llvm/lib/ExecutionEngine/PerfJITEvents/PerfJITEventListener.cpp

   1 //===-- PerfJITEventListener.cpp - Tell Linux's perf about JITted code ----===//
   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 defines a JITEventListener object that tells perf about JITted
  10 // functions, including source line information.
  11 //
  12 // Documentation for perf jit integration is available at:
  13 // https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/tools/perf/Documentation/jitdump-specification.txt
  14 // https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/tools/perf/Documentation/jit-interface.txt
  15 //
  16 //===----------------------------------------------------------------------===//
  17
  18 #include "llvm/ADT/Twine.h"
  19 #include "llvm/Config/config.h"
  20 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
  21 #include "llvm/ExecutionEngine/JITEventListener.h"
  22 #include "llvm/Object/ObjectFile.h"
  23 #include "llvm/Object/SymbolSize.h"
  24 #include "llvm/Support/Debug.h"
  25 #include "llvm/Support/Errno.h"
  26 #include "llvm/Support/FileSystem.h"
  27 #include "llvm/Support/ManagedStatic.h"
  28 #include "llvm/Support/MemoryBuffer.h"
  29 #include "llvm/Support/Mutex.h"
  30 #include "llvm/Support/Path.h"
  31 #include "llvm/Support/Process.h"
  32 #include "llvm/Support/Threading.h"
  33 #include "llvm/Support/raw_ostream.h"
  34 #include <mutex>
  35
  36 #include <sys/mman.h>  // mmap()
  37 #include <time.h>      // clock_gettime(), time(), localtime_r() */
  38 #include <unistd.h>    // for read(), close()
  39
  40 using namespace llvm;
  41 using namespace llvm::object;
  42 typedef DILineInfoSpecifier::FileLineInfoKind FileLineInfoKind;
  43
  44 namespace {
  45
  46 // language identifier (XXX: should we generate something better from debug
  47 // info?)
  48 #define JIT_LANG "llvm-IR"
  49 #define LLVM_PERF_JIT_MAGIC                                                    \
  50   ((uint32_t)'J' << 24 | (uint32_t)'i' << 16 | (uint32_t)'T' << 8 |            \
  51    (uint32_t)'D')
  52 #define LLVM_PERF_JIT_VERSION 1
  53
  54 // bit 0: set if the jitdump file is using an architecture-specific timestamp
  55 // clock source
  56 #define JITDUMP_FLAGS_ARCH_TIMESTAMP (1ULL << 0)
  57
  58 struct LLVMPerfJitHeader;
  59
  60 class PerfJITEventListener : public JITEventListener {
  61 public:
  62   PerfJITEventListener();
  63   ~PerfJITEventListener() {
  64     if (MarkerAddr)
  65       CloseMarker();
  66   }
  67
  68   void notifyObjectLoaded(ObjectKey K, const ObjectFile &Obj,
  69                           const RuntimeDyld::LoadedObjectInfo &L) override;
  70   void notifyFreeingObject(ObjectKey K) override;
  71
  72 private:
  73   bool InitDebuggingDir();
  74   bool OpenMarker();
  75   void CloseMarker();
  76   static bool FillMachine(LLVMPerfJitHeader &hdr);
  77
  78   void NotifyCode(Expected<llvm::StringRef> &Symbol, uint64_t CodeAddr,
  79                   uint64_t CodeSize);
  80   void NotifyDebug(uint64_t CodeAddr, DILineInfoTable Lines);
  81
  82   // cache lookups
  83   sys::Process::Pid Pid;
  84
  85   // base directory for output data
  86   std::string JitPath;
  87
  88   // output data stream, closed via Dumpstream
  89   int DumpFd = -1;
  90
  91   // output data stream
  92   std::unique_ptr<raw_fd_ostream> Dumpstream;
  93
  94   // prevent concurrent dumps from messing up the output file
  95   sys::Mutex Mutex;
  96
  97   // perf mmap marker
  98   void *MarkerAddr = NULL;
  99
 100   // perf support ready
 101   bool SuccessfullyInitialized = false;
 102
 103   // identifier for functions, primarily to identify when moving them around
 104   uint64_t CodeGeneration = 1;
 105 };
 106
