1 //===- Trace.cpp - XRay Trace Loading implementation. ---------------------===//
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 // XRay log reader implementation.
10 //
11 //===----------------------------------------------------------------------===//
12 #include "llvm/XRay/Trace.h"
13 #include "llvm/ADT/STLExtras.h"
14 #include "llvm/Support/DataExtractor.h"
15 #include "llvm/Support/Error.h"
16 #include "llvm/Support/FileSystem.h"
17 #include "llvm/XRay/BlockIndexer.h"
18 #include "llvm/XRay/BlockVerifier.h"
19 #include "llvm/XRay/FDRRecordConsumer.h"
20 #include "llvm/XRay/FDRRecordProducer.h"
21 #include "llvm/XRay/FDRRecords.h"
22 #include "llvm/XRay/FDRTraceExpander.h"
23 #include "llvm/XRay/FileHeaderReader.h"
24 #include "llvm/XRay/YAMLXRayRecord.h"
25 #include <memory>
26 #include <vector>
27
28 using namespace llvm;
29 using namespace llvm::xray;
30 using llvm::yaml::Input;
31
32 namespace {
33 using XRayRecordStorage =
34 std::aligned_storage<sizeof(XRayRecord), alignof(XRayRecord)>::type;
35
loadNaiveFormatLog(StringRef Data,bool IsLittleEndian,XRayFileHeader & FileHeader,std::vector<XRayRecord> & Records)36 Error loadNaiveFormatLog(StringRef Data, bool IsLittleEndian,
37 XRayFileHeader &FileHeader,
38 std::vector<XRayRecord> &Records) {
39 if (Data.size() < 32)
40 return make_error<StringError>(
41 "Not enough bytes for an XRay log.",
42 std::make_error_code(std::errc::invalid_argument));
43
44 if (Data.size() - 32 == 0 || Data.size() % 32 != 0)
45 return make_error<StringError>(
46 "Invalid-sized XRay data.",
47 std::make_error_code(std::errc::invalid_argument));
48
49 DataExtractor Reader(Data, IsLittleEndian, 8);
50 uint64_t OffsetPtr = 0;
51 auto FileHeaderOrError = readBinaryFormatHeader(Reader, OffsetPtr);
52 if (!FileHeaderOrError)
53 return FileHeaderOrError.takeError();
54 FileHeader = std::move(FileHeaderOrError.get());
55
56 // Each record after the header will be 32 bytes, in the following format:
57 //
58 // (2) uint16 : record type
59 // (1) uint8 : cpu id
60 // (1) uint8 : type
61 // (4) sint32 : function id
62 // (8) uint64 : tsc
63 // (4) uint32 : thread id
64 // (4) uint32 : process id
65 // (8) - : padding
66 while (Reader.isValidOffset(OffsetPtr)) {
67 if (!Reader.isValidOffsetForDataOfSize(OffsetPtr, 32))
68 return createStringError(
69 std::make_error_code(std::errc::executable_format_error),
70 "Not enough bytes to read a full record at offset %" PRId64 ".",
71 OffsetPtr);
72 auto PreReadOffset = OffsetPtr;
73 auto RecordType = Reader.getU16(&OffsetPtr);
74 if (OffsetPtr == PreReadOffset)
75 return createStringError(
76 std::make_error_code(std::errc::executable_format_error),
77 "Failed reading record type at offset %" PRId64 ".", OffsetPtr);
78
79 switch (RecordType) {
80 case 0: { // Normal records.
