Vendor import of compiler-rt trunk r307894:vendor/compiler-rt/compiler-rt-trunk-r307894

https://llvm.org/svn/llvm-project/compiler-rt/trunk@307894

21 files changed, 311 insertions, 162 deletions

diff --git a/lib/tsan/rtl/tsan_clock.cc b/lib/tsan/rtl/tsan_clock.cc
index 32435adfdf33..9ee9104283f8 100644
--- a/lib/tsan/rtl/tsan_clock.cc
+++ b/lib/tsan/rtl/tsan_clock.cc
@@ -101,6 +101,9 @@ ThreadClock::ThreadClock(unsigned tid, unsigned reused)
   clk_[tid_].reused = reused_;
 }
 
+void ThreadClock::ResetCached(ClockCache *c) {
+}
+
 void ThreadClock::acquire(ClockCache *c, const SyncClock *src) {
   DCHECK_LE(nclk_, kMaxTid);
   DCHECK_LE(src->size_, kMaxTid);
@@ -116,9 +119,7 @@ void ThreadClock::acquire(ClockCache *c, const SyncClock *src) {
   // Check if we've already acquired src after the last release operation on src
   bool acquired = false;
   if (nclk > tid_) {
-    CPP_STAT_INC(StatClockAcquireLarge);
     if (src->elem(tid_).reused == reused_) {
-      CPP_STAT_INC(StatClockAcquireRepeat);
       for (unsigned i = 0; i < kDirtyTids; i++) {
         unsigned tid = src->dirty_tids_[i];
         if (tid != kInvalidTid) {
@@ -266,11 +267,11 @@ void ThreadClock::UpdateCurrentThread(SyncClock *dst) const {
 
   for (unsigned i = 0; i < kDirtyTids; i++) {
     if (dst->dirty_tids_[i] == tid_) {
-      CPP_STAT_INC(StatClockReleaseFast1);
+      CPP_STAT_INC(StatClockReleaseFast);
       return;
     }
     if (dst->dirty_tids_[i] == kInvalidTid) {
-      CPP_STAT_INC(StatClockReleaseFast2);
+      CPP_STAT_INC(StatClockReleaseFast);
       dst->dirty_tids_[i] = tid_;
       return;
     }
@@ -297,56 +298,9 @@ bool ThreadClock::IsAlreadyAcquired(const SyncClock *src) const {
   return true;
 }
 
-void SyncClock::Resize(ClockCache *c, uptr nclk) {
-  CPP_STAT_INC(StatClockReleaseResize);
-  if (RoundUpTo(nclk, ClockBlock::kClockCount) <=
-      RoundUpTo(size_, ClockBlock::kClockCount)) {
-    // Growing within the same block.
-    // Memory is already allocated, just increase the size.
-    size_ = nclk;
-    return;
-  }
-  if (nclk <= ClockBlock::kClockCount) {
-    // Grow from 0 to one-level table.
-    CHECK_EQ(size_, 0);
-    CHECK_EQ(tab_, 0);
-    CHECK_EQ(tab_idx_, 0);
-    size_ = nclk;
-    tab_idx_ = ctx->clock_alloc.Alloc(c);
-    tab_ = ctx->clock_alloc.Map(tab_idx_);
-    internal_memset(tab_, 0, sizeof(*tab_));
-    return;
-  }
-  // Growing two-level table.
-  if (size_ == 0) {
-    // Allocate first level table.
-    tab_idx_ = ctx->clock_alloc.Alloc(c);
-    tab_ = ctx->clock_alloc.Map(tab_idx_);
-    internal_memset(tab_, 0, sizeof(*tab_));
-  } else if (size_ <= ClockBlock::kClockCount) {
-    // Transform one-level table to two-level table.
-    u32 old = tab_idx_;
-    tab_idx_ = ctx->clock_alloc.Alloc(c);
-    tab_ = ctx->clock_alloc.Map(tab_idx_);
-    internal_memset(tab_, 0, sizeof(*tab_));
-    tab_->table[0] = old;
-  }
-  // At this point we have first level table allocated.
-  // Add second level tables as necessary.
-  for (uptr i = RoundUpTo(size_, ClockBlock::kClockCount);
-      i < nclk; i += ClockBlock::kClockCount) {
-    u32 idx = ctx->clock_alloc.Alloc(c);
-    ClockBlock *cb = ctx->clock_alloc.Map(idx);
-    internal_memset(cb, 0, sizeof(*cb));
-    CHECK_EQ(tab_->table[i/ClockBlock::kClockCount], 0);
-    tab_->table[i/ClockBlock::kClockCount] = idx;
-  }
-  size_ = nclk;
-}
-
 // Sets a single element in the vector clock.
 // This function is called only from weird places like AcquireGlobal.
-void ThreadClock::set(unsigned tid, u64 v) {
+void ThreadClock::set(ClockCache *c, unsigned tid, u64 v) {
   DCHECK_LT(tid, kMaxTid);
   DCHECK_GE(v, clk_[tid].epoch);
   clk_[tid].epoch = v;
@@ -366,14 +320,8 @@ void ThreadClock::DebugDump(int(*printf)(const char *s, ...)) {
       tid_, reused_, last_acquire_);
 }
 
-SyncClock::SyncClock()
-    : release_store_tid_(kInvalidTid)
-    , release_store_reused_()
-    , tab_()
-    , tab_idx_()
-    , size_() {
-  for (uptr i = 0; i < kDirtyTids; i++)
-    dirty_tids_[i] = kInvalidTid;
+SyncClock::SyncClock() {
+  ResetImpl();
 }
 
 SyncClock::~SyncClock() {
@@ -395,6 +343,10 @@ void SyncClock::Reset(ClockCache *c) {
       ctx->clock_alloc.Free(c, tab_->table[i / ClockBlock::kClockCount]);
     ctx->clock_alloc.Free(c, tab_idx_);
   }
+  ResetImpl();
+}
+
+void SyncClock::ResetImpl() {
   tab_ = 0;
   tab_idx_ = 0;
   size_ = 0;
@@ -404,6 +356,53 @@ void SyncClock::Reset(ClockCache *c) {
     dirty_tids_[i] = kInvalidTid;
 }
 