 107 // The following are POD struct definitions from the perf jit specification
 108
 109 enum LLVMPerfJitRecordType {
 110   JIT_CODE_LOAD = 0,
 111   JIT_CODE_MOVE = 1, // not emitted, code isn't moved
 112   JIT_CODE_DEBUG_INFO = 2,
 113   JIT_CODE_CLOSE = 3,          // not emitted, unnecessary
 114   JIT_CODE_UNWINDING_INFO = 4, // not emitted
 115
 116   JIT_CODE_MAX
 117 };
 118
 119 struct LLVMPerfJitHeader {
 120   uint32_t Magic;     // characters "JiTD"
 121   uint32_t Version;   // header version
 122   uint32_t TotalSize; // total size of header
 123   uint32_t ElfMach;   // elf mach target
 124   uint32_t Pad1;      // reserved
 125   uint32_t Pid;
 126   uint64_t Timestamp; // timestamp
 127   uint64_t Flags;     // flags
 128 };
 129
 130 // record prefix (mandatory in each record)
 131 struct LLVMPerfJitRecordPrefix {
 132   uint32_t Id; // record type identifier
 133   uint32_t TotalSize;
 134   uint64_t Timestamp;
 135 };
 136
 137 struct LLVMPerfJitRecordCodeLoad {
 138   LLVMPerfJitRecordPrefix Prefix;
 139
 140   uint32_t Pid;
 141   uint32_t Tid;
 142   uint64_t Vma;
 143   uint64_t CodeAddr;
 144   uint64_t CodeSize;
 145   uint64_t CodeIndex;
 146 };
 147
 148 struct LLVMPerfJitDebugEntry {
 149   uint64_t Addr;
 150   int Lineno;  // source line number starting at 1
 151   int Discrim; // column discriminator, 0 is default
 152   // followed by null terminated filename, \xff\0 if same as previous entry
 153 };
 154
 155 struct LLVMPerfJitRecordDebugInfo {
 156   LLVMPerfJitRecordPrefix Prefix;
 157
 158   uint64_t CodeAddr;
 159   uint64_t NrEntry;
 160   // followed by NrEntry LLVMPerfJitDebugEntry records
 161 };
 162
 163 static inline uint64_t timespec_to_ns(const struct timespec *ts) {
 164   const uint64_t NanoSecPerSec = 1000000000;
 165   return ((uint64_t)ts->tv_sec * NanoSecPerSec) + ts->tv_nsec;
 166 }
 167
 168 static inline uint64_t perf_get_timestamp(void) {
 169   struct timespec ts;
 170   int ret;
 171
 172   ret = clock_gettime(CLOCK_MONOTONIC, &ts);
 173   if (ret)
 174     return 0;
 175
 176   return timespec_to_ns(&ts);
 177 }
 178
 179 PerfJITEventListener::PerfJITEventListener()
 180     : Pid(sys::Process::getProcessId()) {
 181   // check if clock-source is supported
 182   if (!perf_get_timestamp()) {
 183     errs() << "kernel does not support CLOCK_MONOTONIC\n";
 184     return;
 185   }
 186
 187   if (!InitDebuggingDir()) {
 188     errs() << "could not initialize debugging directory\n";
 189     return;
 190   }
 191
 192   std::string Filename;
 193   raw_string_ostream FilenameBuf(Filename);
 194   FilenameBuf << JitPath << "/jit-" << Pid << ".dump";
 195
 196   // Need to open ourselves, because we need to hand the FD to OpenMarker() and
 197   // raw_fd_ostream doesn't expose the FD.
 198   using sys::fs::openFileForWrite;
 199   if (auto EC =
 200           openFileForReadWrite(FilenameBuf.str(), DumpFd,
 201                                sys::fs::CD_CreateNew, sys::fs::OF_None)) {
 202     errs() << "could not open JIT dump file " << FilenameBuf.str() << ": "
 203            << EC.message() << "\n";
 204     return;
 205   }
 206
 207   Dumpstream = std::make_unique<raw_fd_ostream>(DumpFd, true);
 208
 209   LLVMPerfJitHeader Header = {0};
 210   if (!FillMachine(Header))
 211     return;
 212
 213   // signal this process emits JIT information
 214   if (!OpenMarker())
 215     return;
 216
 217   // emit dumpstream header
 218   Header.Magic = LLVM_PERF_JIT_MAGIC;
 219   Header.Version = LLVM_PERF_JIT_VERSION;
 220   Header.TotalSize = sizeof(Header);
 221   Header.Pid = Pid;
 222   Header.Timestamp = perf_get_timestamp();
 223   Dumpstream->write(reinterpret_cast<const char *>(&Header), sizeof(Header));
 224
 225   // Everything initialized, can do profiling now.
