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 #if !SANITIZER_GO
  54 struct MapUnmapCallback;
  55 #if defined(__mips64) || defined(__aarch64__) || defined(__powerpc__)
  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 struct AP64 {  // Allocator64 parameters. Deliberately using a short name.
  68   static const uptr kSpaceBeg = Mapping::kHeapMemBeg;
  69   static const uptr kSpaceSize = Mapping::kHeapMemEnd - Mapping::kHeapMemBeg;
  70   static const uptr kMetadataSize = 0;
  71   typedef DefaultSizeClassMap SizeClassMap;
  72   typedef __tsan::MapUnmapCallback MapUnmapCallback;
  73   static const uptr kFlags = 0;
  74 };
  75 typedef SizeClassAllocator64<AP64> PrimaryAllocator;
  76 #endif
  77 typedef SizeClassAllocatorLocalCache<PrimaryAllocator> AllocatorCache;
  78 typedef LargeMmapAllocator<MapUnmapCallback> SecondaryAllocator;
  79 typedef CombinedAllocator<PrimaryAllocator, AllocatorCache,
  80     SecondaryAllocator> Allocator;
  81 Allocator *allocator();
  82 #endif
  83
  84 void TsanCheckFailed(const char *file, int line, const char *cond,
  85                      u64 v1, u64 v2);
  86
  87 const u64 kShadowRodata = (u64)-1;  // .rodata shadow marker
  88
  89 // FastState (from most significant bit):
  90 //   ignore          : 1
  91 //   tid             : kTidBits
  92 //   unused          : -
  93 //   history_size    : 3
  94 //   epoch           : kClkBits
  95 class FastState {
  96  public:
  97   FastState(u64 tid, u64 epoch) {
  98     x_ = tid << kTidShift;
  99     x_ |= epoch;
 100     DCHECK_EQ(tid, this->tid());
 101     DCHECK_EQ(epoch, this->epoch());
 102     DCHECK_EQ(GetIgnoreBit(), false);
 103   }
 104
 105   explicit FastState(u64 x)
 106       : x_(x) {
 107   }
 108
 109   u64 raw() const {
 110     return x_;
 111   }
 112
 113   u64 tid() const {
 114     u64 res = (x_ & ~kIgnoreBit) >> kTidShift;
 115     return res;
 116   }
 117
 118   u64 TidWithIgnore() const {
 119     u64 res = x_ >> kTidShift;
 120     return res;
 121   }
 122
 123   u64 epoch() const {
 124     u64 res = x_ & ((1ull << kClkBits) - 1);
 125     return res;
 126   }
 127
 128   void IncrementEpoch() {
 129     u64 old_epoch = epoch();
 130     x_ += 1;
 131     DCHECK_EQ(old_epoch + 1, epoch());
 132     (void)old_epoch;
 133   }
 134
 135   void SetIgnoreBit() { x_ |= kIgnoreBit; }
 136   void ClearIgnoreBit() { x_ &= ~kIgnoreBit; }
 137   bool GetIgnoreBit() const { return (s64)x_ < 0; }
 138
 139   void SetHistorySize(int hs) {
 140     CHECK_GE(hs, 0);
 141     CHECK_LE(hs, 7);
 142     x_ = (x_ & ~(kHistoryMask << kHistoryShift)) | (u64(hs) << kHistoryShift);
 143   }
 144
 145   ALWAYS_INLINE
 146   int GetHistorySize() const {
 147     return (int)((x_ >> kHistoryShift) & kHistoryMask);
 148   }
 149
 150   void ClearHistorySize() {
 151     SetHistorySize(0);
 152   }
 153
 154   ALWAYS_INLINE
 155   u64 GetTracePos() const {
 156     const int hs = GetHistorySize();
 157     // When hs == 0, the trace consists of 2 parts.
