Update compiler-rt to 3.9.0 release, and update the build glue for

libcompiler_rt and libclang_rt.

35 files changed, 1849 insertions, 630 deletions

diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_debugging.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_debugging.cc
new file mode 100644
index 000000000000..ac24c89be7da
--- /dev/null
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_debugging.cc
@@ -0,0 +1,162 @@
+//===-- tsan_debugging.cc -------------------------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of ThreadSanitizer (TSan), a race detector.
+//
+// TSan debugging API implementation.
+//===----------------------------------------------------------------------===//
+#include "tsan_interface.h"
+#include "tsan_report.h"
+#include "tsan_rtl.h"
+
+using namespace __tsan;
+
+static const char *ReportTypeDescription(ReportType typ) {
+  if (typ == ReportTypeRace) return "data-race";
+  if (typ == ReportTypeVptrRace) return "data-race-vptr";
+  if (typ == ReportTypeUseAfterFree) return "heap-use-after-free";
+  if (typ == ReportTypeVptrUseAfterFree) return "heap-use-after-free-vptr";
+  if (typ == ReportTypeThreadLeak) return "thread-leak";
+  if (typ == ReportTypeMutexDestroyLocked) return "locked-mutex-destroy";
+  if (typ == ReportTypeMutexDoubleLock) return "mutex-double-lock";
+  if (typ == ReportTypeMutexInvalidAccess) return "mutex-invalid-access";
+  if (typ == ReportTypeMutexBadUnlock) return "mutex-bad-unlock";
+  if (typ == ReportTypeMutexBadReadLock) return "mutex-bad-read-lock";
+  if (typ == ReportTypeMutexBadReadUnlock) return "mutex-bad-read-unlock";
+  if (typ == ReportTypeSignalUnsafe) return "signal-unsafe-call";
+  if (typ == ReportTypeErrnoInSignal) return "errno-in-signal-handler";
+  if (typ == ReportTypeDeadlock) return "lock-order-inversion";
+  return "";
+}
+
+static const char *ReportLocationTypeDescription(ReportLocationType typ) {
+  if (typ == ReportLocationGlobal) return "global";
+  if (typ == ReportLocationHeap) return "heap";
+  if (typ == ReportLocationStack) return "stack";
+  if (typ == ReportLocationTLS) return "tls";
+  if (typ == ReportLocationFD) return "fd";
+  return "";
+}
+
+static void CopyTrace(SymbolizedStack *first_frame, void **trace,
+                      uptr trace_size) {
+  uptr i = 0;
+  for (SymbolizedStack *frame = first_frame; frame != nullptr;
+       frame = frame->next) {
+    trace[i++] = (void *)frame->info.address;
+    if (i >= trace_size) break;
+  }
+}
+
+// Meant to be called by the debugger.
+SANITIZER_INTERFACE_ATTRIBUTE
+void *__tsan_get_current_report() {
+  return const_cast<ReportDesc*>(cur_thread()->current_report);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_get_report_data(void *report, const char **description, int *count,
+                           int *stack_count, int *mop_count, int *loc_count,
+                           int *mutex_count, int *thread_count,
+                           int *unique_tid_count, void **sleep_trace,
+                           uptr trace_size) {
+  const ReportDesc *rep = (ReportDesc *)report;
+  *description = ReportTypeDescription(rep->typ);
+  *count = rep->count;
+  *stack_count = rep->stacks.Size();
+  *mop_count = rep->mops.Size();
+  *loc_count = rep->locs.Size();
+  *mutex_count = rep->mutexes.Size();
+  *thread_count = rep->threads.Size();
+  *unique_tid_count = rep->unique_tids.Size();
+  if (rep->sleep) CopyTrace(rep->sleep->frames, sleep_trace, trace_size);
+  return 1;
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_get_report_stack(void *report, uptr idx, void **trace,
+                            uptr trace_size) {
+  const ReportDesc *rep = (ReportDesc *)report;
+  CHECK_LT(idx, rep->stacks.Size());
+  ReportStack *stack = rep->stacks[idx];
+  if (stack) CopyTrace(stack->frames, trace, trace_size);
+  return stack ? 1 : 0;
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_get_report_mop(void *report, uptr idx, int *tid, void **addr,
+                          int *size, int *write, int *atomic, void **trace,
+                          uptr trace_size) {
+  const ReportDesc *rep = (ReportDesc *)report;
+  CHECK_LT(idx, rep->mops.Size());
+  ReportMop *mop = rep->mops[idx];
+  *tid = mop->tid;
+  *addr = (void *)mop->addr;
+  *size = mop->size;
+  *write = mop->write ? 1 : 0;
+  *atomic = mop->atomic ? 1 : 0;
+  if (mop->stack) CopyTrace(mop->stack->frames, trace, trace_size);
+  return 1;
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_get_report_loc(void *report, uptr idx, const char **type,
+                          void **addr, uptr *start, uptr *size, int *tid,
+                          int *fd, int *suppressable, void **trace,
+                          uptr trace_size) {
+  const ReportDesc *rep = (ReportDesc *)report;
+  CHECK_LT(idx, rep->locs.Size());
+  ReportLocation *loc = rep->locs[idx];
+  *type = ReportLocationTypeDescription(loc->type);
+  *addr = (void *)loc->global.start;
+  *start = loc->heap_chunk_start;
+  *size = loc->heap_chunk_size;
+  *tid = loc->tid;
+  *fd = loc->fd;
+  *suppressable = loc->suppressable;
+  if (loc->stack) CopyTrace(loc->stack->frames, trace, trace_size);
+  return 1;
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_get_report_mutex(void *report, uptr idx, uptr *mutex_id, void **addr,
+                            int *destroyed, void **trace, uptr trace_size) {
+  const ReportDesc *rep = (ReportDesc *)report;
+  CHECK_LT(idx, rep->mutexes.Size());
+  ReportMutex *mutex = rep->mutexes[idx];
+  *mutex_id = mutex->id;
+  *addr = (void *)mutex->addr;
+  *destroyed = mutex->destroyed;
+  if (mutex->stack) CopyTrace(mutex->stack->frames, trace, trace_size);
+  return 1;
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_get_report_thread(void *report, uptr idx, int *tid, uptr *os_id,
+                             int *running, const char **name, int *parent_tid,
+                             void **trace, uptr trace_size) {
+  const ReportDesc *rep = (ReportDesc *)report;
+  CHECK_LT(idx, rep->threads.Size());
+  ReportThread *thread = rep->threads[idx];
+  *tid = thread->id;
+  *os_id = thread->os_id;
+  *running = thread->running;
+  *name = thread->name;
+  *parent_tid = thread->parent_tid;
+  if (thread->stack) CopyTrace(thread->stack->frames, trace, trace_size);
+  return 1;
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_get_report_unique_tid(void *report, uptr idx, int *tid) {
+  const ReportDesc *rep = (ReportDesc *)report;
+  CHECK_LT(idx, rep->unique_tids.Size());
+  *tid = rep->unique_tids[idx];
+  return 1;
+}
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_defs.h b/contrib/compiler-rt/lib/tsan/rtl/tsan_defs.h
index 9c7b329dcf00..cdc23d0a7e49 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_defs.h
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_defs.h
@@ -29,7 +29,11 @@
 #endif
 
 #ifndef TSAN_CONTAINS_UBSAN
-# define TSAN_CONTAINS_UBSAN (CAN_SANITIZE_UB && !defined(SANITIZER_GO))
+# if CAN_SANITIZE_UB && !defined(SANITIZER_GO)
+#  define TSAN_CONTAINS_UBSAN 1
+# else
+#  define TSAN_CONTAINS_UBSAN 0
+# endif
 #endif
 
 namespace __tsan {
@@ -145,6 +149,7 @@ struct MD5Hash {
 
 MD5Hash md5_hash(const void *data, uptr size);
 
+struct Processor;
 struct ThreadState;
 class ThreadContext;
 struct Context;
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_flags.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_flags.cc
index 761523171c77..93f598616f3a 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_flags.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_flags.cc
@@ -71,6 +71,7 @@ void InitializeFlags(Flags *f, const char *env) {
     cf.print_suppressions = false;
     cf.stack_trace_format = "    #%n %f %S %M";
     cf.exitcode = 66;
+    cf.intercept_tls_get_addr = true;
     OverrideCommonFlags(cf);
   }
 
@@ -108,7 +109,7 @@ void InitializeFlags(Flags *f, const char *env) {
     f->report_signal_unsafe = false;
   }
 
-  SetVerbosity(common_flags()->verbosity);
+  InitializeCommonFlags();
 
   if (Verbosity()) ReportUnrecognizedFlags();
 
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_flags.inc b/contrib/compiler-rt/lib/tsan/rtl/tsan_flags.inc
index ab9ca9924936..4fb443612c42 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_flags.inc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_flags.inc
@@ -76,3 +76,7 @@ TSAN_FLAG(int, io_sync, 1,
 TSAN_FLAG(bool, die_after_fork, true,
           "Die after multi-threaded fork if the child creates new threads.")
 TSAN_FLAG(const char *, suppressions, "", "Suppressions file name.")
+TSAN_FLAG(bool, ignore_interceptors_accesses, false,
+          "Ignore reads and writes from all interceptors.")
+TSAN_FLAG(bool, shared_ptr_interceptor, true,
+          "Track atomic reference counting in libc++ shared_ptr and weak_ptr.")
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_interceptors.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_interceptors.cc
index 7c835c6dc7df..fb6227651d21 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_interceptors.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_interceptors.cc
@@ -19,6 +19,7 @@
 #include "sanitizer_common/sanitizer_platform_limits_posix.h"
 #include "sanitizer_common/sanitizer_placement_new.h"
 #include "sanitizer_common/sanitizer_stacktrace.h"
+#include "sanitizer_common/sanitizer_tls_get_addr.h"
 #include "interception/interception.h"
 #include "tsan_interceptors.h"
 #include "tsan_interface.h"
@@ -40,20 +41,8 @@ using namespace __tsan;  // NOLINT
 #define stderr __stderrp
 #endif
 
-#if SANITIZER_FREEBSD
-#define __libc_realloc __realloc
-#define __libc_calloc __calloc
-#elif SANITIZER_MAC
-#define __libc_malloc REAL(malloc)
-#define __libc_realloc REAL(realloc)
-#define __libc_calloc REAL(calloc)
-#define __libc_free REAL(free)
-#elif SANITIZER_ANDROID
+#if SANITIZER_ANDROID
 #define __errno_location __errno
-#define __libc_malloc REAL(malloc)
-#define __libc_realloc REAL(realloc)
-#define __libc_calloc REAL(calloc)
-#define __libc_free REAL(free)
 #define mallopt(a, b)
 #endif
 
@@ -86,11 +75,9 @@ struct ucontext_t {
 };
 #endif
 
-#if defined(__x86_64__) || defined(__mips__) \
-  || (defined(__powerpc64__) && defined(__BIG_ENDIAN__))
+#if defined(__x86_64__) || defined(__mips__) || SANITIZER_PPC64V1
 #define PTHREAD_ABI_BASE  "GLIBC_2.3.2"
-#elif defined(__aarch64__) || (defined(__powerpc64__) \
-  && defined(__LITTLE_ENDIAN__))
+#elif defined(__aarch64__) || SANITIZER_PPC64V2
 #define PTHREAD_ABI_BASE  "GLIBC_2.17"
 #endif
 
@@ -103,8 +90,6 @@ extern "C" int pthread_setspecific(unsigned key, const void *v);
 DECLARE_REAL(int, pthread_mutexattr_gettype, void *, void *)
 extern "C" int pthread_sigmask(int how, const __sanitizer_sigset_t *set,
                                __sanitizer_sigset_t *oldset);
-// REAL(sigfillset) defined in common interceptors.
-DECLARE_REAL(int, sigfillset, __sanitizer_sigset_t *set)
 DECLARE_REAL(int, fflush, __sanitizer_FILE *fp)
 DECLARE_REAL_AND_INTERCEPTOR(void *, malloc, uptr size)
 DECLARE_REAL_AND_INTERCEPTOR(void, free, void *ptr)
@@ -112,21 +97,22 @@ extern "C" void *pthread_self();
 extern "C" void _exit(int status);
 extern "C" int *__errno_location();
 extern "C" int fileno_unlocked(void *stream);
-#if !SANITIZER_ANDROID
-extern "C" void *__libc_calloc(uptr size, uptr n);
-extern "C" void *__libc_realloc(void *ptr, uptr size);
-#endif
 extern "C" int dirfd(void *dirp);
 #if !SANITIZER_FREEBSD && !SANITIZER_ANDROID
 extern "C" int mallopt(int param, int value);
 #endif
 extern __sanitizer_FILE *stdout, *stderr;
+#if !SANITIZER_FREEBSD && !SANITIZER_MAC
 const int PTHREAD_MUTEX_RECURSIVE = 1;
 const int PTHREAD_MUTEX_RECURSIVE_NP = 1;
+#else
+const int PTHREAD_MUTEX_RECURSIVE = 2;
+const int PTHREAD_MUTEX_RECURSIVE_NP = 2;
+#endif
 const int EINVAL = 22;
 const int EBUSY = 16;
 const int EOWNERDEAD = 130;
-#if !SANITIZER_MAC
+#if !SANITIZER_FREEBSD && !SANITIZER_MAC
 const int EPOLL_CTL_ADD = 1;
 #endif
 const int SIGILL = 4;
@@ -135,7 +121,7 @@ const int SIGFPE = 8;
 const int SIGSEGV = 11;
 const int SIGPIPE = 13;
 const int SIGTERM = 15;
-#if defined(__mips__) || SANITIZER_MAC
+#if defined(__mips__) || SANITIZER_FREEBSD || SANITIZER_MAC
 const int SIGBUS = 10;
 const int SIGSYS = 12;
 #else
@@ -165,7 +151,7 @@ struct sigaction_t {
   u32 sa_flags;
   union {
     sighandler_t sa_handler;
-    sigactionhandler_t sa_sgiaction;
+    sigactionhandler_t sa_sigaction;
   };
   __sanitizer_sigset_t sa_mask;
   void (*sa_restorer)();
@@ -271,19 +257,24 @@ ScopedInterceptor::ScopedInterceptor(ThreadState *thr, const char *fname,
     : thr_(thr)
     , pc_(pc)
     , in_ignored_lib_(false) {
-  if (!thr_->ignore_interceptors) {
-    Initialize(thr);
+  Initialize(thr);
+  if (!thr_->is_inited)
+    return;
+  if (!thr_->ignore_interceptors)
     FuncEntry(thr, pc);
-  }
   DPrintf("#%d: intercept %s()\n", thr_->tid, fname);
   if (!thr_->in_ignored_lib && libignore()->IsIgnored(pc)) {
     in_ignored_lib_ = true;
     thr_->in_ignored_lib = true;
     ThreadIgnoreBegin(thr_, pc_);
   }
+  if (flags()->ignore_interceptors_accesses) ThreadIgnoreBegin(thr_, pc_);
 }
 
 ScopedInterceptor::~ScopedInterceptor() {
+  if (!thr_->is_inited)
+    return;
+  if (flags()->ignore_interceptors_accesses) ThreadIgnoreEnd(thr_, pc_);
   if (in_ignored_lib_) {
     thr_->in_ignored_lib = false;
     ThreadIgnoreEnd(thr_, pc_);
@@ -296,6 +287,7 @@ ScopedInterceptor::~ScopedInterceptor() {
 }
 
 void ScopedInterceptor::UserCallbackStart() {
+  if (flags()->ignore_interceptors_accesses) ThreadIgnoreEnd(thr_, pc_);
   if (in_ignored_lib_) {
     thr_->in_ignored_lib = false;
     ThreadIgnoreEnd(thr_, pc_);
@@ -307,6 +299,7 @@ void ScopedInterceptor::UserCallbackEnd() {
     thr_->in_ignored_lib = true;
     ThreadIgnoreBegin(thr_, pc_);
   }
+  if (flags()->ignore_interceptors_accesses) ThreadIgnoreBegin(thr_, pc_);
 }
 
 #define TSAN_INTERCEPT(func) INTERCEPT_FUNCTION(func)
@@ -387,7 +380,7 @@ static void at_exit_wrapper(void *arg) {
   Acquire(thr, pc, (uptr)arg);
   AtExitCtx *ctx = (AtExitCtx*)arg;
   ((void(*)(void *arg))ctx->f)(ctx->arg);
-  __libc_free(ctx);
+  InternalFree(ctx);
 }
 
 static int setup_at_exit_wrapper(ThreadState *thr, uptr pc, void(*f)(),
@@ -413,7 +406,7 @@ TSAN_INTERCEPTOR(int, __cxa_atexit, void (*f)(void *a), void *arg, void *dso) {
 
 static int setup_at_exit_wrapper(ThreadState *thr, uptr pc, void(*f)(),
       void *arg, void *dso) {
-  AtExitCtx *ctx = (AtExitCtx*)__libc_malloc(sizeof(AtExitCtx));
+  AtExitCtx *ctx = (AtExitCtx*)InternalAlloc(sizeof(AtExitCtx));
   ctx->f = f;
   ctx->arg = arg;
   Release(thr, pc, (uptr)ctx);
@@ -432,14 +425,14 @@ static void on_exit_wrapper(int status, void *arg) {
   Acquire(thr, pc, (uptr)arg);
   AtExitCtx *ctx = (AtExitCtx*)arg;
   ((void(*)(int status, void *arg))ctx->f)(status, ctx->arg);
-  __libc_free(ctx);
+  InternalFree(ctx);
 }
 
 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);
-  AtExitCtx *ctx = (AtExitCtx*)__libc_malloc(sizeof(AtExitCtx));
+  AtExitCtx *ctx = (AtExitCtx*)InternalAlloc(sizeof(AtExitCtx));
   ctx->f = (void(*)())f;
   ctx->arg = arg;
   Release(thr, pc, (uptr)ctx);
@@ -571,8 +564,11 @@ DEFINE_REAL(int, __sigsetjmp, void *env)
 #endif  // SANITIZER_MAC
 
 TSAN_INTERCEPTOR(void, longjmp, uptr *env, int val) {
+  // Note: if we call REAL(longjmp) in the context of ScopedInterceptor,
+  // bad things will happen. We will jump over ScopedInterceptor dtor and can
+  // leave thr->in_ignored_lib set.
   {
-    SCOPED_TSAN_INTERCEPTOR(longjmp, env, val);
+    SCOPED_INTERCEPTOR_RAW(longjmp, env, val);
   }
   LongJmp(cur_thread(), env);
   REAL(longjmp)(env, val);
@@ -580,7 +576,7 @@ TSAN_INTERCEPTOR(void, longjmp, uptr *env, int val) {
 
 TSAN_INTERCEPTOR(void, siglongjmp, uptr *env, int val) {
   {
-    SCOPED_TSAN_INTERCEPTOR(siglongjmp, env, val);
+    SCOPED_INTERCEPTOR_RAW(siglongjmp, env, val);
   }
   LongJmp(cur_thread(), env);
   REAL(siglongjmp)(env, val);
@@ -589,7 +585,7 @@ TSAN_INTERCEPTOR(void, siglongjmp, uptr *env, int val) {
 #if !SANITIZER_MAC
 TSAN_INTERCEPTOR(void*, malloc, uptr size) {
   if (cur_thread()->in_symbolizer)
-    return __libc_malloc(size);
+    return InternalAlloc(size);
   void *p = 0;
   {
     SCOPED_INTERCEPTOR_RAW(malloc, size);
@@ -606,7 +602,7 @@ 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);
+    return InternalCalloc(size, n);
   void *p = 0;
   {
     SCOPED_INTERCEPTOR_RAW(calloc, size, n);
@@ -618,7 +614,7 @@ 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);
+    return InternalRealloc(p, size);
   if (p)
     invoke_free_hook(p);
   {
@@ -633,7 +629,7 @@ TSAN_INTERCEPTOR(void, free, void *p) {
   if (p == 0)
     return;
   if (cur_thread()->in_symbolizer)
-    return __libc_free(p);
+    return InternalFree(p);
   invoke_free_hook(p);
   SCOPED_INTERCEPTOR_RAW(free, p);
   user_free(thr, pc, p);
@@ -643,7 +639,7 @@ TSAN_INTERCEPTOR(void, cfree, void *p) {
   if (p == 0)
     return;
   if (cur_thread()->in_symbolizer)
-    return __libc_free(p);
+    return InternalFree(p);
   invoke_free_hook(p);
   SCOPED_INTERCEPTOR_RAW(cfree, p);
   user_free(thr, pc, p);
@@ -655,69 +651,6 @@ TSAN_INTERCEPTOR(uptr, malloc_usable_size, void *p) {
 }
 #endif
 
