lib/XRay/Trace.cpp

   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
  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   uint32_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 %d.", OffsetPtr);
  71     auto PreReadOffset = OffsetPtr;
  72     auto RecordType = Reader.getU16(&OffsetPtr);
  73     if (OffsetPtr == PreReadOffset)
  74       return createStringError(
  75           std::make_error_code(std::errc::executable_format_error),
  76           "Failed reading record type at offset %d.", OffsetPtr);
  77
  78     switch (RecordType) {
  79     case 0: { // Normal records.
  80       Records.emplace_back();
  81       auto &Record = Records.back();
  82       Record.RecordType = RecordType;
  83
  84       PreReadOffset = OffsetPtr;
  85       Record.CPU = Reader.getU8(&OffsetPtr);
  86       if (OffsetPtr == PreReadOffset)
  87         return createStringError(
  88             std::make_error_code(std::errc::executable_format_error),
  89             "Failed reading CPU field at offset %d.", OffsetPtr);
  90
  91       PreReadOffset = OffsetPtr;
  92       auto Type = Reader.getU8(&OffsetPtr);
  93       if (OffsetPtr == PreReadOffset)
  94         return createStringError(
  95             std::make_error_code(std::errc::executable_format_error),
  96             "Failed reading record type field at offset %d.", OffsetPtr);
  97
  98       switch (Type) {
  99       case 0:
 100         Record.Type = RecordTypes::ENTER;
 101         break;
 102       case 1:
 103         Record.Type = RecordTypes::EXIT;
 104         break;
 105       case 2:
 106         Record.Type = RecordTypes::TAIL_EXIT;
 107         break;
 108       case 3:
 109         Record.Type = RecordTypes::ENTER_ARG;
 110         break;
 111       default:
 112         return createStringError(
 113             std::make_error_code(std::errc::executable_format_error),
 114             "Unknown record type '%d' at offset %d.", Type, OffsetPtr);
 115       }
 116
 117       PreReadOffset = OffsetPtr;
 118       Record.FuncId = Reader.getSigned(&OffsetPtr, sizeof(int32_t));
 119       if (OffsetPtr == PreReadOffset)
 120         return createStringError(
 121             std::make_error_code(std::errc::executable_format_error),
 122             "Failed reading function id field at offset %d.", OffsetPtr);
 123
 124       PreReadOffset = OffsetPtr;
 125       Record.TSC = Reader.getU64(&OffsetPtr);
 126       if (OffsetPtr == PreReadOffset)
 127         return createStringError(
 128             std::make_error_code(std::errc::executable_format_error),
 129             "Failed reading TSC field at offset %d.", OffsetPtr);
 130
 131       PreReadOffset = OffsetPtr;
 132       Record.TId = Reader.getU32(&OffsetPtr);
 133       if (OffsetPtr == PreReadOffset)
 134         return createStringError(
 135             std::make_error_code(std::errc::executable_format_error),
 136             "Failed reading thread id field at offset %d.", OffsetPtr);
 137
 138       PreReadOffset = OffsetPtr;
 139       Record.PId = Reader.getU32(&OffsetPtr);
 140       if (OffsetPtr == PreReadOffset)
 141         return createStringError(
 142             std::make_error_code(std::errc::executable_format_error),
 143             "Failed reading process id at offset %d.", OffsetPtr);
 144
 145       break;
 146     }
 147     case 1: { // Arg payload record.
 148       auto &Record = Records.back();
 149
 150       // We skip the next two bytes of the record, because we don't need the
 151       // type and the CPU record for arg payloads.
