libsanitizer merge from upstream r175049

From-SVN: r196009

30 files changed, 675 insertions, 190 deletions

diff --git a/libsanitizer/tsan/tsan_defs.h b/libsanitizer/tsan/tsan_defs.h
index 6683a4e1abb..b2937a428f3 100644
--- a/libsanitizer/tsan/tsan_defs.h
+++ b/libsanitizer/tsan/tsan_defs.h
@@ -26,16 +26,19 @@ namespace __tsan {
 const bool kGoMode = true;
 const bool kCppMode = false;
 const char *const kTsanOptionsEnv = "GORACE";
+// Go linker does not support weak symbols.
+#define CPP_WEAK
 #else
 const bool kGoMode = false;
 const bool kCppMode = true;
 const char *const kTsanOptionsEnv = "TSAN_OPTIONS";
+#define CPP_WEAK WEAK
 #endif
 
 const int kTidBits = 13;
 const unsigned kMaxTid = 1 << kTidBits;
 const unsigned kMaxTidInClock = kMaxTid * 2;  // This includes msb 'freed' bit.
-const int kClkBits = 43;
+const int kClkBits = 42;
 #ifndef TSAN_GO
 const int kShadowStackSize = 4 * 1024;
 const int kTraceStackSize = 256;
diff --git a/libsanitizer/tsan/tsan_fd.cc b/libsanitizer/tsan/tsan_fd.cc
index f640c4f893e..a75d9bde08a 100644
--- a/libsanitizer/tsan/tsan_fd.cc
+++ b/libsanitizer/tsan/tsan_fd.cc
@@ -148,7 +148,7 @@ void FdAcquire(ThreadState *thr, uptr pc, int fd) {
   FdDesc *d = fddesc(thr, pc, fd);
   FdSync *s = d->sync;
   DPrintf("#%d: FdAcquire(%d) -> %p\n", thr->tid, fd, s);
-  MemoryRead8Byte(thr, pc, (uptr)d);
+  MemoryRead(thr, pc, (uptr)d, kSizeLog8);
   if (s)
     Acquire(thr, pc, (uptr)s);
 }
@@ -159,20 +159,20 @@ void FdRelease(ThreadState *thr, uptr pc, int fd) {
   DPrintf("#%d: FdRelease(%d) -> %p\n", thr->tid, fd, s);
   if (s)
     Release(thr, pc, (uptr)s);
-  MemoryRead8Byte(thr, pc, (uptr)d);
+  MemoryRead(thr, pc, (uptr)d, kSizeLog8);
 }
 
 void FdAccess(ThreadState *thr, uptr pc, int fd) {
   DPrintf("#%d: FdAccess(%d)\n", thr->tid, fd);
   FdDesc *d = fddesc(thr, pc, fd);
-  MemoryRead8Byte(thr, pc, (uptr)d);
+  MemoryRead(thr, pc, (uptr)d, kSizeLog8);
 }
 
 void FdClose(ThreadState *thr, uptr pc, int fd) {
   DPrintf("#%d: FdClose(%d)\n", thr->tid, fd);
   FdDesc *d = fddesc(thr, pc, fd);
   // To catch races between fd usage and close.
-  MemoryWrite8Byte(thr, pc, (uptr)d);
+  MemoryWrite(thr, pc, (uptr)d, kSizeLog8);
   // We need to clear it, because if we do not intercept any call out there
   // that creates fd, we will hit false postives.
   MemoryResetRange(thr, pc, (uptr)d, 8);
@@ -191,7 +191,7 @@ void FdDup(ThreadState *thr, uptr pc, int oldfd, int newfd) {
   DPrintf("#%d: FdDup(%d, %d)\n", thr->tid, oldfd, newfd);
   // Ignore the case when user dups not yet connected socket.
   FdDesc *od = fddesc(thr, pc, oldfd);
-  MemoryRead8Byte(thr, pc, (uptr)od);
+  MemoryRead(thr, pc, (uptr)od, kSizeLog8);
   FdClose(thr, pc, newfd);
   init(thr, pc, newfd, ref(od->sync));
 }
diff --git a/libsanitizer/tsan/tsan_flags.cc b/libsanitizer/tsan/tsan_flags.cc
index 630bd75769b..ae748a13e15 100644
--- a/libsanitizer/tsan/tsan_flags.cc
+++ b/libsanitizer/tsan/tsan_flags.cc
@@ -43,6 +43,7 @@ void InitializeFlags(Flags *f, const char *env) {
   f->report_thread_leaks = true;
   f->report_destroy_locked = true;
   f->report_signal_unsafe = true;
+  f->report_atomic_races = true;
   f->force_seq_cst_atomics = false;
   f->strip_path_prefix = "";
   f->suppressions = "";
@@ -70,6 +71,7 @@ void InitializeFlags(Flags *f, const char *env) {
   ParseFlag(env, &f->report_thread_leaks, "report_thread_leaks");
   ParseFlag(env, &f->report_destroy_locked, "report_destroy_locked");
   ParseFlag(env, &f->report_signal_unsafe, "report_signal_unsafe");
+  ParseFlag(env, &f->report_atomic_races, "report_atomic_races");
   ParseFlag(env, &f->force_seq_cst_atomics, "force_seq_cst_atomics");
   ParseFlag(env, &f->strip_path_prefix, "strip_path_prefix");
   ParseFlag(env, &f->suppressions, "suppressions");
diff --git a/libsanitizer/tsan/tsan_flags.h b/libsanitizer/tsan/tsan_flags.h
index ed27363c2ff..480b41538f9 100644
--- a/libsanitizer/tsan/tsan_flags.h
+++ b/libsanitizer/tsan/tsan_flags.h
@@ -41,6 +41,8 @@ struct Flags {
   // Report violations of async signal-safety
   // (e.g. malloc() call from a signal handler).
   bool report_signal_unsafe;
+  // Report races between atomic and plain memory accesses.
+  bool report_atomic_races;
   // If set, all atomics are effectively sequentially consistent (seq_cst),
   // regardless of what user actually specified.
   bool force_seq_cst_atomics;
@@ -84,6 +86,6 @@ struct Flags {
 
 Flags *flags();
 void InitializeFlags(Flags *flags, const char *env);
-}
+}  // namespace __tsan
 
 #endif  // TSAN_FLAGS_H
diff --git a/libsanitizer/tsan/tsan_interceptors.cc b/libsanitizer/tsan/tsan_interceptors.cc
index d8f66de2327..8a54511b6e0 100644
--- a/libsanitizer/tsan/tsan_interceptors.cc
+++ b/libsanitizer/tsan/tsan_interceptors.cc
@@ -51,9 +51,12 @@ extern "C" int pthread_sigmask(int how, const sigset_t *set, sigset_t *oldset);
 extern "C" int sigfillset(sigset_t *set);
 extern "C" void *pthread_self();
 extern "C" void _exit(int status);
-extern "C" int __cxa_atexit(void (*func)(void *arg), void *arg, void *dso);
 extern "C" int *__errno_location();
 extern "C" int fileno_unlocked(void *stream);
+extern "C" void *__libc_malloc(uptr size);
+extern "C" void *__libc_calloc(uptr size, uptr n);
+extern "C" void *__libc_realloc(void *ptr, uptr size);
+extern "C" void __libc_free(void *ptr);
 const int PTHREAD_MUTEX_RECURSIVE = 1;
 const int PTHREAD_MUTEX_RECURSIVE_NP = 1;
 const int kPthreadAttrSize = 56;
@@ -122,7 +125,7 @@ struct SignalContext {
   int pending_signal_count;
   SignalDesc pending_signals[kSigCount];
 };
-}
+}  // namespace __tsan
 
 static SignalContext *SigCtx(ThreadState *thr) {
   SignalContext *ctx = (SignalContext*)thr->signal_ctx;
@@ -238,12 +241,15 @@ class AtExitContext {
 
   typedef void(*atexit_t)();
 
-  int atexit(ThreadState *thr, uptr pc, atexit_t f) {
+  int atexit(ThreadState *thr, uptr pc, bool is_on_exit,
+             atexit_t f, void *arg) {
     Lock l(&mtx_);
     if (pos_ == kMaxAtExit)
       return 1;
     Release(thr, pc, (uptr)this);
     stack_[pos_] = f;
+    args_[pos_] = arg;
+    is_on_exits_[pos_] = is_on_exit;
     pos_++;
     return 0;
   }
@@ -252,11 +258,15 @@ class AtExitContext {
     CHECK_EQ(thr->in_rtl, 0);
     for (;;) {
       atexit_t f = 0;
+      void *arg = 0;
+      bool is_on_exit = false;
       {
         Lock l(&mtx_);
         if (pos_) {
           pos_--;
           f = stack_[pos_];
+          arg = args_[pos_];
+          is_on_exit = is_on_exits_[pos_];
           ScopedInRtl in_rtl;
           Acquire(thr, pc, (uptr)this);
         }
@@ -265,7 +275,10 @@ class AtExitContext {
         break;
       DPrintf("#%d: executing atexit func %p\n", thr->tid, f);
       CHECK_EQ(thr->in_rtl, 0);
-      f();
+      if (is_on_exit)
+        ((void(*)(int status, void *arg))f)(0, arg);
+      else
+        ((void(*)(void *arg, void *dso))f)(arg, 0);
     }
   }
 
@@ -273,6 +286,8 @@ class AtExitContext {
   static const int kMaxAtExit = 128;
   Mutex mtx_;
   atexit_t stack_[kMaxAtExit];
+  void *args_[kMaxAtExit];
+  bool is_on_exits_[kMaxAtExit];
   int pos_;
 };
 
@@ -282,18 +297,32 @@ static void finalize(void *arg) {
   ThreadState * thr = cur_thread();
   uptr pc = 0;
   atexit_ctx->exit(thr, pc);
-  {
-    ScopedInRtl in_rtl;
-    DestroyAndFree(atexit_ctx);
-  }
   int status = Finalize(cur_thread());
   if (status)
     _exit(status);
 }
 
 TSAN_INTERCEPTOR(int, atexit, void (*f)()) {
+  if (cur_thread()->in_symbolizer)
+    return 0;
   SCOPED_TSAN_INTERCEPTOR(atexit, f);
-  return atexit_ctx->atexit(thr, pc, f);
+  return atexit_ctx->atexit(thr, pc, false, (void(*)())f, 0);
+}
+
+TSAN_INTERCEPTOR(int, on_exit, void(*f)(int, void*), void *arg) {
+  if (cur_thread()->in_symbolizer)
+    return 0;
+  SCOPED_TSAN_INTERCEPTOR(on_exit, f, arg);
+  return atexit_ctx->atexit(thr, pc, true, (void(*)())f, arg);
+}
+
+TSAN_INTERCEPTOR(int, __cxa_atexit, void (*f)(void *a), void *arg, void *dso) {
+  if (cur_thread()->in_symbolizer)
+    return 0;
+  SCOPED_TSAN_INTERCEPTOR(__cxa_atexit, f, arg, dso);
+  if (dso)
+    return REAL(__cxa_atexit)(f, arg, dso);
+  return atexit_ctx->atexit(thr, pc, false, (void(*)())f, arg);
 }
 
 TSAN_INTERCEPTOR(void, longjmp, void *env, int val) {
@@ -309,6 +338,8 @@ TSAN_INTERCEPTOR(void, siglongjmp, void *env, int val) {
 }
 
 TSAN_INTERCEPTOR(void*, malloc, uptr size) {
+  if (cur_thread()->in_symbolizer)
+    return __libc_malloc(size);
   void *p = 0;
   {
     SCOPED_INTERCEPTOR_RAW(malloc, size);
@@ -324,6 +355,9 @@ TSAN_INTERCEPTOR(void*, __libc_memalign, uptr align, uptr sz) {
 }
 