-TSAN_INTERCEPTOR(uptr, strlen, const char *s) {
-  SCOPED_TSAN_INTERCEPTOR(strlen, s);
-  uptr len = internal_strlen(s);
-  MemoryAccessRange(thr, pc, (uptr)s, len + 1, false);
-  return len;
-}
-
-TSAN_INTERCEPTOR(void*, memset, void *dst, int v, uptr size) {
-  // On FreeBSD we get here from libthr internals on thread initialization.
-  if (!COMMON_INTERCEPTOR_NOTHING_IS_INITIALIZED) {
-    SCOPED_TSAN_INTERCEPTOR(memset, dst, v, size);
-    MemoryAccessRange(thr, pc, (uptr)dst, size, true);
-  }
-  return internal_memset(dst, v, size);
-}
-
-TSAN_INTERCEPTOR(void*, memcpy, void *dst, const void *src, uptr size) {
-  // On FreeBSD we get here from libthr internals on thread initialization.
-  if (!COMMON_INTERCEPTOR_NOTHING_IS_INITIALIZED) {
-    SCOPED_TSAN_INTERCEPTOR(memcpy, dst, src, size);
-    MemoryAccessRange(thr, pc, (uptr)dst, size, true);
-    MemoryAccessRange(thr, pc, (uptr)src, size, false);
-  }
-  // On OS X, calling internal_memcpy here will cause memory corruptions,
-  // because memcpy and memmove are actually aliases of the same implementation.
-  // We need to use internal_memmove here.
-  return internal_memmove(dst, src, size);
-}
-
-TSAN_INTERCEPTOR(void*, memmove, void *dst, void *src, uptr n) {
-  if (!COMMON_INTERCEPTOR_NOTHING_IS_INITIALIZED) {
-    SCOPED_TSAN_INTERCEPTOR(memmove, dst, src, n);
-    MemoryAccessRange(thr, pc, (uptr)dst, n, true);
-    MemoryAccessRange(thr, pc, (uptr)src, n, false);
-  }
-  return REAL(memmove)(dst, src, n);
-}
-
-TSAN_INTERCEPTOR(char*, strchr, char *s, int c) {
-  SCOPED_TSAN_INTERCEPTOR(strchr, s, c);
-  char *res = REAL(strchr)(s, c);
-  uptr len = internal_strlen(s);
-  uptr n = res ? (char*)res - (char*)s + 1 : len + 1;
-  READ_STRING_OF_LEN(thr, pc, s, len, n);
-  return res;
-}
-
-#if !SANITIZER_MAC
-TSAN_INTERCEPTOR(char*, strchrnul, char *s, int c) {
-  SCOPED_TSAN_INTERCEPTOR(strchrnul, s, c);
-  char *res = REAL(strchrnul)(s, c);
-  uptr len = (char*)res - (char*)s + 1;
-  READ_STRING(thr, pc, s, len);
-  return res;
-}
-#endif
-
-TSAN_INTERCEPTOR(char*, strrchr, char *s, int c) {
-  SCOPED_TSAN_INTERCEPTOR(strrchr, s, c);
-  MemoryAccessRange(thr, pc, (uptr)s, internal_strlen(s) + 1, false);
-  return REAL(strrchr)(s, c);
-}
-
 TSAN_INTERCEPTOR(char*, strcpy, char *dst, const char *src) {  // NOLINT
   SCOPED_TSAN_INTERCEPTOR(strcpy, dst, src);  // NOLINT
   uptr srclen = internal_strlen(src);
@@ -763,7 +696,11 @@ TSAN_INTERCEPTOR(void *, mmap, void *addr, SIZE_T sz, int prot, int flags,
   if (res != MAP_FAILED) {
     if (fd > 0)
       FdAccess(thr, pc, fd);
-    MemoryRangeImitateWrite(thr, pc, (uptr)res, sz);
+
+    if (thr->ignore_reads_and_writes == 0)
+      MemoryRangeImitateWrite(thr, pc, (uptr)res, sz);
+    else
+      MemoryResetRange(thr, pc, (uptr)res, sz);
   }
   return res;
 }
@@ -778,7 +715,11 @@ TSAN_INTERCEPTOR(void *, mmap64, void *addr, SIZE_T sz, int prot, int flags,
   if (res != MAP_FAILED) {
     if (fd > 0)
       FdAccess(thr, pc, fd);
-    MemoryRangeImitateWrite(thr, pc, (uptr)res, sz);
+
+    if (thr->ignore_reads_and_writes == 0)
+      MemoryRangeImitateWrite(thr, pc, (uptr)res, sz);
+    else
+      MemoryResetRange(thr, pc, (uptr)res, sz);
   }
   return res;
 }
@@ -792,7 +733,8 @@ TSAN_INTERCEPTOR(int, munmap, void *addr, long_t sz) {
   if (sz != 0) {
     // If sz == 0, munmap will return EINVAL and don't unmap any memory.
     DontNeedShadowFor((uptr)addr, sz);
-    ctx->metamap.ResetRange(thr, pc, (uptr)addr, (uptr)sz);
+    ScopedGlobalProcessor sgp;
+    ctx->metamap.ResetRange(thr->proc(), (uptr)addr, (uptr)sz);
   }
   int res = REAL(munmap)(addr, sz);
   return res;
@@ -887,12 +829,16 @@ STDCXX_INTERCEPTOR(void, __cxa_guard_abort, atomic_uint32_t *g) {
 namespace __tsan {
 void DestroyThreadState() {
   ThreadState *thr = cur_thread();
+  Processor *proc = thr->proc();
   ThreadFinish(thr);
+  ProcUnwire(proc, thr);
+  ProcDestroy(proc);
   ThreadSignalContext *sctx = thr->signal_ctx;
   if (sctx) {
     thr->signal_ctx = 0;
     UnmapOrDie(sctx, sizeof(*sctx));
   }
+  DTLS_Destroy();
   cur_thread_finalize();
 }
 }  // namespace __tsan
@@ -938,6 +884,8 @@ extern "C" void *__tsan_thread_start_func(void *arg) {
 #endif
     while ((tid = atomic_load(&p->tid, memory_order_acquire)) == 0)
       internal_sched_yield();
+    Processor *proc = ProcCreate();
+    ProcWire(proc, thr);
     ThreadStart(thr, tid, GetTid());
     atomic_store(&p->tid, 0, memory_order_release);
   }
@@ -1095,12 +1043,12 @@ INTERCEPTOR(int, pthread_cond_init, void *c, void *a) {
   return REAL(pthread_cond_init)(cond, a);
 }
 
-INTERCEPTOR(int, pthread_cond_wait, void *c, void *m) {
-  void *cond = init_cond(c);
-  SCOPED_TSAN_INTERCEPTOR(pthread_cond_wait, cond, m);
+static int cond_wait(ThreadState *thr, uptr pc, ScopedInterceptor *si,
+                     int (*fn)(void *c, void *m, void *abstime), void *c,
+                     void *m, void *t) {
   MemoryAccessRange(thr, pc, (uptr)c, sizeof(uptr), false);
   MutexUnlock(thr, pc, (uptr)m);
-  CondMutexUnlockCtx arg = {&si, thr, pc, m};
+  CondMutexUnlockCtx arg = {si, thr, pc, m};
   int res = 0;
   // This ensures that we handle mutex lock even in case of pthread_cancel.
   // See test/tsan/cond_cancel.cc.
@@ -1108,35 +1056,37 @@ INTERCEPTOR(int, pthread_cond_wait, void *c, void *m) {
     // Enable signal delivery while the thread is blocked.
     BlockingCall bc(thr);
     res = call_pthread_cancel_with_cleanup(
-        (int(*)(void *c, void *m, void *abstime))REAL(pthread_cond_wait),
-        cond, m, 0, (void(*)(void *arg))cond_mutex_unlock, &arg);
+        fn, c, m, t, (void (*)(void *arg))cond_mutex_unlock, &arg);
   }
-  if (res == errno_EOWNERDEAD)
-    MutexRepair(thr, pc, (uptr)m);
+  if (res == errno_EOWNERDEAD) MutexRepair(thr, pc, (uptr)m);
   MutexLock(thr, pc, (uptr)m);
   return res;
 }
 
+INTERCEPTOR(int, pthread_cond_wait, void *c, void *m) {
+  void *cond = init_cond(c);
+  SCOPED_TSAN_INTERCEPTOR(pthread_cond_wait, cond, m);
+  return cond_wait(thr, pc, &si, (int (*)(void *c, void *m, void *abstime))REAL(
+                                     pthread_cond_wait),
+                   cond, m, 0);
+}
+
 INTERCEPTOR(int, pthread_cond_timedwait, void *c, void *m, void *abstime) {
   void *cond = init_cond(c);
   SCOPED_TSAN_INTERCEPTOR(pthread_cond_timedwait, cond, m, abstime);
-  MemoryAccessRange(thr, pc, (uptr)c, sizeof(uptr), false);
-  MutexUnlock(thr, pc, (uptr)m);
-  CondMutexUnlockCtx arg = {&si, thr, pc, m};
-  int res = 0;
-  // This ensures that we handle mutex lock even in case of pthread_cancel.
-  // See test/tsan/cond_cancel.cc.
-  {
-    BlockingCall bc(thr);
-    res = call_pthread_cancel_with_cleanup(
-        REAL(pthread_cond_timedwait), cond, m, abstime,
-        (void(*)(void *arg))cond_mutex_unlock, &arg);
-  }
-  if (res == errno_EOWNERDEAD)
-    MutexRepair(thr, pc, (uptr)m);
-  MutexLock(thr, pc, (uptr)m);
-  return res;
+  return cond_wait(thr, pc, &si, REAL(pthread_cond_timedwait), cond, m,
+                   abstime);
+}
+
+#if SANITIZER_MAC
+INTERCEPTOR(int, pthread_cond_timedwait_relative_np, void *c, void *m,
+            void *reltime) {
+  void *cond = init_cond(c);
+  SCOPED_TSAN_INTERCEPTOR(pthread_cond_timedwait_relative_np, cond, m, reltime);
+  return cond_wait(thr, pc, &si, REAL(pthread_cond_timedwait_relative_np), cond,
+                   m, reltime);
 }
+#endif
 
 INTERCEPTOR(int, pthread_cond_signal, void *c) {
   void *cond = init_cond(c);
@@ -1395,96 +1345,6 @@ TSAN_INTERCEPTOR(int, pthread_once, void *o, void (*f)()) {
 }
 
 #if SANITIZER_LINUX && !SANITIZER_ANDROID
-TSAN_INTERCEPTOR(int, __xstat, int version, const char *path, void *buf) {
-  SCOPED_TSAN_INTERCEPTOR(__xstat, version, path, buf);
-  READ_STRING(thr, pc, path, 0);
-  return REAL(__xstat)(version, path, buf);
-}
-#define TSAN_MAYBE_INTERCEPT___XSTAT TSAN_INTERCEPT(__xstat)
-#else
-#define TSAN_MAYBE_INTERCEPT___XSTAT
-#endif
-
-TSAN_INTERCEPTOR(int, stat, const char *path, void *buf) {
-#if SANITIZER_FREEBSD || SANITIZER_MAC || SANITIZER_ANDROID
-  SCOPED_TSAN_INTERCEPTOR(stat, path, buf);
-  READ_STRING(thr, pc, path, 0);
-  return REAL(stat)(path, buf);
-#else
-  SCOPED_TSAN_INTERCEPTOR(__xstat, 0, path, buf);
-  READ_STRING(thr, pc, path, 0);
-  return REAL(__xstat)(0, path, buf);
-#endif
-}
-
-#if SANITIZER_LINUX && !SANITIZER_ANDROID
-TSAN_INTERCEPTOR(int, __xstat64, int version, const char *path, void *buf) {
-  SCOPED_TSAN_INTERCEPTOR(__xstat64, version, path, buf);
-  READ_STRING(thr, pc, path, 0);
-  return REAL(__xstat64)(version, path, buf);
-}
-#define TSAN_MAYBE_INTERCEPT___XSTAT64 TSAN_INTERCEPT(__xstat64)
-#else
-#define TSAN_MAYBE_INTERCEPT___XSTAT64
-#endif
-
-#if SANITIZER_LINUX && !SANITIZER_ANDROID
-TSAN_INTERCEPTOR(int, stat64, const char *path, void *buf) {
-  SCOPED_TSAN_INTERCEPTOR(__xstat64, 0, path, buf);
-  READ_STRING(thr, pc, path, 0);
-  return REAL(__xstat64)(0, path, buf);
-}
-#define TSAN_MAYBE_INTERCEPT_STAT64 TSAN_INTERCEPT(stat64)
-#else
-#define TSAN_MAYBE_INTERCEPT_STAT64
-#endif
-
-#if SANITIZER_LINUX && !SANITIZER_ANDROID
-TSAN_INTERCEPTOR(int, __lxstat, int version, const char *path, void *buf) {
-  SCOPED_TSAN_INTERCEPTOR(__lxstat, version, path, buf);
-  READ_STRING(thr, pc, path, 0);
-  return REAL(__lxstat)(version, path, buf);
-}
-#define TSAN_MAYBE_INTERCEPT___LXSTAT TSAN_INTERCEPT(__lxstat)
-#else
-#define TSAN_MAYBE_INTERCEPT___LXSTAT
-#endif
-
-TSAN_INTERCEPTOR(int, lstat, const char *path, void *buf) {
-#if SANITIZER_FREEBSD || SANITIZER_MAC || SANITIZER_ANDROID
-  SCOPED_TSAN_INTERCEPTOR(lstat, path, buf);
-  READ_STRING(thr, pc, path, 0);
-  return REAL(lstat)(path, buf);
-#else
-  SCOPED_TSAN_INTERCEPTOR(__lxstat, 0, path, buf);
-  READ_STRING(thr, pc, path, 0);
-  return REAL(__lxstat)(0, path, buf);
-#endif
-}
-
-#if SANITIZER_LINUX && !SANITIZER_ANDROID
-TSAN_INTERCEPTOR(int, __lxstat64, int version, const char *path, void *buf) {
-  SCOPED_TSAN_INTERCEPTOR(__lxstat64, version, path, buf);
-  READ_STRING(thr, pc, path, 0);
-  return REAL(__lxstat64)(version, path, buf);
-}
-#define TSAN_MAYBE_INTERCEPT___LXSTAT64 TSAN_INTERCEPT(__lxstat64)
-#else
-#define TSAN_MAYBE_INTERCEPT___LXSTAT64
-#endif
-
-#if SANITIZER_LINUX && !SANITIZER_ANDROID
-TSAN_INTERCEPTOR(int, lstat64, const char *path, void *buf) {
-  SCOPED_TSAN_INTERCEPTOR(__lxstat64, 0, path, buf);
-  READ_STRING(thr, pc, path, 0);
-  return REAL(__lxstat64)(0, path, buf);
-}
-#define TSAN_MAYBE_INTERCEPT_LSTAT64 TSAN_INTERCEPT(lstat64)
-#else
-#define TSAN_MAYBE_INTERCEPT_LSTAT64
-#endif
-
-#if SANITIZER_LINUX && !SANITIZER_ANDROID
 TSAN_INTERCEPTOR(int, __fxstat, int version, int fd, void *buf) {
   SCOPED_TSAN_INTERCEPTOR(__fxstat, version, fd, buf);
   if (fd > 0)
@@ -1701,32 +1561,6 @@ TSAN_INTERCEPTOR(int, listen, int fd, int backlog) {
   return res;
 }
 
-#if SANITIZER_LINUX
-TSAN_INTERCEPTOR(int, epoll_create, int size) {
-  SCOPED_TSAN_INTERCEPTOR(epoll_create, size);
-  int fd = REAL(epoll_create)(size);
-  if (fd >= 0)
-    FdPollCreate(thr, pc, fd);
-  return fd;
-}
-#define TSAN_MAYBE_INTERCEPT_EPOLL_CREATE TSAN_INTERCEPT(epoll_create)
-#else
-#define TSAN_MAYBE_INTERCEPT_EPOLL_CREATE
-#endif
-
-#if SANITIZER_LINUX
-TSAN_INTERCEPTOR(int, epoll_create1, int flags) {
-  SCOPED_TSAN_INTERCEPTOR(epoll_create1, flags);
-  int fd = REAL(epoll_create1)(flags);
-  if (fd >= 0)
-    FdPollCreate(thr, pc, fd);
-  return fd;
-}
-#define TSAN_MAYBE_INTERCEPT_EPOLL_CREATE1 TSAN_INTERCEPT(epoll_create1)
-#else
-#define TSAN_MAYBE_INTERCEPT_EPOLL_CREATE1
-#endif
-
 TSAN_INTERCEPTOR(int, close, int fd) {
   SCOPED_TSAN_INTERCEPTOR(close, fd);
   if (fd >= 0)
@@ -1781,37 +1615,6 @@ TSAN_INTERCEPTOR(int, pipe2, int *pipefd, int flags) {
 }
 #endif
 
-TSAN_INTERCEPTOR(long_t, send, int fd, void *buf, long_t len, int flags) {
-  SCOPED_TSAN_INTERCEPTOR(send, fd, buf, len, flags);
-  if (fd >= 0) {
-    FdAccess(thr, pc, fd);
-    FdRelease(thr, pc, fd);
-  }
-  int res = REAL(send)(fd, buf, len, flags);
-  return res;
-}
-
-TSAN_INTERCEPTOR(long_t, sendmsg, int fd, void *msg, int flags) {
-  SCOPED_TSAN_INTERCEPTOR(sendmsg, fd, msg, flags);
-  if (fd >= 0) {
-    FdAccess(thr, pc, fd);
-    FdRelease(thr, pc, fd);
-  }
-  int res = REAL(sendmsg)(fd, msg, flags);
-  return res;
-}
-
-TSAN_INTERCEPTOR(long_t, recv, int fd, void *buf, long_t len, int flags) {
-  SCOPED_TSAN_INTERCEPTOR(recv, fd, buf, len, flags);
-  if (fd >= 0)
-    FdAccess(thr, pc, fd);
-  int res = REAL(recv)(fd, buf, len, flags);
-  if (res >= 0 && fd >= 0) {
-    FdAcquire(thr, pc, fd);
-  }
-  return res;
-}
-
 TSAN_INTERCEPTOR(int, unlink, char *path) {
   SCOPED_TSAN_INTERCEPTOR(unlink, path);
   Release(thr, pc, File2addr(path));
@@ -1900,6 +1703,22 @@ TSAN_INTERCEPTOR(int, closedir, void *dirp) {
 }
 
 #if SANITIZER_LINUX
+TSAN_INTERCEPTOR(int, epoll_create, int size) {
+  SCOPED_TSAN_INTERCEPTOR(epoll_create, size);
+  int fd = REAL(epoll_create)(size);
+  if (fd >= 0)
+    FdPollCreate(thr, pc, fd);
+  return fd;
+}
+
+TSAN_INTERCEPTOR(int, epoll_create1, int flags) {
+  SCOPED_TSAN_INTERCEPTOR(epoll_create1, flags);
+  int fd = REAL(epoll_create1)(flags);
+  if (fd >= 0)
+    FdPollCreate(thr, pc, fd);
+  return fd;
+}
+
 TSAN_INTERCEPTOR(int, epoll_ctl, int epfd, int op, int fd, void *ev) {
   SCOPED_TSAN_INTERCEPTOR(epoll_ctl, epfd, op, fd, ev);
   if (epfd >= 0)
@@ -1911,12 +1730,7 @@ TSAN_INTERCEPTOR(int, epoll_ctl, int epfd, int op, int fd, void *ev) {
   int res = REAL(epoll_ctl)(epfd, op, fd, ev);
   return res;
 }
-#define TSAN_MAYBE_INTERCEPT_EPOLL_CTL TSAN_INTERCEPT(epoll_ctl)
-#else
-#define TSAN_MAYBE_INTERCEPT_EPOLL_CTL
-#endif
 
-#if SANITIZER_LINUX
 TSAN_INTERCEPTOR(int, epoll_wait, int epfd, void *ev, int cnt, int timeout) {
   SCOPED_TSAN_INTERCEPTOR(epoll_wait, epfd, ev, cnt, timeout);
   if (epfd >= 0)
@@ -1926,9 +1740,26 @@ TSAN_INTERCEPTOR(int, epoll_wait, int epfd, void *ev, int cnt, int timeout) {
     FdAcquire(thr, pc, epfd);
   return res;
 }
-#define TSAN_MAYBE_INTERCEPT_EPOLL_WAIT TSAN_INTERCEPT(epoll_wait)
+
+TSAN_INTERCEPTOR(int, epoll_pwait, int epfd, void *ev, int cnt, int timeout,
+                 void *sigmask) {
+  SCOPED_TSAN_INTERCEPTOR(epoll_pwait, epfd, ev, cnt, timeout, sigmask);
+  if (epfd >= 0)
+    FdAccess(thr, pc, epfd);
+  int res = BLOCK_REAL(epoll_pwait)(epfd, ev, cnt, timeout, sigmask);
+  if (res > 0 && epfd >= 0)
+    FdAcquire(thr, pc, epfd);
+  return res;
+}
+
+#define TSAN_MAYBE_INTERCEPT_EPOLL \
+    TSAN_INTERCEPT(epoll_create); \
+    TSAN_INTERCEPT(epoll_create1); \
+    TSAN_INTERCEPT(epoll_ctl); \
+    TSAN_INTERCEPT(epoll_wait); \
+    TSAN_INTERCEPT(epoll_pwait)
 #else
-#define TSAN_MAYBE_INTERCEPT_EPOLL_WAIT
+#define TSAN_MAYBE_INTERCEPT_EPOLL
 #endif
 
 namespace __tsan {
@@ -1937,6 +1768,19 @@ static void CallUserSignalHandler(ThreadState *thr, bool sync, bool acquire,
     bool sigact, int sig, my_siginfo_t *info, void *uctx) {
   if (acquire)
     Acquire(thr, 0, (uptr)&sigactions[sig]);
+  // Signals are generally asynchronous, so if we receive a signals when
+  // ignores are enabled we should disable ignores. This is critical for sync
+  // and interceptors, because otherwise we can miss syncronization and report
+  // false races.
+  int ignore_reads_and_writes = thr->ignore_reads_and_writes;
+  int ignore_interceptors = thr->ignore_interceptors;
+  int ignore_sync = thr->ignore_sync;
+  if (!ctx->after_multithreaded_fork) {
+    thr->ignore_reads_and_writes = 0;
+    thr->fast_state.ClearIgnoreBit();
+    thr->ignore_interceptors = 0;
+    thr->ignore_sync = 0;
+  }
   // Ensure that the handler does not spoil errno.
   const int saved_errno = errno;
   errno = 99;
@@ -1952,6 +1796,13 @@ static void CallUserSignalHandler(ThreadState *thr, bool sync, bool acquire,
     else
       ((sighandler_t)pc)(sig);
   }
+  if (!ctx->after_multithreaded_fork) {
+    thr->ignore_reads_and_writes = ignore_reads_and_writes;
+    if (ignore_reads_and_writes)
+      thr->fast_state.SetIgnoreBit();
+    thr->ignore_interceptors = ignore_interceptors;
+    thr->ignore_sync = ignore_sync;
+  }
   // We do not detect errno spoiling for SIGTERM,
   // because some SIGTERM handlers do spoil errno but reraise SIGTERM,
   // tsan reports false positive in such case.
@@ -1981,7 +1832,7 @@ void ProcessPendingSignals(ThreadState *thr) {
     return;
   atomic_store(&sctx->have_pending_signals, 0, memory_order_relaxed);
   atomic_fetch_add(&thr->in_signal_handler, 1, memory_order_relaxed);
-  CHECK_EQ(0, REAL(sigfillset)(&sctx->emptyset));
+  internal_sigfillset(&sctx->emptyset);
   CHECK_EQ(0, pthread_sigmask(SIG_SETMASK, &sctx->emptyset, &sctx->oldset));
   for (int sig = 0; sig < kSigCount; sig++) {
     SignalDesc *signal = &sctx->pending_signals[sig];
@@ -2021,13 +1872,8 @@ void ALWAYS_INLINE rtl_generic_sighandler(bool sigact, int sig,
       (sctx && atomic_load(&sctx->in_blocking_func, memory_order_relaxed))) {
     atomic_fetch_add(&thr->in_signal_handler, 1, memory_order_relaxed);
     if (sctx && atomic_load(&sctx->in_blocking_func, memory_order_relaxed)) {
-      // We ignore interceptors in blocking functions,
-      // temporary enbled them again while we are calling user function.
-      int const i = thr->ignore_interceptors;
-      thr->ignore_interceptors = 0;
       atomic_store(&sctx->in_blocking_func, 0, memory_order_relaxed);
       CallUserSignalHandler(thr, sync, true, sigact, sig, info, ctx);
-      thr->ignore_interceptors = i;
       atomic_store(&sctx->in_blocking_func, 1, memory_order_relaxed);
     } else {
       // Be very conservative with when we do acquire in this case.
@@ -2065,7 +1911,10 @@ static void rtl_sigaction(int sig, my_siginfo_t *info, void *ctx) {
 }
 