 152       OffsetPtr += 2;
 153       PreReadOffset = OffsetPtr;
 154       int32_t FuncId = Reader.getSigned(&OffsetPtr, sizeof(int32_t));
 155       if (OffsetPtr == PreReadOffset)
 156         return createStringError(
 157             std::make_error_code(std::errc::executable_format_error),
 158             "Failed reading function id field at offset %d.", OffsetPtr);
 159
 160       PreReadOffset = OffsetPtr;
 161       auto TId = Reader.getU32(&OffsetPtr);
 162       if (OffsetPtr == PreReadOffset)
 163         return createStringError(
 164             std::make_error_code(std::errc::executable_format_error),
 165             "Failed reading thread id field at offset %d.", OffsetPtr);
 166
 167       PreReadOffset = OffsetPtr;
 168       auto PId = Reader.getU32(&OffsetPtr);
 169       if (OffsetPtr == PreReadOffset)
 170         return createStringError(
 171             std::make_error_code(std::errc::executable_format_error),
 172             "Failed reading process id field at offset %d.", OffsetPtr);
 173
 174       // Make a check for versions above 3 for the Pid field
 175       if (Record.FuncId != FuncId || Record.TId != TId ||
 176           (FileHeader.Version >= 3 ? Record.PId != PId : false))
 177         return createStringError(
 178             std::make_error_code(std::errc::executable_format_error),
 179             "Corrupted log, found arg payload following non-matching "
 180             "function+thread record. Record for function %d != %d at offset "
 181             "%d",
 182             Record.FuncId, FuncId, OffsetPtr);
 183
 184       PreReadOffset = OffsetPtr;
 185       auto Arg = Reader.getU64(&OffsetPtr);
 186       if (OffsetPtr == PreReadOffset)
 187         return createStringError(
 188             std::make_error_code(std::errc::executable_format_error),
 189             "Failed reading argument payload at offset %d.", OffsetPtr);
 190
 191       Record.CallArgs.push_back(Arg);
 192       break;
 193     }
 194     default:
 195       return createStringError(
 196           std::make_error_code(std::errc::executable_format_error),
 197           "Unknown record type '%d' at offset %d.", RecordType, OffsetPtr);
 198     }
 199     // Advance the offset pointer enough bytes to align to 32-byte records for
 200     // basic mode logs.
 201     OffsetPtr += 8;
 202   }
 203   return Error::success();
 204 }
 205
 206 /// Reads a log in FDR mode for version 1 of this binary format. FDR mode is
 207 /// defined as part of the compiler-rt project in xray_fdr_logging.h, and such
 208 /// a log consists of the familiar 32 bit XRayHeader, followed by sequences of
 209 /// of interspersed 16 byte Metadata Records and 8 byte Function Records.
 210 ///
 211 /// The following is an attempt to document the grammar of the format, which is
 212 /// parsed by this function for little-endian machines. Since the format makes
 213 /// use of BitFields, when we support big-endian architectures, we will need to
 214 /// adjust not only the endianness parameter to llvm's RecordExtractor, but also
 215 /// the bit twiddling logic, which is consistent with the little-endian
 216 /// convention that BitFields within a struct will first be packed into the
 217 /// least significant bits the address they belong to.
 218 ///
 219 /// We expect a format complying with the grammar in the following pseudo-EBNF
 220 /// in Version 1 of the FDR log.
 221 ///
 222 /// FDRLog: XRayFileHeader ThreadBuffer*
 223 /// XRayFileHeader: 32 bytes to identify the log as FDR with machine metadata.
 224 ///     Includes BufferSize
 225 /// ThreadBuffer: NewBuffer WallClockTime NewCPUId FunctionSequence EOB
 226 /// BufSize: 8 byte unsigned integer indicating how large the buffer is.
 227 /// NewBuffer: 16 byte metadata record with Thread Id.
 228 /// WallClockTime: 16 byte metadata record with human readable time.
 229 /// Pid: 16 byte metadata record with Pid
 230 /// NewCPUId: 16 byte metadata record with CPUId and a 64 bit TSC reading.
 231 /// EOB: 16 byte record in a thread buffer plus mem garbage to fill BufSize.
 232 /// FunctionSequence: NewCPUId | TSCWrap | FunctionRecord
 233 /// TSCWrap: 16 byte metadata record with a full 64 bit TSC reading.
 234 /// FunctionRecord: 8 byte record with FunctionId, entry/exit, and TSC delta.
 235 ///
 236 /// In Version 2, we make the following changes:
 237 ///
 238 /// ThreadBuffer: BufferExtents NewBuffer WallClockTime NewCPUId
 239 ///               FunctionSequence
 240 /// BufferExtents: 16 byte metdata record describing how many usable bytes are
 241 ///                in the buffer. This is measured from the start of the buffer
 242 ///                and must always be at least 48 (bytes).
 243 ///
 244 /// In Version 3, we make the following changes:
 245 ///
 246 /// ThreadBuffer: BufferExtents NewBuffer WallClockTime Pid NewCPUId
 247 ///               FunctionSequence
 248 /// EOB: *deprecated*
 249 ///
 250 /// In Version 4, we make the following changes:
 251 ///
 252 /// CustomEventRecord now includes the CPU data.
 253 ///
 254 /// In Version 5, we make the following changes:
 255 ///
 256 /// CustomEventRecord and TypedEventRecord now use TSC delta encoding similar to
 257 /// what FunctionRecord instances use, and we no longer need to include the CPU
 258 /// id in the CustomEventRecord.