81 Records.emplace_back();
82 auto &Record = Records.back();
83 Record.RecordType = RecordType;
84
85 PreReadOffset = OffsetPtr;
86 Record.CPU = Reader.getU8(&OffsetPtr);
87 if (OffsetPtr == PreReadOffset)
88 return createStringError(
89 std::make_error_code(std::errc::executable_format_error),
90 "Failed reading CPU field at offset %" PRId64 ".", OffsetPtr);
91
92 PreReadOffset = OffsetPtr;
93 auto Type = Reader.getU8(&OffsetPtr);
94 if (OffsetPtr == PreReadOffset)
95 return createStringError(
96 std::make_error_code(std::errc::executable_format_error),
97 "Failed reading record type field at offset %" PRId64 ".",
98 OffsetPtr);
99
100 switch (Type) {
101 case 0:
102 Record.Type = RecordTypes::ENTER;
103 break;
104 case 1:
105 Record.Type = RecordTypes::EXIT;
106 break;
107 case 2:
108 Record.Type = RecordTypes::TAIL_EXIT;
109 break;
110 case 3:
111 Record.Type = RecordTypes::ENTER_ARG;
112 break;
113 default:
114 return createStringError(
115 std::make_error_code(std::errc::executable_format_error),
116 "Unknown record type '%d' at offset %" PRId64 ".", Type, OffsetPtr);
117 }
118
119 PreReadOffset = OffsetPtr;
120 Record.FuncId = Reader.getSigned(&OffsetPtr, sizeof(int32_t));
121 if (OffsetPtr == PreReadOffset)
122 return createStringError(
123 std::make_error_code(std::errc::executable_format_error),
124 "Failed reading function id field at offset %" PRId64 ".",
125 OffsetPtr);
126
127 PreReadOffset = OffsetPtr;
128 Record.TSC = Reader.getU64(&OffsetPtr);
129 if (OffsetPtr == PreReadOffset)
130 return createStringError(
131 std::make_error_code(std::errc::executable_format_error),
132 "Failed reading TSC field at offset %" PRId64 ".", OffsetPtr);
133
134 PreReadOffset = OffsetPtr;
135 Record.TId = Reader.getU32(&OffsetPtr);
136 if (OffsetPtr == PreReadOffset)
137 return createStringError(
138 std::make_error_code(std::errc::executable_format_error),
139 "Failed reading thread id field at offset %" PRId64 ".", OffsetPtr);
140
141 PreReadOffset = OffsetPtr;
142 Record.PId = Reader.getU32(&OffsetPtr);
143 if (OffsetPtr == PreReadOffset)
144 return createStringError(
145 std::make_error_code(std::errc::executable_format_error),
146 "Failed reading process id at offset %" PRId64 ".", OffsetPtr);
147
148 break;
149 }
150 case 1: { // Arg payload record.
151 auto &Record = Records.back();
152
153 // We skip the next two bytes of the record, because we don't need the
154 // type and the CPU record for arg payloads.
155 OffsetPtr += 2;
156 PreReadOffset = OffsetPtr;
157 int32_t FuncId = Reader.getSigned(&OffsetPtr, sizeof(int32_t));
158 if (OffsetPtr == PreReadOffset)
159 return createStringError(
160 std::make_error_code(std::errc::executable_format_error),
161 "Failed reading function id field at offset %" PRId64 ".",
162 OffsetPtr);
163
164 PreReadOffset = OffsetPtr;
165 auto TId = Reader.getU32(&OffsetPtr);
166 if (OffsetPtr == PreReadOffset)
167 return createStringError(
168 std::make_error_code(std::errc::executable_format_error),
169 "Failed reading thread id field at offset %" PRId64 ".", OffsetPtr);
170
171 PreReadOffset = OffsetPtr;
172 auto PId = Reader.getU32(&OffsetPtr);
173 if (OffsetPtr == PreReadOffset)
174 return createStringError(
175 std::make_error_code(std::errc::executable_format_error),
176 "Failed reading process id field at offset %" PRId64 ".",
177 OffsetPtr);
178
179 // Make a check for versions above 3 for the Pid field
180 if (Record.FuncId != FuncId || Record.TId != TId ||
181 (FileHeader.Version >= 3 ? Record.PId != PId : false))
182 return createStringError(
183 std::make_error_code(std::errc::executable_format_error),
184 "Corrupted log, found arg payload following non-matching "
185 "function+thread record. Record for function %d != %d at offset "
186 "%" PRId64 ".",
187 Record.FuncId, FuncId, OffsetPtr);
188
189 PreReadOffset = OffsetPtr;
190 auto Arg = Reader.getU64(&OffsetPtr);
191 if (OffsetPtr == PreReadOffset)
192 return createStringError(
193 std::make_error_code(std::errc::executable_format_error),
194 "Failed reading argument payload at offset %" PRId64 ".",
195 OffsetPtr);
196
197 Record.CallArgs.push_back(Arg);
198 break;
199 }
200 default:
201 return createStringError(
202 std::make_error_code(std::errc::executable_format_error),
203 "Unknown record type '%d' at offset %" PRId64 ".", RecordType,
204 OffsetPtr);
205 }
206 // Advance the offset pointer enough bytes to align to 32-byte records for
207 // basic mode logs.