 TSAN_INTERCEPTOR(int, sigaction, int sig, sigaction_t *act, sigaction_t *old) {
-  SCOPED_TSAN_INTERCEPTOR(sigaction, sig, act, old);
+  // Note: if we call REAL(sigaction) directly for any reason without proxying
+  // the signal handler through rtl_sigaction, very bad things will happen.
+  // The handler will run synchronously and corrupt tsan per-thread state.
+  SCOPED_INTERCEPTOR_RAW(sigaction, sig, act, old);
   if (old)
     internal_memcpy(old, &sigactions[sig], sizeof(*old));
   if (act == 0)
@@ -2085,7 +1934,7 @@ TSAN_INTERCEPTOR(int, sigaction, int sig, sigaction_t *act, sigaction_t *old) {
 #endif
   sigaction_t newact;
   internal_memcpy(&newact, act, sizeof(newact));
-  REAL(sigfillset)(&newact.sa_mask);
+  internal_sigfillset(&newact.sa_mask);
   if (act->sa_handler != SIG_IGN && act->sa_handler != SIG_DFL) {
     if (newact.sa_flags & SA_SIGINFO)
       newact.sa_sigaction = rtl_sigaction;
@@ -2100,7 +1949,7 @@ TSAN_INTERCEPTOR(int, sigaction, int sig, sigaction_t *act, sigaction_t *old) {
 TSAN_INTERCEPTOR(sighandler_t, signal, int sig, sighandler_t h) {
   sigaction_t act;
   act.sa_handler = h;
-  REAL(memset)(&act.sa_mask, -1, sizeof(act.sa_mask));
+  internal_memset(&act.sa_mask, -1, sizeof(act.sa_mask));
   act.sa_flags = 0;
   sigaction_t old;
   int res = sigaction(sig, &act, &old);
@@ -2181,7 +2030,13 @@ TSAN_INTERCEPTOR(int, fork, int fake) {
     return REAL(fork)(fake);
   SCOPED_INTERCEPTOR_RAW(fork, fake);
   ForkBefore(thr, pc);
-  int pid = REAL(fork)(fake);
+  int pid;
+  {
+    // On OS X, REAL(fork) can call intercepted functions (OSSpinLockLock), and
+    // we'll assert in CheckNoLocks() unless we ignore interceptors.
+    ScopedIgnoreInterceptors ignore;
+    pid = REAL(fork)(fake);
+  }
   if (pid == 0) {
     // child
     ForkChildAfter(thr, pc);
@@ -2296,18 +2151,15 @@ static void HandleRecvmsg(ThreadState *thr, uptr pc,
 #undef SANITIZER_INTERCEPT_FGETPWENT
 #undef SANITIZER_INTERCEPT_GETPWNAM_AND_FRIENDS
 #undef SANITIZER_INTERCEPT_GETPWNAM_R_AND_FRIENDS
-// __tls_get_addr can be called with mis-aligned stack due to:
-// https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58066
-// There are two potential issues:
-// 1. Sanitizer code contains a MOVDQA spill (it does not seem to be the case
-// right now). or 2. ProcessPendingSignal calls user handler which contains
-// MOVDQA spill (this happens right now).
-// Since the interceptor only initializes memory for msan, the simplest solution
-// is to disable the interceptor in tsan (other sanitizers do not call
-// signal handlers from COMMON_INTERCEPTOR_ENTER).
+// We define our own.
+#if SANITIZER_INTERCEPT_TLS_GET_ADDR
+#define NEED_TLS_GET_ADDR
+#endif
 #undef SANITIZER_INTERCEPT_TLS_GET_ADDR
 
 #define COMMON_INTERCEPT_FUNCTION(name) INTERCEPT_FUNCTION(name)
+#define COMMON_INTERCEPT_FUNCTION_VER(name, ver)                          \
+  INTERCEPT_FUNCTION_VER(name, ver)
 
 #define COMMON_INTERCEPTOR_WRITE_RANGE(ctx, ptr, size)                    \
   MemoryAccessRange(((TsanInterceptorContext *)ctx)->thr,                 \
@@ -2394,6 +2246,10 @@ static void HandleRecvmsg(ThreadState *thr, uptr pc,
   MutexRepair(((TsanInterceptorContext *)ctx)->thr, \
             ((TsanInterceptorContext *)ctx)->pc, (uptr)m)
 
+#define COMMON_INTERCEPTOR_MUTEX_INVALID(ctx, m) \
+  MutexInvalidAccess(((TsanInterceptorContext *)ctx)->thr, \
+                     ((TsanInterceptorContext *)ctx)->pc, (uptr)m)
+
 #if !SANITIZER_MAC
 #define COMMON_INTERCEPTOR_HANDLE_RECVMSG(ctx, msg) \
   HandleRecvmsg(((TsanInterceptorContext *)ctx)->thr, \
@@ -2408,6 +2264,12 @@ static void HandleRecvmsg(ThreadState *thr, uptr pc,
     *begin = *end = 0;                                                         \
   }
 
+#define COMMON_INTERCEPTOR_USER_CALLBACK_START() \
+  SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START()
+
+#define COMMON_INTERCEPTOR_USER_CALLBACK_END() \
+  SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END()
+
 #include "sanitizer_common/sanitizer_common_interceptors.inc"
 
 #define TSAN_SYSCALL() \
@@ -2430,7 +2292,7 @@ struct ScopedSyscall {
   }
 };
 
-#if !SANITIZER_MAC
+#if !SANITIZER_FREEBSD && !SANITIZER_MAC
 static void syscall_access_range(uptr pc, uptr p, uptr s, bool write) {
   TSAN_SYSCALL();
   MemoryAccessRange(thr, pc, p, s, write);
@@ -2524,6 +2386,31 @@ static void syscall_post_fork(uptr pc, int pid) {
 
 #include "sanitizer_common/sanitizer_common_syscalls.inc"
 
+#ifdef NEED_TLS_GET_ADDR
+// Define own interceptor instead of sanitizer_common's for three reasons:
+// 1. It must not process pending signals.
+//    Signal handlers may contain MOVDQA instruction (see below).
+// 2. It must be as simple as possible to not contain MOVDQA.
+// 3. Sanitizer_common version uses COMMON_INTERCEPTOR_INITIALIZE_RANGE which
+//    is empty for tsan (meant only for msan).
+// Note: __tls_get_addr can be called with mis-aligned stack due to:
+// https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58066
+// So the interceptor must work with mis-aligned stack, in particular, does not
+// execute MOVDQA with stack addresses.
+TSAN_INTERCEPTOR(void *, __tls_get_addr, void *arg) {
+  void *res = REAL(__tls_get_addr)(arg);
+  ThreadState *thr = cur_thread();
+  if (!thr)
+    return res;
+  DTLS::DTV *dtv = DTLS_on_tls_get_addr(arg, res, thr->tls_addr, thr->tls_size);
+  if (!dtv)
+    return res;
+  // New DTLS block has been allocated.
+  MemoryResetRange(thr, 0, dtv->beg, dtv->size);
+  return res;
+}
+#endif
+
 namespace __tsan {
 
 static void finalize(void *arg) {
@@ -2584,13 +2471,6 @@ void InitializeInterceptors() {
   TSAN_MAYBE_INTERCEPT_PVALLOC;
   TSAN_INTERCEPT(posix_memalign);
 
-  TSAN_INTERCEPT(strlen);
-  TSAN_INTERCEPT(memset);
-  TSAN_INTERCEPT(memcpy);
-  TSAN_INTERCEPT(memmove);
-  TSAN_INTERCEPT(strchr);
-  TSAN_INTERCEPT(strchrnul);
-  TSAN_INTERCEPT(strrchr);
   TSAN_INTERCEPT(strcpy);  // NOLINT
   TSAN_INTERCEPT(strncpy);
   TSAN_INTERCEPT(strdup);
@@ -2633,14 +2513,6 @@ void InitializeInterceptors() {
 
   TSAN_INTERCEPT(pthread_once);
 
-  TSAN_INTERCEPT(stat);
-  TSAN_MAYBE_INTERCEPT___XSTAT;
-  TSAN_MAYBE_INTERCEPT_STAT64;
-  TSAN_MAYBE_INTERCEPT___XSTAT64;
-  TSAN_INTERCEPT(lstat);
-  TSAN_MAYBE_INTERCEPT___LXSTAT;
-  TSAN_MAYBE_INTERCEPT_LSTAT64;
-  TSAN_MAYBE_INTERCEPT___LXSTAT64;
   TSAN_INTERCEPT(fstat);
   TSAN_MAYBE_INTERCEPT___FXSTAT;
   TSAN_MAYBE_INTERCEPT_FSTAT64;
@@ -2661,18 +2533,13 @@ void InitializeInterceptors() {
   TSAN_INTERCEPT(connect);
   TSAN_INTERCEPT(bind);
   TSAN_INTERCEPT(listen);
-  TSAN_MAYBE_INTERCEPT_EPOLL_CREATE;
-  TSAN_MAYBE_INTERCEPT_EPOLL_CREATE1;
+  TSAN_MAYBE_INTERCEPT_EPOLL;
   TSAN_INTERCEPT(close);
   TSAN_MAYBE_INTERCEPT___CLOSE;
   TSAN_MAYBE_INTERCEPT___RES_ICLOSE;
   TSAN_INTERCEPT(pipe);
   TSAN_INTERCEPT(pipe2);
 
-  TSAN_INTERCEPT(send);
-  TSAN_INTERCEPT(sendmsg);
-  TSAN_INTERCEPT(recv);
-
   TSAN_INTERCEPT(unlink);
   TSAN_INTERCEPT(tmpfile);
   TSAN_MAYBE_INTERCEPT_TMPFILE64;
@@ -2683,9 +2550,6 @@ void InitializeInterceptors() {
   TSAN_INTERCEPT(rmdir);
   TSAN_INTERCEPT(closedir);
 
-  TSAN_MAYBE_INTERCEPT_EPOLL_CTL;
-  TSAN_MAYBE_INTERCEPT_EPOLL_WAIT;
-
   TSAN_INTERCEPT(sigaction);
   TSAN_INTERCEPT(signal);
   TSAN_INTERCEPT(sigsuspend);
@@ -2707,6 +2571,10 @@ void InitializeInterceptors() {
   TSAN_INTERCEPT(__cxa_atexit);
   TSAN_INTERCEPT(_exit);
 
+#ifdef NEED_TLS_GET_ADDR
+  TSAN_INTERCEPT(__tls_get_addr);
+#endif
+
 #if !SANITIZER_MAC && !SANITIZER_ANDROID
   // 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.
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_interceptors.h b/contrib/compiler-rt/lib/tsan/rtl/tsan_interceptors.h
index d831620cfafe..a0f9a0753a63 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_interceptors.h
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_interceptors.h
@@ -34,7 +34,7 @@ class ScopedInterceptor {
       Report("FATAL: ThreadSanitizer: failed to intercept %s\n", #func); \
       Die(); \
     }                                                    \
-    if (thr->ignore_interceptors || thr->in_ignored_lib) \
+    if (!thr->is_inited || thr->ignore_interceptors || thr->in_ignored_lib) \
       return REAL(func)(__VA_ARGS__); \
 /**/
 
@@ -46,12 +46,4 @@ class ScopedInterceptor {
 
 #define TSAN_INTERCEPTOR(ret, func, ...) INTERCEPTOR(ret, func, __VA_ARGS__)
 
-#if SANITIZER_FREEBSD
-#define __libc_free __free
-#define __libc_malloc __malloc
-#endif
-
-extern "C" void __libc_free(void *ptr);
-extern "C" void *__libc_malloc(uptr size);
-
 #endif  // TSAN_INTERCEPTORS_H
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_interceptors_mac.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_interceptors_mac.cc
index 2bf7ad9861c4..593963825c58 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_interceptors_mac.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_interceptors_mac.cc
@@ -17,11 +17,161 @@
 
 #include "interception/interception.h"
 #include "tsan_interceptors.h"
+#include "tsan_interface.h"
+#include "tsan_interface_ann.h"
 
 #include <libkern/OSAtomic.h>
+#include <xpc/xpc.h>
+
+typedef long long_t;  // NOLINT
 
 namespace __tsan {
 
+// The non-barrier versions of OSAtomic* functions are semantically mo_relaxed,
+// but the two variants (e.g. OSAtomicAdd32 and OSAtomicAdd32Barrier) are
+// actually aliases of each other, and we cannot have different interceptors for
+// them, because they're actually the same function.  Thus, we have to stay
+// conservative and treat the non-barrier versions as mo_acq_rel.
+static const morder kMacOrderBarrier = mo_acq_rel;
+static const morder kMacOrderNonBarrier = mo_acq_rel;
+
+#define OSATOMIC_INTERCEPTOR(return_t, t, tsan_t, f, tsan_atomic_f, mo) \
+  TSAN_INTERCEPTOR(return_t, f, t x, volatile t *ptr) {                 \
+    SCOPED_TSAN_INTERCEPTOR(f, x, ptr);                                 \
+    return tsan_atomic_f((volatile tsan_t *)ptr, x, mo);                \
+  }
+
+#define OSATOMIC_INTERCEPTOR_PLUS_X(return_t, t, tsan_t, f, tsan_atomic_f, mo) \
+  TSAN_INTERCEPTOR(return_t, f, t x, volatile t *ptr) {                        \
+    SCOPED_TSAN_INTERCEPTOR(f, x, ptr);                                        \
+    return tsan_atomic_f((volatile tsan_t *)ptr, x, mo) + x;                   \
+  }
+
+#define OSATOMIC_INTERCEPTOR_PLUS_1(return_t, t, tsan_t, f, tsan_atomic_f, mo) \
+  TSAN_INTERCEPTOR(return_t, f, volatile t *ptr) {                             \
+    SCOPED_TSAN_INTERCEPTOR(f, ptr);                                           \
+    return tsan_atomic_f((volatile tsan_t *)ptr, 1, mo) + 1;                   \
+  }
+
+#define OSATOMIC_INTERCEPTOR_MINUS_1(return_t, t, tsan_t, f, tsan_atomic_f, \
+                                     mo)                                    \
+  TSAN_INTERCEPTOR(return_t, f, volatile t *ptr) {                          \
+    SCOPED_TSAN_INTERCEPTOR(f, ptr);                                        \
+    return tsan_atomic_f((volatile tsan_t *)ptr, 1, mo) - 1;                \
+  }
+
+#define OSATOMIC_INTERCEPTORS_ARITHMETIC(f, tsan_atomic_f, m)                  \
+  m(int32_t, int32_t, a32, f##32, __tsan_atomic32_##tsan_atomic_f,             \
+    kMacOrderNonBarrier)                                                       \
+  m(int32_t, int32_t, a32, f##32##Barrier, __tsan_atomic32_##tsan_atomic_f,    \
+    kMacOrderBarrier)                                                          \
+  m(int64_t, int64_t, a64, f##64, __tsan_atomic64_##tsan_atomic_f,             \
+    kMacOrderNonBarrier)                                                       \
+  m(int64_t, int64_t, a64, f##64##Barrier, __tsan_atomic64_##tsan_atomic_f,    \
+    kMacOrderBarrier)
+
+#define OSATOMIC_INTERCEPTORS_BITWISE(f, tsan_atomic_f, m, m_orig)             \
+  m(int32_t, uint32_t, a32, f##32, __tsan_atomic32_##tsan_atomic_f,            \
+    kMacOrderNonBarrier)                                                       \
+  m(int32_t, uint32_t, a32, f##32##Barrier, __tsan_atomic32_##tsan_atomic_f,   \
+    kMacOrderBarrier)                                                          \
+  m_orig(int32_t, uint32_t, a32, f##32##Orig, __tsan_atomic32_##tsan_atomic_f, \
+    kMacOrderNonBarrier)                                                       \
+  m_orig(int32_t, uint32_t, a32, f##32##OrigBarrier,                           \
+    __tsan_atomic32_##tsan_atomic_f, kMacOrderBarrier)
+
+OSATOMIC_INTERCEPTORS_ARITHMETIC(OSAtomicAdd, fetch_add,
+                                 OSATOMIC_INTERCEPTOR_PLUS_X)
+OSATOMIC_INTERCEPTORS_ARITHMETIC(OSAtomicIncrement, fetch_add,
+                                 OSATOMIC_INTERCEPTOR_PLUS_1)
+OSATOMIC_INTERCEPTORS_ARITHMETIC(OSAtomicDecrement, fetch_sub,
+                                 OSATOMIC_INTERCEPTOR_MINUS_1)
+OSATOMIC_INTERCEPTORS_BITWISE(OSAtomicOr, fetch_or, OSATOMIC_INTERCEPTOR_PLUS_X,
+                              OSATOMIC_INTERCEPTOR)
+OSATOMIC_INTERCEPTORS_BITWISE(OSAtomicAnd, fetch_and,
+                              OSATOMIC_INTERCEPTOR_PLUS_X, OSATOMIC_INTERCEPTOR)
+OSATOMIC_INTERCEPTORS_BITWISE(OSAtomicXor, fetch_xor,
+                              OSATOMIC_INTERCEPTOR_PLUS_X, OSATOMIC_INTERCEPTOR)
+
+#define OSATOMIC_INTERCEPTORS_CAS(f, tsan_atomic_f, tsan_t, t)              \
+  TSAN_INTERCEPTOR(bool, f, t old_value, t new_value, t volatile *ptr) {    \
+    SCOPED_TSAN_INTERCEPTOR(f, old_value, new_value, ptr);                  \
+    return tsan_atomic_f##_compare_exchange_strong(                         \
+        (tsan_t *)ptr, (tsan_t *)&old_value, (tsan_t)new_value,             \
+        kMacOrderNonBarrier, kMacOrderNonBarrier);                          \
+  }                                                                         \
+                                                                            \
+  TSAN_INTERCEPTOR(bool, f##Barrier, t old_value, t new_value,              \
+                   t volatile *ptr) {                                       \
+    SCOPED_TSAN_INTERCEPTOR(f##Barrier, old_value, new_value, ptr);         \
+    return tsan_atomic_f##_compare_exchange_strong(                         \
+        (tsan_t *)ptr, (tsan_t *)&old_value, (tsan_t)new_value,             \
+        kMacOrderBarrier, kMacOrderNonBarrier);                             \
+  }
+
+OSATOMIC_INTERCEPTORS_CAS(OSAtomicCompareAndSwapInt, __tsan_atomic32, a32, int)
+OSATOMIC_INTERCEPTORS_CAS(OSAtomicCompareAndSwapLong, __tsan_atomic64, a64,
+                          long_t)
+OSATOMIC_INTERCEPTORS_CAS(OSAtomicCompareAndSwapPtr, __tsan_atomic64, a64,
+                          void *)
+OSATOMIC_INTERCEPTORS_CAS(OSAtomicCompareAndSwap32, __tsan_atomic32, a32,
+                          int32_t)
+OSATOMIC_INTERCEPTORS_CAS(OSAtomicCompareAndSwap64, __tsan_atomic64, a64,
+                          int64_t)
+
+#define OSATOMIC_INTERCEPTOR_BITOP(f, op, m, mo)              \
+  TSAN_INTERCEPTOR(bool, f, uint32_t n, volatile void *ptr) { \
+    SCOPED_TSAN_INTERCEPTOR(f, n, ptr);                       \
+    char *byte_ptr = ((char *)ptr) + (n >> 3);                \
+    char bit_index = n & 7;                                   \
+    char mask = m;                                            \
+    char orig_byte = op((a8 *)byte_ptr, mask, mo);            \
+    return orig_byte & mask;                                  \
+  }
+
+#define OSATOMIC_INTERCEPTORS_BITOP(f, op, m)                     \
+  OSATOMIC_INTERCEPTOR_BITOP(f, op, m, kMacOrderNonBarrier)       \
+  OSATOMIC_INTERCEPTOR_BITOP(f##Barrier, op, m, kMacOrderBarrier)
+
+OSATOMIC_INTERCEPTORS_BITOP(OSAtomicTestAndSet, __tsan_atomic8_fetch_or,
+                            0x80u >> bit_index)
+OSATOMIC_INTERCEPTORS_BITOP(OSAtomicTestAndClear, __tsan_atomic8_fetch_and,
+                            ~(0x80u >> bit_index))
+
+TSAN_INTERCEPTOR(void, OSAtomicEnqueue, OSQueueHead *list, void *item,
+                 size_t offset) {
+  SCOPED_TSAN_INTERCEPTOR(OSAtomicEnqueue, list, item, offset);
+  __tsan_release(item);
+  REAL(OSAtomicEnqueue)(list, item, offset);
+}
+
+TSAN_INTERCEPTOR(void *, OSAtomicDequeue, OSQueueHead *list, size_t offset) {
+  SCOPED_TSAN_INTERCEPTOR(OSAtomicDequeue, list, offset);
+  void *item = REAL(OSAtomicDequeue)(list, offset);
+  if (item) __tsan_acquire(item);
+  return item;
+}
+
+// OSAtomicFifoEnqueue and OSAtomicFifoDequeue are only on OS X.
+#if !SANITIZER_IOS
+
+TSAN_INTERCEPTOR(void, OSAtomicFifoEnqueue, OSFifoQueueHead *list, void *item,
+                 size_t offset) {
+  SCOPED_TSAN_INTERCEPTOR(OSAtomicFifoEnqueue, list, item, offset);
+  __tsan_release(item);
+  REAL(OSAtomicFifoEnqueue)(list, item, offset);
+}
+
+TSAN_INTERCEPTOR(void *, OSAtomicFifoDequeue, OSFifoQueueHead *list,
+                 size_t offset) {
+  SCOPED_TSAN_INTERCEPTOR(OSAtomicFifoDequeue, list, offset);
+  void *item = REAL(OSAtomicFifoDequeue)(list, offset);
+  if (item) __tsan_acquire(item);
+  return item;
+}
+
+#endif
+
 TSAN_INTERCEPTOR(void, OSSpinLockLock, volatile OSSpinLock *lock) {
   CHECK(!cur_thread()->is_dead);
   if (!cur_thread()->is_inited) {
@@ -86,6 +236,98 @@ TSAN_INTERCEPTOR(void, os_lock_unlock, void *lock) {
   REAL(os_lock_unlock)(lock);
 }
 