 158     const u64 mask = (1ull << (kTracePartSizeBits + hs + 1)) - 1;
 159     return epoch() & mask;
 160   }
 161
 162  private:
 163   friend class Shadow;
 164   static const int kTidShift = 64 - kTidBits - 1;
 165   static const u64 kIgnoreBit = 1ull << 63;
 166   static const u64 kFreedBit = 1ull << 63;
 167   static const u64 kHistoryShift = kClkBits;
 168   static const u64 kHistoryMask = 7;
 169   u64 x_;
 170 };
 171
 172 // Shadow (from most significant bit):
 173 //   freed           : 1
 174 //   tid             : kTidBits
 175 //   is_atomic       : 1
 176 //   is_read         : 1
 177 //   size_log        : 2
 178 //   addr0           : 3
 179 //   epoch           : kClkBits
 180 class Shadow : public FastState {
 181  public:
 182   explicit Shadow(u64 x)
 183       : FastState(x) {
 184   }
 185
 186   explicit Shadow(const FastState &s)
 187       : FastState(s.x_) {
 188     ClearHistorySize();
 189   }
 190
 191   void SetAddr0AndSizeLog(u64 addr0, unsigned kAccessSizeLog) {
 192     DCHECK_EQ((x_ >> kClkBits) & 31, 0);
 193     DCHECK_LE(addr0, 7);
 194     DCHECK_LE(kAccessSizeLog, 3);
 195     x_ |= ((kAccessSizeLog << 3) | addr0) << kClkBits;
 196     DCHECK_EQ(kAccessSizeLog, size_log());
 197     DCHECK_EQ(addr0, this->addr0());
 198   }
 199
 200   void SetWrite(unsigned kAccessIsWrite) {
 201     DCHECK_EQ(x_ & kReadBit, 0);
 202     if (!kAccessIsWrite)
 203       x_ |= kReadBit;
 204     DCHECK_EQ(kAccessIsWrite, IsWrite());
 205   }
 206
 207   void SetAtomic(bool kIsAtomic) {
 208     DCHECK(!IsAtomic());
 209     if (kIsAtomic)
 210       x_ |= kAtomicBit;
 211     DCHECK_EQ(IsAtomic(), kIsAtomic);
 212   }
 213
 214   bool IsAtomic() const {
 215     return x_ & kAtomicBit;
 216   }
 217
 218   bool IsZero() const {
 219     return x_ == 0;
 220   }
 221
 222   static inline bool TidsAreEqual(const Shadow s1, const Shadow s2) {
 223     u64 shifted_xor = (s1.x_ ^ s2.x_) >> kTidShift;
 224     DCHECK_EQ(shifted_xor == 0, s1.TidWithIgnore() == s2.TidWithIgnore());
 225     return shifted_xor == 0;
 226   }
 227
 228   static ALWAYS_INLINE
 229   bool Addr0AndSizeAreEqual(const Shadow s1, const Shadow s2) {
 230     u64 masked_xor = ((s1.x_ ^ s2.x_) >> kClkBits) & 31;
 231     return masked_xor == 0;
 232   }
 233
 234   static ALWAYS_INLINE bool TwoRangesIntersect(Shadow s1, Shadow s2,
 235       unsigned kS2AccessSize) {
 236     bool res = false;
 237     u64 diff = s1.addr0() - s2.addr0();
 238     if ((s64)diff < 0) {  // s1.addr0 < s2.addr0  // NOLINT
 239       // if (s1.addr0() + size1) > s2.addr0()) return true;
 240       if (s1.size() > -diff)
 241         res = true;
 242     } else {
 243       // if (s2.addr0() + kS2AccessSize > s1.addr0()) return true;
 244       if (kS2AccessSize > diff)
 245         res = true;
 246     }
 247     DCHECK_EQ(res, TwoRangesIntersectSlow(s1, s2));
 248     DCHECK_EQ(res, TwoRangesIntersectSlow(s2, s1));
 249     return res;
 250   }
 251
 252   u64 ALWAYS_INLINE addr0() const { return (x_ >> kClkBits) & 7; }
 253   u64 ALWAYS_INLINE size() const { return 1ull << size_log(); }
 254   bool ALWAYS_INLINE IsWrite() const { return !IsRead(); }
 255   bool ALWAYS_INLINE IsRead() const { return x_ & kReadBit; }
 256
 257   // The idea behind the freed bit is as follows.
 258   // When the memory is freed (or otherwise unaccessible) we write to the shadow
 259   // values with tid/epoch related to the free and the freed bit set.
 260   // During memory accesses processing the freed bit is considered
 261   // as msb of tid. So any access races with shadow with freed bit set
 262   // (it is as if write from a thread with which we never synchronized before).