+void SyncClock::Resize(ClockCache *c, uptr nclk) {
+  CPP_STAT_INC(StatClockReleaseResize);
+  if (RoundUpTo(nclk, ClockBlock::kClockCount) <=
+      RoundUpTo(size_, ClockBlock::kClockCount)) {
+    // Growing within the same block.
+    // Memory is already allocated, just increase the size.
+    size_ = nclk;
+    return;
+  }
+  if (nclk <= ClockBlock::kClockCount) {
+    // Grow from 0 to one-level table.
+    CHECK_EQ(size_, 0);
+    CHECK_EQ(tab_, 0);
+    CHECK_EQ(tab_idx_, 0);
+    size_ = nclk;
+    tab_idx_ = ctx->clock_alloc.Alloc(c);
+    tab_ = ctx->clock_alloc.Map(tab_idx_);
+    internal_memset(tab_, 0, sizeof(*tab_));
+    return;
+  }
+  // Growing two-level table.
+  if (size_ == 0) {
+    // Allocate first level table.
+    tab_idx_ = ctx->clock_alloc.Alloc(c);
+    tab_ = ctx->clock_alloc.Map(tab_idx_);
+    internal_memset(tab_, 0, sizeof(*tab_));
+  } else if (size_ <= ClockBlock::kClockCount) {
+    // Transform one-level table to two-level table.
+    u32 old = tab_idx_;
+    tab_idx_ = ctx->clock_alloc.Alloc(c);
+    tab_ = ctx->clock_alloc.Map(tab_idx_);
+    internal_memset(tab_, 0, sizeof(*tab_));
+    tab_->table[0] = old;
+  }
+  // At this point we have first level table allocated.
+  // Add second level tables as necessary.
+  for (uptr i = RoundUpTo(size_, ClockBlock::kClockCount);
+      i < nclk; i += ClockBlock::kClockCount) {
+    u32 idx = ctx->clock_alloc.Alloc(c);
+    ClockBlock *cb = ctx->clock_alloc.Map(idx);
+    internal_memset(cb, 0, sizeof(*cb));
+    CHECK_EQ(tab_->table[i/ClockBlock::kClockCount], 0);
+    tab_->table[i/ClockBlock::kClockCount] = idx;
+  }
+  size_ = nclk;
+}
+
 ClockElem &SyncClock::elem(unsigned tid) const {
   DCHECK_LT(tid, size_);
   if (size_ <= ClockBlock::kClockCount)
diff --git a/lib/tsan/rtl/tsan_clock.h b/lib/tsan/rtl/tsan_clock.h
index 4e352cb81d11..378b550fd11b 100644
--- a/lib/tsan/rtl/tsan_clock.h
+++ b/lib/tsan/rtl/tsan_clock.h
@@ -74,6 +74,7 @@ class SyncClock {
   u32 tab_idx_;
   u32 size_;
 
+  void ResetImpl();
   ClockElem &elem(unsigned tid) const;
 };
 
@@ -89,7 +90,7 @@ struct ThreadClock {
     return clk_[tid].epoch;
   }
 
-  void set(unsigned tid, u64 v);
+  void set(ClockCache *c, unsigned tid, u64 v);
 
   void set(u64 v) {
     DCHECK_GE(v, clk_[tid_].epoch);
@@ -108,6 +109,7 @@ struct ThreadClock {
   void release(ClockCache *c, SyncClock *dst) const;
   void acq_rel(ClockCache *c, SyncClock *dst);
   void ReleaseStore(ClockCache *c, SyncClock *dst) const;
+  void ResetCached(ClockCache *c);
 
   void DebugReset();
   void DebugDump(int(*printf)(const char *s, ...));
diff --git a/lib/tsan/rtl/tsan_dense_alloc.h b/lib/tsan/rtl/tsan_dense_alloc.h
index e9815c90a953..16dbdf391085 100644
--- a/lib/tsan/rtl/tsan_dense_alloc.h
+++ b/lib/tsan/rtl/tsan_dense_alloc.h
@@ -39,7 +39,7 @@ class DenseSlabAlloc {
   typedef DenseSlabAllocCache Cache;
   typedef typename Cache::IndexT IndexT;
 
-  DenseSlabAlloc() {
+  explicit DenseSlabAlloc(const char *name) {
     // Check that kL1Size and kL2Size are sane.
     CHECK_EQ(kL1Size & (kL1Size - 1), 0);
     CHECK_EQ(kL2Size & (kL2Size - 1), 0);
@@ -49,6 +49,7 @@ class DenseSlabAlloc {
     internal_memset(map_, 0, sizeof(map_));
     freelist_ = 0;
     fillpos_ = 0;
+    name_ = name;
   }
 
   ~DenseSlabAlloc() {
@@ -96,15 +97,19 @@ class DenseSlabAlloc {
   SpinMutex mtx_;
   IndexT freelist_;
   uptr fillpos_;
+  const char *name_;
 
   void Refill(Cache *c) {
     SpinMutexLock lock(&mtx_);
     if (freelist_ == 0) {
       if (fillpos_ == kL1Size) {
-        Printf("ThreadSanitizer: DenseSlabAllocator overflow. Dying.\n");
+        Printf("ThreadSanitizer: %s overflow (%zu*%zu). Dying.\n",
+            name_, kL1Size, kL2Size);
         Die();
       }
-      T *batch = (T*)MmapOrDie(kL2Size * sizeof(T), "DenseSlabAllocator");
+      VPrintf(2, "ThreadSanitizer: growing %s: %zu out of %zu*%zu\n",
+          name_, fillpos_, kL1Size, kL2Size);
+      T *batch = (T*)MmapOrDie(kL2Size * sizeof(T), name_);
       // Reserve 0 as invalid index.
       IndexT start = fillpos_ == 0 ? 1 : 0;
       for (IndexT i = start; i < kL2Size; i++) {
diff --git a/lib/tsan/rtl/tsan_interceptors.cc b/lib/tsan/rtl/tsan_interceptors.cc
index 334cc326daf6..001123f4941e 100644
--- a/lib/tsan/rtl/tsan_interceptors.cc
+++ b/lib/tsan/rtl/tsan_interceptors.cc
@@ -14,6 +14,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "sanitizer_common/sanitizer_atomic.h"
+#include "sanitizer_common/sanitizer_errno.h"
 #include "sanitizer_common/sanitizer_libc.h"
 #include "sanitizer_common/sanitizer_linux.h"
 #include "sanitizer_common/sanitizer_platform_limits_posix.h"
@@ -34,13 +35,11 @@
 using namespace __tsan;  // NOLINT
 
 #if SANITIZER_FREEBSD || SANITIZER_MAC
-#define __errno_location __error
 #define stdout __stdoutp
 #define stderr __stderrp
 #endif
 
 #if SANITIZER_ANDROID
-#define __errno_location __errno
 #define mallopt(a, b)
 #endif
 
@@ -84,7 +83,6 @@ DECLARE_REAL_AND_INTERCEPTOR(void *, malloc, uptr size)
 DECLARE_REAL_AND_INTERCEPTOR(void, free, void *ptr)
 extern "C" void *pthread_self();
 extern "C" void _exit(int status);
-extern "C" int *__errno_location();
 extern "C" int fileno_unlocked(void *stream);
 extern "C" int dirfd(void *dirp);
 #if !SANITIZER_FREEBSD && !SANITIZER_ANDROID
@@ -98,9 +96,6 @@ const int PTHREAD_MUTEX_RECURSIVE_NP = 1;
 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_FREEBSD && !SANITIZER_MAC
 const int EPOLL_CTL_ADD = 1;
 #endif
@@ -130,8 +125,6 @@ typedef long long_t;  // NOLINT
 # define F_TLOCK 2      /* Test and lock a region for exclusive use.  */
 # define F_TEST  3      /* Test a region for other processes locks.  */
 