+TSAN_INTERCEPTOR(void, xpc_connection_set_event_handler,
+                 xpc_connection_t connection, xpc_handler_t handler) {
+  SCOPED_TSAN_INTERCEPTOR(xpc_connection_set_event_handler, connection,
+                          handler);
+  Release(thr, pc, (uptr)connection);
+  xpc_handler_t new_handler = ^(xpc_object_t object) {
+    {
+      SCOPED_INTERCEPTOR_RAW(xpc_connection_set_event_handler);
+      Acquire(thr, pc, (uptr)connection);
+    }
+    handler(object);
+  };
+  REAL(xpc_connection_set_event_handler)(connection, new_handler);
+}
+
+TSAN_INTERCEPTOR(void, xpc_connection_send_barrier, xpc_connection_t connection,
+                 dispatch_block_t barrier) {
+  SCOPED_TSAN_INTERCEPTOR(xpc_connection_send_barrier, connection, barrier);
+  Release(thr, pc, (uptr)connection);
+  dispatch_block_t new_barrier = ^() {
+    {
+      SCOPED_INTERCEPTOR_RAW(xpc_connection_send_barrier);
+      Acquire(thr, pc, (uptr)connection);
+    }
+    barrier();
+  };
+  REAL(xpc_connection_send_barrier)(connection, new_barrier);
+}
+
+TSAN_INTERCEPTOR(void, xpc_connection_send_message_with_reply,
+                 xpc_connection_t connection, xpc_object_t message,
+                 dispatch_queue_t replyq, xpc_handler_t handler) {
+  SCOPED_TSAN_INTERCEPTOR(xpc_connection_send_message_with_reply, connection,
+                          message, replyq, handler);
+  Release(thr, pc, (uptr)connection);
+  xpc_handler_t new_handler = ^(xpc_object_t object) {
+    {
+      SCOPED_INTERCEPTOR_RAW(xpc_connection_send_message_with_reply);
+      Acquire(thr, pc, (uptr)connection);
+    }
+    handler(object);
+  };
+  REAL(xpc_connection_send_message_with_reply)
+  (connection, message, replyq, new_handler);
+}
+
+// On macOS, libc++ is always linked dynamically, so intercepting works the
+// usual way.
+#define STDCXX_INTERCEPTOR TSAN_INTERCEPTOR
+
+namespace {
+struct fake_shared_weak_count {
+  volatile a64 shared_owners;
+  volatile a64 shared_weak_owners;
+  virtual void _unused_0x0() = 0;
+  virtual void _unused_0x8() = 0;
+  virtual void on_zero_shared() = 0;
+  virtual void _unused_0x18() = 0;
+  virtual void on_zero_shared_weak() = 0;
+};
+}  // namespace
+
+// This adds a libc++ interceptor for:
+//     void __shared_weak_count::__release_shared() _NOEXCEPT;
+// Shared and weak pointers in C++ maintain reference counts via atomics in
+// libc++.dylib, which are TSan-invisible, and this leads to false positives in
+// destructor code.  This interceptor re-implements the whole function so that
+// the mo_acq_rel semantics of the atomic decrement are visible.
+//
+// Unfortunately, this interceptor cannot simply Acquire/Release some sync
+// object and call the original function, because it would have a race between
+// the sync and the destruction of the object.  Calling both under a lock will
+// not work because the destructor can invoke this interceptor again (and even
+// in a different thread, so recursive locks don't help).
+STDCXX_INTERCEPTOR(void, _ZNSt3__119__shared_weak_count16__release_sharedEv,
+                   fake_shared_weak_count *o) {
+  if (!flags()->shared_ptr_interceptor)
+    return REAL(_ZNSt3__119__shared_weak_count16__release_sharedEv)(o);
+
+  SCOPED_TSAN_INTERCEPTOR(_ZNSt3__119__shared_weak_count16__release_sharedEv,
+                          o);
+  if (__tsan_atomic64_fetch_add(&o->shared_owners, -1, mo_release) == 0) {
+    Acquire(thr, pc, (uptr)&o->shared_owners);
+    o->on_zero_shared();
+    if (__tsan_atomic64_fetch_add(&o->shared_weak_owners, -1, mo_release) ==
+        0) {
+      Acquire(thr, pc, (uptr)&o->shared_weak_owners);
+      o->on_zero_shared_weak();
+    }
+  }
+}
+
 }  // namespace __tsan
 
 #endif  // SANITIZER_MAC
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_interface.h b/contrib/compiler-rt/lib/tsan/rtl/tsan_interface.h
index 41e084b82ab1..fbb099d07764 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_interface.h
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_interface.h
@@ -25,6 +25,8 @@
 extern "C" {
 #endif
 
+#ifndef SANITIZER_GO
+
 // This function should be called at the very beginning of the process,
 // before any instrumented code is executed and before any call to malloc.
 SANITIZER_INTERFACE_ATTRIBUTE void __tsan_init();
@@ -75,8 +77,296 @@ void __tsan_read_range(void *addr, unsigned long size);  // NOLINT
 SANITIZER_INTERFACE_ATTRIBUTE
 void __tsan_write_range(void *addr, unsigned long size);  // NOLINT
 
+// User may provide function that would be called right when TSan detects
+// an error. The argument 'report' is an opaque pointer that can be used to
+// gather additional information using other TSan report API functions.
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_on_report(void *report);
+
+// If TSan is currently reporting a detected issue on the current thread,
+// returns an opaque pointer to the current report. Otherwise returns NULL.
+SANITIZER_INTERFACE_ATTRIBUTE
+void *__tsan_get_current_report();
+
+// Returns a report's description (issue type), number of duplicate issues
+// found, counts of array data (stack traces, memory operations, locations,
+// mutexes, threads, unique thread IDs) and a stack trace of a sleep() call (if
+// one was involved in the issue).
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_get_report_data(void *report, const char **description, int *count,
+                           int *stack_count, int *mop_count, int *loc_count,
+                           int *mutex_count, int *thread_count,
+                           int *unique_tid_count, void **sleep_trace,
+                           uptr trace_size);
+
+// Returns information about stack traces included in the report.
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_get_report_stack(void *report, uptr idx, void **trace,
+                            uptr trace_size);
+
+// Returns information about memory operations included in the report.
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_get_report_mop(void *report, uptr idx, int *tid, void **addr,
+                          int *size, int *write, int *atomic, void **trace,
+                          uptr trace_size);
+
+// Returns information about locations included in the report.
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_get_report_loc(void *report, uptr idx, const char **type,
+                          void **addr, uptr *start, uptr *size, int *tid,
+                          int *fd, int *suppressable, void **trace,
+                          uptr trace_size);
+
+// Returns information about mutexes included in the report.
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_get_report_mutex(void *report, uptr idx, uptr *mutex_id, void **addr,
+                            int *destroyed, void **trace, uptr trace_size);
+
+// Returns information about threads included in the report.
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_get_report_thread(void *report, uptr idx, int *tid, uptr *os_id,
+                             int *running, const char **name, int *parent_tid,
+                             void **trace, uptr trace_size);
+
+// Returns information about unique thread IDs included in the report.
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_get_report_unique_tid(void *report, uptr idx, int *tid);
+
+#endif  // SANITIZER_GO
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
 
+namespace __tsan {
+
+// These should match declarations from public tsan_interface_atomic.h header.
+typedef unsigned char      a8;
+typedef unsigned short     a16;  // NOLINT
+typedef unsigned int       a32;
+typedef unsigned long long a64;  // NOLINT
+#if !defined(SANITIZER_GO) && (defined(__SIZEOF_INT128__) \
+    || (__clang_major__ * 100 + __clang_minor__ >= 302)) && !defined(__mips64)
+__extension__ typedef __int128 a128;
+# define __TSAN_HAS_INT128 1
+#else
+# define __TSAN_HAS_INT128 0
+#endif
+
+// Part of ABI, do not change.
+// http://llvm.org/viewvc/llvm-project/libcxx/trunk/include/atomic?view=markup
+typedef enum {
+  mo_relaxed,
+  mo_consume,
+  mo_acquire,
+  mo_release,
+  mo_acq_rel,
+  mo_seq_cst
+} morder;
+
+struct ThreadState;
+
+extern "C" {
+SANITIZER_INTERFACE_ATTRIBUTE
+a8 __tsan_atomic8_load(const volatile a8 *a, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a16 __tsan_atomic16_load(const volatile a16 *a, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a32 __tsan_atomic32_load(const volatile a32 *a, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a64 __tsan_atomic64_load(const volatile a64 *a, morder mo);
+#if __TSAN_HAS_INT128
+SANITIZER_INTERFACE_ATTRIBUTE
+a128 __tsan_atomic128_load(const volatile a128 *a, morder mo);
+#endif
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_atomic8_store(volatile a8 *a, a8 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_atomic16_store(volatile a16 *a, a16 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_atomic32_store(volatile a32 *a, a32 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_atomic64_store(volatile a64 *a, a64 v, morder mo);
+#if __TSAN_HAS_INT128
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_atomic128_store(volatile a128 *a, a128 v, morder mo);
+#endif
+
+SANITIZER_INTERFACE_ATTRIBUTE
+a8 __tsan_atomic8_exchange(volatile a8 *a, a8 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a16 __tsan_atomic16_exchange(volatile a16 *a, a16 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a32 __tsan_atomic32_exchange(volatile a32 *a, a32 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a64 __tsan_atomic64_exchange(volatile a64 *a, a64 v, morder mo);
+#if __TSAN_HAS_INT128
+SANITIZER_INTERFACE_ATTRIBUTE
+a128 __tsan_atomic128_exchange(volatile a128 *a, a128 v, morder mo);
+#endif
+
+SANITIZER_INTERFACE_ATTRIBUTE
+a8 __tsan_atomic8_fetch_add(volatile a8 *a, a8 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a16 __tsan_atomic16_fetch_add(volatile a16 *a, a16 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a32 __tsan_atomic32_fetch_add(volatile a32 *a, a32 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a64 __tsan_atomic64_fetch_add(volatile a64 *a, a64 v, morder mo);
+#if __TSAN_HAS_INT128
+SANITIZER_INTERFACE_ATTRIBUTE
+a128 __tsan_atomic128_fetch_add(volatile a128 *a, a128 v, morder mo);
+#endif
+
+SANITIZER_INTERFACE_ATTRIBUTE
+a8 __tsan_atomic8_fetch_sub(volatile a8 *a, a8 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a16 __tsan_atomic16_fetch_sub(volatile a16 *a, a16 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a32 __tsan_atomic32_fetch_sub(volatile a32 *a, a32 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a64 __tsan_atomic64_fetch_sub(volatile a64 *a, a64 v, morder mo);
+#if __TSAN_HAS_INT128
+SANITIZER_INTERFACE_ATTRIBUTE
+a128 __tsan_atomic128_fetch_sub(volatile a128 *a, a128 v, morder mo);
+#endif
+
+SANITIZER_INTERFACE_ATTRIBUTE
+a8 __tsan_atomic8_fetch_and(volatile a8 *a, a8 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a16 __tsan_atomic16_fetch_and(volatile a16 *a, a16 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a32 __tsan_atomic32_fetch_and(volatile a32 *a, a32 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a64 __tsan_atomic64_fetch_and(volatile a64 *a, a64 v, morder mo);
+#if __TSAN_HAS_INT128
+SANITIZER_INTERFACE_ATTRIBUTE
+a128 __tsan_atomic128_fetch_and(volatile a128 *a, a128 v, morder mo);
+#endif
+
+SANITIZER_INTERFACE_ATTRIBUTE
+a8 __tsan_atomic8_fetch_or(volatile a8 *a, a8 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a16 __tsan_atomic16_fetch_or(volatile a16 *a, a16 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a32 __tsan_atomic32_fetch_or(volatile a32 *a, a32 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a64 __tsan_atomic64_fetch_or(volatile a64 *a, a64 v, morder mo);
+#if __TSAN_HAS_INT128
+SANITIZER_INTERFACE_ATTRIBUTE
+a128 __tsan_atomic128_fetch_or(volatile a128 *a, a128 v, morder mo);
+#endif
+
+SANITIZER_INTERFACE_ATTRIBUTE
+a8 __tsan_atomic8_fetch_xor(volatile a8 *a, a8 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a16 __tsan_atomic16_fetch_xor(volatile a16 *a, a16 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a32 __tsan_atomic32_fetch_xor(volatile a32 *a, a32 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a64 __tsan_atomic64_fetch_xor(volatile a64 *a, a64 v, morder mo);
+#if __TSAN_HAS_INT128
+SANITIZER_INTERFACE_ATTRIBUTE
+a128 __tsan_atomic128_fetch_xor(volatile a128 *a, a128 v, morder mo);
+#endif
+
+SANITIZER_INTERFACE_ATTRIBUTE
+a8 __tsan_atomic8_fetch_nand(volatile a8 *a, a8 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a16 __tsan_atomic16_fetch_nand(volatile a16 *a, a16 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a32 __tsan_atomic32_fetch_nand(volatile a32 *a, a32 v, morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a64 __tsan_atomic64_fetch_nand(volatile a64 *a, a64 v, morder mo);
+#if __TSAN_HAS_INT128
+SANITIZER_INTERFACE_ATTRIBUTE
+a128 __tsan_atomic128_fetch_nand(volatile a128 *a, a128 v, morder mo);
+#endif
+
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_atomic8_compare_exchange_strong(volatile a8 *a, a8 *c, a8 v,
+                                           morder mo, morder fmo);
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_atomic16_compare_exchange_strong(volatile a16 *a, a16 *c, a16 v,
+                                            morder mo, morder fmo);
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_atomic32_compare_exchange_strong(volatile a32 *a, a32 *c, a32 v,
+                                            morder mo, morder fmo);
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_atomic64_compare_exchange_strong(volatile a64 *a, a64 *c, a64 v,
+                                            morder mo, morder fmo);
+#if __TSAN_HAS_INT128
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_atomic128_compare_exchange_strong(volatile a128 *a, a128 *c, a128 v,
+                                             morder mo, morder fmo);
+#endif
+
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_atomic8_compare_exchange_weak(volatile a8 *a, a8 *c, a8 v, morder mo,
+                                         morder fmo);
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_atomic16_compare_exchange_weak(volatile a16 *a, a16 *c, a16 v,
+                                          morder mo, morder fmo);
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_atomic32_compare_exchange_weak(volatile a32 *a, a32 *c, a32 v,
+                                          morder mo, morder fmo);
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_atomic64_compare_exchange_weak(volatile a64 *a, a64 *c, a64 v,
+                                          morder mo, morder fmo);
+#if __TSAN_HAS_INT128
+SANITIZER_INTERFACE_ATTRIBUTE
+int __tsan_atomic128_compare_exchange_weak(volatile a128 *a, a128 *c, a128 v,
+                                           morder mo, morder fmo);
+#endif
+
+SANITIZER_INTERFACE_ATTRIBUTE
+a8 __tsan_atomic8_compare_exchange_val(volatile a8 *a, a8 c, a8 v, morder mo,
+                                       morder fmo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a16 __tsan_atomic16_compare_exchange_val(volatile a16 *a, a16 c, a16 v,
+                                         morder mo, morder fmo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a32 __tsan_atomic32_compare_exchange_val(volatile a32 *a, a32 c, a32 v,
+                                         morder mo, morder fmo);
+SANITIZER_INTERFACE_ATTRIBUTE
+a64 __tsan_atomic64_compare_exchange_val(volatile a64 *a, a64 c, a64 v,
+                                         morder mo, morder fmo);
+#if __TSAN_HAS_INT128
+SANITIZER_INTERFACE_ATTRIBUTE
+a128 __tsan_atomic128_compare_exchange_val(volatile a128 *a, a128 c, a128 v,
+                                           morder mo, morder fmo);
+#endif
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_atomic_thread_fence(morder mo);
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_atomic_signal_fence(morder mo);
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic32_load(ThreadState *thr, uptr cpc, uptr pc, u8 *a);
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic64_load(ThreadState *thr, uptr cpc, uptr pc, u8 *a);
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic32_store(ThreadState *thr, uptr cpc, uptr pc, u8 *a);
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic64_store(ThreadState *thr, uptr cpc, uptr pc, u8 *a);
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic32_fetch_add(ThreadState *thr, uptr cpc, uptr pc, u8 *a);
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic64_fetch_add(ThreadState *thr, uptr cpc, uptr pc, u8 *a);
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic32_exchange(ThreadState *thr, uptr cpc, uptr pc, u8 *a);
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic64_exchange(ThreadState *thr, uptr cpc, uptr pc, u8 *a);
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic32_compare_exchange(ThreadState *thr, uptr cpc, uptr pc,
+                                         u8 *a);
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic64_compare_exchange(ThreadState *thr, uptr cpc, uptr pc,
+                                         u8 *a);
+}  // extern "C"
+
+}  // namespace __tsan
+
 #endif  // TSAN_INTERFACE_H
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_interface_atomic.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_interface_atomic.cc
index 27031991438c..dc0873f7948b 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_interface_atomic.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_interface_atomic.cc
@@ -23,39 +23,16 @@
 #include "sanitizer_common/sanitizer_stacktrace.h"
 #include "sanitizer_common/sanitizer_mutex.h"
 #include "tsan_flags.h"
+#include "tsan_interface.h"
 #include "tsan_rtl.h"
 
 using namespace __tsan;  // NOLINT
 
-// These should match declarations from public tsan_interface_atomic.h header.
-typedef unsigned char      a8;
-typedef unsigned short     a16;  // NOLINT
-typedef unsigned int       a32;
-typedef unsigned long long a64;  // NOLINT
-#if !defined(SANITIZER_GO) && (defined(__SIZEOF_INT128__) \
-    || (__clang_major__ * 100 + __clang_minor__ >= 302)) && !defined(__mips64)
-__extension__ typedef __int128 a128;
-# define __TSAN_HAS_INT128 1
-#else
-# define __TSAN_HAS_INT128 0
-#endif
-
 #if !defined(SANITIZER_GO) && __TSAN_HAS_INT128
 // Protects emulation of 128-bit atomic operations.
 static StaticSpinMutex mutex128;
 #endif
 
-// Part of ABI, do not change.
-// http://llvm.org/viewvc/llvm-project/libcxx/trunk/include/atomic?view=markup
-typedef enum {
-  mo_relaxed,
-  mo_consume,
-  mo_acquire,
-  mo_release,
-  mo_acq_rel,
-  mo_seq_cst
-} morder;
-
 static bool IsLoadOrder(morder mo) {
   return mo == mo_relaxed || mo == mo_consume
       || mo == mo_acquire || mo == mo_seq_cst;
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_interface_java.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_interface_java.cc
index 0aea63d11671..95be85994c64 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_interface_java.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_interface_java.cc
@@ -111,7 +111,7 @@ void __tsan_java_free(jptr ptr, jptr size) {
   CHECK_GE(ptr, jctx->heap_begin);
   CHECK_LE(ptr + size, jctx->heap_begin + jctx->heap_size);
 
-  ctx->metamap.FreeRange(thr, pc, ptr, size);
+  ctx->metamap.FreeRange(thr->proc(), ptr, size);
 }
 
 void __tsan_java_move(jptr src, jptr dst, jptr size) {
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_libdispatch_mac.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_libdispatch_mac.cc
index 617dc91b33d0..529cedba4d2b 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_libdispatch_mac.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_libdispatch_mac.cc
@@ -33,8 +33,10 @@ typedef struct {
   dispatch_queue_t queue;
   void *orig_context;
   dispatch_function_t orig_work;
-  uptr object_to_acquire;
-  dispatch_object_t object_to_release;
+  bool free_context_in_callback;
+  bool submitted_synchronously;
+  bool is_barrier_block;
+  uptr non_queue_sync_object;
 } tsan_block_context_t;
 
 // The offsets of different fields of the dispatch_queue_t structure, exported
@@ -66,6 +68,15 @@ static bool IsQueueSerial(dispatch_queue_t q) {
   return width == 1;
 }
 
+static dispatch_queue_t GetTargetQueueFromSource(dispatch_source_t source) {
+  CHECK_EQ(dispatch_queue_offsets.dqo_target_queue_size, 8);
+  dispatch_queue_t target_queue =
+      *(dispatch_queue_t *)(((uptr)source) +
+                            dispatch_queue_offsets.dqo_target_queue);
+  CHECK_NE(target_queue, 0);
+  return target_queue;
+}
+
 static tsan_block_context_t *AllocContext(ThreadState *thr, uptr pc,
                                           dispatch_queue_t queue,
                                           void *orig_context,
@@ -75,30 +86,40 @@ static tsan_block_context_t *AllocContext(ThreadState *thr, uptr pc,
   new_context->queue = queue;
   new_context->orig_context = orig_context;
   new_context->orig_work = orig_work;
-  new_context->object_to_acquire = (uptr)new_context;
-  new_context->object_to_release = nullptr;
+  new_context->free_context_in_callback = true;
+  new_context->submitted_synchronously = false;
+  new_context->is_barrier_block = false;
   return new_context;
 }
 
-static void dispatch_callback_wrap_acquire(void *param) {
-  SCOPED_INTERCEPTOR_RAW(dispatch_async_f_callback_wrap);
+static void dispatch_callback_wrap(void *param) {
+  SCOPED_INTERCEPTOR_RAW(dispatch_callback_wrap);
   tsan_block_context_t *context = (tsan_block_context_t *)param;
-  Acquire(thr, pc, context->object_to_acquire);
+  bool is_queue_serial = context->queue && IsQueueSerial(context->queue);
+  uptr sync_ptr = (uptr)context->queue ?: context->non_queue_sync_object;
 
-  // Extra retain/release is required for dispatch groups. We use the group
-  // itself to synchronize, but in a notification (dispatch_group_notify
-  // callback), it may be disposed already. To solve this, we retain the group
-  // and release it here.
-  if (context->object_to_release) dispatch_release(context->object_to_release);
+  uptr serial_sync = (uptr)sync_ptr;
+  uptr concurrent_sync = ((uptr)sync_ptr) + sizeof(uptr);
+  uptr submit_sync = (uptr)context;
+  bool serial_task = context->is_barrier_block || is_queue_serial;
+
+  Acquire(thr, pc, submit_sync);
+  Acquire(thr, pc, serial_sync);
+  if (serial_task) Acquire(thr, pc, concurrent_sync);
 
-  // In serial queues, work items can be executed on different threads, we need
-  // to explicitly synchronize on the queue itself.
-  if (IsQueueSerial(context->queue)) Acquire(thr, pc, (uptr)context->queue);
   SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();
   context->orig_work(context->orig_context);
   SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END();
-  if (IsQueueSerial(context->queue)) Release(thr, pc, (uptr)context->queue);
-  user_free(thr, pc, context);
+
+  Release(thr, pc, serial_task ? serial_sync : concurrent_sync);
+  if (context->submitted_synchronously) Release(thr, pc, submit_sync);
+
+  if (context->free_context_in_callback) user_free(thr, pc, context);
+}
+
+static void invoke_block(void *param) {
+  dispatch_block_t block = (dispatch_block_t)param;
+  block();
 }
 
 static void invoke_and_release_block(void *param) {
@@ -107,44 +128,97 @@ static void invoke_and_release_block(void *param) {
   Block_release(block);
 }
 