 TSAN_INTERCEPTOR(void*, calloc, uptr size, uptr n) {
+  if (cur_thread()->in_symbolizer)
+    return __libc_calloc(size, n);
+  if (__sanitizer::CallocShouldReturnNullDueToOverflow(size, n)) return 0;
   void *p = 0;
   {
     SCOPED_INTERCEPTOR_RAW(calloc, size, n);
@@ -335,6 +369,8 @@ TSAN_INTERCEPTOR(void*, calloc, uptr size, uptr n) {
 }
 
 TSAN_INTERCEPTOR(void*, realloc, void *p, uptr size) {
+  if (cur_thread()->in_symbolizer)
+    return __libc_realloc(p, size);
   if (p)
     invoke_free_hook(p);
   {
@@ -348,6 +384,8 @@ TSAN_INTERCEPTOR(void*, realloc, void *p, uptr size) {
 TSAN_INTERCEPTOR(void, free, void *p) {
   if (p == 0)
     return;
+  if (cur_thread()->in_symbolizer)
+    return __libc_free(p);
   invoke_free_hook(p);
   SCOPED_INTERCEPTOR_RAW(free, p);
   user_free(thr, pc, p);
@@ -356,12 +394,16 @@ TSAN_INTERCEPTOR(void, free, void *p) {
 TSAN_INTERCEPTOR(void, cfree, void *p) {
   if (p == 0)
     return;
+  if (cur_thread()->in_symbolizer)
+    return __libc_free(p);
   invoke_free_hook(p);
   SCOPED_INTERCEPTOR_RAW(cfree, p);
   user_free(thr, pc, p);
 }
 
 #define OPERATOR_NEW_BODY(mangled_name) \
+  if (cur_thread()->in_symbolizer) \
+    return __libc_malloc(size); \
   void *p = 0; \
   {  \
     SCOPED_INTERCEPTOR_RAW(mangled_name, size); \
@@ -385,6 +427,8 @@ void *operator new[](__sanitizer::uptr size, std::nothrow_t const&) {
 
 #define OPERATOR_DELETE_BODY(mangled_name) \
   if (ptr == 0) return;  \
+  if (cur_thread()->in_symbolizer) \
+    return __libc_free(ptr); \
   invoke_free_hook(ptr);  \
   SCOPED_INTERCEPTOR_RAW(mangled_name, ptr);  \
   user_free(thr, pc, ptr);
@@ -549,6 +593,8 @@ TSAN_INTERCEPTOR(void*, mmap, void *addr, long_t sz, int prot,
     return MAP_FAILED;
   void *res = REAL(mmap)(addr, sz, prot, flags, fd, off);
   if (res != MAP_FAILED) {
+    if (fd > 0)
+      FdAccess(thr, pc, fd);
     MemoryResetRange(thr, pc, (uptr)res, sz);
   }
   return res;
@@ -561,6 +607,8 @@ TSAN_INTERCEPTOR(void*, mmap64, void *addr, long_t sz, int prot,
     return MAP_FAILED;
   void *res = REAL(mmap64)(addr, sz, prot, flags, fd, off);
   if (res != MAP_FAILED) {
+    if (fd > 0)
+      FdAccess(thr, pc, fd);
     MemoryResetRange(thr, pc, (uptr)res, sz);
   }
   return res;
@@ -958,14 +1006,14 @@ TSAN_INTERCEPTOR(int, pthread_cond_timedwait, void *c, void *m, void *abstime) {
 
 TSAN_INTERCEPTOR(int, pthread_barrier_init, void *b, void *a, unsigned count) {
   SCOPED_TSAN_INTERCEPTOR(pthread_barrier_init, b, a, count);
-  MemoryWrite1Byte(thr, pc, (uptr)b);
+  MemoryWrite(thr, pc, (uptr)b, kSizeLog1);
   int res = REAL(pthread_barrier_init)(b, a, count);
   return res;
 }
 
 TSAN_INTERCEPTOR(int, pthread_barrier_destroy, void *b) {
   SCOPED_TSAN_INTERCEPTOR(pthread_barrier_destroy, b);
-  MemoryWrite1Byte(thr, pc, (uptr)b);
+  MemoryWrite(thr, pc, (uptr)b, kSizeLog1);
   int res = REAL(pthread_barrier_destroy)(b);
   return res;
 }
@@ -973,9 +1021,9 @@ TSAN_INTERCEPTOR(int, pthread_barrier_destroy, void *b) {
 TSAN_INTERCEPTOR(int, pthread_barrier_wait, void *b) {
   SCOPED_TSAN_INTERCEPTOR(pthread_barrier_wait, b);
   Release(thr, pc, (uptr)b);
-  MemoryRead1Byte(thr, pc, (uptr)b);
+  MemoryRead(thr, pc, (uptr)b, kSizeLog1);
   int res = REAL(pthread_barrier_wait)(b);
-  MemoryRead1Byte(thr, pc, (uptr)b);
+  MemoryRead(thr, pc, (uptr)b, kSizeLog1);
   if (res == 0 || res == PTHREAD_BARRIER_SERIAL_THREAD) {
     Acquire(thr, pc, (uptr)b);
   }
@@ -1062,6 +1110,74 @@ TSAN_INTERCEPTOR(int, sem_getvalue, void *s, int *sval) {
   return res;
 }
 
+TSAN_INTERCEPTOR(int, __xstat, int version, const char *path, void *buf) {
+  SCOPED_TSAN_INTERCEPTOR(__xstat, version, path, buf);
+  return REAL(__xstat)(version, path, buf);
+}
+
+TSAN_INTERCEPTOR(int, stat, const char *path, void *buf) {
+  SCOPED_TSAN_INTERCEPTOR(__xstat, 0, path, buf);
+  return REAL(__xstat)(0, path, buf);
+}
+
+TSAN_INTERCEPTOR(int, __xstat64, int version, const char *path, void *buf) {
+  SCOPED_TSAN_INTERCEPTOR(__xstat64, version, path, buf);
+  return REAL(__xstat64)(version, path, buf);
+}
+
+TSAN_INTERCEPTOR(int, stat64, const char *path, void *buf) {
+  SCOPED_TSAN_INTERCEPTOR(__xstat64, 0, path, buf);
+  return REAL(__xstat64)(0, path, buf);
+}
+
+TSAN_INTERCEPTOR(int, __lxstat, int version, const char *path, void *buf) {
+  SCOPED_TSAN_INTERCEPTOR(__lxstat, version, path, buf);
+  return REAL(__lxstat)(version, path, buf);
+}
+
+TSAN_INTERCEPTOR(int, lstat, const char *path, void *buf) {
+  SCOPED_TSAN_INTERCEPTOR(__lxstat, 0, path, buf);
+  return REAL(__lxstat)(0, path, buf);
+}
+
+TSAN_INTERCEPTOR(int, __lxstat64, int version, const char *path, void *buf) {
+  SCOPED_TSAN_INTERCEPTOR(__lxstat64, version, path, buf);
+  return REAL(__lxstat64)(version, path, buf);
+}
+
+TSAN_INTERCEPTOR(int, lstat64, const char *path, void *buf) {
+  SCOPED_TSAN_INTERCEPTOR(__lxstat64, 0, path, buf);
+  return REAL(__lxstat64)(0, path, buf);
+}
+
+TSAN_INTERCEPTOR(int, __fxstat, int version, int fd, void *buf) {
+  SCOPED_TSAN_INTERCEPTOR(__fxstat, version, fd, buf);
+  if (fd > 0)
+    FdAccess(thr, pc, fd);
+  return REAL(__fxstat)(version, fd, buf);
+}
+
+TSAN_INTERCEPTOR(int, fstat, int fd, void *buf) {
+  SCOPED_TSAN_INTERCEPTOR(__fxstat, 0, fd, buf);
+  if (fd > 0)
+    FdAccess(thr, pc, fd);
+  return REAL(__fxstat)(0, fd, buf);
+}
+
+TSAN_INTERCEPTOR(int, __fxstat64, int version, int fd, void *buf) {
+  SCOPED_TSAN_INTERCEPTOR(__fxstat64, version, fd, buf);
+  if (fd > 0)
+    FdAccess(thr, pc, fd);
+  return REAL(__fxstat64)(version, fd, buf);
+}
+
+TSAN_INTERCEPTOR(int, fstat64, int fd, void *buf) {
+  SCOPED_TSAN_INTERCEPTOR(__fxstat64, 0, fd, buf);
+  if (fd > 0)
+    FdAccess(thr, pc, fd);
+  return REAL(__fxstat64)(0, fd, buf);
+}
+
 TSAN_INTERCEPTOR(int, open, const char *name, int flags, int mode) {
   SCOPED_TSAN_INTERCEPTOR(open, name, flags, mode);
   int fd = REAL(open)(name, flags, mode);
@@ -1177,6 +1293,22 @@ TSAN_INTERCEPTOR(int, connect, int fd, void *addr, unsigned addrlen) {
   return res;
 }
 
+TSAN_INTERCEPTOR(int, bind, int fd, void *addr, unsigned addrlen) {
+  SCOPED_TSAN_INTERCEPTOR(bind, fd, addr, addrlen);
+  int res = REAL(bind)(fd, addr, addrlen);
+  if (fd > 0 && res == 0)
+    FdAccess(thr, pc, fd);
+  return res;
+}
+
+TSAN_INTERCEPTOR(int, listen, int fd, int backlog) {
+  SCOPED_TSAN_INTERCEPTOR(listen, fd, backlog);
+  int res = REAL(listen)(fd, backlog);
+  if (fd > 0 && res == 0)
+    FdAccess(thr, pc, fd);
+  return res;
+}
+
 TSAN_INTERCEPTOR(int, accept, int fd, void *addr, unsigned *addrlen) {
   SCOPED_TSAN_INTERCEPTOR(accept, fd, addr, addrlen);
   int fd2 = REAL(accept)(fd, addr, addrlen);
@@ -1223,6 +1355,18 @@ TSAN_INTERCEPTOR(int, __close, int fd) {
   return REAL(__close)(fd);
 }
 
+// glibc guts
+TSAN_INTERCEPTOR(void, __res_iclose, void *state, bool free_addr) {
+  SCOPED_TSAN_INTERCEPTOR(__res_iclose, state, free_addr);
+  int fds[64];
+  int cnt = ExtractResolvFDs(state, fds, ARRAY_SIZE(fds));
+  for (int i = 0; i < cnt; i++) {
+    if (fds[i] > 0)
+      FdClose(thr, pc, fds[i]);
+  }
+  REAL(__res_iclose)(state, free_addr);
+}
+
 TSAN_INTERCEPTOR(int, pipe, int *pipefd) {
   SCOPED_TSAN_INTERCEPTOR(pipe, pipefd);
   int res = REAL(pipe)(pipefd);
@@ -1765,6 +1909,18 @@ void InitializeInterceptors() {
   TSAN_INTERCEPT(sem_post);
   TSAN_INTERCEPT(sem_getvalue);
 