-#define errno (*__errno_location())
-
 typedef void (*sighandler_t)(int sig);
 typedef void (*sigactionhandler_t)(int sig, my_siginfo_t *siginfo, void *uctx);
 
@@ -268,7 +261,7 @@ ScopedInterceptor::~ScopedInterceptor() {
 
 void ScopedInterceptor::EnableIgnores() {
   if (ignoring_) {
-    ThreadIgnoreBegin(thr_, pc_, false);
+    ThreadIgnoreBegin(thr_, pc_, /*save_stack=*/false);
     if (flags()->ignore_noninstrumented_modules) thr_->suppress_reports++;
     if (in_ignored_lib_) {
       DCHECK(!thr_->in_ignored_lib);
@@ -466,8 +459,14 @@ static void SetJmp(ThreadState *thr, uptr sp, uptr mangled_sp) {
 static void LongJmp(ThreadState *thr, uptr *env) {
 #ifdef __powerpc__
   uptr mangled_sp = env[0];
-#elif SANITIZER_FREEBSD || SANITIZER_MAC
+#elif SANITIZER_FREEBSD
   uptr mangled_sp = env[2];
+#elif SANITIZER_MAC
+# ifdef __aarch64__
+    uptr mangled_sp = env[13];
+# else
+    uptr mangled_sp = env[2];
+# endif
 #elif defined(SANITIZER_LINUX)
 # ifdef __aarch64__
   uptr mangled_sp = env[13];
@@ -665,7 +664,7 @@ static bool fix_mmap_addr(void **addr, long_t sz, int flags) {
   if (*addr) {
     if (!IsAppMem((uptr)*addr) || !IsAppMem((uptr)*addr + sz - 1)) {
       if (flags & MAP_FIXED) {
-        errno = EINVAL;
+        errno = errno_EINVAL;
         return false;
       } else {
         *addr = 0;
@@ -1122,7 +1121,7 @@ TSAN_INTERCEPTOR(int, pthread_mutex_init, void *m, void *a) {
 TSAN_INTERCEPTOR(int, pthread_mutex_destroy, void *m) {
   SCOPED_TSAN_INTERCEPTOR(pthread_mutex_destroy, m);
   int res = REAL(pthread_mutex_destroy)(m);
-  if (res == 0 || res == EBUSY) {
+  if (res == 0 || res == errno_EBUSY) {
     MutexDestroy(thr, pc, (uptr)m);
   }
   return res;
@@ -1131,9 +1130,9 @@ TSAN_INTERCEPTOR(int, pthread_mutex_destroy, void *m) {
 TSAN_INTERCEPTOR(int, pthread_mutex_trylock, void *m) {
   SCOPED_TSAN_INTERCEPTOR(pthread_mutex_trylock, m);
   int res = REAL(pthread_mutex_trylock)(m);
-  if (res == EOWNERDEAD)
+  if (res == errno_EOWNERDEAD)
     MutexRepair(thr, pc, (uptr)m);
-  if (res == 0 || res == EOWNERDEAD)
+  if (res == 0 || res == errno_EOWNERDEAD)
     MutexPostLock(thr, pc, (uptr)m, MutexFlagTryLock);
   return res;
 }
@@ -1311,7 +1310,7 @@ TSAN_INTERCEPTOR(int, pthread_barrier_wait, void *b) {
 TSAN_INTERCEPTOR(int, pthread_once, void *o, void (*f)()) {
   SCOPED_INTERCEPTOR_RAW(pthread_once, o, f);
   if (o == 0 || f == 0)
-    return EINVAL;
+    return errno_EINVAL;
   atomic_uint32_t *a;
   if (!SANITIZER_MAC)
     a = static_cast<atomic_uint32_t*>(o);
diff --git a/lib/tsan/rtl/tsan_interceptors_mac.cc b/lib/tsan/rtl/tsan_interceptors_mac.cc
index f6bf8a0e586b..4f1079467331 100644
--- a/lib/tsan/rtl/tsan_interceptors_mac.cc
+++ b/lib/tsan/rtl/tsan_interceptors_mac.cc
@@ -21,7 +21,10 @@
 #include "tsan_interface_ann.h"
 
 #include <libkern/OSAtomic.h>
+
+#if defined(__has_include) && __has_include(<xpc/xpc.h>)
 #include <xpc/xpc.h>
+#endif  // #if defined(__has_include) && __has_include(<xpc/xpc.h>)
 
 typedef long long_t;  // NOLINT
 
@@ -235,6 +238,8 @@ TSAN_INTERCEPTOR(void, os_lock_unlock, void *lock) {
   REAL(os_lock_unlock)(lock);
 }
 
+#if defined(__has_include) && __has_include(<xpc/xpc.h>)
+
 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,
@@ -287,6 +292,8 @@ TSAN_INTERCEPTOR(void, xpc_connection_cancel, xpc_connection_t connection) {
   REAL(xpc_connection_cancel)(connection);
 }
 
+#endif  // #if defined(__has_include) && __has_include(<xpc/xpc.h>)
+
 // On macOS, libc++ is always linked dynamically, so intercepting works the
 // usual way.
 #define STDCXX_INTERCEPTOR TSAN_INTERCEPTOR
diff --git a/lib/tsan/rtl/tsan_interface_ann.cc b/lib/tsan/rtl/tsan_interface_ann.cc
index 45ec45bbdbbe..f68a0468de53 100644
--- a/lib/tsan/rtl/tsan_interface_ann.cc
+++ b/lib/tsan/rtl/tsan_interface_ann.cc
@@ -483,8 +483,8 @@ void __tsan_mutex_pre_lock(void *m, unsigned flagz) {
     else
       MutexPreLock(thr, pc, (uptr)m);
   }
-  ThreadIgnoreBegin(thr, pc, false);
-  ThreadIgnoreSyncBegin(thr, pc, false);
+  ThreadIgnoreBegin(thr, pc, /*save_stack=*/false);
+  ThreadIgnoreSyncBegin(thr, pc, /*save_stack=*/false);
 }
 
 INTERFACE_ATTRIBUTE
@@ -510,8 +510,8 @@ int __tsan_mutex_pre_unlock(void *m, unsigned flagz) {
   } else {
     ret = MutexUnlock(thr, pc, (uptr)m, flagz);
   }
-  ThreadIgnoreBegin(thr, pc, false);
-  ThreadIgnoreSyncBegin(thr, pc, false);
+  ThreadIgnoreBegin(thr, pc, /*save_stack=*/false);
+  ThreadIgnoreSyncBegin(thr, pc, /*save_stack=*/false);
   return ret;
 }
 