-#define DISPATCH_INTERCEPT_B(name)                                           \
+#define DISPATCH_INTERCEPT_B(name, barrier)                                  \
   TSAN_INTERCEPTOR(void, name, dispatch_queue_t q, dispatch_block_t block) { \
     SCOPED_TSAN_INTERCEPTOR(name, q, block);                                 \
-    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START(); \
+    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();                           \
     dispatch_block_t heap_block = Block_copy(block);                         \
-    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END(); \
+    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END();                             \
     tsan_block_context_t *new_context =                                      \
         AllocContext(thr, pc, q, heap_block, &invoke_and_release_block);     \
+    new_context->is_barrier_block = barrier;                                 \
     Release(thr, pc, (uptr)new_context);                                     \
-    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START(); \
-    REAL(name##_f)(q, new_context, dispatch_callback_wrap_acquire);          \
-    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END(); \
+    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();                           \
+    REAL(name##_f)(q, new_context, dispatch_callback_wrap);                  \
+    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END();                             \
+  }
+
+#define DISPATCH_INTERCEPT_SYNC_B(name, barrier)                             \
+  TSAN_INTERCEPTOR(void, name, dispatch_queue_t q, dispatch_block_t block) { \
+    SCOPED_TSAN_INTERCEPTOR(name, q, block);                                 \
+    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();                           \
+    dispatch_block_t heap_block = Block_copy(block);                         \
+    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END();                             \
+    tsan_block_context_t new_context = {                                     \
+        q, heap_block, &invoke_and_release_block, false, true, barrier, 0};  \
+    Release(thr, pc, (uptr)&new_context);                                    \
+    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();                           \
+    REAL(name##_f)(q, &new_context, dispatch_callback_wrap);                 \
+    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END();                             \
+    Acquire(thr, pc, (uptr)&new_context);                                    \
   }
 
-#define DISPATCH_INTERCEPT_F(name)                                \
+#define DISPATCH_INTERCEPT_F(name, barrier)                       \
   TSAN_INTERCEPTOR(void, name, dispatch_queue_t q, void *context, \
                    dispatch_function_t work) {                    \
     SCOPED_TSAN_INTERCEPTOR(name, q, context, work);              \
     tsan_block_context_t *new_context =                           \
         AllocContext(thr, pc, q, context, work);                  \
+    new_context->is_barrier_block = barrier;                      \
     Release(thr, pc, (uptr)new_context);                          \
-    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START(); \
-    REAL(name)(q, new_context, dispatch_callback_wrap_acquire);   \
-    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END(); \
+    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();                \
+    REAL(name)(q, new_context, dispatch_callback_wrap);           \
+    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END();                  \
+  }
+
+#define DISPATCH_INTERCEPT_SYNC_F(name, barrier)                              \
+  TSAN_INTERCEPTOR(void, name, dispatch_queue_t q, void *context,             \
+                   dispatch_function_t work) {                                \
+    SCOPED_TSAN_INTERCEPTOR(name, q, context, work);                          \
+    tsan_block_context_t new_context = {                                      \
+        q, context, work, false, true, barrier, 0};                           \
+    Release(thr, pc, (uptr)&new_context);                                     \
+    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();                            \
+    REAL(name)(q, &new_context, dispatch_callback_wrap);                      \
+    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END();                              \
+    Acquire(thr, pc, (uptr)&new_context);                                     \
   }
 
 // We wrap dispatch_async, dispatch_sync and friends where we allocate a new
 // context, which is used to synchronize (we release the context before
 // submitting, and the callback acquires it before executing the original
 // callback).
-DISPATCH_INTERCEPT_B(dispatch_async)
-DISPATCH_INTERCEPT_B(dispatch_barrier_async)
-DISPATCH_INTERCEPT_F(dispatch_async_f)
-DISPATCH_INTERCEPT_F(dispatch_barrier_async_f)
-DISPATCH_INTERCEPT_B(dispatch_sync)
-DISPATCH_INTERCEPT_B(dispatch_barrier_sync)
-DISPATCH_INTERCEPT_F(dispatch_sync_f)
-DISPATCH_INTERCEPT_F(dispatch_barrier_sync_f)
+DISPATCH_INTERCEPT_B(dispatch_async, false)
+DISPATCH_INTERCEPT_B(dispatch_barrier_async, true)
+DISPATCH_INTERCEPT_F(dispatch_async_f, false)
+DISPATCH_INTERCEPT_F(dispatch_barrier_async_f, true)
+DISPATCH_INTERCEPT_SYNC_B(dispatch_sync, false)
+DISPATCH_INTERCEPT_SYNC_B(dispatch_barrier_sync, true)
+DISPATCH_INTERCEPT_SYNC_F(dispatch_sync_f, false)
+DISPATCH_INTERCEPT_SYNC_F(dispatch_barrier_sync_f, true)
+
+TSAN_INTERCEPTOR(void, dispatch_after, dispatch_time_t when,
+                 dispatch_queue_t queue, dispatch_block_t block) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_after, when, queue, block);
+  SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();
+  dispatch_block_t heap_block = Block_copy(block);
+  SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END();
+  tsan_block_context_t *new_context =
+      AllocContext(thr, pc, queue, heap_block, &invoke_and_release_block);
+  Release(thr, pc, (uptr)new_context);
+  SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();
+  REAL(dispatch_after_f)(when, queue, new_context, dispatch_callback_wrap);
+  SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END();
+}
+
+TSAN_INTERCEPTOR(void, dispatch_after_f, dispatch_time_t when,
+                 dispatch_queue_t queue, void *context,
+                 dispatch_function_t work) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_after_f, when, queue, context, work);
+  WRAP(dispatch_after)(when, queue, ^(void) {
+    work(context);
+  });
+}
 
 // GCD's dispatch_once implementation has a fast path that contains a racy read
 // and it's inlined into user's code. Furthermore, this fast path doesn't
@@ -161,7 +235,7 @@ DISPATCH_INTERCEPT_F(dispatch_barrier_sync_f)
 #undef dispatch_once
 TSAN_INTERCEPTOR(void, dispatch_once, dispatch_once_t *predicate,
                  dispatch_block_t block) {
-  SCOPED_TSAN_INTERCEPTOR(dispatch_once, predicate, block);
+  SCOPED_INTERCEPTOR_RAW(dispatch_once, predicate, block);
   atomic_uint32_t *a = reinterpret_cast<atomic_uint32_t *>(predicate);
   u32 v = atomic_load(a, memory_order_acquire);
   if (v == 0 &&
@@ -183,7 +257,7 @@ TSAN_INTERCEPTOR(void, dispatch_once, dispatch_once_t *predicate,
 #undef dispatch_once_f
 TSAN_INTERCEPTOR(void, dispatch_once_f, dispatch_once_t *predicate,
                  void *context, dispatch_function_t function) {
-  SCOPED_TSAN_INTERCEPTOR(dispatch_once_f, predicate, context, function);
+  SCOPED_INTERCEPTOR_RAW(dispatch_once_f, predicate, context, function);
   SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();
   WRAP(dispatch_once)(predicate, ^(void) {
     function(context);
@@ -216,6 +290,7 @@ TSAN_INTERCEPTOR(long_t, dispatch_group_wait, dispatch_group_t group,
 
 TSAN_INTERCEPTOR(void, dispatch_group_leave, dispatch_group_t group) {
   SCOPED_TSAN_INTERCEPTOR(dispatch_group_leave, group);
+  // Acquired in the group noticifaction callback in dispatch_group_notify[_f].
   Release(thr, pc, (uptr)group);
   REAL(dispatch_group_leave)(group);
 }
@@ -225,8 +300,10 @@ TSAN_INTERCEPTOR(void, dispatch_group_async, dispatch_group_t group,
   SCOPED_TSAN_INTERCEPTOR(dispatch_group_async, group, queue, block);
   dispatch_retain(group);
   dispatch_group_enter(group);
+  __block dispatch_block_t block_copy = (dispatch_block_t)_Block_copy(block);
   WRAP(dispatch_async)(queue, ^(void) {
-    block();
+    block_copy();
+    _Block_release(block_copy);
     WRAP(dispatch_group_leave)(group);
     dispatch_release(group);
   });
@@ -248,35 +325,355 @@ TSAN_INTERCEPTOR(void, dispatch_group_async_f, dispatch_group_t group,
 TSAN_INTERCEPTOR(void, dispatch_group_notify, dispatch_group_t group,
                  dispatch_queue_t q, dispatch_block_t block) {
   SCOPED_TSAN_INTERCEPTOR(dispatch_group_notify, group, q, block);
+
+  // To make sure the group is still available in the callback (otherwise
+  // it can be already destroyed).  Will be released in the callback.
+  dispatch_retain(group);
+
   SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();
-  dispatch_block_t heap_block = Block_copy(block);
+  dispatch_block_t heap_block = Block_copy(^(void) {
+    {
+      SCOPED_INTERCEPTOR_RAW(dispatch_read_callback);
+      // Released when leaving the group (dispatch_group_leave).
+      Acquire(thr, pc, (uptr)group);
+    }
+    dispatch_release(group);
+    block();
+  });
   SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END();
   tsan_block_context_t *new_context =
       AllocContext(thr, pc, q, heap_block, &invoke_and_release_block);
-  new_context->object_to_acquire = (uptr)group;
-
-  // Will be released in dispatch_callback_wrap_acquire.
-  new_context->object_to_release = group;
-  dispatch_retain(group);
-
-  Release(thr, pc, (uptr)group);
-  REAL(dispatch_group_notify_f)(group, q, new_context,
-                                dispatch_callback_wrap_acquire);
+  new_context->is_barrier_block = true;
+  Release(thr, pc, (uptr)new_context);
+  REAL(dispatch_group_notify_f)(group, q, new_context, dispatch_callback_wrap);
 }
 
 TSAN_INTERCEPTOR(void, dispatch_group_notify_f, dispatch_group_t group,
                  dispatch_queue_t q, void *context, dispatch_function_t work) {
-  SCOPED_TSAN_INTERCEPTOR(dispatch_group_notify_f, group, q, context, work);
-  tsan_block_context_t *new_context = AllocContext(thr, pc, q, context, work);
-  new_context->object_to_acquire = (uptr)group;
+  WRAP(dispatch_group_notify)(group, q, ^(void) { work(context); });
+}
 
-  // Will be released in dispatch_callback_wrap_acquire.
-  new_context->object_to_release = group;
-  dispatch_retain(group);
+TSAN_INTERCEPTOR(void, dispatch_source_set_event_handler,
+                 dispatch_source_t source, dispatch_block_t handler) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_source_set_event_handler, source, handler);
+  if (handler == nullptr)
+    return REAL(dispatch_source_set_event_handler)(source, nullptr);
+  dispatch_queue_t q = GetTargetQueueFromSource(source);
+  __block tsan_block_context_t new_context = {
+      q, handler, &invoke_block, false, false, false, 0 };
+  dispatch_block_t new_handler = Block_copy(^(void) {
+    new_context.orig_context = handler;  // To explicitly capture "handler".
+    dispatch_callback_wrap(&new_context);
+  });
+  uptr submit_sync = (uptr)&new_context;
+  Release(thr, pc, submit_sync);
+  REAL(dispatch_source_set_event_handler)(source, new_handler);
+  Block_release(new_handler);
+}
 
-  Release(thr, pc, (uptr)group);
-  REAL(dispatch_group_notify_f)(group, q, new_context,
-                                dispatch_callback_wrap_acquire);
+TSAN_INTERCEPTOR(void, dispatch_source_set_event_handler_f,
+                 dispatch_source_t source, dispatch_function_t handler) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_source_set_event_handler_f, source, handler);
+  if (handler == nullptr)
+    return REAL(dispatch_source_set_event_handler)(source, nullptr);
+  dispatch_block_t block = ^(void) {
+    handler(dispatch_get_context(source));
+  };
+  WRAP(dispatch_source_set_event_handler)(source, block);
+}
+
+TSAN_INTERCEPTOR(void, dispatch_source_set_cancel_handler,
+                 dispatch_source_t source, dispatch_block_t handler) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_source_set_cancel_handler, source, handler);
+  if (handler == nullptr)
+    return REAL(dispatch_source_set_cancel_handler)(source, nullptr);
+  dispatch_queue_t q = GetTargetQueueFromSource(source);
+  __block tsan_block_context_t new_context = {
+      q, handler, &invoke_block, false, false, false, 0};
+  dispatch_block_t new_handler = Block_copy(^(void) {
+    new_context.orig_context = handler;  // To explicitly capture "handler".
+    dispatch_callback_wrap(&new_context);
+  });
+  uptr submit_sync = (uptr)&new_context;
+  Release(thr, pc, submit_sync);
+  REAL(dispatch_source_set_cancel_handler)(source, new_handler);
+  Block_release(new_handler);
+}
+
+TSAN_INTERCEPTOR(void, dispatch_source_set_cancel_handler_f,
+                 dispatch_source_t source, dispatch_function_t handler) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_source_set_cancel_handler_f, source,
+                          handler);
+  if (handler == nullptr)
+    return REAL(dispatch_source_set_cancel_handler)(source, nullptr);
+  dispatch_block_t block = ^(void) {
+    handler(dispatch_get_context(source));
+  };
+  WRAP(dispatch_source_set_cancel_handler)(source, block);
+}
+
+TSAN_INTERCEPTOR(void, dispatch_source_set_registration_handler,
+                 dispatch_source_t source, dispatch_block_t handler) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_source_set_registration_handler, source,
+                          handler);
+  if (handler == nullptr)
+    return REAL(dispatch_source_set_registration_handler)(source, nullptr);
+  dispatch_queue_t q = GetTargetQueueFromSource(source);
+  __block tsan_block_context_t new_context = {
+      q, handler, &invoke_block, false, false, false, 0};
+  dispatch_block_t new_handler = Block_copy(^(void) {
+    new_context.orig_context = handler;  // To explicitly capture "handler".
+    dispatch_callback_wrap(&new_context);
+  });
+  uptr submit_sync = (uptr)&new_context;
+  Release(thr, pc, submit_sync);
+  REAL(dispatch_source_set_registration_handler)(source, new_handler);
+  Block_release(new_handler);
+}
+
+TSAN_INTERCEPTOR(void, dispatch_source_set_registration_handler_f,
+                 dispatch_source_t source, dispatch_function_t handler) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_source_set_registration_handler_f, source,
+                          handler);
+  if (handler == nullptr)
+    return REAL(dispatch_source_set_registration_handler)(source, nullptr);
+  dispatch_block_t block = ^(void) {
+    handler(dispatch_get_context(source));
+  };
+  WRAP(dispatch_source_set_registration_handler)(source, block);
+}
+
+TSAN_INTERCEPTOR(void, dispatch_apply, size_t iterations,
+                 dispatch_queue_t queue, void (^block)(size_t)) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_apply, iterations, queue, block);
+
+  void *parent_to_child_sync = nullptr;
+  uptr parent_to_child_sync_uptr = (uptr)&parent_to_child_sync;
+  void *child_to_parent_sync = nullptr;
+  uptr child_to_parent_sync_uptr = (uptr)&child_to_parent_sync;
+
+  Release(thr, pc, parent_to_child_sync_uptr);
+  void (^new_block)(size_t) = ^(size_t iteration) {
+    SCOPED_INTERCEPTOR_RAW(dispatch_apply);
+    Acquire(thr, pc, parent_to_child_sync_uptr);
+    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();
+    block(iteration);
+    SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END();
+    Release(thr, pc, child_to_parent_sync_uptr);
+  };
+  SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();
+  REAL(dispatch_apply)(iterations, queue, new_block);
+  SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END();
+  Acquire(thr, pc, child_to_parent_sync_uptr);
+}
+
+TSAN_INTERCEPTOR(void, dispatch_apply_f, size_t iterations,
+                 dispatch_queue_t queue, void *context,
+                 void (*work)(void *, size_t)) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_apply_f, iterations, queue, context, work);
+  void (^new_block)(size_t) = ^(size_t iteration) {
+    work(context, iteration);
+  };
+  WRAP(dispatch_apply)(iterations, queue, new_block);
+}
+
+DECLARE_REAL_AND_INTERCEPTOR(void, free, void *ptr)
+DECLARE_REAL_AND_INTERCEPTOR(int, munmap, void *addr, long_t sz)
+
+TSAN_INTERCEPTOR(dispatch_data_t, dispatch_data_create, const void *buffer,
+                 size_t size, dispatch_queue_t q, dispatch_block_t destructor) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_data_create, buffer, size, q, destructor);
+  if ((q == nullptr) || (destructor == DISPATCH_DATA_DESTRUCTOR_DEFAULT))
+    return REAL(dispatch_data_create)(buffer, size, q, destructor);
+
+  if (destructor == DISPATCH_DATA_DESTRUCTOR_FREE)
+    destructor = ^(void) { WRAP(free)((void *)buffer); };
+  else if (destructor == DISPATCH_DATA_DESTRUCTOR_MUNMAP)
+    destructor = ^(void) { WRAP(munmap)((void *)buffer, size); };
+
+  SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();
+  dispatch_block_t heap_block = Block_copy(destructor);
+  SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_END();
+  tsan_block_context_t *new_context =
+      AllocContext(thr, pc, q, heap_block, &invoke_and_release_block);
+  uptr submit_sync = (uptr)new_context;
+  Release(thr, pc, submit_sync);
+  return REAL(dispatch_data_create)(buffer, size, q, ^(void) {
+    dispatch_callback_wrap(new_context);
+  });
+}
+
+typedef void (^fd_handler_t)(dispatch_data_t data, int error);
+typedef void (^cleanup_handler_t)(int error);
+
+TSAN_INTERCEPTOR(void, dispatch_read, dispatch_fd_t fd, size_t length,
+                 dispatch_queue_t q, fd_handler_t h) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_read, fd, length, q, h);
+  __block tsan_block_context_t new_context = {
+      q, nullptr, &invoke_block, false, false, false, 0};
+  fd_handler_t new_h = Block_copy(^(dispatch_data_t data, int error) {
+    new_context.orig_context = ^(void) {
+      h(data, error);
+    };
+    dispatch_callback_wrap(&new_context);
+  });
+  uptr submit_sync = (uptr)&new_context;
+  Release(thr, pc, submit_sync);
+  REAL(dispatch_read)(fd, length, q, new_h);
+  Block_release(new_h);
+}
+
+TSAN_INTERCEPTOR(void, dispatch_write, dispatch_fd_t fd, dispatch_data_t data,
+                 dispatch_queue_t q, fd_handler_t h) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_write, fd, data, q, h);
+  __block tsan_block_context_t new_context = {
+      q, nullptr, &invoke_block, false, false, false, 0};
+  fd_handler_t new_h = Block_copy(^(dispatch_data_t data, int error) {
+    new_context.orig_context = ^(void) {
+      h(data, error);
+    };
+    dispatch_callback_wrap(&new_context);
+  });
+  uptr submit_sync = (uptr)&new_context;
+  Release(thr, pc, submit_sync);
+  REAL(dispatch_write)(fd, data, q, new_h);
+  Block_release(new_h);
+}
+
+TSAN_INTERCEPTOR(void, dispatch_io_read, dispatch_io_t channel, off_t offset,
+                 size_t length, dispatch_queue_t q, dispatch_io_handler_t h) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_io_read, channel, offset, length, q, h);
+  __block tsan_block_context_t new_context = {
+      q, nullptr, &invoke_block, false, false, false, 0};
+  dispatch_io_handler_t new_h =
+      Block_copy(^(bool done, dispatch_data_t data, int error) {
+        new_context.orig_context = ^(void) {
+          h(done, data, error);
+        };
+        dispatch_callback_wrap(&new_context);
+      });
+  uptr submit_sync = (uptr)&new_context;
+  Release(thr, pc, submit_sync);
+  REAL(dispatch_io_read)(channel, offset, length, q, new_h);
+  Block_release(new_h);
+}
+
+TSAN_INTERCEPTOR(void, dispatch_io_write, dispatch_io_t channel, off_t offset,
+                 dispatch_data_t data, dispatch_queue_t q,
+                 dispatch_io_handler_t h) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_io_write, channel, offset, data, q, h);
+  __block tsan_block_context_t new_context = {
+      q, nullptr, &invoke_block, false, false, false, 0};
+  dispatch_io_handler_t new_h =
+      Block_copy(^(bool done, dispatch_data_t data, int error) {
+        new_context.orig_context = ^(void) {
+          h(done, data, error);
+        };
+        dispatch_callback_wrap(&new_context);
+      });
+  uptr submit_sync = (uptr)&new_context;
+  Release(thr, pc, submit_sync);
+  REAL(dispatch_io_write)(channel, offset, data, q, new_h);
+  Block_release(new_h);
+}
+
+TSAN_INTERCEPTOR(void, dispatch_io_barrier, dispatch_io_t channel,
+                 dispatch_block_t barrier) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_io_barrier, channel, barrier);
+  __block tsan_block_context_t new_context = {
+      nullptr, nullptr, &invoke_block, false, false, false, 0};
+  new_context.non_queue_sync_object = (uptr)channel;
+  new_context.is_barrier_block = true;
+  dispatch_block_t new_block = Block_copy(^(void) {
+    new_context.orig_context = ^(void) {
+      barrier();
+    };
+    dispatch_callback_wrap(&new_context);
+  });
+  uptr submit_sync = (uptr)&new_context;
+  Release(thr, pc, submit_sync);
+  REAL(dispatch_io_barrier)(channel, new_block);
+  Block_release(new_block);
+}
+
+TSAN_INTERCEPTOR(dispatch_io_t, dispatch_io_create, dispatch_io_type_t type,
+                 dispatch_fd_t fd, dispatch_queue_t q, cleanup_handler_t h) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_io_create, type, fd, q, h);
+  __block dispatch_io_t new_channel = nullptr;
+  __block tsan_block_context_t new_context = {
+      q, nullptr, &invoke_block, false, false, false, 0};
+  cleanup_handler_t new_h = Block_copy(^(int error) {
+    {
+      SCOPED_INTERCEPTOR_RAW(dispatch_io_create_callback);
+      Acquire(thr, pc, (uptr)new_channel);  // Release() in dispatch_io_close.
+    }
+    new_context.orig_context = ^(void) {
+      h(error);
+    };
+    dispatch_callback_wrap(&new_context);
+  });
+  uptr submit_sync = (uptr)&new_context;
+  Release(thr, pc, submit_sync);
+  new_channel = REAL(dispatch_io_create)(type, fd, q, new_h);
+  Block_release(new_h);
+  return new_channel;
+}
+
+TSAN_INTERCEPTOR(dispatch_io_t, dispatch_io_create_with_path,
+                 dispatch_io_type_t type, const char *path, int oflag,
+                 mode_t mode, dispatch_queue_t q, cleanup_handler_t h) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_io_create_with_path, type, path, oflag, mode,
+                          q, h);
+  __block dispatch_io_t new_channel = nullptr;
+  __block tsan_block_context_t new_context = {
+      q, nullptr, &invoke_block, false, false, false, 0};
+  cleanup_handler_t new_h = Block_copy(^(int error) {
+    {
+      SCOPED_INTERCEPTOR_RAW(dispatch_io_create_callback);
+      Acquire(thr, pc, (uptr)new_channel);  // Release() in dispatch_io_close.
+    }
+    new_context.orig_context = ^(void) {
+      h(error);
+    };
+    dispatch_callback_wrap(&new_context);
+  });
+  uptr submit_sync = (uptr)&new_context;
+  Release(thr, pc, submit_sync);
+  new_channel =
+      REAL(dispatch_io_create_with_path)(type, path, oflag, mode, q, new_h);
+  Block_release(new_h);
+  return new_channel;
+}
+
+TSAN_INTERCEPTOR(dispatch_io_t, dispatch_io_create_with_io,
+                 dispatch_io_type_t type, dispatch_io_t io, dispatch_queue_t q,
+                 cleanup_handler_t h) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_io_create_with_io, type, io, q, h);
+  __block dispatch_io_t new_channel = nullptr;
+  __block tsan_block_context_t new_context = {
+      q, nullptr, &invoke_block, false, false, false, 0};
+  cleanup_handler_t new_h = Block_copy(^(int error) {
+    {
+      SCOPED_INTERCEPTOR_RAW(dispatch_io_create_callback);
+      Acquire(thr, pc, (uptr)new_channel);  // Release() in dispatch_io_close.
+    }
+    new_context.orig_context = ^(void) {
+      h(error);
+    };
+    dispatch_callback_wrap(&new_context);
+  });
+  uptr submit_sync = (uptr)&new_context;
+  Release(thr, pc, submit_sync);
+  new_channel = REAL(dispatch_io_create_with_io)(type, io, q, new_h);
+  Block_release(new_h);
+  return new_channel;
+}
+
+TSAN_INTERCEPTOR(void, dispatch_io_close, dispatch_io_t channel,
+                 dispatch_io_close_flags_t flags) {
+  SCOPED_TSAN_INTERCEPTOR(dispatch_io_close, channel, flags);
+  Release(thr, pc, (uptr)channel);  // Acquire() in dispatch_io_create[_*].
+  return REAL(dispatch_io_close)(channel, flags);
 }
 