+  TSAN_INTERCEPT(stat);
+  TSAN_INTERCEPT(__xstat);
+  TSAN_INTERCEPT(stat64);
+  TSAN_INTERCEPT(__xstat64);
+  TSAN_INTERCEPT(lstat);
+  TSAN_INTERCEPT(__lxstat);
+  TSAN_INTERCEPT(lstat64);
+  TSAN_INTERCEPT(__lxstat64);
+  TSAN_INTERCEPT(fstat);
+  TSAN_INTERCEPT(__fxstat);
+  TSAN_INTERCEPT(fstat64);
+  TSAN_INTERCEPT(__fxstat64);
   TSAN_INTERCEPT(open);
   TSAN_INTERCEPT(open64);
   TSAN_INTERCEPT(creat);
@@ -1779,11 +1935,15 @@ void InitializeInterceptors() {
   TSAN_INTERCEPT(socket);
   TSAN_INTERCEPT(socketpair);
   TSAN_INTERCEPT(connect);
+  TSAN_INTERCEPT(bind);
+  TSAN_INTERCEPT(listen);
   TSAN_INTERCEPT(accept);
   TSAN_INTERCEPT(accept4);
   TSAN_INTERCEPT(epoll_create);
   TSAN_INTERCEPT(epoll_create1);
   TSAN_INTERCEPT(close);
+  TSAN_INTERCEPT(__close);
+  TSAN_INTERCEPT(__res_iclose);
   TSAN_INTERCEPT(pipe);
   TSAN_INTERCEPT(pipe2);
 
@@ -1826,6 +1986,8 @@ void InitializeInterceptors() {
   TSAN_INTERCEPT(munlockall);
 
   TSAN_INTERCEPT(fork);
+  TSAN_INTERCEPT(on_exit);
+  TSAN_INTERCEPT(__cxa_atexit);
 
   // Need to setup it, because interceptors check that the function is resolved.
   // But atexit is emitted directly into the module, so can't be resolved.
@@ -1833,7 +1995,7 @@ void InitializeInterceptors() {
   atexit_ctx = new(internal_alloc(MBlockAtExit, sizeof(AtExitContext)))
       AtExitContext();
 
-  if (__cxa_atexit(&finalize, 0, 0)) {
+  if (REAL(__cxa_atexit)(&finalize, 0, 0)) {
     Printf("ThreadSanitizer: failed to setup atexit callback\n");
     Die();
   }
diff --git a/libsanitizer/tsan/tsan_interface.cc b/libsanitizer/tsan/tsan_interface.cc
index 08a51651114..992e3834aae 100644
--- a/libsanitizer/tsan/tsan_interface.cc
+++ b/libsanitizer/tsan/tsan_interface.cc
@@ -22,13 +22,13 @@ void __tsan_init() {
 }
 
 void __tsan_read16(void *addr) {
-  MemoryRead8Byte(cur_thread(), CALLERPC, (uptr)addr);
-  MemoryRead8Byte(cur_thread(), CALLERPC, (uptr)addr + 8);
+  MemoryRead(cur_thread(), CALLERPC, (uptr)addr, kSizeLog8);
+  MemoryRead(cur_thread(), CALLERPC, (uptr)addr + 8, kSizeLog8);
 }
 
 void __tsan_write16(void *addr) {
-  MemoryWrite8Byte(cur_thread(), CALLERPC, (uptr)addr);
-  MemoryWrite8Byte(cur_thread(), CALLERPC, (uptr)addr + 8);
+  MemoryWrite(cur_thread(), CALLERPC, (uptr)addr, kSizeLog8);
+  MemoryWrite(cur_thread(), CALLERPC, (uptr)addr + 8, kSizeLog8);
 }
 
 void __tsan_acquire(void *addr) {
diff --git a/libsanitizer/tsan/tsan_interface.h b/libsanitizer/tsan/tsan_interface.h
index 8f265ca27b5..2cfd7684183 100644
--- a/libsanitizer/tsan/tsan_interface.h
+++ b/libsanitizer/tsan/tsan_interface.h
@@ -14,7 +14,7 @@
 #ifndef TSAN_INTERFACE_H
 #define TSAN_INTERFACE_H
 
-#include <sanitizer/common_interface_defs.h>
+#include <sanitizer_common/sanitizer_internal_defs.h>
 
 // This header should NOT include any other headers.
 // All functions in this header are extern "C" and start with __tsan_.
diff --git a/libsanitizer/tsan/tsan_interface_ann.h b/libsanitizer/tsan/tsan_interface_ann.h
index b10264cdd6f..b6500329428 100644
--- a/libsanitizer/tsan/tsan_interface_ann.h
+++ b/libsanitizer/tsan/tsan_interface_ann.h
@@ -12,7 +12,7 @@
 #ifndef TSAN_INTERFACE_ANN_H
 #define TSAN_INTERFACE_ANN_H
 
-#include <sanitizer/common_interface_defs.h>
+#include <sanitizer_common/sanitizer_internal_defs.h>
 
 // This header should NOT include any other headers.
 // All functions in this header are extern "C" and start with __tsan_.
diff --git a/libsanitizer/tsan/tsan_interface_atomic.cc b/libsanitizer/tsan/tsan_interface_atomic.cc
index 770f8bd1014..2c8b2ab049c 100644
--- a/libsanitizer/tsan/tsan_interface_atomic.cc
+++ b/libsanitizer/tsan/tsan_interface_atomic.cc
@@ -18,25 +18,42 @@
 // http://www.hpl.hp.com/personal/Hans_Boehm/c++mm/
 
 #include "sanitizer_common/sanitizer_placement_new.h"
+#include "sanitizer_common/sanitizer_stacktrace.h"
 #include "tsan_interface_atomic.h"
 #include "tsan_flags.h"
 #include "tsan_rtl.h"
 
 using namespace __tsan;  // NOLINT
 
+#define SCOPED_ATOMIC(func, ...) \
+    const uptr callpc = (uptr)__builtin_return_address(0); \
+    uptr pc = __sanitizer::StackTrace::GetCurrentPc(); \
+    pc = __sanitizer::StackTrace::GetPreviousInstructionPc(pc); \
+    mo = ConvertOrder(mo); \
+    mo = flags()->force_seq_cst_atomics ? (morder)mo_seq_cst : mo; \
+    ThreadState *const thr = cur_thread(); \
+    AtomicStatInc(thr, sizeof(*a), mo, StatAtomic##func); \
+    ScopedAtomic sa(thr, callpc, __FUNCTION__); \
+    return Atomic##func(thr, pc, __VA_ARGS__); \
+/**/
+
 class ScopedAtomic {
  public:
   ScopedAtomic(ThreadState *thr, uptr pc, const char *func)
       : thr_(thr) {
-    CHECK_EQ(thr_->in_rtl, 1);  // 1 due to our own ScopedInRtl member.
+    CHECK_EQ(thr_->in_rtl, 0);
+    ProcessPendingSignals(thr);
+    FuncEntry(thr_, pc);
     DPrintf("#%d: %s\n", thr_->tid, func);
+    thr_->in_rtl++;
   }
   ~ScopedAtomic() {
-    CHECK_EQ(thr_->in_rtl, 1);
+    thr_->in_rtl--;
+    CHECK_EQ(thr_->in_rtl, 0);
+    FuncExit(thr_);
   }
  private:
   ThreadState *thr_;
-  ScopedInRtl in_rtl_;
 };
 
 // Some shortcuts.
@@ -210,16 +227,19 @@ a128 func_cas(volatile a128 *v, a128 cmp, a128 xch) {
 }
 #endif
 
-#define SCOPED_ATOMIC(func, ...) \
-    mo = ConvertOrder(mo); \
-    mo = flags()->force_seq_cst_atomics ? (morder)mo_seq_cst : mo; \
-    ThreadState *const thr = cur_thread(); \
-    ProcessPendingSignals(thr); \
-    const uptr pc = (uptr)__builtin_return_address(0); \
-    AtomicStatInc(thr, sizeof(*a), mo, StatAtomic##func); \
-    ScopedAtomic sa(thr, pc, __FUNCTION__); \
-    return Atomic##func(thr, pc, __VA_ARGS__); \
-/**/
+template<typename T>
+static int SizeLog() {
+  if (sizeof(T) <= 1)
+    return kSizeLog1;
+  else if (sizeof(T) <= 2)
+    return kSizeLog2;
+  else if (sizeof(T) <= 4)
+    return kSizeLog4;
+  else
+    return kSizeLog8;
+  // For 16-byte atomics we also use 8-byte memory access,
+  // this leads to false negatives only in very obscure cases.
+}
 
 template<typename T>
 static T AtomicLoad(ThreadState *thr, uptr pc, const volatile T *a,
@@ -227,14 +247,17 @@ static T AtomicLoad(ThreadState *thr, uptr pc, const volatile T *a,
   CHECK(IsLoadOrder(mo));
   // This fast-path is critical for performance.
   // Assume the access is atomic.
-  if (!IsAcquireOrder(mo) && sizeof(T) <= sizeof(a))
+  if (!IsAcquireOrder(mo) && sizeof(T) <= sizeof(a)) {
+    MemoryReadAtomic(thr, pc, (uptr)a, SizeLog<T>());
     return *a;
+  }
   SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, (uptr)a, false);
   thr->clock.set(thr->tid, thr->fast_state.epoch());
   thr->clock.acquire(&s->clock);
   T v = *a;
   s->mtx.ReadUnlock();
   __sync_synchronize();
+  MemoryReadAtomic(thr, pc, (uptr)a, SizeLog<T>());
   return v;
 }
 
@@ -242,6 +265,7 @@ template<typename T>
 static void AtomicStore(ThreadState *thr, uptr pc, volatile T *a, T v,
     morder mo) {
   CHECK(IsStoreOrder(mo));
+  MemoryWriteAtomic(thr, pc, (uptr)a, SizeLog<T>());
   // This fast-path is critical for performance.
   // Assume the access is atomic.
   // Strictly saying even relaxed store cuts off release sequence,
@@ -263,6 +287,7 @@ static void AtomicStore(ThreadState *thr, uptr pc, volatile T *a, T v,
 
 template<typename T, T (*F)(volatile T *v, T op)>
 static T AtomicRMW(ThreadState *thr, uptr pc, volatile T *a, T v, morder mo) {
+  MemoryWriteAtomic(thr, pc, (uptr)a, SizeLog<T>());
   SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, (uptr)a, true);
   thr->clock.set(thr->tid, thr->fast_state.epoch());
   if (IsAcqRelOrder(mo))
@@ -322,6 +347,7 @@ template<typename T>
 static bool AtomicCAS(ThreadState *thr, uptr pc,
     volatile T *a, T *c, T v, morder mo, morder fmo) {
   (void)fmo;  // Unused because llvm does not pass it yet.
+  MemoryWriteAtomic(thr, pc, (uptr)a, SizeLog<T>());
   SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, (uptr)a, true);
   thr->clock.set(thr->tid, thr->fast_state.epoch());
   if (IsAcqRelOrder(mo))
diff --git a/libsanitizer/tsan/tsan_interface_inl.h b/libsanitizer/tsan/tsan_interface_inl.h
index 133348a942c..92796d1178f 100644
--- a/libsanitizer/tsan/tsan_interface_inl.h
+++ b/libsanitizer/tsan/tsan_interface_inl.h
@@ -17,41 +17,41 @@
 using namespace __tsan;  // NOLINT
 
 void __tsan_read1(void *addr) {
-  MemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 0, 0);
+  MemoryRead(cur_thread(), CALLERPC, (uptr)addr, kSizeLog1);
 }
 
 void __tsan_read2(void *addr) {
-  MemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 1, 0);
+  MemoryRead(cur_thread(), CALLERPC, (uptr)addr, kSizeLog2);
 }
 
 void __tsan_read4(void *addr) {
-  MemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 2, 0);
+  MemoryRead(cur_thread(), CALLERPC, (uptr)addr, kSizeLog4);
 }
 
 void __tsan_read8(void *addr) {
-  MemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 3, 0);
+  MemoryRead(cur_thread(), CALLERPC, (uptr)addr, kSizeLog8);
 }
 
 void __tsan_write1(void *addr) {
-  MemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 0, 1);
+  MemoryWrite(cur_thread(), CALLERPC, (uptr)addr, kSizeLog1);
 }
 
 void __tsan_write2(void *addr) {
-  MemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 1, 1);
+  MemoryWrite(cur_thread(), CALLERPC, (uptr)addr, kSizeLog2);
 }
 
 void __tsan_write4(void *addr) {
-  MemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 2, 1);
+  MemoryWrite(cur_thread(), CALLERPC, (uptr)addr, kSizeLog4);
 }
 
 void __tsan_write8(void *addr) {
-  MemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 3, 1);
+  MemoryWrite(cur_thread(), CALLERPC, (uptr)addr, kSizeLog8);
 }
 
 void __tsan_vptr_update(void **vptr_p, void *new_val) {
   CHECK_EQ(sizeof(vptr_p), 8);
   if (*vptr_p != new_val)
-    MemoryAccess(cur_thread(), CALLERPC, (uptr)vptr_p, 3, 1);
+    MemoryWrite(cur_thread(), CALLERPC, (uptr)vptr_p, kSizeLog8);
 }
 
 void __tsan_func_entry(void *pc) {
diff --git a/libsanitizer/tsan/tsan_interface_java.cc b/libsanitizer/tsan/tsan_interface_java.cc
index d7325dcb2c4..f8c0b4eb635 100644
--- a/libsanitizer/tsan/tsan_interface_java.cc
+++ b/libsanitizer/tsan/tsan_interface_java.cc
@@ -150,7 +150,7 @@ SyncVar* GetAndRemoveJavaSync(ThreadState *thr, uptr pc, uptr addr) {
   return 0;
 }
 