208 OffsetPtr += 8;
209 }
210 return Error::success();
211 }
212
213 /// Reads a log in FDR mode for version 1 of this binary format. FDR mode is
214 /// defined as part of the compiler-rt project in xray_fdr_logging.h, and such
215 /// a log consists of the familiar 32 bit XRayHeader, followed by sequences of
216 /// of interspersed 16 byte Metadata Records and 8 byte Function Records.
217 ///
218 /// The following is an attempt to document the grammar of the format, which is
219 /// parsed by this function for little-endian machines. Since the format makes
220 /// use of BitFields, when we support big-endian architectures, we will need to
221 /// adjust not only the endianness parameter to llvm's RecordExtractor, but also
222 /// the bit twiddling logic, which is consistent with the little-endian
223 /// convention that BitFields within a struct will first be packed into the
224 /// least significant bits the address they belong to.
225 ///
226 /// We expect a format complying with the grammar in the following pseudo-EBNF
227 /// in Version 1 of the FDR log.
228 ///
229 /// FDRLog: XRayFileHeader ThreadBuffer*
230 /// XRayFileHeader: 32 bytes to identify the log as FDR with machine metadata.
231 /// Includes BufferSize
232 /// ThreadBuffer: NewBuffer WallClockTime NewCPUId FunctionSequence EOB
233 /// BufSize: 8 byte unsigned integer indicating how large the buffer is.
234 /// NewBuffer: 16 byte metadata record with Thread Id.
235 /// WallClockTime: 16 byte metadata record with human readable time.
236 /// Pid: 16 byte metadata record with Pid
237 /// NewCPUId: 16 byte metadata record with CPUId and a 64 bit TSC reading.
238 /// EOB: 16 byte record in a thread buffer plus mem garbage to fill BufSize.
239 /// FunctionSequence: NewCPUId | TSCWrap | FunctionRecord
240 /// TSCWrap: 16 byte metadata record with a full 64 bit TSC reading.
241 /// FunctionRecord: 8 byte record with FunctionId, entry/exit, and TSC delta.
242 ///
243 /// In Version 2, we make the following changes:
244 ///
245 /// ThreadBuffer: BufferExtents NewBuffer WallClockTime NewCPUId
246 /// FunctionSequence
247 /// BufferExtents: 16 byte metdata record describing how many usable bytes are
248 /// in the buffer. This is measured from the start of the buffer
249 /// and must always be at least 48 (bytes).
250 ///
251 /// In Version 3, we make the following changes:
252 ///
253 /// ThreadBuffer: BufferExtents NewBuffer WallClockTime Pid NewCPUId
254 /// FunctionSequence
255 /// EOB: *deprecated*
256 ///
257 /// In Version 4, we make the following changes:
258 ///
259 /// CustomEventRecord now includes the CPU data.
260 ///
261 /// In Version 5, we make the following changes:
262 ///
263 /// CustomEventRecord and TypedEventRecord now use TSC delta encoding similar to
264 /// what FunctionRecord instances use, and we no longer need to include the CPU
265 /// id in the CustomEventRecord.
266 ///
loadFDRLog(StringRef Data,bool IsLittleEndian,XRayFileHeader & FileHeader,std::vector<XRayRecord> & Records)267 Error loadFDRLog(StringRef Data, bool IsLittleEndian,
268 XRayFileHeader &FileHeader, std::vector<XRayRecord> &Records) {
269
270 if (Data.size() < 32)
271 return createStringError(std::make_error_code(std::errc::invalid_argument),
272 "Not enough bytes for an XRay FDR log.");
273 DataExtractor DE(Data, IsLittleEndian, 8);
274
275 uint64_t OffsetPtr = 0;
276 auto FileHeaderOrError = readBinaryFormatHeader(DE, OffsetPtr);
277 if (!FileHeaderOrError)
278 return FileHeaderOrError.takeError();
279 FileHeader = std::move(FileHeaderOrError.get());
280
281 // First we load the records into memory.