 }  // namespace __tsan
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_malloc_mac.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_malloc_mac.cc
index 7fd94273c314..8d31ccbcaaca 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_malloc_mac.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_malloc_mac.cc
@@ -27,33 +27,28 @@ using namespace __tsan;
 #define COMMON_MALLOC_MEMALIGN(alignment, size) \
   void *p =                                     \
       user_alloc(cur_thread(), StackTrace::GetCurrentPc(), size, alignment)
-#define COMMON_MALLOC_MALLOC(size)      \
-  if (cur_thread()->in_symbolizer)      \
-    return REAL(malloc)(size);          \
-  SCOPED_INTERCEPTOR_RAW(malloc, size); \
+#define COMMON_MALLOC_MALLOC(size)                             \
+  if (cur_thread()->in_symbolizer) return InternalAlloc(size); \
+  SCOPED_INTERCEPTOR_RAW(malloc, size);                        \
   void *p = user_alloc(thr, pc, size)
-#define COMMON_MALLOC_REALLOC(ptr, size)      \
-  if (cur_thread()->in_symbolizer)            \
-    return REAL(realloc)(ptr, size);          \
-  SCOPED_INTERCEPTOR_RAW(realloc, ptr, size); \
+#define COMMON_MALLOC_REALLOC(ptr, size)                              \
+  if (cur_thread()->in_symbolizer) return InternalRealloc(ptr, size); \
+  SCOPED_INTERCEPTOR_RAW(realloc, ptr, size);                         \
   void *p = user_realloc(thr, pc, ptr, size)
-#define COMMON_MALLOC_CALLOC(count, size)      \
-  if (cur_thread()->in_symbolizer)             \
-    return REAL(calloc)(count, size);          \
-  SCOPED_INTERCEPTOR_RAW(calloc, size, count); \
+#define COMMON_MALLOC_CALLOC(count, size)                              \
+  if (cur_thread()->in_symbolizer) return InternalCalloc(count, size); \
+  SCOPED_INTERCEPTOR_RAW(calloc, size, count);                         \
   void *p = user_calloc(thr, pc, size, count)
-#define COMMON_MALLOC_VALLOC(size)                          \
-  if (cur_thread()->in_symbolizer)                          \
-    return REAL(valloc)(size);                              \
-  SCOPED_INTERCEPTOR_RAW(valloc, size);                     \
+#define COMMON_MALLOC_VALLOC(size)                            \
+  if (cur_thread()->in_symbolizer)                            \
+    return InternalAlloc(size, nullptr, GetPageSizeCached()); \
+  SCOPED_INTERCEPTOR_RAW(valloc, size);                       \
   void *p = user_alloc(thr, pc, size, GetPageSizeCached())
-#define COMMON_MALLOC_FREE(ptr)      \
-  if (cur_thread()->in_symbolizer)   \
-    return REAL(free)(ptr);          \
-  SCOPED_INTERCEPTOR_RAW(free, ptr); \
+#define COMMON_MALLOC_FREE(ptr)                              \
+  if (cur_thread()->in_symbolizer) return InternalFree(ptr); \
+  SCOPED_INTERCEPTOR_RAW(free, ptr);                         \
   user_free(thr, pc, ptr)
-#define COMMON_MALLOC_SIZE(ptr) \
-  uptr size = user_alloc_usable_size(ptr);
+#define COMMON_MALLOC_SIZE(ptr) uptr size = user_alloc_usable_size(ptr);
 #define COMMON_MALLOC_FILL_STATS(zone, stats)
 #define COMMON_MALLOC_REPORT_UNKNOWN_REALLOC(ptr, zone_ptr, zone_name) \
   (void)zone_name; \
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_mman.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_mman.cc
index 7247c6e00035..7693077f622b 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_mman.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_mman.cc
@@ -63,18 +63,69 @@ Allocator *allocator() {
   return reinterpret_cast<Allocator*>(&allocator_placeholder);
 }
 
+struct GlobalProc {
+  Mutex mtx;
+  Processor *proc;
+
+  GlobalProc()
+      : mtx(MutexTypeGlobalProc, StatMtxGlobalProc)
+      , proc(ProcCreate()) {
+  }
+};
+
+static char global_proc_placeholder[sizeof(GlobalProc)] ALIGNED(64);
+GlobalProc *global_proc() {
+  return reinterpret_cast<GlobalProc*>(&global_proc_placeholder);
+}
+
+ScopedGlobalProcessor::ScopedGlobalProcessor() {
+  GlobalProc *gp = global_proc();
+  ThreadState *thr = cur_thread();
+  if (thr->proc())
+    return;
+  // If we don't have a proc, use the global one.
+  // There are currently only two known case where this path is triggered:
+  //   __interceptor_free
+  //   __nptl_deallocate_tsd
+  //   start_thread
+  //   clone
+  // and:
+  //   ResetRange
+  //   __interceptor_munmap
+  //   __deallocate_stack
+  //   start_thread
+  //   clone
+  // Ideally, we destroy thread state (and unwire proc) when a thread actually
+  // exits (i.e. when we join/wait it). Then we would not need the global proc
+  gp->mtx.Lock();
+  ProcWire(gp->proc, thr);
+}
+
+ScopedGlobalProcessor::~ScopedGlobalProcessor() {
+  GlobalProc *gp = global_proc();
+  ThreadState *thr = cur_thread();
+  if (thr->proc() != gp->proc)
+    return;
+  ProcUnwire(gp->proc, thr);
+  gp->mtx.Unlock();
+}
+
 void InitializeAllocator() {
   allocator()->Init(common_flags()->allocator_may_return_null);
 }
 
-void AllocatorThreadStart(ThreadState *thr) {
-  allocator()->InitCache(&thr->alloc_cache);
-  internal_allocator()->InitCache(&thr->internal_alloc_cache);
+void InitializeAllocatorLate() {
+  new(global_proc()) GlobalProc();
+}
+
+void AllocatorProcStart(Processor *proc) {
+  allocator()->InitCache(&proc->alloc_cache);
+  internal_allocator()->InitCache(&proc->internal_alloc_cache);
 }
 
-void AllocatorThreadFinish(ThreadState *thr) {
-  allocator()->DestroyCache(&thr->alloc_cache);
-  internal_allocator()->DestroyCache(&thr->internal_alloc_cache);
+void AllocatorProcFinish(Processor *proc) {
+  allocator()->DestroyCache(&proc->alloc_cache);
+  internal_allocator()->DestroyCache(&proc->internal_alloc_cache);
 }
 
 void AllocatorPrintStats() {
@@ -98,7 +149,7 @@ static void SignalUnsafeCall(ThreadState *thr, uptr pc) {
 void *user_alloc(ThreadState *thr, uptr pc, uptr sz, uptr align, bool signal) {
   if ((sz >= (1ull << 40)) || (align >= (1ull << 40)))
     return allocator()->ReturnNullOrDie();
-  void *p = allocator()->Allocate(&thr->alloc_cache, sz, align);
+  void *p = allocator()->Allocate(&thr->proc()->alloc_cache, sz, align);
   if (p == 0)
     return 0;
   if (ctx && ctx->initialized)
@@ -118,9 +169,10 @@ void *user_calloc(ThreadState *thr, uptr pc, uptr size, uptr n) {
 }
 
 void user_free(ThreadState *thr, uptr pc, void *p, bool signal) {
+  ScopedGlobalProcessor sgp;
   if (ctx && ctx->initialized)
     OnUserFree(thr, pc, (uptr)p, true);
-  allocator()->Deallocate(&thr->alloc_cache, p);
+  allocator()->Deallocate(&thr->proc()->alloc_cache, p);
   if (signal)
     SignalUnsafeCall(thr, pc);
 }
@@ -136,7 +188,7 @@ void OnUserAlloc(ThreadState *thr, uptr pc, uptr p, uptr sz, bool write) {
 
 void OnUserFree(ThreadState *thr, uptr pc, uptr p, bool write) {
   CHECK_NE(p, (void*)0);
-  uptr sz = ctx->metamap.FreeBlock(thr, pc, p);
+  uptr sz = ctx->metamap.FreeBlock(thr->proc(), p);
   DPrintf("#%d: free(%p, %zu)\n", thr->tid, p, sz);
   if (write && thr->ignore_reads_and_writes == 0)
     MemoryRangeFreed(thr, pc, (uptr)p, sz);
@@ -164,7 +216,11 @@ uptr user_alloc_usable_size(const void *p) {
   if (p == 0)
     return 0;
   MBlock *b = ctx->metamap.GetBlock((uptr)p);
-  return b ? b->siz : 0;
+  if (!b)
+    return 0;  // Not a valid pointer.
+  if (b->siz == 0)
+    return 1;  // Zero-sized allocations are actually 1 byte.
+  return b->siz;
 }
 
 void invoke_malloc_hook(void *ptr, uptr size) {
@@ -172,6 +228,7 @@ void invoke_malloc_hook(void *ptr, uptr size) {
   if (ctx == 0 || !ctx->initialized || thr->ignore_interceptors)
     return;
   __sanitizer_malloc_hook(ptr, size);
+  RunMallocHooks(ptr, size);
 }
 
 void invoke_free_hook(void *ptr) {
@@ -179,6 +236,7 @@ void invoke_free_hook(void *ptr) {
   if (ctx == 0 || !ctx->initialized || thr->ignore_interceptors)
     return;
   __sanitizer_free_hook(ptr);
+  RunFreeHooks(ptr);
 }
 
 void *internal_alloc(MBlockType typ, uptr sz) {
@@ -187,7 +245,7 @@ void *internal_alloc(MBlockType typ, uptr sz) {
     thr->nomalloc = 0;  // CHECK calls internal_malloc().
     CHECK(0);
   }
-  return InternalAlloc(sz, &thr->internal_alloc_cache);
+  return InternalAlloc(sz, &thr->proc()->internal_alloc_cache);
 }
 
 void internal_free(void *p) {
@@ -196,7 +254,7 @@ void internal_free(void *p) {
     thr->nomalloc = 0;  // CHECK calls internal_malloc().
     CHECK(0);
   }
-  InternalFree(p, &thr->internal_alloc_cache);
+  InternalFree(p, &thr->proc()->internal_alloc_cache);
 }
 
 }  // namespace __tsan
@@ -238,8 +296,8 @@ uptr __sanitizer_get_allocated_size(const void *p) {
 
 void __tsan_on_thread_idle() {
   ThreadState *thr = cur_thread();
-  allocator()->SwallowCache(&thr->alloc_cache);
-  internal_allocator()->SwallowCache(&thr->internal_alloc_cache);
-  ctx->metamap.OnThreadIdle(thr);
+  allocator()->SwallowCache(&thr->proc()->alloc_cache);
+  internal_allocator()->SwallowCache(&thr->proc()->internal_alloc_cache);
+  ctx->metamap.OnProcIdle(thr->proc());
 }
 }  // extern "C"
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_mman.h b/contrib/compiler-rt/lib/tsan/rtl/tsan_mman.h
index b419b58ca457..8cdeeb35aa6b 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_mman.h
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_mman.h
@@ -20,9 +20,10 @@ namespace __tsan {
 const uptr kDefaultAlignment = 16;
 
 void InitializeAllocator();
+void InitializeAllocatorLate();
 void ReplaceSystemMalloc();
-void AllocatorThreadStart(ThreadState *thr);
-void AllocatorThreadFinish(ThreadState *thr);
+void AllocatorProcStart(Processor *proc);
+void AllocatorProcFinish(Processor *proc);
 void AllocatorPrintStats();
 
 // For user allocations.
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_mutex.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_mutex.cc
index 9dd24803b183..22afefce3384 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_mutex.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_mutex.cc
@@ -43,6 +43,7 @@ static MutexType CanLockTab[MutexTypeCount][MutexTypeCount] = {
   /*11 MutexTypeDDetector*/   {},
   /*12 MutexTypeFired*/       {MutexTypeLeaf},
   /*13 MutexTypeRacy*/        {MutexTypeLeaf},
+  /*14 MutexTypeGlobalProc*/  {},
 };
 
 static bool CanLockAdj[MutexTypeCount][MutexTypeCount];
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_mutex.h b/contrib/compiler-rt/lib/tsan/rtl/tsan_mutex.h
index 27f55385c959..22ee2f33fcc7 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_mutex.h
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_mutex.h
@@ -34,6 +34,7 @@ enum MutexType {
   MutexTypeDDetector,
   MutexTypeFired,
   MutexTypeRacy,
+  MutexTypeGlobalProc,
 
   // This must be the last.
   MutexTypeCount
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_new_delete.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_new_delete.cc
index ebb422cf2023..b6478bb08c57 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_new_delete.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_new_delete.cc
@@ -23,14 +23,10 @@ struct nothrow_t {};
 
 DECLARE_REAL(void *, malloc, uptr size)
 DECLARE_REAL(void, free, void *ptr)
-#if SANITIZER_MAC || SANITIZER_ANDROID
-#define __libc_malloc REAL(malloc)
-#define __libc_free REAL(free)
-#endif
 
 #define OPERATOR_NEW_BODY(mangled_name) \
   if (cur_thread()->in_symbolizer) \
-    return __libc_malloc(size); \
+    return InternalAlloc(size); \
   void *p = 0; \
   {  \
     SCOPED_INTERCEPTOR_RAW(mangled_name, size); \
@@ -66,7 +62,7 @@ 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); \
+    return InternalFree(ptr); \
   invoke_free_hook(ptr);  \
   SCOPED_INTERCEPTOR_RAW(mangled_name, ptr);  \
   user_free(thr, pc, ptr);
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_platform.h b/contrib/compiler-rt/lib/tsan/rtl/tsan_platform.h
index c2b487155300..2bd6637d0ea3 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_platform.h
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_platform.h
@@ -297,7 +297,7 @@ struct Mapping {
   static const uptr kShadowEnd     = 0x050000000000ull;
   static const uptr kAppMemBeg     = 0x000000001000ull;
   static const uptr kAppMemEnd     = 0x00e000000000ull;
-}
+};
 
 #else
 # error "Unknown platform"
@@ -587,7 +587,11 @@ uptr MemToShadowImpl(uptr x) {
   return (((x) & ~(Mapping::kAppMemMsk | (kShadowCell - 1)))
       ^ Mapping::kAppMemXor) * kShadowCnt;
 #else
+# ifndef SANITIZER_WINDOWS
   return ((x & ~(kShadowCell - 1)) * kShadowCnt) | Mapping::kShadowBeg;
+# else
+  return ((x & ~(kShadowCell - 1)) * kShadowCnt) + Mapping::kShadowBeg;
+# endif
 #endif
 }
 
@@ -662,7 +666,6 @@ uptr ShadowToMemImpl(uptr s) {
 # ifndef SANITIZER_WINDOWS
   return (s & ~Mapping::kShadowBeg) / kShadowCnt;
 # else
-  // FIXME(dvyukov): this is most likely wrong as the mapping is not bijection.
   return (s - Mapping::kShadowBeg) / kShadowCnt;
 # endif // SANITIZER_WINDOWS
 #endif
@@ -754,10 +757,6 @@ void CheckAndProtect();
 void InitializeShadowMemoryPlatform();
 void FlushShadowMemory();
 void WriteMemoryProfile(char *buf, uptr buf_size, uptr nthread, uptr nlive);
-
-// Says whether the addr relates to a global var.
-// Guesses with high probability, may yield both false positives and negatives.
-bool IsGlobalVar(uptr addr);
 int ExtractResolvFDs(void *state, int *fds, int nfd);
 int ExtractRecvmsgFDs(void *msg, int *fds, int nfd);
 
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_platform_linux.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_platform_linux.cc
index 6602561186ce..d7182fdc1a4d 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_platform_linux.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_platform_linux.cc
@@ -18,6 +18,8 @@
 
 #include "sanitizer_common/sanitizer_common.h"
 #include "sanitizer_common/sanitizer_libc.h"
+#include "sanitizer_common/sanitizer_linux.h"
+#include "sanitizer_common/sanitizer_platform_limits_posix.h"
 #include "sanitizer_common/sanitizer_posix.h"
 #include "sanitizer_common/sanitizer_procmaps.h"
 #include "sanitizer_common/sanitizer_stoptheworld.h"
@@ -34,6 +36,9 @@
 #include <string.h>
 #include <stdarg.h>
 #include <sys/mman.h>
+#if SANITIZER_LINUX
+#include <sys/personality.h>
+#endif
 #include <sys/syscall.h>
 #include <sys/socket.h>
 #include <sys/time.h>
@@ -64,9 +69,6 @@ void *__libc_stack_end = 0;
 
 namespace __tsan {
 
-static uptr g_data_start;
-static uptr g_data_end;
-
 #ifdef TSAN_RUNTIME_VMA
 // Runtime detected VMA size.
 uptr vmaSize;
@@ -199,46 +201,6 @@ void InitializeShadowMemoryPlatform() {
   MapRodata();
 }
 
-static void InitDataSeg() {
-  MemoryMappingLayout proc_maps(true);
-  uptr start, end, offset;
-  char name[128];
-#if SANITIZER_FREEBSD
-  // On FreeBSD BSS is usually the last block allocated within the
-  // low range and heap is the last block allocated within the range
-  // 0x800000000-0x8ffffffff.
-  while (proc_maps.Next(&start, &end, &offset, name, ARRAY_SIZE(name),
-                        /*protection*/ 0)) {
-    DPrintf("%p-%p %p %s\n", start, end, offset, name);
-    if ((start & 0xffff00000000ULL) == 0 && (end & 0xffff00000000ULL) == 0 &&
-        name[0] == '\0') {
-      g_data_start = start;
-      g_data_end = end;
-    }
-  }
-#else
-  bool prev_is_data = false;
-  while (proc_maps.Next(&start, &end, &offset, name, ARRAY_SIZE(name),
-                        /*protection*/ 0)) {
-    DPrintf("%p-%p %p %s\n", start, end, offset, name);
-    bool is_data = offset != 0 && name[0] != 0;
-    // BSS may get merged with [heap] in /proc/self/maps. This is not very
-    // reliable.
-    bool is_bss = offset == 0 &&
-      (name[0] == 0 || internal_strcmp(name, "[heap]") == 0) && prev_is_data;
-    if (g_data_start == 0 && is_data)
-      g_data_start = start;
-    if (is_bss)
-      g_data_end = end;
-    prev_is_data = is_data;
-  }
-#endif
-  DPrintf("guessed data_start=%p data_end=%p\n",  g_data_start, g_data_end);
-  CHECK_LT(g_data_start, g_data_end);
-  CHECK_GE((uptr)&g_data_start, g_data_start);
-  CHECK_LT((uptr)&g_data_start, g_data_end);
-}
-
 #endif  // #ifndef SANITIZER_GO
 
 void InitializePlatformEarly() {
@@ -289,6 +251,20 @@ void InitializePlatform() {
       SetAddressSpaceUnlimited();
       reexec = true;
     }
+#if SANITIZER_LINUX && defined(__aarch64__)
+    // After patch "arm64: mm: support ARCH_MMAP_RND_BITS." is introduced in
+    // linux kernel, the random gap between stack and mapped area is increased
+    // from 128M to 36G on 39-bit aarch64. As it is almost impossible to cover
+    // this big range, we should disable randomized virtual space on aarch64.
+    int old_personality = personality(0xffffffff);
+    if (old_personality != -1 && (old_personality & ADDR_NO_RANDOMIZE) == 0) {
+      VReport(1, "WARNING: Program is run with randomized virtual address "
+              "space, which wouldn't work with ThreadSanitizer.\n"
+              "Re-execing with fixed virtual address space.\n");
+      CHECK_NE(personality(old_personality | ADDR_NO_RANDOMIZE), -1);
+      reexec = true;
+    }
+#endif
     if (reexec)
       ReExec();
   }
@@ -296,14 +272,9 @@ void InitializePlatform() {
 #ifndef SANITIZER_GO
   CheckAndProtect();
   InitTlsSize();
-  InitDataSeg();
 #endif
 }
 
-bool IsGlobalVar(uptr addr) {
-  return g_data_start && addr >= g_data_start && addr < g_data_end;
-}
-
 #ifndef SANITIZER_GO
 // Extract file descriptors passed to glibc internal __res_iclose function.
 // This is required to properly "close" the fds, because we do not see internal
@@ -361,6 +332,69 @@ int call_pthread_cancel_with_cleanup(int(*fn)(void *c, void *m,
 void ReplaceSystemMalloc() { }
 #endif
 