-}  // namespace __tsan {
+}  // namespace __tsan
 
 #define SCOPED_JAVA_FUNC(func) \
   ThreadState *thr = cur_thread(); \
diff --git a/libsanitizer/tsan/tsan_md5.cc b/libsanitizer/tsan/tsan_md5.cc
index 6df823dae98..883239c2e71 100644
--- a/libsanitizer/tsan/tsan_md5.cc
+++ b/libsanitizer/tsan/tsan_md5.cc
@@ -240,4 +240,4 @@ MD5Hash md5_hash(const void *data, uptr size) {
   MD5_Final((unsigned char*)&res.hash[0], &ctx);
   return res;
 }
-}
+}  // namespace __tsan
diff --git a/libsanitizer/tsan/tsan_mman.cc b/libsanitizer/tsan/tsan_mman.cc
index f4fafaf77f0..23c73c50cc1 100644
--- a/libsanitizer/tsan/tsan_mman.cc
+++ b/libsanitizer/tsan/tsan_mman.cc
@@ -36,8 +36,16 @@ void InitializeAllocator() {
   allocator()->Init();
 }
 
-void AlloctorThreadFinish(ThreadState *thr) {
-  allocator()->SwallowCache(&thr->alloc_cache);
+void AllocatorThreadStart(ThreadState *thr) {
+  allocator()->InitCache(&thr->alloc_cache);
+}
+
+void AllocatorThreadFinish(ThreadState *thr) {
+  allocator()->DestroyCache(&thr->alloc_cache);
+}
+
+void AllocatorPrintStats() {
+  allocator()->PrintStats();
 }
 
 static void SignalUnsafeCall(ThreadState *thr, uptr pc) {
@@ -162,3 +170,49 @@ void internal_free(void *p) {
 }
 
 }  // namespace __tsan
+
+using namespace __tsan;
+
+extern "C" {
+uptr __tsan_get_current_allocated_bytes() {
+  u64 stats[AllocatorStatCount];
+  allocator()->GetStats(stats);
+  u64 m = stats[AllocatorStatMalloced];
+  u64 f = stats[AllocatorStatFreed];
+  return m >= f ? m - f : 1;
+}
+
+uptr __tsan_get_heap_size() {
+  u64 stats[AllocatorStatCount];
+  allocator()->GetStats(stats);
+  u64 m = stats[AllocatorStatMmapped];
+  u64 f = stats[AllocatorStatUnmapped];
+  return m >= f ? m - f : 1;
+}
+
+uptr __tsan_get_free_bytes() {
+  return 1;
+}
+
+uptr __tsan_get_unmapped_bytes() {
+  return 1;
+}
+
+uptr __tsan_get_estimated_allocated_size(uptr size) {
+  return size;
+}
+
+bool __tsan_get_ownership(void *p) {
+  return allocator()->GetBlockBegin(p) != 0;
+}
+
+uptr __tsan_get_allocated_size(void *p) {
+  if (p == 0)
+    return 0;
+  p = allocator()->GetBlockBegin(p);
+  if (p == 0)
+    return 0;
+  MBlock *b = (MBlock*)allocator()->GetMetaData(p);
+  return b->size;
+}
+}  // extern "C"
diff --git a/libsanitizer/tsan/tsan_mman.h b/libsanitizer/tsan/tsan_mman.h
index 8697d228730..7a657e124bf 100644
--- a/libsanitizer/tsan/tsan_mman.h
+++ b/libsanitizer/tsan/tsan_mman.h
@@ -18,7 +18,9 @@ namespace __tsan {
 const uptr kDefaultAlignment = 16;
 
 void InitializeAllocator();
-void AlloctorThreadFinish(ThreadState *thr);
+void AllocatorThreadStart(ThreadState *thr);
+void AllocatorThreadFinish(ThreadState *thr);
+void AllocatorPrintStats();
 
 // For user allocations.
 void *user_alloc(ThreadState *thr, uptr pc, uptr sz,
diff --git a/libsanitizer/tsan/tsan_platform.h b/libsanitizer/tsan/tsan_platform.h
index 9fdc4dd46e7..78c1a7d8195 100644
--- a/libsanitizer/tsan/tsan_platform.h
+++ b/libsanitizer/tsan/tsan_platform.h
@@ -149,6 +149,7 @@ bool IsGlobalVar(uptr addr);
 uptr GetTlsSize();
 void GetThreadStackAndTls(bool main, uptr *stk_addr, uptr *stk_size,
                           uptr *tls_addr, uptr *tls_size);
+int ExtractResolvFDs(void *state, int *fds, int nfd);
 
 }  // namespace __tsan
 
diff --git a/libsanitizer/tsan/tsan_platform_linux.cc b/libsanitizer/tsan/tsan_platform_linux.cc
index 2e7cd5138d6..def91559d5e 100644
--- a/libsanitizer/tsan/tsan_platform_linux.cc
+++ b/libsanitizer/tsan/tsan_platform_linux.cc
@@ -38,6 +38,8 @@
 #include <errno.h>
 #include <sched.h>
 #include <dlfcn.h>
+#define __need_res_state
+#include <resolv.h>
 
 extern "C" int arch_prctl(int code, __sanitizer::uptr *addr);
 
@@ -287,6 +289,19 @@ bool IsGlobalVar(uptr addr) {
   return g_data_start && addr >= g_data_start && addr < g_data_end;
 }
 
+#ifndef TSAN_GO
+int ExtractResolvFDs(void *state, int *fds, int nfd) {
+  int cnt = 0;
+  __res_state *statp = (__res_state*)state;
+  for (int i = 0; i < MAXNS && cnt < nfd; i++) {
+    if (statp->_u._ext.nsaddrs[i] && statp->_u._ext.nssocks[i] != -1)
+      fds[cnt++] = statp->_u._ext.nssocks[i];
+  }
+  return cnt;
+}
+#endif
+
+
 }  // namespace __tsan
 
 #endif  // #ifdef __linux__
diff --git a/libsanitizer/tsan/tsan_report.cc b/libsanitizer/tsan/tsan_report.cc
index f99fd2ea105..098d8262ba1 100644
--- a/libsanitizer/tsan/tsan_report.cc
+++ b/libsanitizer/tsan/tsan_report.cc
@@ -41,23 +41,20 @@ const char *thread_name(char *buf, int tid) {
   return buf;
 }
 
-static void PrintHeader(ReportType typ) {
-  Printf("WARNING: ThreadSanitizer: ");
-
+static const char *ReportTypeString(ReportType typ) {
   if (typ == ReportTypeRace)
-    Printf("data race");
-  else if (typ == ReportTypeUseAfterFree)
-    Printf("heap-use-after-free");
-  else if (typ == ReportTypeThreadLeak)
-    Printf("thread leak");
-  else if (typ == ReportTypeMutexDestroyLocked)
-    Printf("destroy of a locked mutex");
-  else if (typ == ReportTypeSignalUnsafe)
-    Printf("signal-unsafe call inside of a signal");
-  else if (typ == ReportTypeErrnoInSignal)
-    Printf("signal handler spoils errno");
-
-  Printf(" (pid=%d)\n", GetPid());
+    return "data race";
+  if (typ == ReportTypeUseAfterFree)
+    return "heap-use-after-free";
+  if (typ == ReportTypeThreadLeak)
+    return "thread leak";
+  if (typ == ReportTypeMutexDestroyLocked)
+    return "destroy of a locked mutex";
+  if (typ == ReportTypeSignalUnsafe)
+    return "signal-unsafe call inside of a signal";
+  if (typ == ReportTypeErrnoInSignal)
+    return "signal handler spoils errno";
+  return "";
 }
 
 void PrintStack(const ReportStack *ent) {
@@ -87,11 +84,17 @@ static void PrintMutexSet(Vector<ReportMopMutex> const& mset) {
   }
 }
 
+static const char *MopDesc(bool first, bool write, bool atomic) {
+  return atomic ? (first ? (write ? "Atomic write" : "Atomic read")
+                : (write ? "Previous atomic write" : "Previous atomic read"))
+                : (first ? (write ? "Write" : "Read")
+                : (write ? "Previous write" : "Previous read"));
+}
+
 static void PrintMop(const ReportMop *mop, bool first) {
   char thrbuf[kThreadBufSize];
   Printf("  %s of size %d at %p by %s",
-      (first ? (mop->write ? "Write" : "Read")
-             : (mop->write ? "Previous write" : "Previous read")),
+      MopDesc(first, mop->write, mop->atomic),
       mop->size, (void*)mop->addr,
       thread_name(thrbuf, mop->tid));
   PrintMutexSet(mop->mset);
@@ -150,9 +153,28 @@ static void PrintSleep(const ReportStack *s) {
   PrintStack(s);
 }
 
+static ReportStack *ChooseSummaryStack(const ReportDesc *rep) {
+  if (rep->mops.Size())
+    return rep->mops[0]->stack;
+  if (rep->stacks.Size())
+    return rep->stacks[0];
+  if (rep->mutexes.Size())
+    return rep->mutexes[0]->stack;
+  if (rep->threads.Size())
+    return rep->threads[0]->stack;
+  return 0;
+}
+
+ReportStack *SkipTsanInternalFrames(ReportStack *ent) {
+  while (FrameIsInternal(ent) && ent->next)
+    ent = ent->next;
+  return ent;
+}
+
 void PrintReport(const ReportDesc *rep) {
   Printf("==================\n");
-  PrintHeader(rep->typ);
+  const char *rep_typ_str = ReportTypeString(rep->typ);
+  Printf("WARNING: ThreadSanitizer: %s (pid=%d)\n", rep_typ_str, GetPid());
 
   for (uptr i = 0; i < rep->stacks.Size(); i++) {
     if (i)
@@ -175,6 +197,9 @@ void PrintReport(const ReportDesc *rep) {
   for (uptr i = 0; i < rep->threads.Size(); i++)
     PrintThread(rep->threads[i]);
 