@@ -525,8 +525,8 @@ void __tsan_mutex_post_unlock(void *m, unsigned flagz) {
 INTERFACE_ATTRIBUTE
 void __tsan_mutex_pre_signal(void *addr, unsigned flagz) {
   SCOPED_ANNOTATION(__tsan_mutex_pre_signal);
-  ThreadIgnoreBegin(thr, pc, false);
-  ThreadIgnoreSyncBegin(thr, pc, false);
+  ThreadIgnoreBegin(thr, pc, /*save_stack=*/false);
+  ThreadIgnoreSyncBegin(thr, pc, /*save_stack=*/false);
 }
 
 INTERFACE_ATTRIBUTE
@@ -547,7 +547,7 @@ void __tsan_mutex_pre_divert(void *addr, unsigned flagz) {
 INTERFACE_ATTRIBUTE
 void __tsan_mutex_post_divert(void *addr, unsigned flagz) {
   SCOPED_ANNOTATION(__tsan_mutex_post_divert);
-  ThreadIgnoreBegin(thr, pc, false);
-  ThreadIgnoreSyncBegin(thr, pc, false);
+  ThreadIgnoreBegin(thr, pc, /*save_stack=*/false);
+  ThreadIgnoreSyncBegin(thr, pc, /*save_stack=*/false);
 }
 }  // extern "C"
diff --git a/lib/tsan/rtl/tsan_interface_atomic.cc b/lib/tsan/rtl/tsan_interface_atomic.cc
index b22d5c1ecef8..d334394f5303 100644
--- a/lib/tsan/rtl/tsan_interface_atomic.cc
+++ b/lib/tsan/rtl/tsan_interface_atomic.cc
@@ -220,8 +220,7 @@ static a128 NoTsanAtomicLoad(const volatile a128 *a, morder mo) {
 #endif
 
 template<typename T>
-static T AtomicLoad(ThreadState *thr, uptr pc, const volatile T *a,
-    morder mo) {
+static T AtomicLoad(ThreadState *thr, uptr pc, const volatile T *a, morder mo) {
   CHECK(IsLoadOrder(mo));
   // This fast-path is critical for performance.
   // Assume the access is atomic.
@@ -229,10 +228,17 @@ static T AtomicLoad(ThreadState *thr, uptr pc, const volatile T *a,
     MemoryReadAtomic(thr, pc, (uptr)a, SizeLog<T>());
     return NoTsanAtomicLoad(a, mo);
   }
-  SyncVar *s = ctx->metamap.GetOrCreateAndLock(thr, pc, (uptr)a, false);
-  AcquireImpl(thr, pc, &s->clock);
+  // Don't create sync object if it does not exist yet. For example, an atomic
+  // pointer is initialized to nullptr and then periodically acquire-loaded.
   T v = NoTsanAtomicLoad(a, mo);
-  s->mtx.ReadUnlock();
+  SyncVar *s = ctx->metamap.GetIfExistsAndLock((uptr)a, false);
+  if (s) {
+    AcquireImpl(thr, pc, &s->clock);
+    // Re-read under sync mutex because we need a consistent snapshot
+    // of the value and the clock we acquire.
+    v = NoTsanAtomicLoad(a, mo);
+    s->mtx.ReadUnlock();
+  }
   MemoryReadAtomic(thr, pc, (uptr)a, SizeLog<T>());
   return v;
 }
diff --git a/lib/tsan/rtl/tsan_mman.cc b/lib/tsan/rtl/tsan_mman.cc
index 7169d5b02c04..1434cf688ce9 100644
--- a/lib/tsan/rtl/tsan_mman.cc
+++ b/lib/tsan/rtl/tsan_mman.cc
@@ -294,6 +294,8 @@ uptr __sanitizer_get_allocated_size(const void *p) {
 
 void __tsan_on_thread_idle() {
   ThreadState *thr = cur_thread();
+  thr->clock.ResetCached(&thr->proc()->clock_cache);
+  thr->last_sleep_clock.ResetCached(&thr->proc()->clock_cache);
   allocator()->SwallowCache(&thr->proc()->alloc_cache);
   internal_allocator()->SwallowCache(&thr->proc()->internal_alloc_cache);
   ctx->metamap.OnProcIdle(thr->proc());
diff --git a/lib/tsan/rtl/tsan_platform.h b/lib/tsan/rtl/tsan_platform.h
index 60d9b9d8c452..bea1daba3952 100644
--- a/lib/tsan/rtl/tsan_platform.h
+++ b/lib/tsan/rtl/tsan_platform.h
@@ -100,6 +100,37 @@ struct Mapping {
 };
 
 #define TSAN_MID_APP_RANGE 1
+#elif defined(__aarch64__) && defined(__APPLE__)
+/*
+C/C++ on Darwin/iOS/ARM64 (36-bit VMA, 64 GB VM)
+0000 0000 00 - 0100 0000 00: -                                    (4 GB)
+0100 0000 00 - 0200 0000 00: main binary, modules, thread stacks  (4 GB)
+0200 0000 00 - 0300 0000 00: heap                                 (4 GB)
+0300 0000 00 - 0400 0000 00: -                                    (4 GB)
+0400 0000 00 - 0c00 0000 00: shadow memory                       (32 GB)
+0c00 0000 00 - 0d00 0000 00: -                                    (4 GB)
+0d00 0000 00 - 0e00 0000 00: metainfo                             (4 GB)
+0e00 0000 00 - 0f00 0000 00: -                                    (4 GB)
+0f00 0000 00 - 1000 0000 00: traces                               (4 GB)
+*/
+struct Mapping {
+  static const uptr kLoAppMemBeg   = 0x0100000000ull;
+  static const uptr kLoAppMemEnd   = 0x0200000000ull;
+  static const uptr kHeapMemBeg    = 0x0200000000ull;
+  static const uptr kHeapMemEnd    = 0x0300000000ull;
+  static const uptr kShadowBeg     = 0x0400000000ull;
+  static const uptr kShadowEnd     = 0x0c00000000ull;
+  static const uptr kMetaShadowBeg = 0x0d00000000ull;
+  static const uptr kMetaShadowEnd = 0x0e00000000ull;
+  static const uptr kTraceMemBeg   = 0x0f00000000ull;
+  static const uptr kTraceMemEnd   = 0x1000000000ull;
+  static const uptr kHiAppMemBeg   = 0x1000000000ull;
+  static const uptr kHiAppMemEnd   = 0x1000000000ull;
+  static const uptr kAppMemMsk     =          0x0ull;
+  static const uptr kAppMemXor     =          0x0ull;
+  static const uptr kVdsoBeg       = 0x7000000000000000ull;
+};
+
 #elif defined(__aarch64__)
 // AArch64 supports multiple VMA which leads to multiple address transformation
 // functions.  To support these multiple VMAS transformations and mappings TSAN
@@ -389,7 +420,7 @@ uptr MappingImpl(void) {
 
 template<int Type>
 uptr MappingArchImpl(void) {
-#ifdef __aarch64__
+#if defined(__aarch64__) && !defined(__APPLE__)
   switch (vmaSize) {
     case 39: return MappingImpl<Mapping39, Type>();
     case 42: return MappingImpl<Mapping42, Type>();
@@ -542,7 +573,7 @@ bool IsAppMemImpl(uptr mem) {
 