 259 ///
 260 Error loadFDRLog(StringRef Data, bool IsLittleEndian,
 261                  XRayFileHeader &FileHeader, std::vector<XRayRecord> &Records) {
 262
 263   if (Data.size() < 32)
 264     return createStringError(std::make_error_code(std::errc::invalid_argument),
 265                              "Not enough bytes for an XRay FDR log.");
 266   DataExtractor DE(Data, IsLittleEndian, 8);
 267
 268   uint32_t OffsetPtr = 0;
 269   auto FileHeaderOrError = readBinaryFormatHeader(DE, OffsetPtr);
 270   if (!FileHeaderOrError)
 271     return FileHeaderOrError.takeError();
 272   FileHeader = std::move(FileHeaderOrError.get());
 273
 274   // First we load the records into memory.
 275   std::vector<std::unique_ptr<Record>> FDRRecords;
 276
 277   {
 278     FileBasedRecordProducer P(FileHeader, DE, OffsetPtr);
 279     LogBuilderConsumer C(FDRRecords);
 280     while (DE.isValidOffsetForDataOfSize(OffsetPtr, 1)) {
 281       auto R = P.produce();
 282       if (!R)
 283         return R.takeError();
 284       if (auto E = C.consume(std::move(R.get())))
 285         return E;
 286     }
 287   }
 288
 289   // Next we index the records into blocks.
 290   BlockIndexer::Index Index;
 291   {
 292     BlockIndexer Indexer(Index);
 293     for (auto &R : FDRRecords)
 294       if (auto E = R->apply(Indexer))
 295         return E;
 296     if (auto E = Indexer.flush())
 297       return E;
 298   }
 299
 300   // Then we verify the consistency of the blocks.
 301   {
 302     for (auto &PTB : Index) {
 303       auto &Blocks = PTB.second;
 304       for (auto &B : Blocks) {
 305         BlockVerifier Verifier;
 306         for (auto *R : B.Records)
 307           if (auto E = R->apply(Verifier))
 308             return E;
 309         if (auto E = Verifier.verify())
 310           return E;
 311       }
 312     }
 313   }
 314
 315   // This is now the meat of the algorithm. Here we sort the blocks according to
 316   // the Walltime record in each of the blocks for the same thread. This allows
 317   // us to more consistently recreate the execution trace in temporal order.
 318   // After the sort, we then reconstitute `Trace` records using a stateful
 319   // visitor associated with a single process+thread pair.
 320   {
 321     for (auto &PTB : Index) {
 322       auto &Blocks = PTB.second;
 323       llvm::sort(Blocks, [](const BlockIndexer::Block &L,
 324                             const BlockIndexer::Block &R) {
 325         return (L.WallclockTime->seconds() < R.WallclockTime->seconds() &&
 326                 L.WallclockTime->nanos() < R.WallclockTime->nanos());
 327       });
 328       auto Adder = [&](const XRayRecord &R) { Records.push_back(R); };
 329       TraceExpander Expander(Adder, FileHeader.Version);
 330       for (auto &B : Blocks) {
 331         for (auto *R : B.Records)
 332           if (auto E = R->apply(Expander))
 333             return E;
 334       }
 335       if (auto E = Expander.flush())
 336         return E;
 337     }
 338   }
 339
 340   return Error::success();
 341 }
 342
 343 Error loadYAMLLog(StringRef Data, XRayFileHeader &FileHeader,
 344                   std::vector<XRayRecord> &Records) {
 345   YAMLXRayTrace Trace;
 346   Input In(Data);
 347   In >> Trace;
 348   if (In.error())
 349     return make_error<StringError>("Failed loading YAML Data.", In.error());
 350
 351   FileHeader.Version = Trace.Header.Version;
 352   FileHeader.Type = Trace.Header.Type;
 353   FileHeader.ConstantTSC = Trace.Header.ConstantTSC;
 354   FileHeader.NonstopTSC = Trace.Header.NonstopTSC;
 355   FileHeader.CycleFrequency = Trace.Header.CycleFrequency;
 356
 357   if (FileHeader.Version != 1)
 358     return make_error<StringError>(
 359         Twine("Unsupported XRay file version: ") + Twine(FileHeader.Version),
 360         std::make_error_code(std::errc::invalid_argument));
 361
 362   Records.clear();
 363   std::transform(Trace.Records.begin(), Trace.Records.end(),
 364                  std::back_inserter(Records), [&](const YAMLXRayRecord &R) {
 365                    return XRayRecord{R.RecordType, R.CPU,      R.Type,
 366                                      R.FuncId,     R.TSC,      R.TId,
 367                                      R.PId,        R.CallArgs, R.Data};
 368                  });
 369   return Error::success();
 370 }
 371 } // namespace
 372
 373 Expected<Trace> llvm::xray::loadTraceFile(StringRef Filename, bool Sort) {
 374   Expected<sys::fs::file_t> FdOrErr = sys::fs::openNativeFileForRead(Filename);
 375   if (!FdOrErr)
 376     return FdOrErr.takeError();
 377
 378   uint64_t FileSize;
 379   if (auto EC = sys::fs::file_size(Filename, FileSize)) {
 380     return make_error<StringError>(
 381         Twine("Cannot read log from '") + Filename + "'", EC);
 382   }
 383   if (FileSize < 4) {
 384     return make_error<StringError>(
 385         Twine("File '") + Filename + "' too small for XRay.",
 386         std::make_error_code(std::errc::executable_format_error));
 387   }
 388
 389   // Map the opened file into memory and use a StringRef to access it later.
