libsanitizer/tsan/tsan_rtl.h

   1 //===-- tsan_rtl.h ----------------------------------------------*- C++ -*-===//
   2 //
   3 // This file is distributed under the University of Illinois Open Source
   4 // License. See LICENSE.TXT for details.
   5 //
   6 //===----------------------------------------------------------------------===//
   7 //
   8 // This file is a part of ThreadSanitizer (TSan), a race detector.
   9 //
  10 // Main internal TSan header file.
  11 //
  12 // Ground rules:
  13 //   - C++ run-time should not be used (static CTORs, RTTI, exceptions, static
  14 //     function-scope locals)
  15 //   - All functions/classes/etc reside in namespace __tsan, except for those
  16 //     declared in tsan_interface.h.
  17 //   - Platform-specific files should be used instead of ifdefs (*).
  18 //   - No system headers included in header files (*).
  19 //   - Platform specific headres included only into platform-specific files (*).
  20 //
  21 //  (*) Except when inlining is critical for performance.
  22 //===----------------------------------------------------------------------===//
  23
  24 #ifndef TSAN_RTL_H
  25 #define TSAN_RTL_H
  26
  27 #include "sanitizer_common/sanitizer_allocator.h"
  28 #include "sanitizer_common/sanitizer_allocator_internal.h"
  29 #include "sanitizer_common/sanitizer_asm.h"
  30 #include "sanitizer_common/sanitizer_common.h"
  31 #include "sanitizer_common/sanitizer_deadlock_detector_interface.h"
  32 #include "sanitizer_common/sanitizer_libignore.h"
  33 #include "sanitizer_common/sanitizer_suppressions.h"
  34 #include "sanitizer_common/sanitizer_thread_registry.h"
  35 #include "tsan_clock.h"
  36 #include "tsan_defs.h"
  37 #include "tsan_flags.h"
  38 #include "tsan_sync.h"
  39 #include "tsan_trace.h"
  40 #include "tsan_vector.h"
  41 #include "tsan_report.h"
  42 #include "tsan_platform.h"
  43 #include "tsan_mutexset.h"
  44 #include "tsan_ignoreset.h"
  45 #include "tsan_stack_trace.h"
  46
  47 #if SANITIZER_WORDSIZE != 64
  48 # error "ThreadSanitizer is supported only on 64-bit platforms"
  49 #endif
  50
  51 namespace __tsan {
  52
  53 #ifndef SANITIZER_GO
  54 struct MapUnmapCallback;
  55 #if defined(__mips64) || defined(__aarch64__)
  56 static const uptr kAllocatorSpace = 0;
  57 static const uptr kAllocatorSize = SANITIZER_MMAP_RANGE_SIZE;
  58 static const uptr kAllocatorRegionSizeLog = 20;
  59 static const uptr kAllocatorNumRegions =
  60     kAllocatorSize >> kAllocatorRegionSizeLog;
  61 typedef TwoLevelByteMap<(kAllocatorNumRegions >> 12), 1 << 12,
  62     MapUnmapCallback> ByteMap;
  63 typedef SizeClassAllocator32<kAllocatorSpace, kAllocatorSize, 0,
  64     CompactSizeClassMap, kAllocatorRegionSizeLog, ByteMap,
  65     MapUnmapCallback> PrimaryAllocator;
  66 #else
  67 typedef SizeClassAllocator64<kHeapMemBeg, kHeapMemEnd - kHeapMemBeg, 0,
  68     DefaultSizeClassMap, MapUnmapCallback> PrimaryAllocator;
  69 #endif
  70 typedef SizeClassAllocatorLocalCache<PrimaryAllocator> AllocatorCache;
  71 typedef LargeMmapAllocator<MapUnmapCallback> SecondaryAllocator;
  72 typedef CombinedAllocator<PrimaryAllocator, AllocatorCache,
  73     SecondaryAllocator> Allocator;
  74 Allocator *allocator();
  75 #endif
  76
  77 void TsanCheckFailed(const char *file, int line, const char *cond,
  78                      u64 v1, u64 v2);
  79
  80 const u64 kShadowRodata = (u64)-1;  // .rodata shadow marker
  81
  82 // FastState (from most significant bit):
  83 //   ignore          : 1