 ALWAYS_INLINE
 bool IsAppMem(uptr mem) {
-#ifdef __aarch64__
+#if defined(__aarch64__) && !defined(__APPLE__)
   switch (vmaSize) {
     case 39: return IsAppMemImpl<Mapping39>(mem);
     case 42: return IsAppMemImpl<Mapping42>(mem);
@@ -569,7 +600,7 @@ bool IsShadowMemImpl(uptr mem) {
 
 ALWAYS_INLINE
 bool IsShadowMem(uptr mem) {
-#ifdef __aarch64__
+#if defined(__aarch64__) && !defined(__APPLE__)
   switch (vmaSize) {
     case 39: return IsShadowMemImpl<Mapping39>(mem);
     case 42: return IsShadowMemImpl<Mapping42>(mem);
@@ -596,7 +627,7 @@ bool IsMetaMemImpl(uptr mem) {
 
 ALWAYS_INLINE
 bool IsMetaMem(uptr mem) {
-#ifdef __aarch64__
+#if defined(__aarch64__) && !defined(__APPLE__)
   switch (vmaSize) {
     case 39: return IsMetaMemImpl<Mapping39>(mem);
     case 42: return IsMetaMemImpl<Mapping42>(mem);
@@ -633,7 +664,7 @@ uptr MemToShadowImpl(uptr x) {
 
 ALWAYS_INLINE
 uptr MemToShadow(uptr x) {
-#ifdef __aarch64__
+#if defined(__aarch64__) && !defined(__APPLE__)
   switch (vmaSize) {
     case 39: return MemToShadowImpl<Mapping39>(x);
     case 42: return MemToShadowImpl<Mapping42>(x);
@@ -672,7 +703,7 @@ u32 *MemToMetaImpl(uptr x) {
 
 ALWAYS_INLINE
 u32 *MemToMeta(uptr x) {
-#ifdef __aarch64__
+#if defined(__aarch64__) && !defined(__APPLE__)
   switch (vmaSize) {
     case 39: return MemToMetaImpl<Mapping39>(x);
     case 42: return MemToMetaImpl<Mapping42>(x);
@@ -724,7 +755,7 @@ uptr ShadowToMemImpl(uptr s) {
 
 ALWAYS_INLINE
 uptr ShadowToMem(uptr s) {
-#ifdef __aarch64__
+#if defined(__aarch64__) && !defined(__APPLE__)
   switch (vmaSize) {
     case 39: return ShadowToMemImpl<Mapping39>(s);
     case 42: return ShadowToMemImpl<Mapping42>(s);
@@ -759,7 +790,7 @@ uptr GetThreadTraceImpl(int tid) {
 
 ALWAYS_INLINE
 uptr GetThreadTrace(int tid) {
-#ifdef __aarch64__
+#if defined(__aarch64__) && !defined(__APPLE__)
   switch (vmaSize) {
     case 39: return GetThreadTraceImpl<Mapping39>(tid);
     case 42: return GetThreadTraceImpl<Mapping42>(tid);
@@ -789,7 +820,7 @@ uptr GetThreadTraceHeaderImpl(int tid) {
 
 ALWAYS_INLINE
 uptr GetThreadTraceHeader(int tid) {
-#ifdef __aarch64__
+#if defined(__aarch64__) && !defined(__APPLE__)
   switch (vmaSize) {
     case 39: return GetThreadTraceHeaderImpl<Mapping39>(tid);
     case 42: return GetThreadTraceHeaderImpl<Mapping42>(tid);
diff --git a/lib/tsan/rtl/tsan_platform_linux.cc b/lib/tsan/rtl/tsan_platform_linux.cc
index d05c0e701e72..0ba01babe69a 100644
--- a/lib/tsan/rtl/tsan_platform_linux.cc
+++ b/lib/tsan/rtl/tsan_platform_linux.cc
@@ -47,7 +47,6 @@
 #include <sys/resource.h>
 #include <sys/stat.h>
 #include <unistd.h>
-#include <errno.h>
 #include <sched.h>
 #include <dlfcn.h>
 #if SANITIZER_LINUX
@@ -182,17 +181,15 @@ static void MapRodata() {
   }
   // Map the file into shadow of .rodata sections.
   MemoryMappingLayout proc_maps(/*cache_enabled*/true);
-  uptr start, end, offset, prot;
   // Reusing the buffer 'name'.
-  while (proc_maps.Next(&start, &end, &offset, name, ARRAY_SIZE(name), &prot)) {
-    if (name[0] != 0 && name[0] != '['
-        && (prot & MemoryMappingLayout::kProtectionRead)
-        && (prot & MemoryMappingLayout::kProtectionExecute)
-        && !(prot & MemoryMappingLayout::kProtectionWrite)
-        && IsAppMem(start)) {
+  MemoryMappedSegment segment(name, ARRAY_SIZE(name));
+  while (proc_maps.Next(&segment)) {
+    if (segment.filename[0] != 0 && segment.filename[0] != '[' &&
+        segment.IsReadable() && segment.IsExecutable() &&
+        !segment.IsWritable() && IsAppMem(segment.start)) {
       // Assume it's .rodata
-      char *shadow_start = (char*)MemToShadow(start);
-      char *shadow_end = (char*)MemToShadow(end);
+      char *shadow_start = (char *)MemToShadow(segment.start);
+      char *shadow_end = (char *)MemToShadow(segment.end);
       for (char *p = shadow_start; p < shadow_end; p += marker.size()) {
         internal_mmap(p, Min<uptr>(marker.size(), shadow_end - p),
                       PROT_READ, MAP_PRIVATE | MAP_FIXED, fd, 0);
diff --git a/lib/tsan/rtl/tsan_platform_mac.cc b/lib/tsan/rtl/tsan_platform_mac.cc
index a82bcd01bbf4..73a656ffca5e 100644
--- a/lib/tsan/rtl/tsan_platform_mac.cc
+++ b/lib/tsan/rtl/tsan_platform_mac.cc
@@ -230,6 +230,14 @@ static void my_pthread_introspection_hook(unsigned int event, pthread_t thread,
 #endif
 