+  if (ReportStack *ent = SkipTsanInternalFrames(ChooseSummaryStack(rep)))
+    ReportErrorSummary(rep_typ_str, ent->file, ent->line, ent->func);
+
   Printf("==================\n");
 }
 
diff --git a/libsanitizer/tsan/tsan_report.h b/libsanitizer/tsan/tsan_report.h
index 42f52af9e37..eae2b3c721f 100644
--- a/libsanitizer/tsan/tsan_report.h
+++ b/libsanitizer/tsan/tsan_report.h
@@ -46,6 +46,7 @@ struct ReportMop {
   uptr addr;
   int size;
   bool write;
+  bool atomic;
   Vector<ReportMopMutex> mset;
   ReportStack *stack;
 
diff --git a/libsanitizer/tsan/tsan_rtl.cc b/libsanitizer/tsan/tsan_rtl.cc
index 3615a7a9c2f..673a355f1dc 100644
--- a/libsanitizer/tsan/tsan_rtl.cc
+++ b/libsanitizer/tsan/tsan_rtl.cc
@@ -35,6 +35,11 @@ THREADLOCAL char cur_thread_placeholder[sizeof(ThreadState)] ALIGNED(64);
 #endif
 static char ctx_placeholder[sizeof(Context)] ALIGNED(64);
 
+// Can be overriden by a front-end.
+bool CPP_WEAK OnFinalize(bool failed) {
+  return failed;
+}
+
 static Context *ctx;
 Context *CTX() {
   return ctx;
@@ -136,7 +141,7 @@ static void InitializeMemoryProfile() {
   InternalScopedBuffer<char> filename(4096);
   internal_snprintf(filename.data(), filename.size(), "%s.%d",
       flags()->profile_memory, GetPid());
-  fd_t fd = internal_open(filename.data(), true);
+  fd_t fd = OpenFile(filename.data(), true);
   if (fd == kInvalidFd) {
     Printf("Failed to open memory profile file '%s'\n", &filename[0]);
     Die();
@@ -180,6 +185,7 @@ void Initialize(ThreadState *thr) {
   if (is_initialized)
     return;
   is_initialized = true;
+  SanitizerToolName = "ThreadSanitizer";
   // Install tool-specific callbacks in sanitizer_common.
   SetCheckFailedCallback(TsanCheckFailed);
 
@@ -237,7 +243,7 @@ void Initialize(ThreadState *thr) {
     Printf("ThreadSanitizer is suspended at startup (pid %d)."
            " Call __tsan_resume().\n",
            GetPid());
-    while (__tsan_resumed == 0);
+    while (__tsan_resumed == 0) {}
   }
 }
 
@@ -253,6 +259,11 @@ int Finalize(ThreadState *thr) {
   ctx->report_mtx.Lock();
   ctx->report_mtx.Unlock();
 
+#ifndef TSAN_GO
+  if (ctx->flags.verbosity)
+    AllocatorPrintStats();
+#endif
+
   ThreadFinalize(thr);
 
   if (ctx->nreported) {
@@ -270,6 +281,8 @@ int Finalize(ThreadState *thr) {
         ctx->nmissed_expected);
   }
 
+  failed = OnFinalize(failed);
+
   StatAggregate(ctx->stat, thr->stat);
   StatOutput(ctx->stat);
   return failed ? flags()->exitcode : 0;
@@ -356,18 +369,6 @@ static inline void HandleRace(ThreadState *thr, u64 *shadow_mem,
 #endif
 }
 
-static inline bool BothReads(Shadow s, int kAccessIsWrite) {
-  return !kAccessIsWrite && !s.is_write();
-}
-
-static inline bool OldIsRWNotWeaker(Shadow old, int kAccessIsWrite) {
-  return old.is_write() || !kAccessIsWrite;
-}
-
-static inline bool OldIsRWWeakerOrEqual(Shadow old, int kAccessIsWrite) {
-  return !old.is_write() || kAccessIsWrite;
-}
-
 static inline bool OldIsInSameSynchEpoch(Shadow old, ThreadState *thr) {
   return old.epoch() >= thr->fast_synch_epoch;
 }
@@ -378,7 +379,7 @@ static inline bool HappensBefore(Shadow old, ThreadState *thr) {
 
 ALWAYS_INLINE
 void MemoryAccessImpl(ThreadState *thr, uptr addr,
-    int kAccessSizeLog, bool kAccessIsWrite,
+    int kAccessSizeLog, bool kAccessIsWrite, bool kIsAtomic,
     u64 *shadow_mem, Shadow cur) {
   StatInc(thr, StatMop);
   StatInc(thr, kAccessIsWrite ? StatMopWrite : StatMopRead);
@@ -452,7 +453,7 @@ void MemoryAccessImpl(ThreadState *thr, uptr addr,
 
 ALWAYS_INLINE
 void MemoryAccess(ThreadState *thr, uptr pc, uptr addr,
-    int kAccessSizeLog, bool kAccessIsWrite) {
+    int kAccessSizeLog, bool kAccessIsWrite, bool kIsAtomic) {
   u64 *shadow_mem = (u64*)MemToShadow(addr);
   DPrintf2("#%d: MemoryAccess: @%p %p size=%d"
       " is_write=%d shadow_mem=%p {%zx, %zx, %zx, %zx}\n",
@@ -479,12 +480,13 @@ void MemoryAccess(ThreadState *thr, uptr pc, uptr addr,
   Shadow cur(fast_state);
   cur.SetAddr0AndSizeLog(addr & 7, kAccessSizeLog);
   cur.SetWrite(kAccessIsWrite);
+  cur.SetAtomic(kIsAtomic);
 
   // We must not store to the trace if we do not store to the shadow.
   // That is, this call must be moved somewhere below.
   TraceAddEvent(thr, fast_state, EventTypeMop, pc);
 
-  MemoryAccessImpl(thr, addr, kAccessSizeLog, kAccessIsWrite,
+  MemoryAccessImpl(thr, addr, kAccessSizeLog, kAccessIsWrite, kIsAtomic,
       shadow_mem, cur);
 }
 
@@ -531,7 +533,10 @@ void MemoryResetRange(ThreadState *thr, uptr pc, uptr addr, uptr size) {
 }
 
 void MemoryRangeFreed(ThreadState *thr, uptr pc, uptr addr, uptr size) {
+  CHECK_EQ(thr->is_freeing, false);
+  thr->is_freeing = true;
   MemoryAccessRange(thr, pc, addr, size, true);
+  thr->is_freeing = false;
   Shadow s(thr->fast_state);
   s.ClearIgnoreBit();
   s.MarkAsFreed();
diff --git a/libsanitizer/tsan/tsan_rtl.h b/libsanitizer/tsan/tsan_rtl.h
index bb2fe56bfe7..717f4599705 100644
--- a/libsanitizer/tsan/tsan_rtl.h
+++ b/libsanitizer/tsan/tsan_rtl.h
@@ -171,7 +171,8 @@ class FastState {
 //   freed           : 1
 //   tid             : kTidBits
 //   epoch           : kClkBits
-//   is_write        : 1
+//   is_atomic       : 1
+//   is_read         : 1
 //   size_log        : 2
 //   addr0           : 3
 class Shadow : public FastState {
@@ -195,13 +196,26 @@ class Shadow : public FastState {
   }
 
   void SetWrite(unsigned kAccessIsWrite) {
-    DCHECK_EQ(x_ & 32, 0);
-    if (kAccessIsWrite)
-      x_ |= 32;
-    DCHECK_EQ(kAccessIsWrite, is_write());
+    DCHECK_EQ(x_ & kReadBit, 0);
+    if (!kAccessIsWrite)
+      x_ |= kReadBit;
+    DCHECK_EQ(kAccessIsWrite, IsWrite());
   }
 
-  bool IsZero() const { return x_ == 0; }
+  void SetAtomic(bool kIsAtomic) {
+    DCHECK(!IsAtomic());
+    if (kIsAtomic)
+      x_ |= kAtomicBit;
+    DCHECK_EQ(IsAtomic(), kIsAtomic);
+  }
+
+  bool IsAtomic() const {
+    return x_ & kAtomicBit;
+  }
+
+  bool IsZero() const {
+    return x_ == 0;
+  }
 
   static inline bool TidsAreEqual(const Shadow s1, const Shadow s2) {
     u64 shifted_xor = (s1.x_ ^ s2.x_) >> kTidShift;
@@ -248,7 +262,8 @@ class Shadow : public FastState {
   }
   u64 addr0() const { return x_ & 7; }
   u64 size() const { return 1ull << size_log(); }
-  bool is_write() const { return x_ & 32; }
+  bool IsWrite() const { return !IsRead(); }
+  bool IsRead() const { return x_ & kReadBit; }
 
   // The idea behind the freed bit is as follows.
   // When the memory is freed (or otherwise unaccessible) we write to the shadow
@@ -263,13 +278,46 @@ class Shadow : public FastState {
      x_ |= kFreedBit;
   }
 
+  bool IsFreed() const {
+    return x_ & kFreedBit;
+  }
+
   bool GetFreedAndReset() {
     bool res = x_ & kFreedBit;
     x_ &= ~kFreedBit;
     return res;
   }
 
+  bool IsBothReadsOrAtomic(bool kIsWrite, bool kIsAtomic) const {
+    // analyzes 5-th bit (is_read) and 6-th bit (is_atomic)
+    bool v = x_ & u64(((kIsWrite ^ 1) << kReadShift)
+        | (kIsAtomic << kAtomicShift));
+    DCHECK_EQ(v, (!IsWrite() && !kIsWrite) || (IsAtomic() && kIsAtomic));
+    return v;
+  }
+
+  bool IsRWNotWeaker(bool kIsWrite, bool kIsAtomic) const {
+    bool v = ((x_ >> kReadShift) & 3)
+        <= u64((kIsWrite ^ 1) | (kIsAtomic << 1));
+    DCHECK_EQ(v, (IsAtomic() < kIsAtomic) ||
+        (IsAtomic() == kIsAtomic && !IsWrite() <= !kIsWrite));
+    return v;
+  }
+
+  bool IsRWWeakerOrEqual(bool kIsWrite, bool kIsAtomic) const {
+    bool v = ((x_ >> kReadShift) & 3)
+        >= u64((kIsWrite ^ 1) | (kIsAtomic << 1));
+    DCHECK_EQ(v, (IsAtomic() > kIsAtomic) ||
+        (IsAtomic() == kIsAtomic && !IsWrite() >= !kIsWrite));
+    return v;
+  }
+
  private:
+  static const u64 kReadShift   = 5;
+  static const u64 kReadBit     = 1ull << kReadShift;
+  static const u64 kAtomicShift = 6;
+  static const u64 kAtomicBit   = 1ull << kAtomicShift;
+
   u64 size_log() const { return (x_ >> 3) & 3; }
 
   static bool TwoRangesIntersectSLOW(const Shadow s1, const Shadow s2) {
@@ -324,7 +372,9 @@ struct ThreadState {
   const int tid;
   const int unique_id;
   int in_rtl;
+  bool in_symbolizer;
   bool is_alive;
+  bool is_freeing;
   const uptr stk_addr;
   const uptr stk_size;
   const uptr tls_addr;
@@ -501,11 +551,14 @@ void InitializeDynamicAnnotations();
 void ReportRace(ThreadState *thr);
 bool OutputReport(Context *ctx,
                   const ScopedReport &srep,
-                  const ReportStack *suppress_stack = 0);
+                  const ReportStack *suppress_stack1 = 0,
+                  const ReportStack *suppress_stack2 = 0);
 bool IsFiredSuppression(Context *ctx,
                         const ScopedReport &srep,
                         const StackTrace &trace);
 bool IsExpectedReport(uptr addr, uptr size);
+bool FrameIsInternal(const ReportStack *frame);
+ReportStack *SkipTsanInternalFrames(ReportStack *ent);
 