+#ifndef SANITIZER_GO
+#if SANITIZER_ANDROID
+
+#if defined(__aarch64__)
+# define __get_tls() \
+    ({ void** __val; __asm__("mrs %0, tpidr_el0" : "=r"(__val)); __val; })
+#elif defined(__x86_64__)
+# define __get_tls() \
+    ({ void** __val; __asm__("mov %%fs:0, %0" : "=r"(__val)); __val; })
+#else
+#error unsupported architecture
+#endif
+
+// On Android, __thread is not supported. So we store the pointer to ThreadState
+// in TLS_SLOT_TSAN, which is the tls slot allocated by Android bionic for tsan.
+static const int TLS_SLOT_TSAN = 8;
+// On Android, one thread can call intercepted functions after
+// DestroyThreadState(), so add a fake thread state for "dead" threads.
+static ThreadState *dead_thread_state = nullptr;
+
+ThreadState *cur_thread() {
+  ThreadState* thr = (ThreadState*)__get_tls()[TLS_SLOT_TSAN];
+  if (thr == nullptr) {
+    __sanitizer_sigset_t emptyset;
+    internal_sigfillset(&emptyset);
+    __sanitizer_sigset_t oldset;
+    CHECK_EQ(0, internal_sigprocmask(SIG_SETMASK, &emptyset, &oldset));
+    thr = reinterpret_cast<ThreadState*>(__get_tls()[TLS_SLOT_TSAN]);
+    if (thr == nullptr) {
+      thr = reinterpret_cast<ThreadState*>(MmapOrDie(sizeof(ThreadState),
+                                                     "ThreadState"));
+      __get_tls()[TLS_SLOT_TSAN] = thr;
+      if (dead_thread_state == nullptr) {
+        dead_thread_state = reinterpret_cast<ThreadState*>(
+            MmapOrDie(sizeof(ThreadState), "ThreadState"));
+        dead_thread_state->fast_state.SetIgnoreBit();
+        dead_thread_state->ignore_interceptors = 1;
+        dead_thread_state->is_dead = true;
+        *const_cast<int*>(&dead_thread_state->tid) = -1;
+        CHECK_EQ(0, internal_mprotect(dead_thread_state, sizeof(ThreadState),
+                                      PROT_READ));
+      }
+    }
+    CHECK_EQ(0, internal_sigprocmask(SIG_SETMASK, &oldset, nullptr));
+  }
+  return thr;
+}
+
+void cur_thread_finalize() {
+  __sanitizer_sigset_t emptyset;
+  internal_sigfillset(&emptyset);
+  __sanitizer_sigset_t oldset;
+  CHECK_EQ(0, internal_sigprocmask(SIG_SETMASK, &emptyset, &oldset));
+  ThreadState* thr = (ThreadState*)__get_tls()[TLS_SLOT_TSAN];
+  if (thr != dead_thread_state) {
+    __get_tls()[TLS_SLOT_TSAN] = dead_thread_state;
+    UnmapOrDie(thr, sizeof(ThreadState));
+  }
+  CHECK_EQ(0, internal_sigprocmask(SIG_SETMASK, &oldset, nullptr));
+}
+#endif  // SANITIZER_ANDROID
+#endif  // ifndef SANITIZER_GO
+
 }  // namespace __tsan
 
 #endif  // SANITIZER_LINUX || SANITIZER_FREEBSD
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_platform_mac.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_platform_mac.cc
index 31caf37dee5a..0cc02ab87a3e 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_platform_mac.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_platform_mac.cc
@@ -67,20 +67,18 @@ static void *SignalSafeGetOrAllocate(uptr *dst, uptr size) {
 // when TLVs are not accessible (early process startup, thread cleanup, ...).
 // The following provides a "poor man's TLV" implementation, where we use the
 // shadow memory of the pointer returned by pthread_self() to store a pointer to
-// the ThreadState object. The main thread's ThreadState pointer is stored
-// separately in a static variable, because we need to access it even before the
+// the ThreadState object. The main thread's ThreadState is stored separately
+// in a static variable, because we need to access it even before the
 // shadow memory is set up.
 static uptr main_thread_identity = 0;
-static ThreadState *main_thread_state = nullptr;
+ALIGNED(64) static char main_thread_state[sizeof(ThreadState)];
 
 ThreadState *cur_thread() {
-  ThreadState **fake_tls;
   uptr thread_identity = (uptr)pthread_self();
   if (thread_identity == main_thread_identity || main_thread_identity == 0) {
-    fake_tls = &main_thread_state;
-  } else {
-    fake_tls = (ThreadState **)MemToShadow(thread_identity);
+    return (ThreadState *)&main_thread_state;
   }
+  ThreadState **fake_tls = (ThreadState **)MemToShadow(thread_identity);
   ThreadState *thr = (ThreadState *)SignalSafeGetOrAllocate(
       (uptr *)fake_tls, sizeof(ThreadState));
   return thr;
@@ -91,7 +89,11 @@ ThreadState *cur_thread() {
 // handler will try to access the unmapped ThreadState.
 void cur_thread_finalize() {
   uptr thread_identity = (uptr)pthread_self();
-  CHECK_NE(thread_identity, main_thread_identity);
+  if (thread_identity == main_thread_identity) {
+    // Calling dispatch_main() or xpc_main() actually invokes pthread_exit to
+    // exit the main thread. Let's keep the main thread's ThreadState.
+    return;
+  }
   ThreadState **fake_tls = (ThreadState **)MemToShadow(thread_identity);
   internal_munmap(*fake_tls, sizeof(ThreadState));
   *fake_tls = nullptr;
@@ -131,10 +133,12 @@ static void my_pthread_introspection_hook(unsigned int event, pthread_t thread,
   if (event == PTHREAD_INTROSPECTION_THREAD_CREATE) {
     if (thread == pthread_self()) {
       // The current thread is a newly created GCD worker thread.
+      ThreadState *thr = cur_thread();
+      Processor *proc = ProcCreate();
+      ProcWire(proc, thr);
       ThreadState *parent_thread_state = nullptr;  // No parent.
       int tid = ThreadCreate(parent_thread_state, 0, (uptr)thread, true);
       CHECK_NE(tid, 0);
-      ThreadState *thr = cur_thread();
       ThreadStart(thr, tid, GetTid());
     }
   } else if (event == PTHREAD_INTROSPECTION_THREAD_TERMINATE) {
@@ -183,10 +187,6 @@ int call_pthread_cancel_with_cleanup(int(*fn)(void *c, void *m,
 }
 #endif
 
-bool IsGlobalVar(uptr addr) {
-  return false;
-}
-
 }  // namespace __tsan
 
 #endif  // SANITIZER_MAC
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_platform_posix.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_platform_posix.cc
index 90476cbc5fd5..805ce1be4bde 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_platform_posix.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_platform_posix.cc
@@ -105,7 +105,7 @@ static void ProtectRange(uptr beg, uptr end) {
   CHECK_LE(beg, end);
   if (beg == end)
     return;
-  if (beg != (uptr)MmapNoAccess(beg, end - beg)) {
+  if (beg != (uptr)MmapFixedNoAccess(beg, end - beg)) {
     Printf("FATAL: ThreadSanitizer can not protect [%zx,%zx]\n", beg, end);
     Printf("FATAL: Make sure you are not using unlimited stack\n");
     Die();
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_preinit.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_preinit.cc
new file mode 100644
index 000000000000..a96618d7dc81
--- /dev/null
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_preinit.cc
@@ -0,0 +1,27 @@
+//===-- tsan_preinit.cc ---------------------------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of ThreadSanitizer.
+//
+// Call __tsan_init at the very early stage of process startup.
+//===----------------------------------------------------------------------===//
+
+#include "sanitizer_common/sanitizer_internal_defs.h"
+#include "tsan_interface.h"
+
+#if SANITIZER_CAN_USE_PREINIT_ARRAY
+
+// The symbol is called __local_tsan_preinit, because it's not intended to be
+// exported.
+// This code linked into the main executable when -fsanitize=thread is in
+// the link flags. It can only use exported interface functions.
+__attribute__((section(".preinit_array"), used))
+void (*__local_tsan_preinit)(void) = __tsan_init;
+
+#endif
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_report.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_report.cc
index c1d2fd07c0d9..93604946ac38 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_report.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_report.cc
@@ -96,6 +96,8 @@ static const char *ReportTypeString(ReportType typ) {
     return "destroy of a locked mutex";
   if (typ == ReportTypeMutexDoubleLock)
     return "double lock of a mutex";
+  if (typ == ReportTypeMutexInvalidAccess)
+    return "use of an invalid mutex (e.g. uninitialized or destroyed)";
   if (typ == ReportTypeMutexBadUnlock)
     return "unlock of an unlocked mutex (or by a wrong thread)";
   if (typ == ReportTypeMutexBadReadLock)
@@ -234,7 +236,7 @@ static void PrintThread(const ReportThread *rt) {
     Printf(" '%s'", rt->name);
   char thrbuf[kThreadBufSize];
   Printf(" (tid=%zu, %s) created by %s",
-    rt->pid, rt->running ? "running" : "finished",
+    rt->os_id, rt->running ? "running" : "finished",
     thread_name(thrbuf, rt->parent_tid));
   if (rt->stack)
     Printf(" at:");
@@ -379,9 +381,9 @@ void PrintStack(const ReportStack *ent) {
 
 static void PrintMop(const ReportMop *mop, bool first) {
   Printf("\n");
-  Printf("%s by ",
+  Printf("%s at %p by ",
       (first ? (mop->write ? "Write" : "Read")
-             : (mop->write ? "Previous write" : "Previous read")));
+             : (mop->write ? "Previous write" : "Previous read")), mop->addr);
   if (mop->tid == kMainThreadId)
     Printf("main goroutine:\n");
   else
@@ -389,6 +391,31 @@ static void PrintMop(const ReportMop *mop, bool first) {
   PrintStack(mop->stack);
 }
 
+static void PrintLocation(const ReportLocation *loc) {
+  switch (loc->type) {
+  case ReportLocationHeap: {
+    Printf("\n");
+    Printf("Heap block of size %zu at %p allocated by ",
+        loc->heap_chunk_size, loc->heap_chunk_start);
+    if (loc->tid == kMainThreadId)
+      Printf("main goroutine:\n");
+    else
+      Printf("goroutine %d:\n", loc->tid);
+    PrintStack(loc->stack);
+    break;
+  }
+  case ReportLocationGlobal: {
+    Printf("\n");
+    Printf("Global var %s of size %zu at %p declared at %s:%zu\n",
+        loc->global.name, loc->global.size, loc->global.start,
+        loc->global.file, loc->global.line);
+    break;
+  }
+  default:
+    break;
+  }
+}
+
 static void PrintThread(const ReportThread *rt) {
   if (rt->id == kMainThreadId)
     return;
@@ -404,6 +431,8 @@ void PrintReport(const ReportDesc *rep) {
     Printf("WARNING: DATA RACE");
     for (uptr i = 0; i < rep->mops.Size(); i++)
       PrintMop(rep->mops[i], i == 0);
+    for (uptr i = 0; i < rep->locs.Size(); i++)
+      PrintLocation(rep->locs[i]);
     for (uptr i = 0; i < rep->threads.Size(); i++)
       PrintThread(rep->threads[i]);
   } else if (rep->typ == ReportTypeDeadlock) {
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_report.h b/contrib/compiler-rt/lib/tsan/rtl/tsan_report.h
index 3e344a048e43..d0b9d7458bf8 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_report.h
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_report.h
@@ -27,6 +27,7 @@ enum ReportType {
   ReportTypeThreadLeak,
   ReportTypeMutexDestroyLocked,
   ReportTypeMutexDoubleLock,
+  ReportTypeMutexInvalidAccess,
   ReportTypeMutexBadUnlock,
   ReportTypeMutexBadReadLock,
   ReportTypeMutexBadReadUnlock,
@@ -86,7 +87,7 @@ struct ReportLocation {
 
 struct ReportThread {
   int id;
-  uptr pid;
+  uptr os_id;
   bool running;
   char *name;
   int parent_tid;
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl.cc
index 4df4db557a24..629871ef8f7a 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl.cc
@@ -321,6 +321,7 @@ void Initialize(ThreadState *thr) {
   const char *options = GetEnv(kTsanOptionsEnv);
   CacheBinaryName();
   InitializeFlags(&ctx->flags, options);
+  AvoidCVE_2016_2143();
   InitializePlatformEarly();
 #ifndef SANITIZER_GO
   // Re-exec ourselves if we need to set additional env or command line args.
@@ -329,6 +330,10 @@ void Initialize(ThreadState *thr) {
   InitializeAllocator();
   ReplaceSystemMalloc();
 #endif
+  if (common_flags()->detect_deadlocks)
+    ctx->dd = DDetector::Create(flags());
+  Processor *proc = ProcCreate();
+  ProcWire(proc, thr);
   InitializeInterceptors();
   CheckShadowMapping();
   InitializePlatform();
@@ -336,6 +341,7 @@ void Initialize(ThreadState *thr) {
   InitializeDynamicAnnotations();
 #ifndef SANITIZER_GO
   InitializeShadowMemory();
+  InitializeAllocatorLate();
 #endif
   // Setup correct file descriptor for error reports.
   __sanitizer_set_report_path(common_flags()->log_path);
@@ -351,8 +357,6 @@ void Initialize(ThreadState *thr) {
   SetSandboxingCallback(StopBackgroundThread);
 #endif
 #endif
-  if (common_flags()->detect_deadlocks)
-    ctx->dd = DDetector::Create(flags());
 
   VPrintf(1, "***** Running under ThreadSanitizer v2 (pid %d) *****\n",
           (int)internal_getpid());
@@ -366,6 +370,10 @@ void Initialize(ThreadState *thr) {
 #endif
   ctx->initialized = true;
 
+#ifndef SANITIZER_GO
+  Symbolizer::LateInitialize();
+#endif
+
   if (flags()->stop_on_start) {
     Printf("ThreadSanitizer is suspended at startup (pid %d)."
            " Call __tsan_resume().\n",
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl.h b/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl.h
index 04104b162f98..ff69015660b6 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl.h
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl.h
@@ -325,6 +325,36 @@ struct JmpBuf {
   uptr *shadow_stack_pos;
 };
 
+// A Processor represents a physical thread, or a P for Go.
+// It is used to store internal resources like allocate cache, and does not
+// participate in race-detection logic (invisible to end user).
+// In C++ it is tied to an OS thread just like ThreadState, however ideally
+// it should be tied to a CPU (this way we will have fewer allocator caches).
+// In Go it is tied to a P, so there are significantly fewer Processor's than
+// ThreadState's (which are tied to Gs).
+// A ThreadState must be wired with a Processor to handle events.
+struct Processor {
+  ThreadState *thr; // currently wired thread, or nullptr
+#ifndef SANITIZER_GO
+  AllocatorCache alloc_cache;
+  InternalAllocatorCache internal_alloc_cache;
+#endif
+  DenseSlabAllocCache block_cache;
+  DenseSlabAllocCache sync_cache;
+  DenseSlabAllocCache clock_cache;
+  DDPhysicalThread *dd_pt;
+};
+
+#ifndef SANITIZER_GO
+// ScopedGlobalProcessor temporary setups a global processor for the current
+// thread, if it does not have one. Intended for interceptors that can run
+// at the very thread end, when we already destroyed the thread processor.
+struct ScopedGlobalProcessor {
+  ScopedGlobalProcessor();
+  ~ScopedGlobalProcessor();
+};
+#endif
+
 // This struct is stored in TLS.
 struct ThreadState {
   FastState fast_state;
@@ -360,8 +390,6 @@ struct ThreadState {
   MutexSet mset;
   ThreadClock clock;
 #ifndef SANITIZER_GO
-  AllocatorCache alloc_cache;
-  InternalAllocatorCache internal_alloc_cache;
   Vector<JmpBuf> jmp_bufs;
   int ignore_interceptors;
 #endif
@@ -385,16 +413,19 @@ struct ThreadState {
 #if SANITIZER_DEBUG && !SANITIZER_GO
   InternalDeadlockDetector internal_deadlock_detector;
 #endif
-  DDPhysicalThread *dd_pt;
   DDLogicalThread *dd_lt;
 
+  // Current wired Processor, or nullptr. Required to handle any events.
+  Processor *proc1;
+#ifndef SANITIZER_GO
+  Processor *proc() { return proc1; }
+#else
+  Processor *proc();
+#endif
+
   atomic_uintptr_t in_signal_handler;
   ThreadSignalContext *signal_ctx;
 
-  DenseSlabAllocCache block_cache;
-  DenseSlabAllocCache sync_cache;
-  DenseSlabAllocCache clock_cache;
-
 #ifndef SANITIZER_GO
   u32 last_sleep_stack_id;
   ThreadClock last_sleep_clock;
@@ -404,6 +435,8 @@ struct ThreadState {
   // If set, malloc must not be called.
   int nomalloc;
 
+  const ReportDesc *current_report;
+
   explicit ThreadState(Context *ctx, int tid, int unique_id, u64 epoch,
                        unsigned reuse_count,
                        uptr stk_addr, uptr stk_size,
@@ -411,7 +444,7 @@ struct ThreadState {
 };
 
 #ifndef SANITIZER_GO
-#if SANITIZER_MAC
+#if SANITIZER_MAC || SANITIZER_ANDROID
 ThreadState *cur_thread();
 void cur_thread_finalize();
 #else
@@ -421,7 +454,7 @@ INLINE ThreadState *cur_thread() {
   return reinterpret_cast<ThreadState *>(&cur_thread_placeholder);
 }
 INLINE void cur_thread_finalize() { }
-#endif  // SANITIZER_MAC
+#endif  // SANITIZER_MAC || SANITIZER_ANDROID
 #endif  // SANITIZER_GO
 
 class ThreadContext : public ThreadContextBase {
@@ -683,6 +716,11 @@ void ThreadSetName(ThreadState *thr, const char *name);
 int ThreadCount(ThreadState *thr);
 void ProcessPendingSignals(ThreadState *thr);
 
+Processor *ProcCreate();
+void ProcDestroy(Processor *proc);
+void ProcWire(Processor *proc, ThreadState *thr);
+void ProcUnwire(Processor *proc, ThreadState *thr);
+
 void MutexCreate(ThreadState *thr, uptr pc, uptr addr,
                  bool rw, bool recursive, bool linker_init);
 void MutexDestroy(ThreadState *thr, uptr pc, uptr addr);
@@ -693,6 +731,7 @@ void MutexReadLock(ThreadState *thr, uptr pc, uptr addr, bool try_lock = false);
 void MutexReadUnlock(ThreadState *thr, uptr pc, uptr addr);
 void MutexReadOrWriteUnlock(ThreadState *thr, uptr pc, uptr addr);
 void MutexRepair(ThreadState *thr, uptr pc, uptr addr);  // call on EOWNERDEAD
+void MutexInvalidAccess(ThreadState *thr, uptr pc, uptr addr);
 
 void Acquire(ThreadState *thr, uptr pc, uptr addr);
 // AcquireGlobal synchronizes the current thread with all other threads.
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl_mutex.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl_mutex.cc
index 62ab7aa6b2b4..1806acf063bc 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl_mutex.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl_mutex.cc
@@ -32,7 +32,7 @@ struct Callback : DDCallback {
   Callback(ThreadState *thr, uptr pc)
       : thr(thr)
       , pc(pc) {
-    DDCallback::pt = thr->dd_pt;
+    DDCallback::pt = thr->proc()->dd_pt;
     DDCallback::lt = thr->dd_lt;
   }
 
@@ -84,21 +84,14 @@ void MutexCreate(ThreadState *thr, uptr pc, uptr addr,
 void MutexDestroy(ThreadState *thr, uptr pc, uptr addr) {
   DPrintf("#%d: MutexDestroy %zx\n", thr->tid, addr);
   StatInc(thr, StatMutexDestroy);
-#ifndef SANITIZER_GO
-  // Global mutexes not marked as LINKER_INITIALIZED
-  // cause tons of not interesting reports, so just ignore it.
-  if (IsGlobalVar(addr))
-    return;
-#endif
-  if (IsAppMem(addr)) {
-    CHECK(!thr->is_freeing);
-    thr->is_freeing = true;
-    MemoryWrite(thr, pc, addr, kSizeLog1);
-    thr->is_freeing = false;
-  }
-  SyncVar *s = ctx->metamap.GetIfExistsAndLock(addr);
+  SyncVar *s = ctx->metamap.GetIfExistsAndLock(addr, true);
   if (s == 0)
     return;
+  if (s->is_linker_init) {
+    // Destroy is no-op for linker-initialized mutexes.
+    s->mtx.Unlock();
+    return;
+  }
   if (common_flags()->detect_deadlocks) {
     Callback cb(thr, pc);
     ctx->dd->MutexDestroy(&cb, &s->dd);
@@ -114,7 +107,7 @@ void MutexDestroy(ThreadState *thr, uptr pc, uptr addr) {
   u64 mid = s->GetId();
   u32 last_lock = s->last_lock;
   if (!unlock_locked)
-    s->Reset(thr);  // must not reset it before the report is printed
+    s->Reset(thr->proc());  // must not reset it before the report is printed
   s->mtx.Unlock();
   if (unlock_locked) {
     ThreadRegistryLock l(ctx->thread_registry);
@@ -128,15 +121,23 @@ void MutexDestroy(ThreadState *thr, uptr pc, uptr addr) {
     rep.AddStack(trace, true);
     rep.AddLocation(addr, 1);
     OutputReport(thr, rep);
-  }
-  if (unlock_locked) {
-    SyncVar *s = ctx->metamap.GetIfExistsAndLock(addr);
+
+    SyncVar *s = ctx->metamap.GetIfExistsAndLock(addr, true);
     if (s != 0) {
-      s->Reset(thr);
+      s->Reset(thr->proc());
       s->mtx.Unlock();
     }
   }
   thr->mset.Remove(mid);
+  // Imitate a memory write to catch unlock-destroy races.
+  // Do this outside of sync mutex, because it can report a race which locks
+  // sync mutexes.
+  if (IsAppMem(addr)) {
+    CHECK(!thr->is_freeing);
+    thr->is_freeing = true;
+    MemoryWrite(thr, pc, addr, kSizeLog1);
+    thr->is_freeing = false;
+  }
   // s will be destroyed and freed in MetaMap::FreeBlock.
 }
 
@@ -350,11 +351,21 @@ void MutexRepair(ThreadState *thr, uptr pc, uptr addr) {
   s->mtx.Unlock();
 }
 
+void MutexInvalidAccess(ThreadState *thr, uptr pc, uptr addr) {
+  DPrintf("#%d: MutexInvalidAccess %zx\n", thr->tid, addr);
+  SyncVar *s = ctx->metamap.GetOrCreateAndLock(thr, pc, addr, true);
+  u64 mid = s->GetId();
+  s->mtx.Unlock();
+  ReportMutexMisuse(thr, pc, ReportTypeMutexInvalidAccess, addr, mid);
+}
+
 void Acquire(ThreadState *thr, uptr pc, uptr addr) {
   DPrintf("#%d: Acquire %zx\n", thr->tid, addr);
   if (thr->ignore_sync)
     return;
-  SyncVar *s = ctx->metamap.GetOrCreateAndLock(thr, pc, addr, false);
+  SyncVar *s = ctx->metamap.GetIfExistsAndLock(addr, false);
+  if (!s)
+    return;
   AcquireImpl(thr, pc, &s->clock);
   s->mtx.ReadUnlock();
 }
@@ -426,7 +437,7 @@ void AcquireImpl(ThreadState *thr, uptr pc, SyncClock *c) {
   if (thr->ignore_sync)
     return;
   thr->clock.set(thr->fast_state.epoch());
-  thr->clock.acquire(&thr->clock_cache, c);
+  thr->clock.acquire(&thr->proc()->clock_cache, c);
   StatInc(thr, StatSyncAcquire);
 }
 