 void InitializePlatformEarly() {
+#if defined(__aarch64__)
+  uptr max_vm = GetMaxVirtualAddress() + 1;
+  if (max_vm != Mapping::kHiAppMemEnd) {
+    Printf("ThreadSanitizer: unsupported vm address limit %p, expected %p.\n",
+           max_vm, Mapping::kHiAppMemEnd);
+    Die();
+  }
+#endif
 }
 
 void InitializePlatform() {
diff --git a/lib/tsan/rtl/tsan_platform_posix.cc b/lib/tsan/rtl/tsan_platform_posix.cc
index 0732c83d689d..e4f90a811c35 100644
--- a/lib/tsan/rtl/tsan_platform_posix.cc
+++ b/lib/tsan/rtl/tsan_platform_posix.cc
@@ -46,6 +46,9 @@ void InitializeShadowMemory() {
 #elif defined(__mips64)
   const uptr kMadviseRangeBeg  = 0xff00000000ull;
   const uptr kMadviseRangeSize = 0x0100000000ull;
+#elif defined(__aarch64__) && defined(__APPLE__)
+  uptr kMadviseRangeBeg = LoAppMemBeg();
+  uptr kMadviseRangeSize = LoAppMemEnd() - LoAppMemBeg();
 #elif defined(__aarch64__)
   uptr kMadviseRangeBeg = 0;
   uptr kMadviseRangeSize = 0;
@@ -115,21 +118,24 @@ static void ProtectRange(uptr beg, uptr end) {
 void CheckAndProtect() {
   // Ensure that the binary is indeed compiled with -pie.
   MemoryMappingLayout proc_maps(true);
-  uptr p, end, prot;
-  while (proc_maps.Next(&p, &end, 0, 0, 0, &prot)) {
-    if (IsAppMem(p))
+  MemoryMappedSegment segment;
+  while (proc_maps.Next(&segment)) {
+    if (IsAppMem(segment.start)) continue;
+    if (segment.start >= HeapMemEnd() && segment.start < HeapEnd()) continue;
+    if (segment.protection == 0)  // Zero page or mprotected.
       continue;
-    if (p >= HeapMemEnd() &&
-        p < HeapEnd())
-      continue;
-    if (prot == 0)  // Zero page or mprotected.
-      continue;
-    if (p >= VdsoBeg())  // vdso
+    if (segment.start >= VdsoBeg())  // vdso
       break;
-    Printf("FATAL: ThreadSanitizer: unexpected memory mapping %p-%p\n", p, end);
+    Printf("FATAL: ThreadSanitizer: unexpected memory mapping %p-%p\n",
+           segment.start, segment.end);
     Die();
   }
 
+#if defined(__aarch64__) && defined(__APPLE__)
+  ProtectRange(HeapMemEnd(), ShadowBeg());
+  ProtectRange(ShadowEnd(), MetaShadowBeg());
+  ProtectRange(MetaShadowEnd(), TraceMemBeg());
+#else
   ProtectRange(LoAppMemEnd(), ShadowBeg());
   ProtectRange(ShadowEnd(), MetaShadowBeg());
 #ifdef TSAN_MID_APP_RANGE
@@ -143,6 +149,7 @@ void CheckAndProtect() {
   ProtectRange(TraceMemBeg(), TraceMemEnd());
   ProtectRange(TraceMemEnd(), HeapMemBeg());
   ProtectRange(HeapEnd(), HiAppMemBeg());
+#endif
 }
 #endif
 
diff --git a/lib/tsan/rtl/tsan_rtl.cc b/lib/tsan/rtl/tsan_rtl.cc
index fa60f3247c38..a01525302b02 100644
--- a/lib/tsan/rtl/tsan_rtl.cc
+++ b/lib/tsan/rtl/tsan_rtl.cc
@@ -104,7 +104,8 @@ Context::Context()
   , racy_stacks(MBlockRacyStacks)
   , racy_addresses(MBlockRacyAddresses)
   , fired_suppressions_mtx(MutexTypeFired, StatMtxFired)
-  , fired_suppressions(8) {
+  , fired_suppressions(8)
+  , clock_alloc("clock allocator") {
 }
 
 // The objects are allocated in TLS, so one may rely on zero-initialization.
diff --git a/lib/tsan/rtl/tsan_rtl_aarch64.S b/lib/tsan/rtl/tsan_rtl_aarch64.S
index ef06f0444ae4..61171d635c18 100644
--- a/lib/tsan/rtl/tsan_rtl_aarch64.S
+++ b/lib/tsan/rtl/tsan_rtl_aarch64.S
@@ -1,13 +1,46 @@
+// The content of this file is AArch64-only:
+#if defined(__aarch64__)
+
 #include "sanitizer_common/sanitizer_asm.h"
 
+#if !defined(__APPLE__)
 .section .bss
 .type	__tsan_pointer_chk_guard, %object
-.size	__tsan_pointer_chk_guard, 8
+ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(__tsan_pointer_chk_guard))
 __tsan_pointer_chk_guard:
 .zero	8
+#endif
+
+#if defined(__APPLE__)
+.align  2
+
+.section  __DATA,__nl_symbol_ptr,non_lazy_symbol_pointers
+.long _setjmp$non_lazy_ptr
+_setjmp$non_lazy_ptr:
+.indirect_symbol _setjmp
+.long 0
+
+.section  __DATA,__nl_symbol_ptr,non_lazy_symbol_pointers
+.long __setjmp$non_lazy_ptr
+__setjmp$non_lazy_ptr:
+.indirect_symbol __setjmp
+.long 0
+
+.section  __DATA,__nl_symbol_ptr,non_lazy_symbol_pointers
+.long _sigsetjmp$non_lazy_ptr
+_sigsetjmp$non_lazy_ptr:
+.indirect_symbol _sigsetjmp
+.long 0
+#endif
 
+#if !defined(__APPLE__)
 .section .text
+#else
+.section __TEXT,__text
+.align 3
+#endif
 
+#if !defined(__APPLE__)
 // GLIBC mangles the function pointers in jmp_buf (used in {set,long}*jmp
 // functions) by XORing them with a random guard pointer.  For AArch64 it is a
 // global variable rather than a TCB one (as for x86_64/powerpc) and althought
@@ -16,9 +49,9 @@ __tsan_pointer_chk_guard:
 // not stable). So InitializeGuardPtr obtains the pointer guard value by
 // issuing a setjmp and checking the resulting pointers values against the
 // original ones.
-.hidden _Z18InitializeGuardPtrv
+ASM_HIDDEN(_Z18InitializeGuardPtrv)
 .global _Z18InitializeGuardPtrv
-.type _Z18InitializeGuardPtrv, @function
+ASM_TYPE_FUNCTION(ASM_TSAN_SYMBOL_INTERCEPTOR(_Z18InitializeGuardPtrv))
 _Z18InitializeGuardPtrv:
   CFI_STARTPROC
   // Allocates a jmp_buf for the setjmp call.
@@ -55,12 +88,14 @@ _Z18InitializeGuardPtrv:
   CFI_DEF_CFA (31, 0)
   ret
   CFI_ENDPROC
-.size _Z18InitializeGuardPtrv, .-_Z18InitializeGuardPtrv
+ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(_Z18InitializeGuardPtrv))
+#endif
 