 263   // This allows us to detect accesses to freed memory w/o additional
 264   // overheads in memory access processing and at the same time restore
 265   // tid/epoch of free.
 266   void MarkAsFreed() {
 267      x_ |= kFreedBit;
 268   }
 269
 270   bool IsFreed() const {
 271     return x_ & kFreedBit;
 272   }
 273
 274   bool GetFreedAndReset() {
 275     bool res = x_ & kFreedBit;
 276     x_ &= ~kFreedBit;
 277     return res;
 278   }
 279
 280   bool ALWAYS_INLINE IsBothReadsOrAtomic(bool kIsWrite, bool kIsAtomic) const {
 281     bool v = x_ & ((u64(kIsWrite ^ 1) << kReadShift)
 282         | (u64(kIsAtomic) << kAtomicShift));
 283     DCHECK_EQ(v, (!IsWrite() && !kIsWrite) || (IsAtomic() && kIsAtomic));
 284     return v;
 285   }
 286
 287   bool ALWAYS_INLINE IsRWNotWeaker(bool kIsWrite, bool kIsAtomic) const {
 288     bool v = ((x_ >> kReadShift) & 3)
 289         <= u64((kIsWrite ^ 1) | (kIsAtomic << 1));
 290     DCHECK_EQ(v, (IsAtomic() < kIsAtomic) ||
 291         (IsAtomic() == kIsAtomic && !IsWrite() <= !kIsWrite));
 292     return v;
 293   }
 294
 295   bool ALWAYS_INLINE IsRWWeakerOrEqual(bool kIsWrite, bool kIsAtomic) const {
 296     bool v = ((x_ >> kReadShift) & 3)
 297         >= u64((kIsWrite ^ 1) | (kIsAtomic << 1));
 298     DCHECK_EQ(v, (IsAtomic() > kIsAtomic) ||
 299         (IsAtomic() == kIsAtomic && !IsWrite() >= !kIsWrite));
 300     return v;
 301   }
 302
 303  private:
 304   static const u64 kReadShift   = 5 + kClkBits;
 305   static const u64 kReadBit     = 1ull << kReadShift;
 306   static const u64 kAtomicShift = 6 + kClkBits;
 307   static const u64 kAtomicBit   = 1ull << kAtomicShift;
 308
 309   u64 size_log() const { return (x_ >> (3 + kClkBits)) & 3; }
 310
 311   static bool TwoRangesIntersectSlow(const Shadow s1, const Shadow s2) {
 312     if (s1.addr0() == s2.addr0()) return true;
 313     if (s1.addr0() < s2.addr0() && s1.addr0() + s1.size() > s2.addr0())
 314       return true;
 315     if (s2.addr0() < s1.addr0() && s2.addr0() + s2.size() > s1.addr0())
 316       return true;
 317     return false;
 318   }
 319 };
 320
 321 struct ThreadSignalContext;
 322
 323 struct JmpBuf {
 324   uptr sp;
 325   uptr mangled_sp;
 326   int int_signal_send;
 327   bool in_blocking_func;
 328   uptr in_signal_handler;
 329   uptr *shadow_stack_pos;
 330 };
 331
 332 // A Processor represents a physical thread, or a P for Go.
 333 // It is used to store internal resources like allocate cache, and does not
 334 // participate in race-detection logic (invisible to end user).
 335 // In C++ it is tied to an OS thread just like ThreadState, however ideally
 336 // it should be tied to a CPU (this way we will have fewer allocator caches).
 337 // In Go it is tied to a P, so there are significantly fewer Processor's than
 338 // ThreadState's (which are tied to Gs).
 339 // A ThreadState must be wired with a Processor to handle events.
 340 struct Processor {
 341   ThreadState *thr; // currently wired thread, or nullptr
 342 #if !SANITIZER_GO
 343   AllocatorCache alloc_cache;
 344   InternalAllocatorCache internal_alloc_cache;
 345 #endif
 346   DenseSlabAllocCache block_cache;
 347   DenseSlabAllocCache sync_cache;
 348   DenseSlabAllocCache clock_cache;
 349   DDPhysicalThread *dd_pt;
 350 };
 351
 352 #if !SANITIZER_GO
 353 // ScopedGlobalProcessor temporary setups a global processor for the current
 354 // thread, if it does not have one. Intended for interceptors that can run
 355 // at the very thread end, when we already destroyed the thread processor.