  84 //   tid             : kTidBits
  85 //   unused          : -
  86 //   history_size    : 3
  87 //   epoch           : kClkBits
  88 class FastState {
  89  public:
  90   FastState(u64 tid, u64 epoch) {
  91     x_ = tid << kTidShift;
  92     x_ |= epoch;
  93     DCHECK_EQ(tid, this->tid());
  94     DCHECK_EQ(epoch, this->epoch());
  95     DCHECK_EQ(GetIgnoreBit(), false);
  96   }
  97
  98   explicit FastState(u64 x)
  99       : x_(x) {
 100   }
 101
 102   u64 raw() const {
 103     return x_;
 104   }
 105
 106   u64 tid() const {
 107     u64 res = (x_ & ~kIgnoreBit) >> kTidShift;
 108     return res;
 109   }
 110
 111   u64 TidWithIgnore() const {
 112     u64 res = x_ >> kTidShift;
 113     return res;
 114   }
 115
 116   u64 epoch() const {
 117     u64 res = x_ & ((1ull << kClkBits) - 1);
 118     return res;
 119   }
 120
 121   void IncrementEpoch() {
 122     u64 old_epoch = epoch();
 123     x_ += 1;
 124     DCHECK_EQ(old_epoch + 1, epoch());
 125     (void)old_epoch;
 126   }
 127
 128   void SetIgnoreBit() { x_ |= kIgnoreBit; }
 129   void ClearIgnoreBit() { x_ &= ~kIgnoreBit; }
 130   bool GetIgnoreBit() const { return (s64)x_ < 0; }
 131
 132   void SetHistorySize(int hs) {
 133     CHECK_GE(hs, 0);
 134     CHECK_LE(hs, 7);
 135     x_ = (x_ & ~(kHistoryMask << kHistoryShift)) | (u64(hs) << kHistoryShift);
 136   }
 137
 138   ALWAYS_INLINE
 139   int GetHistorySize() const {
 140     return (int)((x_ >> kHistoryShift) & kHistoryMask);
 141   }
 142
 143   void ClearHistorySize() {
 144     SetHistorySize(0);
 145   }
 146
 147   ALWAYS_INLINE
 148   u64 GetTracePos() const {
 149     const int hs = GetHistorySize();
 150     // When hs == 0, the trace consists of 2 parts.
 151     const u64 mask = (1ull << (kTracePartSizeBits + hs + 1)) - 1;
 152     return epoch() & mask;
 153   }
 154
 155  private:
 156   friend class Shadow;
 157   static const int kTidShift = 64 - kTidBits - 1;
 158   static const u64 kIgnoreBit = 1ull << 63;
 159   static const u64 kFreedBit = 1ull << 63;
 160   static const u64 kHistoryShift = kClkBits;
 161   static const u64 kHistoryMask = 7;
 162   u64 x_;
 163 };
 164
 165 // Shadow (from most significant bit):
 166 //   freed           : 1
 167 //   tid             : kTidBits
 168 //   is_atomic       : 1
 169 //   is_read         : 1
 170 //   size_log        : 2
 171 //   addr0           : 3
 172 //   epoch           : kClkBits
 173 class Shadow : public FastState {
 174  public:
 175   explicit Shadow(u64 x)
 176       : FastState(x) {
 177   }
 178
 179   explicit Shadow(const FastState &s)
 180       : FastState(s.x_) {
 181     ClearHistorySize();
 182   }
 183
 184   void SetAddr0AndSizeLog(u64 addr0, unsigned kAccessSizeLog) {
 185     DCHECK_EQ((x_ >> kClkBits) & 31, 0);
 186     DCHECK_LE(addr0, 7);
 187     DCHECK_LE(kAccessSizeLog, 3);
 188     x_ |= ((kAccessSizeLog << 3) | addr0) << kClkBits;
 189     DCHECK_EQ(kAccessSizeLog, size_log());
 190     DCHECK_EQ(addr0, this->addr0());
 191   }
 192
 193   void SetWrite(unsigned kAccessIsWrite) {
 194     DCHECK_EQ(x_ & kReadBit, 0);
 195     if (!kAccessIsWrite)
 196       x_ |= kReadBit;
 197     DCHECK_EQ(kAccessIsWrite, IsWrite());
 198   }
 199
 200   void SetAtomic(bool kIsAtomic) {
 201     DCHECK(!IsAtomic());
 202     if (kIsAtomic)
 203       x_ |= kAtomicBit;
 204     DCHECK_EQ(IsAtomic(), kIsAtomic);
 205   }
 206
 207   bool IsAtomic() const {
 208     return x_ & kAtomicBit;
 209   }
 210
 211   bool IsZero() const {