-.hidden __tsan_setjmp
+ASM_HIDDEN(__tsan_setjmp)
 .comm _ZN14__interception11real_setjmpE,8,8
-.type setjmp, @function
-setjmp:
+.globl ASM_TSAN_SYMBOL_INTERCEPTOR(setjmp)
+ASM_TYPE_FUNCTION(ASM_TSAN_SYMBOL_INTERCEPTOR(setjmp))
+ASM_TSAN_SYMBOL_INTERCEPTOR(setjmp):
   CFI_STARTPROC
 
   // save env parameters for function call
@@ -78,14 +113,19 @@ setjmp:
   CFI_OFFSET (19, -16)
   mov     x19, x0
 
+#if !defined(__APPLE__)
   // SP pointer mangling (see glibc setjmp)
   adrp    x2, __tsan_pointer_chk_guard
   ldr     x2, [x2, #:lo12:__tsan_pointer_chk_guard]
   add     x0, x29, 32
   eor     x1, x2, x0
+#else
+  add     x0, x29, 32
+  mov     x1, x0
+#endif
 
   // call tsan interceptor
-  bl      __tsan_setjmp
+  bl      ASM_TSAN_SYMBOL(__tsan_setjmp)
 
   // restore env parameter
   mov     x0, x19
@@ -96,18 +136,24 @@ setjmp:
   CFI_DEF_CFA (31, 0)
 
   // tail jump to libc setjmp
+#if !defined(__APPLE__)
   adrp    x1, :got:_ZN14__interception11real_setjmpE
   ldr     x1, [x1, #:got_lo12:_ZN14__interception11real_setjmpE]
   ldr     x1, [x1]
+#else
+  adrp    x1, _setjmp$non_lazy_ptr@page
+  add     x1, x1, _setjmp$non_lazy_ptr@pageoff
+  ldr     x1, [x1]
+#endif
   br      x1
 
   CFI_ENDPROC
-.size setjmp, .-setjmp
+ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(setjmp))
 
 .comm _ZN14__interception12real__setjmpE,8,8
-.globl _setjmp
-.type _setjmp, @function
-_setjmp:
+.globl ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp)
+ASM_TYPE_FUNCTION(ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp))
+ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp):
   CFI_STARTPROC
 
   // save env parameters for function call
@@ -125,14 +171,19 @@ _setjmp:
   CFI_OFFSET (19, -16)
   mov     x19, x0
 
+#if !defined(__APPLE__)
   // SP pointer mangling (see glibc setjmp)
   adrp    x2, __tsan_pointer_chk_guard
   ldr     x2, [x2, #:lo12:__tsan_pointer_chk_guard]
   add     x0, x29, 32
   eor     x1, x2, x0
+#else
+  add     x0, x29, 32
+  mov     x1, x0
+#endif
 
   // call tsan interceptor
-  bl      __tsan_setjmp
+  bl      ASM_TSAN_SYMBOL(__tsan_setjmp)
 
   // Restore jmp_buf parameter
   mov     x0, x19
@@ -143,18 +194,24 @@ _setjmp:
   CFI_DEF_CFA (31, 0)
 
   // tail jump to libc setjmp
+#if !defined(__APPLE__)
   adrp    x1, :got:_ZN14__interception12real__setjmpE
   ldr     x1, [x1, #:got_lo12:_ZN14__interception12real__setjmpE]
   ldr     x1, [x1]
+#else
+  adrp    x1, __setjmp$non_lazy_ptr@page
+  add     x1, x1, __setjmp$non_lazy_ptr@pageoff
+  ldr     x1, [x1]
+#endif
   br      x1
 
   CFI_ENDPROC
-.size _setjmp, .-_setjmp
+ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp))
 
 .comm _ZN14__interception14real_sigsetjmpE,8,8
-.globl sigsetjmp
-.type sigsetjmp, @function
-sigsetjmp:
+.globl ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp)
+ASM_TYPE_FUNCTION(ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp))
+ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp):
   CFI_STARTPROC
 
   // save env parameters for function call
@@ -174,14 +231,19 @@ sigsetjmp:
   mov     w20, w1
   mov     x19, x0
 
+#if !defined(__APPLE__)
   // SP pointer mangling (see glibc setjmp)
   adrp    x2, __tsan_pointer_chk_guard
   ldr     x2, [x2, #:lo12:__tsan_pointer_chk_guard]
   add     x0, x29, 32
   eor     x1, x2, x0
+#else
+  add     x0, x29, 32
+  mov     x1, x0
+#endif
 
   // call tsan interceptor
-  bl      __tsan_setjmp
+  bl      ASM_TSAN_SYMBOL(__tsan_setjmp)
 
   // restore env parameter
   mov     w1, w20
@@ -195,17 +257,24 @@ sigsetjmp:
   CFI_DEF_CFA (31, 0)
 
   // tail jump to libc sigsetjmp
+#if !defined(__APPLE__)
   adrp    x2, :got:_ZN14__interception14real_sigsetjmpE
   ldr     x2, [x2, #:got_lo12:_ZN14__interception14real_sigsetjmpE]
   ldr     x2, [x2]
+#else
+  adrp    x2, _sigsetjmp$non_lazy_ptr@page
+  add     x2, x2, _sigsetjmp$non_lazy_ptr@pageoff
+  ldr     x2, [x2]
+#endif
   br      x2
   CFI_ENDPROC
-.size sigsetjmp, .-sigsetjmp
+ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp))
 
+#if !defined(__APPLE__)
 .comm _ZN14__interception16real___sigsetjmpE,8,8
-.globl __sigsetjmp
-.type __sigsetjmp, @function
-__sigsetjmp:
+.globl ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp)
+ASM_TYPE_FUNCTION(ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp))
+ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp):
   CFI_STARTPROC
 
   // save env parameters for function call
@@ -225,14 +294,16 @@ __sigsetjmp:
   mov     w20, w1
   mov     x19, x0
 
+#if !defined(__APPLE__)
   // SP pointer mangling (see glibc setjmp)
   adrp    x2, __tsan_pointer_chk_guard
   ldr     x2, [x2, #:lo12:__tsan_pointer_chk_guard]
   add     x0, x29, 32
   eor     x1, x2, x0
+#endif
 