 226   if (!Dumpstream->has_error())
 227     SuccessfullyInitialized = true;
 228 }
 229
 230 void PerfJITEventListener::notifyObjectLoaded(
 231     ObjectKey K, const ObjectFile &Obj,
 232     const RuntimeDyld::LoadedObjectInfo &L) {
 233
 234   if (!SuccessfullyInitialized)
 235     return;
 236
 237   OwningBinary<ObjectFile> DebugObjOwner = L.getObjectForDebug(Obj);
 238   const ObjectFile &DebugObj = *DebugObjOwner.getBinary();
 239
 240   // Get the address of the object image for use as a unique identifier
 241   std::unique_ptr<DIContext> Context = DWARFContext::create(DebugObj);
 242
 243   // Use symbol info to iterate over functions in the object.
 244   for (const std::pair<SymbolRef, uint64_t> &P : computeSymbolSizes(DebugObj)) {
 245     SymbolRef Sym = P.first;
 246     std::string SourceFileName;
 247
 248     Expected<SymbolRef::Type> SymTypeOrErr = Sym.getType();
 249     if (!SymTypeOrErr) {
 250       // There's not much we can with errors here
 251       consumeError(SymTypeOrErr.takeError());
 252       continue;
 253     }
 254     SymbolRef::Type SymType = *SymTypeOrErr;
 255     if (SymType != SymbolRef::ST_Function)
 256       continue;
 257
 258     Expected<StringRef> Name = Sym.getName();
 259     if (!Name) {
 260       consumeError(Name.takeError());
 261       continue;
 262     }
 263
 264     Expected<uint64_t> AddrOrErr = Sym.getAddress();
 265     if (!AddrOrErr) {
 266       consumeError(AddrOrErr.takeError());
 267       continue;
 268     }
 269     uint64_t Size = P.second;
 270     object::SectionedAddress Address;
 271     Address.Address = *AddrOrErr;
 272
 273     uint64_t SectionIndex = object::SectionedAddress::UndefSection;
 274     if (auto SectOrErr = Sym.getSection())
 275         if (*SectOrErr != Obj.section_end())
 276             SectionIndex = SectOrErr.get()->getIndex();
 277
 278     // According to spec debugging info has to come before loading the
 279     // corresonding code load.
 280     DILineInfoTable Lines = Context->getLineInfoForAddressRange(
 281         {*AddrOrErr, SectionIndex}, Size, FileLineInfoKind::AbsoluteFilePath);
 282
 283     NotifyDebug(*AddrOrErr, Lines);
 284     NotifyCode(Name, *AddrOrErr, Size);
 285   }
 286
 287   // avoid races with writes
 288   std::lock_guard<sys::Mutex> Guard(Mutex);
 289
 290   Dumpstream->flush();
 291 }
 292
 293 void PerfJITEventListener::notifyFreeingObject(ObjectKey K) {
 294   // perf currently doesn't have an interface for unloading. But munmap()ing the
 295   // code section does, so that's ok.
