Counts.reserve(RawCounts.size());
230     for (uint64_t Count : RawCounts)
231       Counts.push_back(swap(Count));
232     Record.Counts = Counts;
233   } else
234     Record.Counts = RawCounts;
235 
236   // Iterate.
237   ++Data;
238   return success();
239 }
240 
241 namespace llvm {
242 template class RawInstrProfReader<uint32_t>;
243 template class RawInstrProfReader<uint64_t>;
244 }
245 
246 InstrProfLookupTrait::hash_value_type
247 InstrProfLookupTrait::ComputeHash(StringRef K) {
248   return IndexedInstrProf::ComputeHash(HashType, K);
249 }
250 
251 bool IndexedInstrProfReader::hasFormat(const MemoryBuffer &DataBuffer) {
252   if (DataBuffer.getBufferSize() < 8)
253     return false;
254   using namespace support;
255   uint64_t Magic =
256       endian::read<uint64_t, little, aligned>(DataBuffer.getBufferStart());
257   return Magic == IndexedInstrProf::Magic;
258 }
259 
260 error_code IndexedInstrProfReader::readHeader() {
261   const unsigned char *Start = (unsigned char *)DataBuffer->getBufferStart();
262   const unsigned char *Cur = Start;
263   if ((unsigned char *)DataBuffer->getBufferEnd() - Cur < 24)
264     return error(instrprof_error::truncated);
265 
266   using namespace support;
267 
268   // Check the magic number.
269   uint64_t Magic = endian::readNext<uint64_t, little, unaligned>(Cur);
270   if (Magic != IndexedInstrProf::Magic)
271     return error(instrprof_error::bad_magic);
272 
273   // Read the version.
274   uint64_t Version = endian::readNext<uint64_t, little, unaligned>(Cur);
275   if (Version != IndexedInstrProf::Version)
276     return error(instrprof_error::unsupported_version);
277 
278   // Read the maximal function count.
279   MaxFunctionCount = endian::readNext<uint64_t, little, unaligned>(Cur);
280 
281   // Read the hash type and start offset.
282   IndexedInstrProf::HashT HashType = static_cast<IndexedInstrProf::HashT>(
283       endian::readNext<uint64_t, little, unaligned>(Cur));
284   if (HashType > IndexedInstrProf::HashT::Last)
285     return error(instrprof_error::unsupported_hash_type);
286   uint64_t HashOffset = endian::readNext<uint64_t, little, unaligned>(Cur);
287 
288   // The rest of the file is an on disk hash table.
289   Index.reset(InstrProfReaderIndex::Create(Start + HashOffset, Cur, Start,
290                                            InstrProfLookupTrait(HashType)));
291   // Set up our iterator for readNextRecord.
292   RecordIterator = Index->data_begin();
293 
294   return success();
295 }
296 
297 error_code IndexedInstrProfReader::getFunctionCounts(
298     StringRef FuncName, uint64_t &FuncHash, std::vector<uint64_t> &Counts) {
299   const auto &Iter = Index->find(FuncName);
300   if (Iter == Index->end())
301     return error(instrprof_error::unknown_function);
302 
303   // Found it. Make sure it's valid before giving back a result.
304   const InstrProfRecord &Record = *Iter;
305   if (Record.Name.empty())
306     return error(instrprof_error::malformed);
307   FuncHash = Record.Hash;
308   Counts = Record.Counts;
309   return success();
310 }
311 
312 error_code IndexedInstrProfReader::readNextRecord(InstrProfRecord &Record) {
313   // Are we out of records?
314   if (RecordIterator == Index->data_end())
315     return error(instrprof_error::eof);
316 
317   // Read the next one.
318   Record = *RecordIterator;
319   ++RecordIterator;
320   if (Record.Name.empty())
321     return error(instrprof_error::malformed);
322   return success();
323 }
324