 356 struct ScopedGlobalProcessor {
 357   ScopedGlobalProcessor();
 358   ~ScopedGlobalProcessor();
 359 };
 360 #endif
 361
 362 // This struct is stored in TLS.
 363 struct ThreadState {
 364   FastState fast_state;
 365   // Synch epoch represents the threads's epoch before the last synchronization
 366   // action. It allows to reduce number of shadow state updates.
 367   // For example, fast_synch_epoch=100, last write to addr X was at epoch=150,
 368   // if we are processing write to X from the same thread at epoch=200,
 369   // we do nothing, because both writes happen in the same 'synch epoch'.
 370   // That is, if another memory access does not race with the former write,
 371   // it does not race with the latter as well.
 372   // QUESTION: can we can squeeze this into ThreadState::Fast?
 373   // E.g. ThreadState::Fast is a 44-bit, 32 are taken by synch_epoch and 12 are
 374   // taken by epoch between synchs.
 375   // This way we can save one load from tls.
 376   u64 fast_synch_epoch;
 377   // This is a slow path flag. On fast path, fast_state.GetIgnoreBit() is read.
 378   // We do not distinguish beteween ignoring reads and writes
 379   // for better performance.
 380   int ignore_reads_and_writes;
 381   int ignore_sync;
 382   // Go does not support ignores.
 383 #if !SANITIZER_GO
 384   IgnoreSet mop_ignore_set;
 385   IgnoreSet sync_ignore_set;
 386 #endif
 387   // C/C++ uses fixed size shadow stack embed into Trace.
 388   // Go uses malloc-allocated shadow stack with dynamic size.
 389   uptr *shadow_stack;
 390   uptr *shadow_stack_end;
 391   uptr *shadow_stack_pos;
 392   u64 *racy_shadow_addr;
 393   u64 racy_state[2];
 394   MutexSet mset;
 395   ThreadClock clock;
 396 #if !SANITIZER_GO
 397   Vector<JmpBuf> jmp_bufs;
 398   int ignore_interceptors;
 399 #endif
 400 #if TSAN_COLLECT_STATS
 401   u64 stat[StatCnt];
 402 #endif
 403   const int tid;
 404   const int unique_id;
 405   bool in_symbolizer;
 406   bool in_ignored_lib;
 407   bool is_inited;
 408   bool is_dead;
 409   bool is_freeing;
 410   bool is_vptr_access;
 411   const uptr stk_addr;
 412   const uptr stk_size;
 413   const uptr tls_addr;
 414   const uptr tls_size;
 415   ThreadContext *tctx;
 416
 417 #if SANITIZER_DEBUG && !SANITIZER_GO
 418   InternalDeadlockDetector internal_deadlock_detector;
 419 #endif
 420   DDLogicalThread *dd_lt;
 421
 422   // Current wired Processor, or nullptr. Required to handle any events.
 423   Processor *proc1;
 424 #if !SANITIZER_GO
 425   Processor *proc() { return proc1; }
 426 #else
 427   Processor *proc();
 428 #endif
 429
 430   atomic_uintptr_t in_signal_handler;
 431   ThreadSignalContext *signal_ctx;
 432
 433 #if !SANITIZER_GO
 434   u32 last_sleep_stack_id;
 435   ThreadClock last_sleep_clock;
 436 #endif
 437
 438   // Set in regions of runtime that must be signal-safe and fork-safe.
 439   // If set, malloc must not be called.