 296 }
 297
 298 bool PerfJITEventListener::InitDebuggingDir() {
 299   time_t Time;
 300   struct tm LocalTime;
 301   char TimeBuffer[sizeof("YYYYMMDD")];
 302   SmallString<64> Path;
 303
 304   // search for location to dump data to
 305   if (const char *BaseDir = getenv("JITDUMPDIR"))
 306     Path.append(BaseDir);
 307   else if (!sys::path::home_directory(Path))
 308     Path = ".";
 309
 310   // create debug directory
 311   Path += "/.debug/jit/";
 312   if (auto EC = sys::fs::create_directories(Path)) {
 313     errs() << "could not create jit cache directory " << Path << ": "
 314            << EC.message() << "\n";
 315     return false;
 316   }
 317
 318   // create unique directory for dump data related to this process
 319   time(&Time);
 320   localtime_r(&Time, &LocalTime);
 321   strftime(TimeBuffer, sizeof(TimeBuffer), "%Y%m%d", &LocalTime);
 322   Path += JIT_LANG "-jit-";
 323   Path += TimeBuffer;
 324
 325   SmallString<128> UniqueDebugDir;
 326
 327   using sys::fs::createUniqueDirectory;
 328   if (auto EC = createUniqueDirectory(Path, UniqueDebugDir)) {
 329     errs() << "could not create unique jit cache directory " << UniqueDebugDir
 330            << ": " << EC.message() << "\n";
 331     return false;
 332   }
 333
 334   JitPath = std::string(UniqueDebugDir.str());
 335
 336   return true;
 337 }
 338
 339 bool PerfJITEventListener::OpenMarker() {
 340   // We mmap the jitdump to create an MMAP RECORD in perf.data file.  The mmap
 341   // is captured either live (perf record running when we mmap) or in deferred
 342   // mode, via /proc/PID/maps. The MMAP record is used as a marker of a jitdump
 343   // file for more meta data info about the jitted code. Perf report/annotate
 344   // detect this special filename and process the jitdump file.
 345   //
 346   // Mapping must be PROT_EXEC to ensure it is captured by perf record
 347   // even when not using -d option.
 348   MarkerAddr = ::mmap(NULL, sys::Process::getPageSizeEstimate(),
 349                       PROT_READ | PROT_EXEC, MAP_PRIVATE, DumpFd, 0);
 350
 351   if (MarkerAddr == MAP_FAILED) {
 352     errs() << "could not mmap JIT marker\n";
 353     return false;
 354   }
 355   return true;
 356 }
 357
 358 void PerfJITEventListener::CloseMarker() {
 359   if (!MarkerAddr)
 360     return;
 361
 362   munmap(MarkerAddr, sys::Process::getPageSizeEstimate());
 363   MarkerAddr = nullptr;
 364 }
 365
 366 bool PerfJITEventListener::FillMachine(LLVMPerfJitHeader &hdr) {
 367   char id[16];
 368   struct {
 369     uint16_t e_type;
 370     uint16_t e_machine;
 371   } info;
 372
 373   size_t RequiredMemory = sizeof(id) + sizeof(info);
 374
 375   ErrorOr<std::unique_ptr<MemoryBuffer>> MB =
 376     MemoryBuffer::getFileSlice("/proc/self/exe",
 377                                RequiredMemory,
 378                                0);
 379
 380   // This'll not guarantee that enough data was actually read from the
 381   // underlying file. Instead the trailing part of the buffer would be
 382   // zeroed. Given the ELF signature check below that seems ok though,
 383   // it's unlikely that the file ends just after that, and the
 384   // consequence would just be that perf wouldn't recognize the
 385   // signature.
 386   if (auto EC = MB.getError()) {
 387     errs() << "could not open /proc/self/exe: " << EC.message() << "\n";
 388     return false;
 389   }
 390
 391   memcpy(&id, (*MB)->getBufferStart(), sizeof(id));
 392   memcpy(&info, (*MB)->getBufferStart() + sizeof(id), sizeof(info));
 393
 394   // check ELF signature
 395   if (id[0] != 0x7f || id[1] != 'E' || id[2] != 'L' || id[3] != 'F') {
 396     errs() << "invalid elf signature\n";
 397     return false;
 398   }
 399
 400   hdr.ElfMach = info.e_machine;
 401
 402   return true;
 403 }
 404
 405 void PerfJITEventListener::NotifyCode(Expected<llvm::StringRef> &Symbol,
 406                                       uint64_t CodeAddr, uint64_t CodeSize) {
 407   assert(SuccessfullyInitialized);
 408
 409   // 0 length functions can't have samples.