 212     return x_ == 0;
 213   }
 214
 215   static inline bool TidsAreEqual(const Shadow s1, const Shadow s2) {
 216     u64 shifted_xor = (s1.x_ ^ s2.x_) >> kTidShift;
 217     DCHECK_EQ(shifted_xor == 0, s1.TidWithIgnore() == s2.TidWithIgnore());
 218     return shifted_xor == 0;
 219   }
 220
 221   static ALWAYS_INLINE
 222   bool Addr0AndSizeAreEqual(const Shadow s1, const Shadow s2) {
 223     u64 masked_xor = ((s1.x_ ^ s2.x_) >> kClkBits) & 31;
 224     return masked_xor == 0;
 225   }
 226
 227   static ALWAYS_INLINE bool TwoRangesIntersect(Shadow s1, Shadow s2,
 228       unsigned kS2AccessSize) {
 229     bool res = false;
 230     u64 diff = s1.addr0() - s2.addr0();
 231     if ((s64)diff < 0) {  // s1.addr0 < s2.addr0  // NOLINT
 232       // if (s1.addr0() + size1) > s2.addr0()) return true;
 233       if (s1.size() > -diff)
 234         res = true;
 235     } else {
 236       // if (s2.addr0() + kS2AccessSize > s1.addr0()) return true;
 237       if (kS2AccessSize > diff)
 238         res = true;
 239     }
 240     DCHECK_EQ(res, TwoRangesIntersectSlow(s1, s2));
 241     DCHECK_EQ(res, TwoRangesIntersectSlow(s2, s1));
 242     return res;
 243   }
 244
 245   u64 ALWAYS_INLINE addr0() const { return (x_ >> kClkBits) & 7; }
 246   u64 ALWAYS_INLINE size() const { return 1ull << size_log(); }
 247   bool ALWAYS_INLINE IsWrite() const { return !IsRead(); }
 248   bool ALWAYS_INLINE IsRead() const { return x_ & kReadBit; }
 249
 250   // The idea behind the freed bit is as follows.
 251   // When the memory is freed (or otherwise unaccessible) we write to the shadow
 252   // values with tid/epoch related to the free and the freed bit set.
 253   // During memory accesses processing the freed bit is considered
 254   // as msb of tid. So any access races with shadow with freed bit set
 255   // (it is as if write from a thread with which we never synchronized before).
 256   // This allows us to detect accesses to freed memory w/o additional
 257   // overheads in memory access processing and at the same time restore
 258   // tid/epoch of free.
 259   void MarkAsFreed() {
 260      x_ |= kFreedBit;
 261   }
 262
 263   bool IsFreed() const {
 264     return x_ & kFreedBit;
 265   }
 266
 267   bool GetFreedAndReset() {
 268     bool res = x_ & kFreedBit;
 269     x_ &= ~kFreedBit;
 270     return res;
 271   }
 272
 273   bool ALWAYS_INLINE IsBothReadsOrAtomic(bool kIsWrite, bool kIsAtomic) const {
 274     bool v = x_ & ((u64(kIsWrite ^ 1) << kReadShift)
 275         | (u64(kIsAtomic) << kAtomicShift));
 276     DCHECK_EQ(v, (!IsWrite() && !kIsWrite) || (IsAtomic() && kIsAtomic));
 277     return v;
 278   }
 279
 280   bool ALWAYS_INLINE IsRWNotWeaker(bool kIsWrite, bool kIsAtomic) const {
 281     bool v = ((x_ >> kReadShift) & 3)
 282         <= u64((kIsWrite ^ 1) | (kIsAtomic << 1));
 283     DCHECK_EQ(v, (IsAtomic() < kIsAtomic) ||
 284         (IsAtomic() == kIsAtomic && !IsWrite() <= !kIsWrite));
 285     return v;
 286   }
 287
 288   bool ALWAYS_INLINE IsRWWeakerOrEqual(bool kIsWrite, bool kIsAtomic) const {
 289     bool v = ((x_ >> kReadShift) & 3)
 290         >= u64((kIsWrite ^ 1) | (kIsAtomic << 1));
 291     DCHECK_EQ(v, (IsAtomic() > kIsAtomic) ||
 292         (IsAtomic() == kIsAtomic && !IsWrite() >= !kIsWrite));
 293     return v;
 294   }
 295
 296  private:
 297   static const u64 kReadShift   = 5 + kClkBits;
 298   static const u64 kReadBit     = 1ull << kReadShift;
 299   static const u64 kAtomicShift = 6 + kClkBits;