 440   int nomalloc;
 441
 442   const ReportDesc *current_report;
 443
 444   explicit ThreadState(Context *ctx, int tid, int unique_id, u64 epoch,
 445                        unsigned reuse_count,
 446                        uptr stk_addr, uptr stk_size,
 447                        uptr tls_addr, uptr tls_size);
 448 };
 449
 450 #if !SANITIZER_GO
 451 #if SANITIZER_MAC || SANITIZER_ANDROID
 452 ThreadState *cur_thread();
 453 void cur_thread_finalize();
 454 #else
 455 __attribute__((tls_model("initial-exec")))
 456 extern THREADLOCAL char cur_thread_placeholder[];
 457 INLINE ThreadState *cur_thread() {
 458   return reinterpret_cast<ThreadState *>(&cur_thread_placeholder);
 459 }
 460 INLINE void cur_thread_finalize() { }
 461 #endif  // SANITIZER_MAC || SANITIZER_ANDROID
 462 #endif  // SANITIZER_GO
 463
 464 class ThreadContext : public ThreadContextBase {
 465  public:
 466   explicit ThreadContext(int tid);
 467   ~ThreadContext();
 468   ThreadState *thr;
 469   u32 creation_stack_id;
 470   SyncClock sync;
 471   // Epoch at which the thread had started.
 472   // If we see an event from the thread stamped by an older epoch,
 473   // the event is from a dead thread that shared tid with this thread.
 474   u64 epoch0;
 475   u64 epoch1;
 476
 477   // Override superclass callbacks.
 478   void OnDead() override;
 479   void OnJoined(void *arg) override;
 480   void OnFinished() override;
 481   void OnStarted(void *arg) override;
 482   void OnCreated(void *arg) override;
 483   void OnReset() override;
 484   void OnDetached(void *arg) override;
 485 };
 486
 487 struct RacyStacks {
 488   MD5Hash hash[2];
 489   bool operator==(const RacyStacks &other) const {
 490     if (hash[0] == other.hash[0] && hash[1] == other.hash[1])
 491       return true;
 492     if (hash[0] == other.hash[1] && hash[1] == other.hash[0])
 493       return true;
 494     return false;
 495   }
 496 };
 497
 498 struct RacyAddress {
 499   uptr addr_min;
 500   uptr addr_max;
 501 };
 502
 503 struct FiredSuppression {
 504   ReportType type;
 505   uptr pc_or_addr;
 506   Suppression *supp;
 507 };
 508
 509 struct Context {
 510   Context();
 511
 512   bool initialized;
 513   bool after_multithreaded_fork;
 514
 515   MetaMap metamap;
 516
 517   Mutex report_mtx;
 518   int nreported;
 519   int nmissed_expected;
 520   atomic_uint64_t last_symbolize_time_ns;
 521
 522   void *background_thread;
 523   atomic_uint32_t stop_background_thread;
 524
 525   ThreadRegistry *thread_registry;
 526
 527   Mutex racy_mtx;
 528   Vector<RacyStacks> racy_stacks;
 529   Vector<RacyAddress> racy_addresses;
 530   // Number of fired suppressions may be large enough.
 531   Mutex fired_suppressions_mtx;
 532   InternalMmapVector<FiredSuppression> fired_suppressions;
 533   DDetector *dd;
 534
 535   ClockAlloc clock_alloc;
 536
 537   Flags flags;
 538
 539   u64 stat[StatCnt];
 540   u64 int_alloc_cnt[MBlockTypeCount];
 541   u64 int_alloc_siz[MBlockTypeCount];
 542 };
 543
 544 extern Context *ctx;  // The one and the only global runtime context.
 545
 546 struct ScopedIgnoreInterceptors {
 547   ScopedIgnoreInterceptors() {
 548 #if !SANITIZER_GO
 549     cur_thread()->ignore_interceptors++;
 550 #endif
 551   }
 552
 553   ~ScopedIgnoreInterceptors() {
 554 #if !SANITIZER_GO
 555     cur_thread()->ignore_interceptors--;
 556 #endif
 557   }
 558 };
 559
 560 class ScopedReport {
 561  public:
 562   explicit ScopedReport(ReportType typ);
 563   ~ScopedReport();
 564
 565   void AddMemoryAccess(uptr addr, Shadow s, StackTrace stack,
 566                        const MutexSet *mset);
 567   void AddStack(StackTrace stack, bool suppressable = false);
 568   void AddThread(const ThreadContext *tctx, bool suppressable = false);
 569   void AddThread(int unique_tid, bool suppressable = false);
 570   void AddUniqueTid(int unique_tid);
 571   void AddMutex(const SyncVar *s);
 572   u64 AddMutex(u64 id);
 573   void AddLocation(uptr addr, uptr size);
 574   void AddSleep(u32 stack_id);
 575   void SetCount(int count);
 576
 577   const ReportDesc *GetReport() const;
 578
 579  private:
 580   ReportDesc *rep_;
 581   // Symbolizer makes lots of intercepted calls. If we try to process them,
 582   // at best it will cause deadlocks on internal mutexes.