 410   if (CodeSize == 0)
 411     return;
 412
 413   LLVMPerfJitRecordCodeLoad rec;
 414   rec.Prefix.Id = JIT_CODE_LOAD;
 415   rec.Prefix.TotalSize = sizeof(rec) +        // debug record itself
 416                          Symbol->size() + 1 + // symbol name
 417                          CodeSize;            // and code
 418   rec.Prefix.Timestamp = perf_get_timestamp();
 419
 420   rec.CodeSize = CodeSize;
 421   rec.Vma = 0;
 422   rec.CodeAddr = CodeAddr;
 423   rec.Pid = Pid;
 424   rec.Tid = get_threadid();
 425
 426   // avoid interspersing output
 427   std::lock_guard<sys::Mutex> Guard(Mutex);
 428
 429   rec.CodeIndex = CodeGeneration++; // under lock!
 430
 431   Dumpstream->write(reinterpret_cast<const char *>(&rec), sizeof(rec));
 432   Dumpstream->write(Symbol->data(), Symbol->size() + 1);
 433   Dumpstream->write(reinterpret_cast<const char *>(CodeAddr), CodeSize);
 434 }
 435
 436 void PerfJITEventListener::NotifyDebug(uint64_t CodeAddr,
 437                                        DILineInfoTable Lines) {
 438   assert(SuccessfullyInitialized);
 439
 440   // Didn't get useful debug info.
 441   if (Lines.empty())
 442     return;
 443
 444   LLVMPerfJitRecordDebugInfo rec;
 445   rec.Prefix.Id = JIT_CODE_DEBUG_INFO;
 446   rec.Prefix.TotalSize = sizeof(rec); // will be increased further
 447   rec.Prefix.Timestamp = perf_get_timestamp();
 448   rec.CodeAddr = CodeAddr;
 449   rec.NrEntry = Lines.size();
 450
 451   // compute total size size of record (variable due to filenames)
 452   DILineInfoTable::iterator Begin = Lines.begin();
 453   DILineInfoTable::iterator End = Lines.end();
 454   for (DILineInfoTable::iterator It = Begin; It != End; ++It) {
 455     DILineInfo &line = It->second;
 456     rec.Prefix.TotalSize += sizeof(LLVMPerfJitDebugEntry);
 457     rec.Prefix.TotalSize += line.FileName.size() + 1;
 458   }
 459
 460   // The debug_entry describes the source line information. It is defined as
 461   // follows in order:
 462   // * uint64_t code_addr: address of function for which the debug information
 463   // is generated
 464   // * uint32_t line     : source file line number (starting at 1)
 465   // * uint32_t discrim  : column discriminator, 0 is default
 466   // * char name[n]      : source file name in ASCII, including null termination
 467
 468   // avoid interspersing output
 469   std::lock_guard<sys::Mutex> Guard(Mutex);
 470
 471   Dumpstream->write(reinterpret_cast<const char *>(&rec), sizeof(rec));
 472
 473   for (DILineInfoTable::iterator It = Begin; It != End; ++It) {
 474     LLVMPerfJitDebugEntry LineInfo;
 475     DILineInfo &Line = It->second;
 476
 477     LineInfo.Addr = It->first;
 478     // The function re-created by perf is preceded by a elf
 479     // header. Need to adjust for that, otherwise the results are
 480     // wrong.
 481     LineInfo.Addr += 0x40;
 482     LineInfo.Lineno = Line.Line;
 483     LineInfo.Discrim = Line.Discriminator;
 484
 485     Dumpstream->write(reinterpret_cast<const char *>(&LineInfo),
 486                       sizeof(LineInfo));
 487     Dumpstream->write(Line.FileName.c_str(), Line.FileName.size() + 1);
 488   }
 489 }
 490
 491 // There should be only a single event listener per process, otherwise perf gets
 492 // confused.
 493 llvm::ManagedStatic<PerfJITEventListener> PerfListener;
 494
 495 } // end anonymous namespace
 496
 497 namespace llvm {
 498 JITEventListener *JITEventListener::createPerfJITEventListener() {
 499   return &*PerfListener;
 500 }
 501
 502 } // namespace llvm
 503
 504 LLVMJITEventListenerRef LLVMCreatePerfJITEventListener(void)
 505 {
 506   return wrap(JITEventListener::createPerfJITEventListener());
 507 }