 300   static const u64 kAtomicBit   = 1ull << kAtomicShift;
 301
 302   u64 size_log() const { return (x_ >> (3 + kClkBits)) & 3; }
 303
 304   static bool TwoRangesIntersectSlow(const Shadow s1, const Shadow s2) {
 305     if (s1.addr0() == s2.addr0()) return true;
 306     if (s1.addr0() < s2.addr0() && s1.addr0() + s1.size() > s2.addr0())
 307       return true;
 308     if (s2.addr0() < s1.addr0() && s2.addr0() + s2.size() > s1.addr0())
 309       return true;
 310     return false;
 311   }
 312 };
 313
 314 struct ThreadSignalContext;
 315
 316 struct JmpBuf {
 317   uptr sp;
 318   uptr mangled_sp;
 319   int int_signal_send;
 320   bool in_blocking_func;
 321   uptr in_signal_handler;
 322   uptr *shadow_stack_pos;
 323 };
 324
 325 // This struct is stored in TLS.
 326 struct ThreadState {
 327   FastState fast_state;
 328   // Synch epoch represents the threads's epoch before the last synchronization
 329   // action. It allows to reduce number of shadow state updates.
 330   // For example, fast_synch_epoch=100, last write to addr X was at epoch=150,
 331   // if we are processing write to X from the same thread at epoch=200,
 332   // we do nothing, because both writes happen in the same 'synch epoch'.
 333   // That is, if another memory access does not race with the former write,
 334   // it does not race with the latter as well.
 335   // QUESTION: can we can squeeze this into ThreadState::Fast?
 336   // E.g. ThreadState::Fast is a 44-bit, 32 are taken by synch_epoch and 12 are
 337   // taken by epoch between synchs.
 338   // This way we can save one load from tls.
 339   u64 fast_synch_epoch;
 340   // This is a slow path flag. On fast path, fast_state.GetIgnoreBit() is read.
 341   // We do not distinguish beteween ignoring reads and writes
 342   // for better performance.
 343   int ignore_reads_and_writes;
 344   int ignore_sync;
 345   // Go does not support ignores.
 346 #ifndef SANITIZER_GO
 347   IgnoreSet mop_ignore_set;
 348   IgnoreSet sync_ignore_set;
 349 #endif
 350   // C/C++ uses fixed size shadow stack embed into Trace.
 351   // Go uses malloc-allocated shadow stack with dynamic size.
 352   uptr *shadow_stack;
 353   uptr *shadow_stack_end;
 354   uptr *shadow_stack_pos;
 355   u64 *racy_shadow_addr;
 356   u64 racy_state[2];
 357   MutexSet mset;
 358   ThreadClock clock;
 359 #ifndef SANITIZER_GO
 360   AllocatorCache alloc_cache;
 361   InternalAllocatorCache internal_alloc_cache;
 362   Vector<JmpBuf> jmp_bufs;
 363   int ignore_interceptors;
 364 #endif
 365 #if TSAN_COLLECT_STATS
 366   u64 stat[StatCnt];
 367 #endif
 368   const int tid;
 369   const int unique_id;
 370   bool in_symbolizer;
 371   bool in_ignored_lib;
 372   bool is_inited;
 373   bool is_dead;
 374   bool is_freeing;
 375   bool is_vptr_access;
 376   const uptr stk_addr;
 377   const uptr stk_size;
 378   const uptr tls_addr;
 379   const uptr tls_size;
 380   ThreadContext *tctx;
 381
 382 #if SANITIZER_DEBUG && !SANITIZER_GO
 383   InternalDeadlockDetector internal_deadlock_detector;
 384 #endif
 385   DDPhysicalThread *dd_pt;
 386   DDLogicalThread *dd_lt;
 387
 388   atomic_uintptr_t in_signal_handler;
 389   ThreadSignalContext *signal_ctx;
 390
 391   DenseSlabAllocCache block_cache;
 392   DenseSlabAllocCache sync_cache;
 393   DenseSlabAllocCache clock_cache;
 394
 395 #ifndef SANITIZER_GO
 396   u32 last_sleep_stack_id;
 397   ThreadClock last_sleep_clock;
 398 #endif
 399
 400   // Set in regions of runtime that must be signal-safe and fork-safe.