 #if defined(TSAN_DEBUG_OUTPUT) && TSAN_DEBUG_OUTPUT >= 1
 # define DPrintf Printf
@@ -521,6 +574,7 @@ bool IsExpectedReport(uptr addr, uptr size);
 
 u32 CurrentStackId(ThreadState *thr, uptr pc);
 void PrintCurrentStack(ThreadState *thr, uptr pc);
+void PrintCurrentStackSlow();  // uses libunwind
 
 void Initialize(ThreadState *thr);
 int Finalize(ThreadState *thr);
@@ -530,16 +584,38 @@ SyncVar* GetJavaSync(ThreadState *thr, uptr pc, uptr addr,
 SyncVar* GetAndRemoveJavaSync(ThreadState *thr, uptr pc, uptr addr);
 
 void MemoryAccess(ThreadState *thr, uptr pc, uptr addr,
-    int kAccessSizeLog, bool kAccessIsWrite);
+    int kAccessSizeLog, bool kAccessIsWrite, bool kIsAtomic);
 void MemoryAccessImpl(ThreadState *thr, uptr addr,
-    int kAccessSizeLog, bool kAccessIsWrite,
+    int kAccessSizeLog, bool kAccessIsWrite, bool kIsAtomic,
     u64 *shadow_mem, Shadow cur);
-void MemoryRead1Byte(ThreadState *thr, uptr pc, uptr addr);
-void MemoryWrite1Byte(ThreadState *thr, uptr pc, uptr addr);
-void MemoryRead8Byte(ThreadState *thr, uptr pc, uptr addr);
-void MemoryWrite8Byte(ThreadState *thr, uptr pc, uptr addr);
 void MemoryAccessRange(ThreadState *thr, uptr pc, uptr addr,
-                       uptr size, bool is_write);
+    uptr size, bool is_write);
+
+const int kSizeLog1 = 0;
+const int kSizeLog2 = 1;
+const int kSizeLog4 = 2;
+const int kSizeLog8 = 3;
+
+void ALWAYS_INLINE INLINE MemoryRead(ThreadState *thr, uptr pc,
+                                     uptr addr, int kAccessSizeLog) {
+  MemoryAccess(thr, pc, addr, kAccessSizeLog, false, false);
+}
+
+void ALWAYS_INLINE INLINE MemoryWrite(ThreadState *thr, uptr pc,
+                                      uptr addr, int kAccessSizeLog) {
+  MemoryAccess(thr, pc, addr, kAccessSizeLog, true, false);
+}
+
+void ALWAYS_INLINE INLINE MemoryReadAtomic(ThreadState *thr, uptr pc,
+                                           uptr addr, int kAccessSizeLog) {
+  MemoryAccess(thr, pc, addr, kAccessSizeLog, false, true);
+}
+
+void ALWAYS_INLINE INLINE MemoryWriteAtomic(ThreadState *thr, uptr pc,
+                                            uptr addr, int kAccessSizeLog) {
+  MemoryAccess(thr, pc, addr, kAccessSizeLog, true, true);
+}
+
 void MemoryResetRange(ThreadState *thr, uptr pc, uptr addr, uptr size);
 void MemoryRangeFreed(ThreadState *thr, uptr pc, uptr addr, uptr size);
 void MemoryRangeImitateWrite(ThreadState *thr, uptr pc, uptr addr, uptr size);
diff --git a/libsanitizer/tsan/tsan_rtl_mutex.cc b/libsanitizer/tsan/tsan_rtl_mutex.cc
index db97e1d9853..22a71503c5c 100644
--- a/libsanitizer/tsan/tsan_rtl_mutex.cc
+++ b/libsanitizer/tsan/tsan_rtl_mutex.cc
@@ -24,8 +24,12 @@ void MutexCreate(ThreadState *thr, uptr pc, uptr addr,
   CHECK_GT(thr->in_rtl, 0);
   DPrintf("#%d: MutexCreate %zx\n", thr->tid, addr);
   StatInc(thr, StatMutexCreate);
-  if (!linker_init && IsAppMem(addr))
-    MemoryWrite1Byte(thr, pc, addr);
+  if (!linker_init && IsAppMem(addr)) {
+    CHECK(!thr->is_freeing);
+    thr->is_freeing = true;
+    MemoryWrite(thr, pc, addr, kSizeLog1);
+    thr->is_freeing = false;
+  }
   SyncVar *s = ctx->synctab.GetOrCreateAndLock(thr, pc, addr, true);
   s->is_rw = rw;
   s->is_recursive = recursive;
@@ -47,8 +51,12 @@ void MutexDestroy(ThreadState *thr, uptr pc, uptr addr) {
   SyncVar *s = ctx->synctab.GetAndRemove(thr, pc, addr);
   if (s == 0)
     return;
-  if (IsAppMem(addr))
-    MemoryWrite1Byte(thr, pc, addr);
+  if (IsAppMem(addr)) {
+    CHECK(!thr->is_freeing);
+    thr->is_freeing = true;
+    MemoryWrite(thr, pc, addr, kSizeLog1);
+    thr->is_freeing = false;
+  }
   if (flags()->report_destroy_locked
       && s->owner_tid != SyncVar::kInvalidTid
       && !s->is_broken) {
@@ -73,7 +81,7 @@ void MutexLock(ThreadState *thr, uptr pc, uptr addr) {
   CHECK_GT(thr->in_rtl, 0);
   DPrintf("#%d: MutexLock %zx\n", thr->tid, addr);
   if (IsAppMem(addr))
-    MemoryRead1Byte(thr, pc, addr);
+    MemoryReadAtomic(thr, pc, addr, kSizeLog1);
   SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, addr, true);
   thr->fast_state.IncrementEpoch();
   TraceAddEvent(thr, thr->fast_state, EventTypeLock, s->GetId());
@@ -106,7 +114,7 @@ void MutexUnlock(ThreadState *thr, uptr pc, uptr addr) {
   CHECK_GT(thr->in_rtl, 0);
   DPrintf("#%d: MutexUnlock %zx\n", thr->tid, addr);
   if (IsAppMem(addr))
-    MemoryRead1Byte(thr, pc, addr);
+    MemoryReadAtomic(thr, pc, addr, kSizeLog1);
   SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, addr, true);
   thr->fast_state.IncrementEpoch();
   TraceAddEvent(thr, thr->fast_state, EventTypeUnlock, s->GetId());
@@ -144,7 +152,7 @@ void MutexReadLock(ThreadState *thr, uptr pc, uptr addr) {
   DPrintf("#%d: MutexReadLock %zx\n", thr->tid, addr);
   StatInc(thr, StatMutexReadLock);
   if (IsAppMem(addr))
-    MemoryRead1Byte(thr, pc, addr);
+    MemoryReadAtomic(thr, pc, addr, kSizeLog1);
   SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, addr, false);
   thr->fast_state.IncrementEpoch();
   TraceAddEvent(thr, thr->fast_state, EventTypeRLock, s->GetId());
@@ -165,7 +173,7 @@ void MutexReadUnlock(ThreadState *thr, uptr pc, uptr addr) {
   DPrintf("#%d: MutexReadUnlock %zx\n", thr->tid, addr);
   StatInc(thr, StatMutexReadUnlock);
   if (IsAppMem(addr))
-    MemoryRead1Byte(thr, pc, addr);
+    MemoryReadAtomic(thr, pc, addr, kSizeLog1);
   SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, addr, true);
   thr->fast_state.IncrementEpoch();
   TraceAddEvent(thr, thr->fast_state, EventTypeRUnlock, s->GetId());
@@ -186,7 +194,7 @@ void MutexReadOrWriteUnlock(ThreadState *thr, uptr pc, uptr addr) {
   CHECK_GT(thr->in_rtl, 0);
   DPrintf("#%d: MutexReadOrWriteUnlock %zx\n", thr->tid, addr);
   if (IsAppMem(addr))
-    MemoryRead1Byte(thr, pc, addr);
+    MemoryReadAtomic(thr, pc, addr, kSizeLog1);
   SyncVar *s = CTX()->synctab.GetOrCreateAndLock(thr, pc, addr, true);
   bool write = true;
   if (s->owner_tid == SyncVar::kInvalidTid) {
diff --git a/libsanitizer/tsan/tsan_rtl_report.cc b/libsanitizer/tsan/tsan_rtl_report.cc
index c4256da412a..ff1d43bc9e8 100644
--- a/libsanitizer/tsan/tsan_rtl_report.cc
+++ b/libsanitizer/tsan/tsan_rtl_report.cc
@@ -13,6 +13,7 @@
 #include "sanitizer_common/sanitizer_placement_new.h"
 #include "sanitizer_common/sanitizer_stackdepot.h"
 #include "sanitizer_common/sanitizer_common.h"
+#include "sanitizer_common/sanitizer_stacktrace.h"
 #include "tsan_platform.h"
 #include "tsan_rtl.h"
 #include "tsan_suppressions.h"
@@ -27,12 +28,15 @@ namespace __tsan {
 
 using namespace __sanitizer;  // NOLINT
 
+static ReportStack *SymbolizeStack(const StackTrace& trace);
+
 void TsanCheckFailed(const char *file, int line, const char *cond,
                      u64 v1, u64 v2) {
   ScopedInRtl in_rtl;
   Printf("FATAL: ThreadSanitizer CHECK failed: "
          "%s:%d \"%s\" (0x%zx, 0x%zx)\n",
          file, line, cond, (uptr)v1, (uptr)v2);
+  PrintCurrentStackSlow();
   Die();
 }
 
@@ -144,7 +148,8 @@ void ScopedReport::AddMemoryAccess(uptr addr, Shadow s,
   mop->tid = s.tid();
   mop->addr = addr + s.addr0();
   mop->size = s.size();
-  mop->write = s.is_write();
+  mop->write = s.IsWrite();
+  mop->atomic = s.IsAtomic();
   mop->stack = SymbolizeStack(*stack);
   for (uptr i = 0; i < mset->Size(); i++) {
     MutexSet::Desc d = mset->Get(i);
@@ -458,9 +463,12 @@ static void AddRacyStacks(ThreadState *thr, const StackTrace (&traces)[2],
 
 bool OutputReport(Context *ctx,
                   const ScopedReport &srep,
-                  const ReportStack *suppress_stack) {
+                  const ReportStack *suppress_stack1,
+                  const ReportStack *suppress_stack2) {
   const ReportDesc *rep = srep.GetReport();
-  const uptr suppress_pc = IsSuppressed(rep->typ, suppress_stack);
+  uptr suppress_pc = IsSuppressed(rep->typ, suppress_stack1);
+  if (suppress_pc == 0)
+    suppress_pc = IsSuppressed(rep->typ, suppress_stack2);
   if (suppress_pc != 0) {
     FiredSuppression supp = {srep.GetReport()->typ, suppress_pc};
     ctx->fired_suppressions.PushBack(supp);
@@ -486,6 +494,13 @@ bool IsFiredSuppression(Context *ctx,
   return false;
 }
 