282 std::vector<std::unique_ptr<Record>> FDRRecords;
283
284 {
285 FileBasedRecordProducer P(FileHeader, DE, OffsetPtr);
286 LogBuilderConsumer C(FDRRecords);
287 while (DE.isValidOffsetForDataOfSize(OffsetPtr, 1)) {
288 auto R = P.produce();
289 if (!R)
290 return R.takeError();
291 if (auto E = C.consume(std::move(R.get())))
292 return E;
293 }
294 }
295
296 // Next we index the records into blocks.
297 BlockIndexer::Index Index;
298 {
299 BlockIndexer Indexer(Index);
300 for (auto &R : FDRRecords)
301 if (auto E = R->apply(Indexer))
302 return E;
303 if (auto E = Indexer.flush())
304 return E;
305 }
306
307 // Then we verify the consistency of the blocks.
308 {
309 for (auto &PTB : Index) {
310 auto &Blocks = PTB.second;
311 for (auto &B : Blocks) {
312 BlockVerifier Verifier;
313 for (auto *R : B.Records)
314 if (auto E = R->apply(Verifier))
315 return E;
316 if (auto E = Verifier.verify())
317 return E;
318 }
319 }
320 }
321
322 // This is now the meat of the algorithm. Here we sort the blocks according to
323 // the Walltime record in each of the blocks for the same thread. This allows
324 // us to more consistently recreate the execution trace in temporal order.
325 // After the sort, we then reconstitute `Trace` records using a stateful
326 // visitor associated with a single process+thread pair.
327 {
328 for (auto &PTB : Index) {
329 auto &Blocks = PTB.second;
330 llvm::sort(Blocks, [](const BlockIndexer::Block &L,
331 const BlockIndexer::Block &R) {
332 return (L.WallclockTime->seconds() < R.WallclockTime->seconds() &&
333 L.WallclockTime->nanos() < R.WallclockTime->nanos());
334 });
335 auto Adder = [&](const XRayRecord &R) { Records.push_back(R); };
336 TraceExpander Expander(Adder, FileHeader.Version);
337 for (auto &B : Blocks) {
338 for (auto *R : B.Records)
339 if (auto E = R->apply(Expander))
340 return E;
341 }
342 if (auto E = Expander.flush())
343 return E;
344 }
345 }
346
347 return Error::success();
348 }
349
loadYAMLLog(StringRef Data,XRayFileHeader & FileHeader,std::vector<XRayRecord> & Records)350 Error loadYAMLLog(StringRef Data, XRayFileHeader &FileHeader,
351 std::vector<XRayRecord> &Records) {
352 YAMLXRayTrace Trace;
353 Input In(Data);
354 In >> Trace;
355 if (In.error())
356 return make_error<StringError>("Failed loading YAML Data.", In.error());
357
358 FileHeader.Version = Trace.Header.Version;
359 FileHeader.Type = Trace.Header.Type;
360 FileHeader.ConstantTSC = Trace.Header.ConstantTSC;
361 FileHeader.NonstopTSC = Trace.Header.NonstopTSC;
362 FileHeader.CycleFrequency = Trace.Header.CycleFrequency;
363
364 if (FileHeader.Version != 1)
365 return make_error<StringError>(
366 Twine("Unsupported XRay file version: ") + Twine(FileHeader.Version),
367 std::make_error_code(std::errc::invalid_argument));
368
369 Records.clear();
370 std::transform(Trace.Records.begin(), Trace.Records.end(),
371 std::back_inserter(Records), [&](const YAMLXRayRecord &R) {
372 return XRayRecord{R.RecordType, R.CPU, R.Type,
373 R.FuncId, R.TSC, R.TId,
374 R.PId, R.CallArgs, R.Data};
375 });
376 return Error::success();
377 }
378 } // namespace
379
loadTraceFile(StringRef Filename,bool Sort)380 Expected<Trace> llvm::xray::loadTraceFile(StringRef Filename, bool Sort) {
381 Expected<sys::fs::file_t> FdOrErr = sys::fs::openNativeFileForRead(Filename);
382 if (!FdOrErr)
383 return FdOrErr.takeError();
384
385 uint64_t FileSize;
386 if (auto EC = sys::fs::file_size(Filename, FileSize)) {
387 return make_error<StringError>(
388 Twine("Cannot read log from '") + Filename + "'", EC);
389 }
390 if (FileSize < 4) {
391 return make_error<StringError>(
392 Twine("File '") + Filename + "' too small for XRay.",
393 std::make_error_code(std::errc::executable_format_error));
394 }
395
396 // Map the opened file into memory and use a StringRef to access it later.