 401   // If set, malloc must not be called.
 402   int nomalloc;
 403
 404   explicit ThreadState(Context *ctx, int tid, int unique_id, u64 epoch,
 405                        unsigned reuse_count,
 406                        uptr stk_addr, uptr stk_size,
 407                        uptr tls_addr, uptr tls_size);
 408 };
 409
 410 #ifndef SANITIZER_GO
 411 #if SANITIZER_MAC
 412 ThreadState *cur_thread();
 413 void cur_thread_finalize();
 414 #else
 415 __attribute__((tls_model("initial-exec")))
 416 extern THREADLOCAL char cur_thread_placeholder[];
 417 INLINE ThreadState *cur_thread() {
 418   return reinterpret_cast<ThreadState *>(&cur_thread_placeholder);
 419 }
 420 INLINE void cur_thread_finalize() { }
 421 #endif  // SANITIZER_MAC
 422 #endif  // SANITIZER_GO
 423
 424 class ThreadContext : public ThreadContextBase {
 425  public:
 426   explicit ThreadContext(int tid);
 427   ~ThreadContext();
 428   ThreadState *thr;
 429   u32 creation_stack_id;
 430   SyncClock sync;
 431   // Epoch at which the thread had started.
 432   // If we see an event from the thread stamped by an older epoch,
 433   // the event is from a dead thread that shared tid with this thread.
 434   u64 epoch0;
 435   u64 epoch1;
 436
 437   // Override superclass callbacks.
 438   void OnDead() override;
 439   void OnJoined(void *arg) override;
 440   void OnFinished() override;
 441   void OnStarted(void *arg) override;
 442   void OnCreated(void *arg) override;
 443   void OnReset() override;
 444   void OnDetached(void *arg) override;
 445 };
 446
 447 struct RacyStacks {
 448   MD5Hash hash[2];
 449   bool operator==(const RacyStacks &other) const {
 450     if (hash[0] == other.hash[0] && hash[1] == other.hash[1])
 451       return true;
 452     if (hash[0] == other.hash[1] && hash[1] == other.hash[0])
 453       return true;
 454     return false;
 455   }
 456 };
 457
 458 struct RacyAddress {
 459   uptr addr_min;
 460   uptr addr_max;
 461 };
 462
 463 struct FiredSuppression {
 464   ReportType type;
 465   uptr pc_or_addr;
 466   Suppression *supp;
 467 };
 468
 469 struct Context {
 470   Context();
 471
 472   bool initialized;
 473   bool after_multithreaded_fork;
 474
 475   MetaMap metamap;
 476
 477   Mutex report_mtx;
 478   int nreported;
 479   int nmissed_expected;
 480   atomic_uint64_t last_symbolize_time_ns;
 481
 482   void *background_thread;
 483   atomic_uint32_t stop_background_thread;
 484
 485   ThreadRegistry *thread_registry;
 486
 487   Mutex racy_mtx;
 488   Vector<RacyStacks> racy_stacks;
 489   Vector<RacyAddress> racy_addresses;
 490   // Number of fired suppressions may be large enough.
 491   Mutex fired_suppressions_mtx;
 492   InternalMmapVector<FiredSuppression> fired_suppressions;
 493   DDetector *dd;
 494
 495   ClockAlloc clock_alloc;
 496
 497   Flags flags;
 498
 499   u64 stat[StatCnt];
 500   u64 int_alloc_cnt[MBlockTypeCount];
 501   u64 int_alloc_siz[MBlockTypeCount];
 502 };
 503
 504 extern Context *ctx;  // The one and the only global runtime context.