 583   ScopedIgnoreInterceptors ignore_interceptors_;
 584
 585   void AddDeadMutex(u64 id);
 586
 587   ScopedReport(const ScopedReport&);
 588   void operator = (const ScopedReport&);
 589 };
 590
 591 void RestoreStack(int tid, const u64 epoch, VarSizeStackTrace *stk,
 592                   MutexSet *mset);
 593
 594 template<typename StackTraceTy>
 595 void ObtainCurrentStack(ThreadState *thr, uptr toppc, StackTraceTy *stack) {
 596   uptr size = thr->shadow_stack_pos - thr->shadow_stack;
 597   uptr start = 0;
 598   if (size + !!toppc > kStackTraceMax) {
 599     start = size + !!toppc - kStackTraceMax;
 600     size = kStackTraceMax - !!toppc;
 601   }
 602   stack->Init(&thr->shadow_stack[start], size, toppc);
 603 }
 604
 605
 606 #if TSAN_COLLECT_STATS
 607 void StatAggregate(u64 *dst, u64 *src);
 608 void StatOutput(u64 *stat);
 609 #endif
 610
 611 void ALWAYS_INLINE StatInc(ThreadState *thr, StatType typ, u64 n = 1) {
 612 #if TSAN_COLLECT_STATS
 613   thr->stat[typ] += n;
 614 #endif
 615 }
 616 void ALWAYS_INLINE StatSet(ThreadState *thr, StatType typ, u64 n) {
 617 #if TSAN_COLLECT_STATS
 618   thr->stat[typ] = n;
 619 #endif
 620 }
 621
 622 void MapShadow(uptr addr, uptr size);
 623 void MapThreadTrace(uptr addr, uptr size, const char *name);
 624 void DontNeedShadowFor(uptr addr, uptr size);
 625 void InitializeShadowMemory();
 626 void InitializeInterceptors();
 627 void InitializeLibIgnore();
 628 void InitializeDynamicAnnotations();
 629
 630 void ForkBefore(ThreadState *thr, uptr pc);
 631 void ForkParentAfter(ThreadState *thr, uptr pc);
 632 void ForkChildAfter(ThreadState *thr, uptr pc);
 633
 634 void ReportRace(ThreadState *thr);
 635 bool OutputReport(ThreadState *thr, const ScopedReport &srep);
 636 bool IsFiredSuppression(Context *ctx, ReportType type, StackTrace trace);
 637 bool IsExpectedReport(uptr addr, uptr size);
 638 void PrintMatchedBenignRaces();
 639
 640 #if defined(TSAN_DEBUG_OUTPUT) && TSAN_DEBUG_OUTPUT >= 1
 641 # define DPrintf Printf
 642 #else
 643 # define DPrintf(...)
 644 #endif
 645
 646 #if defined(TSAN_DEBUG_OUTPUT) && TSAN_DEBUG_OUTPUT >= 2
 647 # define DPrintf2 Printf
 648 #else
 649 # define DPrintf2(...)