+bool FrameIsInternal(const ReportStack *frame) {
+  return frame != 0 && frame->file != 0
+      && (internal_strstr(frame->file, "tsan_interceptors.cc") ||
+          internal_strstr(frame->file, "sanitizer_common_interceptors.inc") ||
+          internal_strstr(frame->file, "tsan_interface_"));
+}
+
 // On programs that use Java we see weird reports like:
 // WARNING: ThreadSanitizer: data race (pid=22512)
 //   Read of size 8 at 0x7d2b00084318 by thread 100:
@@ -495,22 +510,20 @@ bool IsFiredSuppression(Context *ctx,
 //     #0 strncpy tsan_interceptors.cc:501 (foo+0x00000d8e0919)
 //     #1 <null> <null>:0 (0x7f7ad9b42707)
 static bool IsJavaNonsense(const ReportDesc *rep) {
+#ifndef TSAN_GO
   for (uptr i = 0; i < rep->mops.Size(); i++) {
     ReportMop *mop = rep->mops[i];
     ReportStack *frame = mop->stack;
-    if (frame != 0 && frame->func != 0
-        && (internal_strcmp(frame->func, "memset") == 0
-        || internal_strcmp(frame->func, "memcpy") == 0
-        || internal_strcmp(frame->func, "memmove") == 0
-        || internal_strcmp(frame->func, "strcmp") == 0
-        || internal_strcmp(frame->func, "strncpy") == 0
-        || internal_strcmp(frame->func, "strlen") == 0
-        || internal_strcmp(frame->func, "free") == 0
-        || internal_strcmp(frame->func, "pthread_mutex_lock") == 0)) {
+    if (frame == 0
+        || (frame->func == 0 && frame->file == 0 && frame->line == 0
+          && frame->module == 0)) {
+      return true;
+    }
+    if (FrameIsInternal(frame)) {
       frame = frame->next;
       if (frame == 0
           || (frame->func == 0 && frame->file == 0 && frame->line == 0
-            && frame->module == 0)) {
+          && frame->module == 0)) {
         if (frame) {
           FiredSuppression supp = {rep->typ, frame->pc};
           CTX()->fired_suppressions.PushBack(supp);
@@ -519,6 +532,20 @@ static bool IsJavaNonsense(const ReportDesc *rep) {
       }
     }
   }
+#endif
+  return false;
+}
+
+static bool RaceBetweenAtomicAndFree(ThreadState *thr) {
+  Shadow s0(thr->racy_state[0]);
+  Shadow s1(thr->racy_state[1]);
+  CHECK(!(s0.IsAtomic() && s1.IsAtomic()));
+  if (!s0.IsAtomic() && !s1.IsAtomic())
+    return true;
+  if (s0.IsAtomic() && s1.IsFreed())
+    return true;
+  if (s1.IsAtomic() && thr->is_freeing)
+    return true;
   return false;
 }
 
@@ -527,6 +554,9 @@ void ReportRace(ThreadState *thr) {
     return;
   ScopedInRtl in_rtl;
 
+  if (!flags()->report_atomic_races && !RaceBetweenAtomicAndFree(thr))
+    return;
+
   if (thr->in_signal_handler)
     Printf("ThreadSanitizer: printing report from signal handler."
            " Can crash or hang.\n");
@@ -597,7 +627,8 @@ void ReportRace(ThreadState *thr) {
   }
 #endif
 
-  if (!OutputReport(ctx, rep, rep.GetReport()->mops[0]->stack))
+  if (!OutputReport(ctx, rep, rep.GetReport()->mops[0]->stack,
+                              rep.GetReport()->mops[1]->stack))
     return;
 
   AddRacyStacks(thr, traces, addr_min, addr_max);
@@ -609,4 +640,16 @@ void PrintCurrentStack(ThreadState *thr, uptr pc) {
   PrintStack(SymbolizeStack(trace));
 }
 
+void PrintCurrentStackSlow() {
+#ifndef TSAN_GO
+  __sanitizer::StackTrace *ptrace = new(internal_alloc(MBlockStackTrace,
+      sizeof(__sanitizer::StackTrace))) __sanitizer::StackTrace;
+  ptrace->SlowUnwindStack(__sanitizer::StackTrace::GetCurrentPc(),
+      kStackTraceMax);
+  StackTrace trace;
+  trace.Init(ptrace->trace, ptrace->size);
+  PrintStack(SymbolizeStack(trace));
+#endif
+}
+
 }  // namespace __tsan
diff --git a/libsanitizer/tsan/tsan_rtl_thread.cc b/libsanitizer/tsan/tsan_rtl_thread.cc
index d5b3444be6d..9c89417ad68 100644
--- a/libsanitizer/tsan/tsan_rtl_thread.cc
+++ b/libsanitizer/tsan/tsan_rtl_thread.cc
@@ -207,6 +207,9 @@ void ThreadStart(ThreadState *thr, int tid, uptr os_id) {
   thr->shadow_stack_pos = thr->shadow_stack;
   thr->shadow_stack_end = thr->shadow_stack + kInitStackSize;
 #endif
+#ifndef TSAN_GO
+  AllocatorThreadStart(thr);
+#endif
   tctx->thr = thr;
   thr->fast_synch_epoch = tctx->epoch0;
   thr->clock.set(tid, tctx->epoch0);
@@ -267,7 +270,7 @@ void ThreadFinish(ThreadState *thr) {
   tctx->epoch1 = thr->fast_state.epoch();
 
 #ifndef TSAN_GO
-  AlloctorThreadFinish(thr);
+  AllocatorThreadFinish(thr);
 #endif
   thr->~ThreadState();
   StatAggregate(ctx->stat, thr->stat);
@@ -392,7 +395,7 @@ void MemoryAccessRange(ThreadState *thr, uptr pc, uptr addr,
     Shadow cur(fast_state);
     cur.SetWrite(is_write);
     cur.SetAddr0AndSizeLog(addr & (kShadowCell - 1), kAccessSizeLog);
-    MemoryAccessImpl(thr, addr, kAccessSizeLog, is_write,
+    MemoryAccessImpl(thr, addr, kAccessSizeLog, is_write, false,
         shadow_mem, cur);
   }
   if (unaligned)
@@ -403,7 +406,7 @@ void MemoryAccessRange(ThreadState *thr, uptr pc, uptr addr,
     Shadow cur(fast_state);
     cur.SetWrite(is_write);
     cur.SetAddr0AndSizeLog(0, kAccessSizeLog);
-    MemoryAccessImpl(thr, addr, kAccessSizeLog, is_write,
+    MemoryAccessImpl(thr, addr, kAccessSizeLog, is_write, false,
         shadow_mem, cur);
     shadow_mem += kShadowCnt;
   }
@@ -413,24 +416,8 @@ void MemoryAccessRange(ThreadState *thr, uptr pc, uptr addr,
     Shadow cur(fast_state);
     cur.SetWrite(is_write);
     cur.SetAddr0AndSizeLog(addr & (kShadowCell - 1), kAccessSizeLog);
-    MemoryAccessImpl(thr, addr, kAccessSizeLog, is_write,
+    MemoryAccessImpl(thr, addr, kAccessSizeLog, is_write, false,
         shadow_mem, cur);
   }
 }
-
-void MemoryRead1Byte(ThreadState *thr, uptr pc, uptr addr) {
-  MemoryAccess(thr, pc, addr, 0, 0);
-}
-
-void MemoryWrite1Byte(ThreadState *thr, uptr pc, uptr addr) {
-  MemoryAccess(thr, pc, addr, 0, 1);
-}
-
-void MemoryRead8Byte(ThreadState *thr, uptr pc, uptr addr) {
-  MemoryAccess(thr, pc, addr, 3, 0);
-}
-
-void MemoryWrite8Byte(ThreadState *thr, uptr pc, uptr addr) {
-  MemoryAccess(thr, pc, addr, 3, 1);
-}
 }  // namespace __tsan
diff --git a/libsanitizer/tsan/tsan_stat.cc b/libsanitizer/tsan/tsan_stat.cc
index 105d0bc2375..cd88d2df928 100644
--- a/libsanitizer/tsan/tsan_stat.cc
+++ b/libsanitizer/tsan/tsan_stat.cc
@@ -179,6 +179,18 @@ void StatOutput(u64 *stat) {
   name[StatInt_sem_timedwait]            = "  sem_timedwait                   ";
   name[StatInt_sem_post]                 = "  sem_post                        ";
   name[StatInt_sem_getvalue]             = "  sem_getvalue                    ";
+  name[StatInt_stat]                     = "  stat                            ";
+  name[StatInt___xstat]                  = "  __xstat                         ";
+  name[StatInt_stat64]                   = "  stat64                          ";
+  name[StatInt___xstat64]                = "  __xstat64                       ";
+  name[StatInt_lstat]                    = "  lstat                           ";
+  name[StatInt___lxstat]                 = "  __lxstat                        ";
+  name[StatInt_lstat64]                  = "  lstat64                         ";
+  name[StatInt___lxstat64]               = "  __lxstat64                      ";
+  name[StatInt_fstat]                    = "  fstat                           ";
+  name[StatInt___fxstat]                 = "  __fxstat                        ";
+  name[StatInt_fstat64]                  = "  fstat64                         ";
+  name[StatInt___fxstat64]               = "  __fxstat64                      ";
   name[StatInt_open]                     = "  open                            ";
   name[StatInt_open64]                   = "  open64                          ";
   name[StatInt_creat]                    = "  creat                           ";
@@ -193,12 +205,15 @@ void StatOutput(u64 *stat) {
   name[StatInt_socket]                   = "  socket                          ";
   name[StatInt_socketpair]               = "  socketpair                      ";
   name[StatInt_connect]                  = "  connect                         ";
+  name[StatInt_bind]                     = "  bind                            ";
+  name[StatInt_listen]                   = "  listen                          ";
   name[StatInt_accept]                   = "  accept                          ";
   name[StatInt_accept4]                  = "  accept4                         ";
   name[StatInt_epoll_create]             = "  epoll_create                    ";
   name[StatInt_epoll_create1]            = "  epoll_create1                   ";
   name[StatInt_close]                    = "  close                           ";
   name[StatInt___close]                  = "  __close                         ";
+  name[StatInt___res_iclose]             = "  __res_iclose                    ";
   name[StatInt_pipe]                     = "  pipe                            ";
   name[StatInt_pipe2]                    = "  pipe2                           ";
   name[StatInt_read]                     = "  read                            ";
@@ -240,6 +255,14 @@ void StatOutput(u64 *stat) {
   name[StatInt_scanf]                    = "  scanf                           ";
   name[StatInt_sscanf]                   = "  sscanf                          ";
   name[StatInt_fscanf]                   = "  fscanf                          ";
+  name[StatInt___isoc99_vscanf]          = "  vscanf                          ";
+  name[StatInt___isoc99_vsscanf]         = "  vsscanf                         ";
+  name[StatInt___isoc99_vfscanf]         = "  vfscanf                         ";
+  name[StatInt___isoc99_scanf]           = "  scanf                           ";
+  name[StatInt___isoc99_sscanf]          = "  sscanf                          ";
+  name[StatInt___isoc99_fscanf]          = "  fscanf                          ";
+  name[StatInt_on_exit]                  = "  on_exit                         ";
+  name[StatInt___cxa_atexit]             = "  __cxa_atexit                    ";
 
   name[StatAnnotation]                   = "Dynamic annotations               ";
   name[StatAnnotateHappensBefore]        = "  HappensBefore                   ";
@@ -285,6 +308,7 @@ void StatOutput(u64 *stat) {
   name[StatMtxAnnotations]               = "  Annotations                     ";
   name[StatMtxMBlock]                    = "  MBlock                          ";
   name[StatMtxJavaMBlock]                = "  JavaMBlock                      ";
+  name[StatMtxFD]                        = "  FD                              ";
 