 505
 506 struct ScopedIgnoreInterceptors {
 507   ScopedIgnoreInterceptors() {
 508 #ifndef SANITIZER_GO
 509     cur_thread()->ignore_interceptors++;
 510 #endif
 511   }
 512
 513   ~ScopedIgnoreInterceptors() {
 514 #ifndef SANITIZER_GO
 515     cur_thread()->ignore_interceptors--;
 516 #endif
 517   }
 518 };
 519
 520 class ScopedReport {
 521  public:
 522   explicit ScopedReport(ReportType typ);
 523   ~ScopedReport();
 524
 525   void AddMemoryAccess(uptr addr, Shadow s, StackTrace stack,
 526                        const MutexSet *mset);
 527   void AddStack(StackTrace stack, bool suppressable = false);
 528   void AddThread(const ThreadContext *tctx, bool suppressable = false);
 529   void AddThread(int unique_tid, bool suppressable = false);
 530   void AddUniqueTid(int unique_tid);
 531   void AddMutex(const SyncVar *s);
 532   u64 AddMutex(u64 id);
 533   void AddLocation(uptr addr, uptr size);
 534   void AddSleep(u32 stack_id);
 535   void SetCount(int count);
 536
 537   const ReportDesc *GetReport() const;
 538
 539  private:
 540   ReportDesc *rep_;
 541   // Symbolizer makes lots of intercepted calls. If we try to process them,
 542   // at best it will cause deadlocks on internal mutexes.
 543   ScopedIgnoreInterceptors ignore_interceptors_;
 544
 545   void AddDeadMutex(u64 id);
 546
 547   ScopedReport(const ScopedReport&);
 548   void operator = (const ScopedReport&);
 549 };
 550
 551 void RestoreStack(int tid, const u64 epoch, VarSizeStackTrace *stk,
 552                   MutexSet *mset);
 553
 554 template<typename StackTraceTy>
 555 void ObtainCurrentStack(ThreadState *thr, uptr toppc, StackTraceTy *stack) {
 556   uptr size = thr->shadow_stack_pos - thr->shadow_stack;
 557   uptr start = 0;
 558   if (size + !!toppc > kStackTraceMax) {
 559     start = size + !!toppc - kStackTraceMax;
 560     size = kStackTraceMax - !!toppc;
 561   }
 562   stack->Init(&thr->shadow_stack[start], size, toppc);
 563 }
 564
 565
 566 #if TSAN_COLLECT_STATS
 567 void StatAggregate(u64 *dst, u64 *src);
 568 void StatOutput(u64 *stat);
 569 #endif
 570
 571 void ALWAYS_INLINE StatInc(ThreadState *thr, StatType typ, u64 n = 1) {
 572 #if TSAN_COLLECT_STATS
 573   thr->stat[typ] += n;
 574 #endif
 575 }
 576 void ALWAYS_INLINE StatSet(ThreadState *thr, StatType typ, u64 n) {
 577 #if TSAN_COLLECT_STATS
 578   thr->stat[typ] = n;
 579 #endif
 580 }
 581
 582 void MapShadow(uptr addr, uptr size);
 583 void MapThreadTrace(uptr addr, uptr size, const char *name);
 584 void DontNeedShadowFor(uptr addr, uptr size);
 585 void InitializeShadowMemory();
 586 void InitializeInterceptors();
 587 void InitializeLibIgnore();
 588 void InitializeDynamicAnnotations();
 589
 590 void ForkBefore(ThreadState *thr, uptr pc);
 591 void ForkParentAfter(ThreadState *thr, uptr pc);
 592 void ForkChildAfter(ThreadState *thr, uptr pc);
 593
 594 void ReportRace(ThreadState *thr);
 595 bool OutputReport(ThreadState *thr, const ScopedReport &srep);
 596 bool IsFiredSuppression(Context *ctx, ReportType type, StackTrace trace);
 597 bool IsExpectedReport(uptr addr, uptr size);
 598 void PrintMatchedBenignRaces();
 599
 600 #if defined(TSAN_DEBUG_OUTPUT) && TSAN_DEBUG_OUTPUT >= 1
 601 # define DPrintf Printf
 602 #else
 603 # define DPrintf(...)
 604 #endif
 605
 606 #if defined(TSAN_DEBUG_OUTPUT) && TSAN_DEBUG_OUTPUT >= 2
 607 # define DPrintf2 Printf
 608 #else
 609 # define DPrintf2(...)