 650 #endif
 651
 652 u32 CurrentStackId(ThreadState *thr, uptr pc);
 653 ReportStack *SymbolizeStackId(u32 stack_id);
 654 void PrintCurrentStack(ThreadState *thr, uptr pc);
 655 void PrintCurrentStackSlow(uptr pc);  // uses libunwind
 656
 657 void Initialize(ThreadState *thr);
 658 int Finalize(ThreadState *thr);
 659
 660 void OnUserAlloc(ThreadState *thr, uptr pc, uptr p, uptr sz, bool write);
 661 void OnUserFree(ThreadState *thr, uptr pc, uptr p, bool write);
 662
 663 void MemoryAccess(ThreadState *thr, uptr pc, uptr addr,
 664     int kAccessSizeLog, bool kAccessIsWrite, bool kIsAtomic);
 665 void MemoryAccessImpl(ThreadState *thr, uptr addr,
 666     int kAccessSizeLog, bool kAccessIsWrite, bool kIsAtomic,
 667     u64 *shadow_mem, Shadow cur);
 668 void MemoryAccessRange(ThreadState *thr, uptr pc, uptr addr,
 669     uptr size, bool is_write);
 670 void MemoryAccessRangeStep(ThreadState *thr, uptr pc, uptr addr,
 671     uptr size, uptr step, bool is_write);
 672 void UnalignedMemoryAccess(ThreadState *thr, uptr pc, uptr addr,
 673     int size, bool kAccessIsWrite, bool kIsAtomic);
 674
 675 const int kSizeLog1 = 0;
 676 const int kSizeLog2 = 1;
 677 const int kSizeLog4 = 2;
 678 const int kSizeLog8 = 3;
 679
 680 void ALWAYS_INLINE MemoryRead(ThreadState *thr, uptr pc,
 681                                      uptr addr, int kAccessSizeLog) {
 682   MemoryAccess(thr, pc, addr, kAccessSizeLog, false, false);
 683 }
 684
 685 void ALWAYS_INLINE MemoryWrite(ThreadState *thr, uptr pc,
 686                                       uptr addr, int kAccessSizeLog) {
 687   MemoryAccess(thr, pc, addr, kAccessSizeLog, true, false);
 688 }
 689
 690 void ALWAYS_INLINE MemoryReadAtomic(ThreadState *thr, uptr pc,
 691                                            uptr addr, int kAccessSizeLog) {
 692   MemoryAccess(thr, pc, addr, kAccessSizeLog, false, true);
 693 }
 694
 695 void ALWAYS_INLINE MemoryWriteAtomic(ThreadState *thr, uptr pc,
 696                                             uptr addr, int kAccessSizeLog) {
 697   MemoryAccess(thr, pc, addr, kAccessSizeLog, true, true);
 698 }
 699
 700 void MemoryResetRange(ThreadState *thr, uptr pc, uptr addr, uptr size);
 701 void MemoryRangeFreed(ThreadState *thr, uptr pc, uptr addr, uptr size);
 702 void MemoryRangeImitateWrite(ThreadState *thr, uptr pc, uptr addr, uptr size);
 703
 704 void ThreadIgnoreBegin(ThreadState *thr, uptr pc);
 705 void ThreadIgnoreEnd(ThreadState *thr, uptr pc);
 706 void ThreadIgnoreSyncBegin(ThreadState *thr, uptr pc);
 707 void ThreadIgnoreSyncEnd(ThreadState *thr, uptr pc);
 708
 709 void FuncEntry(ThreadState *thr, uptr pc);
 710 void FuncExit(ThreadState *thr);
 711
 712 int ThreadCreate(ThreadState *thr, uptr pc, uptr uid, bool detached);
 713 void ThreadStart(ThreadState *thr, int tid, uptr os_id);
 714 void ThreadFinish(ThreadState *thr);
 715 int ThreadTid(ThreadState *thr, uptr pc, uptr uid);
 716 void ThreadJoin(ThreadState *thr, uptr pc, int tid);
 717 void ThreadDetach(ThreadState *thr, uptr pc, int tid);
 718 void ThreadFinalize(ThreadState *thr);
 719 void ThreadSetName(ThreadState *thr, const char *name);
 720 int ThreadCount(ThreadState *thr);
 721 void ProcessPendingSignals(ThreadState *thr);
 722
 723 Processor *ProcCreate();
 724 void ProcDestroy(Processor *proc);
 725 void ProcWire(Processor *proc, ThreadState *thr);
 726 void ProcUnwire(Processor *proc, ThreadState *thr);
 727
 728 void MutexCreate(ThreadState *thr, uptr pc, uptr addr,
 729                  bool rw, bool recursive, bool linker_init);
 730 void MutexDestroy(ThreadState *thr, uptr pc, uptr addr);
 731 void MutexLock(ThreadState *thr, uptr pc, uptr addr, int rec = 1,
 732                bool try_lock = false);
 733 int  MutexUnlock(ThreadState *thr, uptr pc, uptr addr, bool all = false);
 734 void MutexReadLock(ThreadState *thr, uptr pc, uptr addr, bool try_lock = false);
 735 void MutexReadUnlock(ThreadState *thr, uptr pc, uptr addr);
 736 void MutexReadOrWriteUnlock(ThreadState *thr, uptr pc, uptr addr);
 737 void MutexRepair(ThreadState *thr, uptr pc, uptr addr);  // call on EOWNERDEAD
 738 void MutexInvalidAccess(ThreadState *thr, uptr pc, uptr addr);
 739
 740 void Acquire(ThreadState *thr, uptr pc, uptr addr);
 741 // AcquireGlobal synchronizes the current thread with all other threads.