   Printf("Statistics:\n");
   for (int i = 0; i < StatCnt; i++)
diff --git a/libsanitizer/tsan/tsan_stat.h b/libsanitizer/tsan/tsan_stat.h
index f40d3a2ac5e..1d6c54cdd4f 100644
--- a/libsanitizer/tsan/tsan_stat.h
+++ b/libsanitizer/tsan/tsan_stat.h
@@ -174,6 +174,18 @@ enum StatType {
   StatInt_sem_timedwait,
   StatInt_sem_post,
   StatInt_sem_getvalue,
+  StatInt_stat,
+  StatInt___xstat,
+  StatInt_stat64,
+  StatInt___xstat64,
+  StatInt_lstat,
+  StatInt___lxstat,
+  StatInt_lstat64,
+  StatInt___lxstat64,
+  StatInt_fstat,
+  StatInt___fxstat,
+  StatInt_fstat64,
+  StatInt___fxstat64,
   StatInt_open,
   StatInt_open64,
   StatInt_creat,
@@ -188,12 +200,15 @@ enum StatType {
   StatInt_socket,
   StatInt_socketpair,
   StatInt_connect,
+  StatInt_bind,
+  StatInt_listen,
   StatInt_accept,
   StatInt_accept4,
   StatInt_epoll_create,
   StatInt_epoll_create1,
   StatInt_close,
   StatInt___close,
+  StatInt___res_iclose,
   StatInt_pipe,
   StatInt_pipe2,
   StatInt_read,
@@ -239,6 +254,14 @@ enum StatType {
   StatInt_scanf,
   StatInt_sscanf,
   StatInt_fscanf,
+  StatInt___isoc99_vscanf,
+  StatInt___isoc99_vsscanf,
+  StatInt___isoc99_vfscanf,
+  StatInt___isoc99_scanf,
+  StatInt___isoc99_sscanf,
+  StatInt___isoc99_fscanf,
+  StatInt_on_exit,
+  StatInt___cxa_atexit,
 
   // Dynamic annotations.
   StatAnnotation,
@@ -287,6 +310,7 @@ enum StatType {
   StatMtxAtExit,
   StatMtxMBlock,
   StatMtxJavaMBlock,
+  StatMtxFD,
 
   // This must be the last.
   StatCnt
diff --git a/libsanitizer/tsan/tsan_suppressions.cc b/libsanitizer/tsan/tsan_suppressions.cc
index 42ba9b509eb..b6c54db2c51 100644
--- a/libsanitizer/tsan/tsan_suppressions.cc
+++ b/libsanitizer/tsan/tsan_suppressions.cc
@@ -17,6 +17,13 @@
 #include "tsan_mman.h"
 #include "tsan_platform.h"
 
+// Can be overriden in frontend.
+#ifndef TSAN_GO
+extern "C" const char *WEAK __tsan_default_suppressions() {
+  return 0;
+}
+#endif
+
 namespace __tsan {
 
 static Suppression *g_suppressions;
@@ -29,7 +36,7 @@ static char *ReadFile(const char *filename) {
     internal_snprintf(tmp.data(), tmp.size(), "%s", filename);
   else
     internal_snprintf(tmp.data(), tmp.size(), "%s/%s", GetPwd(), filename);
-  fd_t fd = internal_open(tmp.data(), false);
+  fd_t fd = OpenFile(tmp.data(), false);
   if (fd == kInvalidFd) {
     Printf("ThreadSanitizer: failed to open suppressions file '%s'\n",
                tmp.data());
@@ -78,8 +85,7 @@ bool SuppressionMatch(char *templ, const char *str) {
   return true;
 }
 
-Suppression *SuppressionParse(const char* supp) {
-  Suppression *head = 0;
+Suppression *SuppressionParse(Suppression *head, const char* supp) {
   const char *line = supp;
   while (line) {
     while (line[0] == ' ' || line[0] == '\t')
@@ -128,8 +134,12 @@ Suppression *SuppressionParse(const char* supp) {
 }
 
 void InitializeSuppressions() {
-  char *supp = ReadFile(flags()->suppressions);
-  g_suppressions = SuppressionParse(supp);
+  const char *supp = ReadFile(flags()->suppressions);
+  g_suppressions = SuppressionParse(0, supp);
+#ifndef TSAN_GO
+  supp = __tsan_default_suppressions();
+  g_suppressions = SuppressionParse(0, supp);
+#endif
 }
 
 uptr IsSuppressed(ReportType typ, const ReportStack *stack) {
@@ -150,7 +160,8 @@ uptr IsSuppressed(ReportType typ, const ReportStack *stack) {
     for (Suppression *supp = g_suppressions; supp; supp = supp->next) {
       if (stype == supp->type &&
           (SuppressionMatch(supp->templ, frame->func) ||
-          SuppressionMatch(supp->templ, frame->file))) {
+           SuppressionMatch(supp->templ, frame->file) ||
+           SuppressionMatch(supp->templ, frame->module))) {
         DPrintf("ThreadSanitizer: matched suppression '%s'\n", supp->templ);
         return frame->pc;
       }
diff --git a/libsanitizer/tsan/tsan_suppressions.h b/libsanitizer/tsan/tsan_suppressions.h
index 4761eaa0cee..d44d8dd3529 100644
--- a/libsanitizer/tsan/tsan_suppressions.h
+++ b/libsanitizer/tsan/tsan_suppressions.h
@@ -33,7 +33,7 @@ struct Suppression {
   char *templ;
 };
 
-Suppression *SuppressionParse(const char* supp);
+Suppression *SuppressionParse(Suppression *head, const char* supp);
 bool SuppressionMatch(char *templ, const char *str);
 
 }  // namespace __tsan
diff --git a/libsanitizer/tsan/tsan_symbolize.cc b/libsanitizer/tsan/tsan_symbolize.cc
index 015b98717f1..65a994670b5 100644
--- a/libsanitizer/tsan/tsan_symbolize.cc
+++ b/libsanitizer/tsan/tsan_symbolize.cc
@@ -16,9 +16,24 @@
 #include "sanitizer_common/sanitizer_symbolizer.h"
 #include "tsan_flags.h"
 #include "tsan_report.h"
+#include "tsan_rtl.h"
 
 namespace __tsan {
 
+struct ScopedInSymbolizer {
+  ScopedInSymbolizer() {
+    ThreadState *thr = cur_thread();
+    CHECK(!thr->in_symbolizer);
+    thr->in_symbolizer = true;
+  }
+
+  ~ScopedInSymbolizer() {
+    ThreadState *thr = cur_thread();
+    CHECK(thr->in_symbolizer);
+    thr->in_symbolizer = false;
+  }
+};
+
 ReportStack *NewReportStackEntry(uptr addr) {
   ReportStack *ent = (ReportStack*)internal_alloc(MBlockReportStack,
                                                   sizeof(ReportStack));
@@ -53,35 +68,36 @@ static ReportStack *NewReportStackEntry(const AddressInfo &info) {
 }
 
 ReportStack *SymbolizeCode(uptr addr) {
-  if (flags()->external_symbolizer_path[0]) {
-    static const uptr kMaxAddrFrames = 16;
-    InternalScopedBuffer<AddressInfo> addr_frames(kMaxAddrFrames);
-    for (uptr i = 0; i < kMaxAddrFrames; i++)
-      new(&addr_frames[i]) AddressInfo();
-    uptr addr_frames_num = __sanitizer::SymbolizeCode(addr, addr_frames.data(),
-                                                      kMaxAddrFrames);
-    if (addr_frames_num == 0)
-      return NewReportStackEntry(addr);
-    ReportStack *top = 0;
-    ReportStack *bottom = 0;
-    for (uptr i = 0; i < addr_frames_num; i++) {
-      ReportStack *cur_entry = NewReportStackEntry(addr_frames[i]);
-      CHECK(cur_entry);
-      addr_frames[i].Clear();
-      if (i == 0)
-        top = cur_entry;
-      else
-        bottom->next = cur_entry;
-      bottom = cur_entry;
-    }
-    return top;
+  if (!IsSymbolizerAvailable())
+    return SymbolizeCodeAddr2Line(addr);
+  ScopedInSymbolizer in_symbolizer;
+  static const uptr kMaxAddrFrames = 16;
+  InternalScopedBuffer<AddressInfo> addr_frames(kMaxAddrFrames);
+  for (uptr i = 0; i < kMaxAddrFrames; i++)
+    new(&addr_frames[i]) AddressInfo();
+  uptr addr_frames_num = __sanitizer::SymbolizeCode(addr, addr_frames.data(),
+                                                    kMaxAddrFrames);
+  if (addr_frames_num == 0)
+    return NewReportStackEntry(addr);
+  ReportStack *top = 0;
+  ReportStack *bottom = 0;
+  for (uptr i = 0; i < addr_frames_num; i++) {
+    ReportStack *cur_entry = NewReportStackEntry(addr_frames[i]);
+    CHECK(cur_entry);
+    addr_frames[i].Clear();
+    if (i == 0)
+      top = cur_entry;
+    else
+      bottom->next = cur_entry;
+    bottom = cur_entry;
   }
-  return SymbolizeCodeAddr2Line(addr);
+  return top;
 }
 
 ReportLocation *SymbolizeData(uptr addr) {
-  if (flags()->external_symbolizer_path[0] == 0)
+  if (!IsSymbolizerAvailable())
     return 0;
+  ScopedInSymbolizer in_symbolizer;
   DataInfo info;
   if (!__sanitizer::SymbolizeData(addr, &info))
     return 0;
diff --git a/libsanitizer/tsan/tsan_update_shadow_word_inl.h b/libsanitizer/tsan/tsan_update_shadow_word_inl.h
index e22859c1dde..b9aa51c7957 100644
--- a/libsanitizer/tsan/tsan_update_shadow_word_inl.h
+++ b/libsanitizer/tsan/tsan_update_shadow_word_inl.h
@@ -32,7 +32,7 @@ do {
     if (Shadow::TidsAreEqual(old, cur)) {
       StatInc(thr, StatShadowSameThread);
       if (OldIsInSameSynchEpoch(old, thr)) {
-        if (OldIsRWNotWeaker(old, kAccessIsWrite)) {
+        if (old.IsRWNotWeaker(kAccessIsWrite, kIsAtomic)) {
           // found a slot that holds effectively the same info
           // (that is, same tid, same sync epoch and same size)
           StatInc(thr, StatMopSame);
@@ -41,7 +41,7 @@ do {
         StoreIfNotYetStored(sp, &store_word);
         break;
       }
-      if (OldIsRWWeakerOrEqual(old, kAccessIsWrite))
+      if (old.IsRWWeakerOrEqual(kAccessIsWrite, kIsAtomic))
         StoreIfNotYetStored(sp, &store_word);
       break;
     }
@@ -50,25 +50,23 @@ do {
       StoreIfNotYetStored(sp, &store_word);
       break;
     }
-    if (BothReads(old, kAccessIsWrite))
+    if (old.IsBothReadsOrAtomic(kAccessIsWrite, kIsAtomic))
       break;
     goto RACE;
   }
-
   // Do the memory access intersect?
-  if (Shadow::TwoRangesIntersect(old, cur, kAccessSize)) {
+  // In Go all memory accesses are 1 byte, so there can be no intersections.
+  if (kCppMode && Shadow::TwoRangesIntersect(old, cur, kAccessSize)) {
     StatInc(thr, StatShadowIntersect);
     if (Shadow::TidsAreEqual(old, cur)) {
       StatInc(thr, StatShadowSameThread);
       break;
     }
     StatInc(thr, StatShadowAnotherThread);
-    if (HappensBefore(old, thr))
+    if (old.IsBothReadsOrAtomic(kAccessIsWrite, kIsAtomic))
       break;
-
-    if (BothReads(old, kAccessIsWrite))
+    if (HappensBefore(old, thr))
       break;
-
     goto RACE;
   }
   // The accesses do not intersect.
diff --git a/libsanitizer/tsan/tsan_vector.h b/libsanitizer/tsan/tsan_vector.h
index d6bb7076a45..99e9f792c20 100644
--- a/libsanitizer/tsan/tsan_vector.h
+++ b/libsanitizer/tsan/tsan_vector.h
@@ -103,6 +103,6 @@ class Vector {
   Vector(const Vector&);
   void operator=(const Vector&);
 };
-}
+}  // namespace __tsan
 
 #endif  // #ifndef TSAN_VECTOR_H