 610 #endif
 611
 612 u32 CurrentStackId(ThreadState *thr, uptr pc);
 613 ReportStack *SymbolizeStackId(u32 stack_id);
 614 void PrintCurrentStack(ThreadState *thr, uptr pc);
 615 void PrintCurrentStackSlow(uptr pc);  // uses libunwind
 616
 617 void Initialize(ThreadState *thr);
 618 int Finalize(ThreadState *thr);
 619
 620 void OnUserAlloc(ThreadState *thr, uptr pc, uptr p, uptr sz, bool write);
 621 void OnUserFree(ThreadState *thr, uptr pc, uptr p, bool write);
 622
 623 void MemoryAccess(ThreadState *thr, uptr pc, uptr addr,
 624     int kAccessSizeLog, bool kAccessIsWrite, bool kIsAtomic);
 625 void MemoryAccessImpl(ThreadState *thr, uptr addr,
 626     int kAccessSizeLog, bool kAccessIsWrite, bool kIsAtomic,
 627     u64 *shadow_mem, Shadow cur);
 628 void MemoryAccessRange(ThreadState *thr, uptr pc, uptr addr,
 629     uptr size, bool is_write);
 630 void MemoryAccessRangeStep(ThreadState *thr, uptr pc, uptr addr,
 631     uptr size, uptr step, bool is_write);
 632 void UnalignedMemoryAccess(ThreadState *thr, uptr pc, uptr addr,
 633     int size, bool kAccessIsWrite, bool kIsAtomic);
 634
 635 const int kSizeLog1 = 0;
 636 const int kSizeLog2 = 1;
 637 const int kSizeLog4 = 2;
 638 const int kSizeLog8 = 3;
 639
 640 void ALWAYS_INLINE MemoryRead(ThreadState *thr, uptr pc,
 641                                      uptr addr, int kAccessSizeLog) {
 642   MemoryAccess(thr, pc, addr, kAccessSizeLog, false, false);
 643 }
 644
 645 void ALWAYS_INLINE MemoryWrite(ThreadState *thr, uptr pc,
 646                                       uptr addr, int kAccessSizeLog) {
 647   MemoryAccess(thr, pc, addr, kAccessSizeLog, true, false);
 648 }
 649
 650 void ALWAYS_INLINE MemoryReadAtomic(ThreadState *thr, uptr pc,
 651                                            uptr addr, int kAccessSizeLog) {
 652   MemoryAccess(thr, pc, addr, kAccessSizeLog, false, true);
 653 }
 654
 655 void ALWAYS_INLINE MemoryWriteAtomic(ThreadState *thr, uptr pc,
 656                                             uptr addr, int kAccessSizeLog) {
 657   MemoryAccess(thr, pc, addr, kAccessSizeLog, true, true);
 658 }
 659
 660 void MemoryResetRange(ThreadState *thr, uptr pc, uptr addr, uptr size);
 661 void MemoryRangeFreed(ThreadState *thr, uptr pc, uptr addr, uptr size);
 662 void MemoryRangeImitateWrite(ThreadState *thr, uptr pc, uptr addr, uptr size);
 663
 664 void ThreadIgnoreBegin(ThreadState *thr, uptr pc);
 665 void ThreadIgnoreEnd(ThreadState *thr, uptr pc);
 666 void ThreadIgnoreSyncBegin(ThreadState *thr, uptr pc);
 667 void ThreadIgnoreSyncEnd(ThreadState *thr, uptr pc);
 668
 669 void FuncEntry(ThreadState *thr, uptr pc);
 670 void FuncExit(ThreadState *thr);
 671
 672 int ThreadCreate(ThreadState *thr, uptr pc, uptr uid, bool detached);
 673 void ThreadStart(ThreadState *thr, int tid, uptr os_id);
 674 void ThreadFinish(ThreadState *thr);
 675 int ThreadTid(ThreadState *thr, uptr pc, uptr uid);
 676 void ThreadJoin(ThreadState *thr, uptr pc, int tid);
 677 void ThreadDetach(ThreadState *thr, uptr pc, int tid);
 678 void ThreadFinalize(ThreadState *thr);
 679 void ThreadSetName(ThreadState *thr, const char *name);
 680 int ThreadCount(ThreadState *thr);
 681 void ProcessPendingSignals(ThreadState *thr);
 682
 683 void MutexCreate(ThreadState *thr, uptr pc, uptr addr,
 684                  bool rw, bool recursive, bool linker_init);
 685 void MutexDestroy(ThreadState *thr, uptr pc, uptr addr);
 686 void MutexLock(ThreadState *thr, uptr pc, uptr addr, int rec = 1,
 687                bool try_lock = false);
 688 int  MutexUnlock(ThreadState *thr, uptr pc, uptr addr, bool all = false);
 689 void MutexReadLock(ThreadState *thr, uptr pc, uptr addr, bool try_lock = false);
 690 void MutexReadUnlock(ThreadState *thr, uptr pc, uptr addr);
 691 void MutexReadOrWriteUnlock(ThreadState *thr, uptr pc, uptr addr);
 692 void MutexRepair(ThreadState *thr, uptr pc, uptr addr);  // call on EOWNERDEAD
 693
 694 void Acquire(ThreadState *thr, uptr pc, uptr addr);
 695 // AcquireGlobal synchronizes the current thread with all other threads.