397 std::error_code EC;
398 sys::fs::mapped_file_region MappedFile(
399 *FdOrErr, sys::fs::mapped_file_region::mapmode::readonly, FileSize, 0,
400 EC);
401 sys::fs::closeFile(*FdOrErr);
402 if (EC) {
403 return make_error<StringError>(
404 Twine("Cannot read log from '") + Filename + "'", EC);
405 }
406 auto Data = StringRef(MappedFile.data(), MappedFile.size());
407
408 // TODO: Lift the endianness and implementation selection here.
409 DataExtractor LittleEndianDE(Data, true, 8);
410 auto TraceOrError = loadTrace(LittleEndianDE, Sort);
411 if (!TraceOrError) {
412 DataExtractor BigEndianDE(Data, false, 8);
413 consumeError(TraceOrError.takeError());
414 TraceOrError = loadTrace(BigEndianDE, Sort);
415 }
416 return TraceOrError;
417 }
418
loadTrace(const DataExtractor & DE,bool Sort)419 Expected<Trace> llvm::xray::loadTrace(const DataExtractor &DE, bool Sort) {
420 // Attempt to detect the file type using file magic. We have a slight bias
421 // towards the binary format, and we do this by making sure that the first 4
422 // bytes of the binary file is some combination of the following byte
423 // patterns: (observe the code loading them assumes they're little endian)
424 //
425 // 0x01 0x00 0x00 0x00 - version 1, "naive" format
426 // 0x01 0x00 0x01 0x00 - version 1, "flight data recorder" format
427 // 0x02 0x00 0x01 0x00 - version 2, "flight data recorder" format
428 //
429 // YAML files don't typically have those first four bytes as valid text so we
430 // try loading assuming YAML if we don't find these bytes.
431 //
432 // Only if we can't load either the binary or the YAML format will we yield an
433 // error.
434 DataExtractor HeaderExtractor(DE.getData(), DE.isLittleEndian(), 8);
435 uint64_t OffsetPtr = 0;
436 uint16_t Version = HeaderExtractor.getU16(&OffsetPtr);
437 uint16_t Type = HeaderExtractor.getU16(&OffsetPtr);
438
439 enum BinaryFormatType { NAIVE_FORMAT = 0, FLIGHT_DATA_RECORDER_FORMAT = 1 };
440
441 Trace T;
442 switch (Type) {
443 case NAIVE_FORMAT:
444 if (Version == 1 || Version == 2 || Version == 3) {
445 if (auto E = loadNaiveFormatLog(DE.getData(), DE.isLittleEndian(),
446 T.FileHeader, T.Records))
447 return std::move(E);
448 } else {
449 return make_error<StringError>(
450 Twine("Unsupported version for Basic/Naive Mode logging: ") +
451 Twine(Version),
452 std::make_error_code(std::errc::executable_format_error));
453 }
454 break;
455 case FLIGHT_DATA_RECORDER_FORMAT:
456 if (Version >= 1 && Version <= 5) {
457 if (auto E = loadFDRLog(DE.getData(), DE.isLittleEndian(), T.FileHeader,
458 T.Records))
459 return std::move(E);
460 } else {
461 return make_error<StringError>(
462 Twine("Unsupported version for FDR Mode logging: ") + Twine(Version),
463 std::make_error_code(std::errc::executable_format_error));
464 }
465 break;
466 default:
467 if (auto E = loadYAMLLog(DE.getData(), T.FileHeader, T.Records))
468 return std::move(E);
469 }
470
471 if (Sort)
472 llvm::stable_sort(T.Records, [&](const XRayRecord &L, const XRayRecord &R) {
473 return L.TSC < R.TSC;
474 });
475
476 return std::move(T);
477 }
478