@@ -435,7 +446,7 @@ void ReleaseImpl(ThreadState *thr, uptr pc, SyncClock *c) {
     return;
   thr->clock.set(thr->fast_state.epoch());
   thr->fast_synch_epoch = thr->fast_state.epoch();
-  thr->clock.release(&thr->clock_cache, c);
+  thr->clock.release(&thr->proc()->clock_cache, c);
   StatInc(thr, StatSyncRelease);
 }
 
@@ -444,7 +455,7 @@ void ReleaseStoreImpl(ThreadState *thr, uptr pc, SyncClock *c) {
     return;
   thr->clock.set(thr->fast_state.epoch());
   thr->fast_synch_epoch = thr->fast_state.epoch();
-  thr->clock.ReleaseStore(&thr->clock_cache, c);
+  thr->clock.ReleaseStore(&thr->proc()->clock_cache, c);
   StatInc(thr, StatSyncRelease);
 }
 
@@ -453,7 +464,7 @@ void AcquireReleaseImpl(ThreadState *thr, uptr pc, SyncClock *c) {
     return;
   thr->clock.set(thr->fast_state.epoch());
   thr->fast_synch_epoch = thr->fast_state.epoch();
-  thr->clock.acq_rel(&thr->clock_cache, c);
+  thr->clock.acq_rel(&thr->proc()->clock_cache, c);
   StatInc(thr, StatSyncAcquire);
   StatInc(thr, StatSyncRelease);
 }
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl_proc.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl_proc.cc
new file mode 100644
index 000000000000..0c838a1388f5
--- /dev/null
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl_proc.cc
@@ -0,0 +1,61 @@
+//===-- tsan_rtl_proc.cc ------------------------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of ThreadSanitizer (TSan), a race detector.
+//
+//===----------------------------------------------------------------------===//
+
+#include "sanitizer_common/sanitizer_placement_new.h"
+#include "tsan_rtl.h"
+#include "tsan_mman.h"
+#include "tsan_flags.h"
+
+namespace __tsan {
+
+Processor *ProcCreate() {
+  void *mem = InternalAlloc(sizeof(Processor));
+  internal_memset(mem, 0, sizeof(Processor));
+  Processor *proc = new(mem) Processor;
+  proc->thr = nullptr;
+#ifndef SANITIZER_GO
+  AllocatorProcStart(proc);
+#endif
+  if (common_flags()->detect_deadlocks)
+    proc->dd_pt = ctx->dd->CreatePhysicalThread();
+  return proc;
+}
+
+void ProcDestroy(Processor *proc) {
+  CHECK_EQ(proc->thr, nullptr);
+#ifndef SANITIZER_GO
+  AllocatorProcFinish(proc);
+#endif
+  ctx->clock_alloc.FlushCache(&proc->clock_cache);
+  ctx->metamap.OnProcIdle(proc);
+  if (common_flags()->detect_deadlocks)
+     ctx->dd->DestroyPhysicalThread(proc->dd_pt);
+  proc->~Processor();
+  InternalFree(proc);
+}
+
+void ProcWire(Processor *proc, ThreadState *thr) {
+  CHECK_EQ(thr->proc1, nullptr);
+  CHECK_EQ(proc->thr, nullptr);
+  thr->proc1 = proc;
+  proc->thr = thr;
+}
+
+void ProcUnwire(Processor *proc, ThreadState *thr) {
+  CHECK_EQ(thr->proc1, proc);
+  CHECK_EQ(proc->thr, thr);
+  thr->proc1 = nullptr;
+  proc->thr = nullptr;
+}
+
+}  // namespace __tsan
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl_report.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl_report.cc
index 5aff6ca56adf..810119b93a7a 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl_report.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl_report.cc
@@ -56,6 +56,11 @@ bool OnReport(const ReportDesc *rep, bool suppressed) {
 }
 #endif
 
+SANITIZER_WEAK_DEFAULT_IMPL
+void __tsan_on_report(const ReportDesc *rep) {
+  (void)rep;
+}
+
 static void StackStripMain(SymbolizedStack *frames) {
   SymbolizedStack *last_frame = nullptr;
   SymbolizedStack *last_frame2 = nullptr;
@@ -189,7 +194,7 @@ void ScopedReport::AddThread(const ThreadContext *tctx, bool suppressable) {
   ReportThread *rt = new(mem) ReportThread;
   rep_->threads.PushBack(rt);
   rt->id = tctx->tid;
-  rt->pid = tctx->os_id;
+  rt->os_id = tctx->os_id;
   rt->running = (tctx->status == ThreadStatusRunning);
   rt->name = internal_strdup(tctx->name);
   rt->parent_tid = tctx->parent_tid;
@@ -268,7 +273,7 @@ u64 ScopedReport::AddMutex(u64 id) {
   u64 uid = 0;
   u64 mid = id;
   uptr addr = SyncVar::SplitId(id, &uid);
-  SyncVar *s = ctx->metamap.GetIfExistsAndLock(addr);
+  SyncVar *s = ctx->metamap.GetIfExistsAndLock(addr, true);
   // Check that the mutex is still alive.
   // Another mutex can be created at the same address,
   // so check uid as well.
@@ -342,12 +347,12 @@ void ScopedReport::AddLocation(uptr addr, uptr size) {
     rep_->locs.PushBack(loc);
     AddThread(tctx);
   }
+#endif
   if (ReportLocation *loc = SymbolizeData(addr)) {
     loc->suppressable = true;
     rep_->locs.PushBack(loc);
     return;
   }
-#endif
 }
 
 #ifndef SANITIZER_GO
@@ -492,6 +497,8 @@ bool OutputReport(ThreadState *thr, const ScopedReport &srep) {
     return false;
   atomic_store_relaxed(&ctx->last_symbolize_time_ns, NanoTime());
   const ReportDesc *rep = srep.GetReport();
+  CHECK_EQ(thr->current_report, nullptr);
+  thr->current_report = rep;
   Suppression *supp = 0;
   uptr pc_or_addr = 0;
   for (uptr i = 0; pc_or_addr == 0 && i < rep->mops.Size(); i++)
@@ -512,13 +519,17 @@ bool OutputReport(ThreadState *thr, const ScopedReport &srep) {
     thr->is_freeing = false;
     bool suppressed = OnReport(rep, pc_or_addr != 0);
     thr->is_freeing = old_is_freeing;
-    if (suppressed)
+    if (suppressed) {
+      thr->current_report = nullptr;
       return false;
+    }
   }
   PrintReport(rep);
+  __tsan_on_report(rep);
   ctx->nreported++;
   if (flags()->halt_on_error)
     Die();
+  thr->current_report = nullptr;
   return true;
 }
 
@@ -669,6 +680,14 @@ void PrintCurrentStack(ThreadState *thr, uptr pc) {
   PrintStack(SymbolizeStack(trace));
 }
 
+// Always inlining PrintCurrentStackSlow, because LocatePcInTrace assumes
+// __sanitizer_print_stack_trace exists in the actual unwinded stack, but
+// tail-call to PrintCurrentStackSlow breaks this assumption because
+// __sanitizer_print_stack_trace disappears after tail-call.
+// However, this solution is not reliable enough, please see dvyukov's comment
+// http://reviews.llvm.org/D19148#406208
+// Also see PR27280 comment 2 and 3 for breaking examples and analysis.
+ALWAYS_INLINE
 void PrintCurrentStackSlow(uptr pc) {
 #ifndef SANITIZER_GO
   BufferedStackTrace *ptrace =
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl_thread.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl_thread.cc
index dcae255f7643..ab8f3c38a960 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl_thread.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_rtl_thread.cc
@@ -42,7 +42,7 @@ void ThreadContext::OnDead() {
 void ThreadContext::OnJoined(void *arg) {
   ThreadState *caller_thr = static_cast<ThreadState *>(arg);
   AcquireImpl(caller_thr, 0, &sync);
-  sync.Reset(&caller_thr->clock_cache);
+  sync.Reset(&caller_thr->proc()->clock_cache);
 }
 
 struct OnCreatedArgs {
@@ -74,7 +74,7 @@ void ThreadContext::OnReset() {
 
 void ThreadContext::OnDetached(void *arg) {
   ThreadState *thr1 = static_cast<ThreadState*>(arg);
-  sync.Reset(&thr1->clock_cache);
+  sync.Reset(&thr1->proc()->clock_cache);
 }
 
 struct OnStartedArgs {
@@ -106,13 +106,8 @@ void ThreadContext::OnStarted(void *arg) {
   thr->shadow_stack_pos = thr->shadow_stack;
   thr->shadow_stack_end = thr->shadow_stack + kInitStackSize;
 #endif
-#ifndef SANITIZER_GO
-  AllocatorThreadStart(thr);
-#endif
-  if (common_flags()->detect_deadlocks) {
-    thr->dd_pt = ctx->dd->CreatePhysicalThread();
+  if (common_flags()->detect_deadlocks)
     thr->dd_lt = ctx->dd->CreateLogicalThread(unique_id);
-  }
   thr->fast_state.SetHistorySize(flags()->history_size);
   // Commit switch to the new part of the trace.
   // TraceAddEvent will reset stack0/mset0 in the new part for us.
@@ -121,7 +116,7 @@ void ThreadContext::OnStarted(void *arg) {
   thr->fast_synch_epoch = epoch0;
   AcquireImpl(thr, 0, &sync);
   StatInc(thr, StatSyncAcquire);
-  sync.Reset(&thr->clock_cache);
+  sync.Reset(&thr->proc()->clock_cache);
   thr->is_inited = true;
   DPrintf("#%d: ThreadStart epoch=%zu stk_addr=%zx stk_size=%zx "
           "tls_addr=%zx tls_size=%zx\n",
@@ -138,15 +133,8 @@ void ThreadContext::OnFinished() {
   }
   epoch1 = thr->fast_state.epoch();
 
-  if (common_flags()->detect_deadlocks) {
-    ctx->dd->DestroyPhysicalThread(thr->dd_pt);
+  if (common_flags()->detect_deadlocks)
     ctx->dd->DestroyLogicalThread(thr->dd_lt);
-  }
-  ctx->clock_alloc.FlushCache(&thr->clock_cache);
-  ctx->metamap.OnThreadIdle(thr);
-#ifndef SANITIZER_GO
-  AllocatorThreadFinish(thr);
-#endif
   thr->~ThreadState();
 #if TSAN_COLLECT_STATS
   StatAggregate(ctx->stat, thr->stat);
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_stat.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_stat.cc
index a5cca9679582..d1d6ed24d991 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_stat.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_stat.cc
@@ -168,6 +168,7 @@ void StatOutput(u64 *stat) {
   name[StatMtxFired]                     = "  FiredSuppressions               ";
   name[StatMtxRacy]                      = "  RacyStacks                      ";
   name[StatMtxFD]                        = "  FD                              ";
+  name[StatMtxGlobalProc]                = "  GlobalProc                      ";
 
   Printf("Statistics:\n");
   for (int i = 0; i < StatCnt; i++)
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_stat.h b/contrib/compiler-rt/lib/tsan/rtl/tsan_stat.h
index 8ea32048e147..8447dd84fc17 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_stat.h
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_stat.h
@@ -173,6 +173,7 @@ enum StatType {
   StatMtxFired,
   StatMtxRacy,
   StatMtxFD,
+  StatMtxGlobalProc,
 
   // This must be the last.
   StatCnt
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_suppressions.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_suppressions.cc
index b992d78f8cb5..aea3cb978cb1 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_suppressions.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_suppressions.cc
@@ -80,6 +80,8 @@ static const char *conv(ReportType typ) {
     return kSuppressionMutex;
   else if (typ == ReportTypeMutexDoubleLock)
     return kSuppressionMutex;
+  else if (typ == ReportTypeMutexInvalidAccess)
+    return kSuppressionMutex;
   else if (typ == ReportTypeMutexBadUnlock)
     return kSuppressionMutex;
   else if (typ == ReportTypeMutexBadReadLock)
@@ -92,7 +94,7 @@ static const char *conv(ReportType typ) {
     return kSuppressionNone;
   else if (typ == ReportTypeDeadlock)
     return kSuppressionDeadlock;
-  Printf("ThreadSanitizer: unknown report type %d\n", typ),
+  Printf("ThreadSanitizer: unknown report type %d\n", typ);
   Die();
 }
 
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_sync.cc b/contrib/compiler-rt/lib/tsan/rtl/tsan_sync.cc
index 4202d30a2e17..58b26802b0e7 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_sync.cc
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_sync.cc
@@ -36,7 +36,7 @@ void SyncVar::Init(ThreadState *thr, uptr pc, uptr addr, u64 uid) {
     DDMutexInit(thr, pc, this);
 }
 
-void SyncVar::Reset(ThreadState *thr) {
+void SyncVar::Reset(Processor *proc) {
   uid = 0;
   creation_stack_id = 0;
   owner_tid = kInvalidTid;
@@ -47,12 +47,12 @@ void SyncVar::Reset(ThreadState *thr) {
   is_broken = 0;
   is_linker_init = 0;
 
-  if (thr == 0) {
+  if (proc == 0) {
     CHECK_EQ(clock.size(), 0);
     CHECK_EQ(read_clock.size(), 0);
   } else {
-    clock.Reset(&thr->clock_cache);
-    read_clock.Reset(&thr->clock_cache);
+    clock.Reset(&proc->clock_cache);
+    read_clock.Reset(&proc->clock_cache);
   }
 }
 
@@ -61,7 +61,7 @@ MetaMap::MetaMap() {
 }
 
 void MetaMap::AllocBlock(ThreadState *thr, uptr pc, uptr p, uptr sz) {
-  u32 idx = block_alloc_.Alloc(&thr->block_cache);
+  u32 idx = block_alloc_.Alloc(&thr->proc()->block_cache);
   MBlock *b = block_alloc_.Map(idx);
   b->siz = sz;
   b->tid = thr->tid;
@@ -71,16 +71,16 @@ void MetaMap::AllocBlock(ThreadState *thr, uptr pc, uptr p, uptr sz) {
   *meta = idx | kFlagBlock;
 }
 
-uptr MetaMap::FreeBlock(ThreadState *thr, uptr pc, uptr p) {
+uptr MetaMap::FreeBlock(Processor *proc, uptr p) {
   MBlock* b = GetBlock(p);
   if (b == 0)
     return 0;
   uptr sz = RoundUpTo(b->siz, kMetaShadowCell);
-  FreeRange(thr, pc, p, sz);
+  FreeRange(proc, p, sz);
   return sz;
 }
 
-bool MetaMap::FreeRange(ThreadState *thr, uptr pc, uptr p, uptr sz) {
+bool MetaMap::FreeRange(Processor *proc, uptr p, uptr sz) {
   bool has_something = false;
   u32 *meta = MemToMeta(p);
   u32 *end = MemToMeta(p + sz);
@@ -96,14 +96,14 @@ bool MetaMap::FreeRange(ThreadState *thr, uptr pc, uptr p, uptr sz) {
     has_something = true;
     while (idx != 0) {
       if (idx & kFlagBlock) {
-        block_alloc_.Free(&thr->block_cache, idx & ~kFlagMask);
+        block_alloc_.Free(&proc->block_cache, idx & ~kFlagMask);
         break;
       } else if (idx & kFlagSync) {
         DCHECK(idx & kFlagSync);
         SyncVar *s = sync_alloc_.Map(idx & ~kFlagMask);
         u32 next = s->next;
-        s->Reset(thr);
-        sync_alloc_.Free(&thr->sync_cache, idx & ~kFlagMask);
+        s->Reset(proc);
+        sync_alloc_.Free(&proc->sync_cache, idx & ~kFlagMask);
         idx = next;
       } else {
         CHECK(0);
@@ -119,24 +119,30 @@ bool MetaMap::FreeRange(ThreadState *thr, uptr pc, uptr p, uptr sz) {
 // which can be huge. The function probes pages one-by-one until it finds a page
 // without meta objects, at this point it stops freeing meta objects. Because
 // thread stacks grow top-down, we do the same starting from end as well.
-void MetaMap::ResetRange(ThreadState *thr, uptr pc, uptr p, uptr sz) {
+void MetaMap::ResetRange(Processor *proc, uptr p, uptr sz) {
+  if (kGoMode) {
+    // UnmapOrDie/MmapFixedNoReserve does not work on Windows,
+    // so we do the optimization only for C/C++.
+    FreeRange(proc, p, sz);
+    return;
+  }
   const uptr kMetaRatio = kMetaShadowCell / kMetaShadowSize;
   const uptr kPageSize = GetPageSizeCached() * kMetaRatio;
   if (sz <= 4 * kPageSize) {
     // If the range is small, just do the normal free procedure.
-    FreeRange(thr, pc, p, sz);
+    FreeRange(proc, p, sz);
     return;
   }
   // First, round both ends of the range to page size.
   uptr diff = RoundUp(p, kPageSize) - p;
   if (diff != 0) {
-    FreeRange(thr, pc, p, diff);
+    FreeRange(proc, p, diff);
     p += diff;
     sz -= diff;
   }
   diff = p + sz - RoundDown(p + sz, kPageSize);
   if (diff != 0) {
-    FreeRange(thr, pc, p + sz - diff, diff);
+    FreeRange(proc, p + sz - diff, diff);
     sz -= diff;
   }
   // Now we must have a non-empty page-aligned range.
@@ -146,18 +152,21 @@ void MetaMap::ResetRange(ThreadState *thr, uptr pc, uptr p, uptr sz) {
   const uptr p0 = p;
   const uptr sz0 = sz;
   // Probe start of the range.
-  while (sz > 0) {
-    bool has_something = FreeRange(thr, pc, p, kPageSize);
+  for (uptr checked = 0; sz > 0; checked += kPageSize) {
+    bool has_something = FreeRange(proc, p, kPageSize);
     p += kPageSize;
     sz -= kPageSize;
-    if (!has_something)
+    if (!has_something && checked > (128 << 10))
       break;
   }
   // Probe end of the range.
-  while (sz > 0) {
-    bool has_something = FreeRange(thr, pc, p - kPageSize, kPageSize);
+  for (uptr checked = 0; sz > 0; checked += kPageSize) {
+    bool has_something = FreeRange(proc, p + sz - kPageSize, kPageSize);
     sz -= kPageSize;
-    if (!has_something)
+    // Stacks grow down, so sync object are most likely at the end of the region
+    // (if it is a stack). The very end of the stack is TLS and tsan increases
+    // TLS by at least 256K, so check at least 512K.
+    if (!has_something && checked > (512 << 10))
       break;
   }
   // Finally, page out the whole range (including the parts that we've just
@@ -189,8 +198,8 @@ SyncVar* MetaMap::GetOrCreateAndLock(ThreadState *thr, uptr pc,
   return GetAndLock(thr, pc, addr, write_lock, true);
 }
 
-SyncVar* MetaMap::GetIfExistsAndLock(uptr addr) {
-  return GetAndLock(0, 0, addr, true, false);
+SyncVar* MetaMap::GetIfExistsAndLock(uptr addr, bool write_lock) {
+  return GetAndLock(0, 0, addr, write_lock, false);
 }
 
 SyncVar* MetaMap::GetAndLock(ThreadState *thr, uptr pc,
@@ -210,8 +219,8 @@ SyncVar* MetaMap::GetAndLock(ThreadState *thr, uptr pc,
       SyncVar * s = sync_alloc_.Map(idx & ~kFlagMask);
       if (s->addr == addr) {
         if (myidx != 0) {
-          mys->Reset(thr);
-          sync_alloc_.Free(&thr->sync_cache, myidx);
+          mys->Reset(thr->proc());
+          sync_alloc_.Free(&thr->proc()->sync_cache, myidx);
         }
         if (write_lock)
           s->mtx.Lock();
@@ -230,7 +239,7 @@ SyncVar* MetaMap::GetAndLock(ThreadState *thr, uptr pc,
 
     if (myidx == 0) {
       const u64 uid = atomic_fetch_add(&uid_gen_, 1, memory_order_relaxed);
-      myidx = sync_alloc_.Alloc(&thr->sync_cache);
+      myidx = sync_alloc_.Alloc(&thr->proc()->sync_cache);
       mys = sync_alloc_.Map(myidx);
       mys->Init(thr, pc, addr, uid);
     }
@@ -279,9 +288,9 @@ void MetaMap::MoveMemory(uptr src, uptr dst, uptr sz) {
   }
 }
 
-void MetaMap::OnThreadIdle(ThreadState *thr) {
-  block_alloc_.FlushCache(&thr->block_cache);
-  sync_alloc_.FlushCache(&thr->sync_cache);
+void MetaMap::OnProcIdle(Processor *proc) {
+  block_alloc_.FlushCache(&proc->block_cache);
+  sync_alloc_.FlushCache(&proc->sync_cache);
 }
 
 }  // namespace __tsan
diff --git a/contrib/compiler-rt/lib/tsan/rtl/tsan_sync.h b/contrib/compiler-rt/lib/tsan/rtl/tsan_sync.h
index f07ea3b9776b..2bc2f41fbe26 100644
--- a/contrib/compiler-rt/lib/tsan/rtl/tsan_sync.h
+++ b/contrib/compiler-rt/lib/tsan/rtl/tsan_sync.h
@@ -47,7 +47,7 @@ struct SyncVar {
   SyncClock clock;
 
   void Init(ThreadState *thr, uptr pc, uptr addr, u64 uid);
-  void Reset(ThreadState *thr);
+  void Reset(Processor *proc);
 
   u64 GetId() const {
     // 47 lsb is addr, then 14 bits is low part of uid, then 3 zero bits.
@@ -72,18 +72,18 @@ class MetaMap {
   MetaMap();
 
   void AllocBlock(ThreadState *thr, uptr pc, uptr p, uptr sz);
-  uptr FreeBlock(ThreadState *thr, uptr pc, uptr p);
-  bool FreeRange(ThreadState *thr, uptr pc, uptr p, uptr sz);
-  void ResetRange(ThreadState *thr, uptr pc, uptr p, uptr sz);
+  uptr FreeBlock(Processor *proc, uptr p);
+  bool FreeRange(Processor *proc, uptr p, uptr sz);
+  void ResetRange(Processor *proc, uptr p, uptr sz);
   MBlock* GetBlock(uptr p);
 
   SyncVar* GetOrCreateAndLock(ThreadState *thr, uptr pc,
                               uptr addr, bool write_lock);
-  SyncVar* GetIfExistsAndLock(uptr addr);
+  SyncVar* GetIfExistsAndLock(uptr addr, bool write_lock);
 
   void MoveMemory(uptr src, uptr dst, uptr sz);
 
-  void OnThreadIdle(ThreadState *thr);
+  void OnProcIdle(Processor *proc);
 
  private:
   static const u32 kFlagMask  = 3u << 30;