 696 // In terms of happens-before relation, it draws a HB edge from all threads
 697 // (where they happen to execute right now) to the current thread. We use it to
 698 // handle Go finalizers. Namely, finalizer goroutine executes AcquireGlobal
 699 // right before executing finalizers. This provides a coarse, but simple
 700 // approximation of the actual required synchronization.
 701 void AcquireGlobal(ThreadState *thr, uptr pc);
 702 void Release(ThreadState *thr, uptr pc, uptr addr);
 703 void ReleaseStore(ThreadState *thr, uptr pc, uptr addr);
 704 void AfterSleep(ThreadState *thr, uptr pc);
 705 void AcquireImpl(ThreadState *thr, uptr pc, SyncClock *c);
 706 void ReleaseImpl(ThreadState *thr, uptr pc, SyncClock *c);
 707 void ReleaseStoreImpl(ThreadState *thr, uptr pc, SyncClock *c);
 708 void AcquireReleaseImpl(ThreadState *thr, uptr pc, SyncClock *c);
 709
 710 // The hacky call uses custom calling convention and an assembly thunk.
 711 // It is considerably faster that a normal call for the caller
 712 // if it is not executed (it is intended for slow paths from hot functions).
 713 // The trick is that the call preserves all registers and the compiler
 714 // does not treat it as a call.
 715 // If it does not work for you, use normal call.
 716 #if !SANITIZER_DEBUG && defined(__x86_64__) && !SANITIZER_MAC
 717 // The caller may not create the stack frame for itself at all,
 718 // so we create a reserve stack frame for it (1024b must be enough).
 719 #define HACKY_CALL(f) \
 720   __asm__ __volatile__("sub $1024, %%rsp;" \
 721                        CFI_INL_ADJUST_CFA_OFFSET(1024) \
 722                        ".hidden " #f "_thunk;" \
 723                        "call " #f "_thunk;" \
 724                        "add $1024, %%rsp;" \
 725                        CFI_INL_ADJUST_CFA_OFFSET(-1024) \
 726                        ::: "memory", "cc");
 727 #else
 728 #define HACKY_CALL(f) f()
 729 #endif
 730
 731 void TraceSwitch(ThreadState *thr);
 732 uptr TraceTopPC(ThreadState *thr);
 733 uptr TraceSize();
 734 uptr TraceParts();
 735 Trace *ThreadTrace(int tid);
 736
 737 extern "C" void __tsan_trace_switch();
 738 void ALWAYS_INLINE TraceAddEvent(ThreadState *thr, FastState fs,
 739                                         EventType typ, u64 addr) {
 740   if (!kCollectHistory)
 741     return;
 742   DCHECK_GE((int)typ, 0);
 743   DCHECK_LE((int)typ, 7);
 744   DCHECK_EQ(GetLsb(addr, 61), addr);
 745   StatInc(thr, StatEvents);
 746   u64 pos = fs.GetTracePos();
 747   if (UNLIKELY((pos % kTracePartSize) == 0)) {
 748 #ifndef SANITIZER_GO
 749     HACKY_CALL(__tsan_trace_switch);
 750 #else
 751     TraceSwitch(thr);
 752 #endif
 753   }
 754   Event *trace = (Event*)GetThreadTrace(fs.tid());
 755   Event *evp = &trace[pos];
 756   Event ev = (u64)addr | ((u64)typ << 61);
 757   *evp = ev;
 758 }
 759
 760 #ifndef SANITIZER_GO
 761 uptr ALWAYS_INLINE HeapEnd() {
 762   return kHeapMemEnd + PrimaryAllocator::AdditionalSize();
 763 }
 764 #endif
 765
 766 }  // namespace __tsan
 767
 768 #endif  // TSAN_RTL_H