 390   std::error_code EC;
 391   sys::fs::mapped_file_region MappedFile(
 392       *FdOrErr, sys::fs::mapped_file_region::mapmode::readonly, FileSize, 0,
 393       EC);
 394   sys::fs::closeFile(*FdOrErr);
 395   if (EC) {
 396     return make_error<StringError>(
 397         Twine("Cannot read log from '") + Filename + "'", EC);
 398   }
 399   auto Data = StringRef(MappedFile.data(), MappedFile.size());
 400
 401   // TODO: Lift the endianness and implementation selection here.
 402   DataExtractor LittleEndianDE(Data, true, 8);
 403   auto TraceOrError = loadTrace(LittleEndianDE, Sort);
 404   if (!TraceOrError) {
 405     DataExtractor BigEndianDE(Data, false, 8);
 406     TraceOrError = loadTrace(BigEndianDE, Sort);
 407   }
 408   return TraceOrError;
 409 }
 410
 411 Expected<Trace> llvm::xray::loadTrace(const DataExtractor &DE, bool Sort) {
 412   // Attempt to detect the file type using file magic. We have a slight bias
 413   // towards the binary format, and we do this by making sure that the first 4
 414   // bytes of the binary file is some combination of the following byte
 415   // patterns: (observe the code loading them assumes they're little endian)
 416   //
 417   //   0x01 0x00 0x00 0x00 - version 1, "naive" format
 418   //   0x01 0x00 0x01 0x00 - version 1, "flight data recorder" format
 419   //   0x02 0x00 0x01 0x00 - version 2, "flight data recorder" format
 420   //
 421   // YAML files don't typically have those first four bytes as valid text so we
 422   // try loading assuming YAML if we don't find these bytes.
 423   //
 424   // Only if we can't load either the binary or the YAML format will we yield an
 425   // error.
 426   DataExtractor HeaderExtractor(DE.getData(), DE.isLittleEndian(), 8);
 427   uint32_t OffsetPtr = 0;
 428   uint16_t Version = HeaderExtractor.getU16(&OffsetPtr);
 429   uint16_t Type = HeaderExtractor.getU16(&OffsetPtr);
 430
 431   enum BinaryFormatType { NAIVE_FORMAT = 0, FLIGHT_DATA_RECORDER_FORMAT = 1 };
 432
 433   Trace T;
 434   switch (Type) {
 435   case NAIVE_FORMAT:
 436     if (Version == 1 || Version == 2 || Version == 3) {
 437       if (auto E = loadNaiveFormatLog(DE.getData(), DE.isLittleEndian(),
 438                                       T.FileHeader, T.Records))
 439         return std::move(E);
 440     } else {
 441       return make_error<StringError>(
 442           Twine("Unsupported version for Basic/Naive Mode logging: ") +
 443               Twine(Version),
 444           std::make_error_code(std::errc::executable_format_error));
 445     }
 446     break;
 447   case FLIGHT_DATA_RECORDER_FORMAT:
 448     if (Version >= 1 && Version <= 5) {
 449       if (auto E = loadFDRLog(DE.getData(), DE.isLittleEndian(), T.FileHeader,
 450                               T.Records))
 451         return std::move(E);
 452     } else {
 453       return make_error<StringError>(
 454           Twine("Unsupported version for FDR Mode logging: ") + Twine(Version),
 455           std::make_error_code(std::errc::executable_format_error));
 456     }
 457     break;
 458   default:
 459     if (auto E = loadYAMLLog(DE.getData(), T.FileHeader, T.Records))
 460       return std::move(E);
 461   }
 462
 463   if (Sort)
 464     llvm::stable_sort(T.Records, [&](const XRayRecord &L, const XRayRecord &R) {
 465       return L.TSC < R.TSC;
 466     });
 467
 468   return std::move(T);
 469 }