 742 // In terms of happens-before relation, it draws a HB edge from all threads
 743 // (where they happen to execute right now) to the current thread. We use it to
 744 // handle Go finalizers. Namely, finalizer goroutine executes AcquireGlobal
 745 // right before executing finalizers. This provides a coarse, but simple
 746 // approximation of the actual required synchronization.
 747 void AcquireGlobal(ThreadState *thr, uptr pc);
 748 void Release(ThreadState *thr, uptr pc, uptr addr);
 749 void ReleaseStore(ThreadState *thr, uptr pc, uptr addr);
 750 void AfterSleep(ThreadState *thr, uptr pc);
 751 void AcquireImpl(ThreadState *thr, uptr pc, SyncClock *c);
 752 void ReleaseImpl(ThreadState *thr, uptr pc, SyncClock *c);
 753 void ReleaseStoreImpl(ThreadState *thr, uptr pc, SyncClock *c);
 754 void AcquireReleaseImpl(ThreadState *thr, uptr pc, SyncClock *c);
 755
 756 // The hacky call uses custom calling convention and an assembly thunk.
 757 // It is considerably faster that a normal call for the caller
 758 // if it is not executed (it is intended for slow paths from hot functions).
 759 // The trick is that the call preserves all registers and the compiler
 760 // does not treat it as a call.
 761 // If it does not work for you, use normal call.
 762 #if !SANITIZER_DEBUG && defined(__x86_64__) && !SANITIZER_MAC
 763 // The caller may not create the stack frame for itself at all,
 764 // so we create a reserve stack frame for it (1024b must be enough).
 765 #define HACKY_CALL(f) \
 766   __asm__ __volatile__("sub $1024, %%rsp;" \
 767                        CFI_INL_ADJUST_CFA_OFFSET(1024) \
 768                        ".hidden " #f "_thunk;" \
 769                        "call " #f "_thunk;" \
 770                        "add $1024, %%rsp;" \
 771                        CFI_INL_ADJUST_CFA_OFFSET(-1024) \
 772                        ::: "memory", "cc");
 773 #else
 774 #define HACKY_CALL(f) f()
 775 #endif
 776
 777 void TraceSwitch(ThreadState *thr);
 778 uptr TraceTopPC(ThreadState *thr);
 779 uptr TraceSize();
 780 uptr TraceParts();
 781 Trace *ThreadTrace(int tid);
 782
 783 extern "C" void __tsan_trace_switch();
 784 void ALWAYS_INLINE TraceAddEvent(ThreadState *thr, FastState fs,
 785                                         EventType typ, u64 addr) {
 786   if (!kCollectHistory)
 787     return;
 788   DCHECK_GE((int)typ, 0);
 789   DCHECK_LE((int)typ, 7);
 790   DCHECK_EQ(GetLsb(addr, 61), addr);
 791   StatInc(thr, StatEvents);
 792   u64 pos = fs.GetTracePos();
 793   if (UNLIKELY((pos % kTracePartSize) == 0)) {
 794 #if !SANITIZER_GO
 795     HACKY_CALL(__tsan_trace_switch);
 796 #else
 797     TraceSwitch(thr);
 798 #endif
 799   }
 800   Event *trace = (Event*)GetThreadTrace(fs.tid());
 801   Event *evp = &trace[pos];
 802   Event ev = (u64)addr | ((u64)typ << 61);
 803   *evp = ev;
 804 }
 805
 806 #if !SANITIZER_GO
 807 uptr ALWAYS_INLINE HeapEnd() {
 808   return HeapMemEnd() + PrimaryAllocator::AdditionalSize();
 809 }
 810 #endif
 811
 812 }  // namespace __tsan
 813
 814 #endif  // TSAN_RTL_H