   // call tsan interceptor
-  bl      __tsan_setjmp
+  bl      ASM_TSAN_SYMBOL(__tsan_setjmp)
 
   mov     w1, w20
   mov     x0, x19
@@ -245,14 +316,22 @@ __sigsetjmp:
   CFI_DEF_CFA (31, 0)
 
   // tail jump to libc __sigsetjmp
+#if !defined(__APPLE__)
   adrp    x2, :got:_ZN14__interception16real___sigsetjmpE
   ldr     x2, [x2, #:got_lo12:_ZN14__interception16real___sigsetjmpE]
   ldr     x2, [x2]
+#else
+  adrp    x2, ASM_TSAN_SYMBOL(__sigsetjmp)@page
+  add     x2, x2, ASM_TSAN_SYMBOL(__sigsetjmp)@pageoff
+#endif
   br      x2
   CFI_ENDPROC
-.size __sigsetjmp, .-__sigsetjmp
+ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp))
+#endif
 
 #if defined(__linux__)
 /* We do not need executable stack.  */
 .section        .note.GNU-stack,"",@progbits
 #endif
+
+#endif
diff --git a/lib/tsan/rtl/tsan_rtl_amd64.S b/lib/tsan/rtl/tsan_rtl_amd64.S
index caa832375e52..98947fd2a1ba 100644
--- a/lib/tsan/rtl/tsan_rtl_amd64.S
+++ b/lib/tsan/rtl/tsan_rtl_amd64.S
@@ -1,4 +1,8 @@
+// The content of this file is x86_64-only:
+#if defined(__x86_64__)
+
 #include "sanitizer_common/sanitizer_asm.h"
+
 #if !defined(__APPLE__)
 .section .text
 #else
@@ -357,3 +361,5 @@ ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp))
 /* We do not need executable stack.  */
 .section        .note.GNU-stack,"",@progbits
 #endif
+
+#endif
diff --git a/lib/tsan/rtl/tsan_rtl_mutex.cc b/lib/tsan/rtl/tsan_rtl_mutex.cc
index 54938f37e243..2f85811620f1 100644
--- a/lib/tsan/rtl/tsan_rtl_mutex.cc
+++ b/lib/tsan/rtl/tsan_rtl_mutex.cc
@@ -413,10 +413,10 @@ void Acquire(ThreadState *thr, uptr pc, uptr addr) {
 static void UpdateClockCallback(ThreadContextBase *tctx_base, void *arg) {
   ThreadState *thr = reinterpret_cast<ThreadState*>(arg);
   ThreadContext *tctx = static_cast<ThreadContext*>(tctx_base);
+  u64 epoch = tctx->epoch1;
   if (tctx->status == ThreadStatusRunning)
-    thr->clock.set(tctx->tid, tctx->thr->fast_state.epoch());
-  else
-    thr->clock.set(tctx->tid, tctx->epoch1);
+    epoch = tctx->thr->fast_state.epoch();
+  thr->clock.set(&thr->proc()->clock_cache, tctx->tid, epoch);
 }
 
 void AcquireGlobal(ThreadState *thr, uptr pc) {
@@ -456,10 +456,10 @@ void ReleaseStore(ThreadState *thr, uptr pc, uptr addr) {
 static void UpdateSleepClockCallback(ThreadContextBase *tctx_base, void *arg) {
   ThreadState *thr = reinterpret_cast<ThreadState*>(arg);
   ThreadContext *tctx = static_cast<ThreadContext*>(tctx_base);
+  u64 epoch = tctx->epoch1;
   if (tctx->status == ThreadStatusRunning)
-    thr->last_sleep_clock.set(tctx->tid, tctx->thr->fast_state.epoch());
-  else
-    thr->last_sleep_clock.set(tctx->tid, tctx->epoch1);
+    epoch = tctx->thr->fast_state.epoch();
+  thr->last_sleep_clock.set(&thr->proc()->clock_cache, tctx->tid, epoch);
 }
 
 void AfterSleep(ThreadState *thr, uptr pc) {
diff --git a/lib/tsan/rtl/tsan_rtl_report.cc b/lib/tsan/rtl/tsan_rtl_report.cc
index 68b9f50308ea..85a982941ed7 100644
--- a/lib/tsan/rtl/tsan_rtl_report.cc
+++ b/lib/tsan/rtl/tsan_rtl_report.cc
@@ -314,7 +314,7 @@ void ScopedReport::AddLocation(uptr addr, uptr size) {
     return;
 #if !SANITIZER_GO
   int fd = -1;
-  int creat_tid = -1;
+  int creat_tid = kInvalidTid;
   u32 creat_stack = 0;
   if (FdLocation(addr, &fd, &creat_tid, &creat_stack)) {
     ReportLocation *loc = ReportLocation::New(ReportLocationFD);
diff --git a/lib/tsan/rtl/tsan_rtl_thread.cc b/lib/tsan/rtl/tsan_rtl_thread.cc
index 67eebf5d0c38..83fab082afe3 100644
--- a/lib/tsan/rtl/tsan_rtl_thread.cc
+++ b/lib/tsan/rtl/tsan_rtl_thread.cc
@@ -142,6 +142,10 @@ void ThreadContext::OnFinished() {
 
   if (common_flags()->detect_deadlocks)
     ctx->dd->DestroyLogicalThread(thr->dd_lt);
+  thr->clock.ResetCached(&thr->proc()->clock_cache);
+#if !SANITIZER_GO
+  thr->last_sleep_clock.ResetCached(&thr->proc()->clock_cache);
+#endif
   thr->~ThreadState();
 #if TSAN_COLLECT_STATS
   StatAggregate(ctx->stat, thr->stat);
diff --git a/lib/tsan/rtl/tsan_stat.cc b/lib/tsan/rtl/tsan_stat.cc
index 2ee688bf5771..18c83d5c6dac 100644
--- a/lib/tsan/rtl/tsan_stat.cc
+++ b/lib/tsan/rtl/tsan_stat.cc
@@ -75,14 +75,11 @@ void StatOutput(u64 *stat) {
   name[StatClockAcquire]                 = "Clock acquire                     ";
   name[StatClockAcquireEmpty]            = "  empty clock                     ";
   name[StatClockAcquireFastRelease]      = "  fast from release-store         ";
-  name[StatClockAcquireLarge]            = "  contains my tid                 ";
-  name[StatClockAcquireRepeat]           = "  repeated (fast)                 ";
   name[StatClockAcquireFull]             = "  full (slow)                     ";
   name[StatClockAcquiredSomething]       = "  acquired something              ";
   name[StatClockRelease]                 = "Clock release                     ";
   name[StatClockReleaseResize]           = "  resize                          ";
-  name[StatClockReleaseFast1]            = "  fast1                           ";
-  name[StatClockReleaseFast2]            = "  fast2                           ";
+  name[StatClockReleaseFast]             = "  fast                            ";
   name[StatClockReleaseSlow]             = "  dirty overflow (slow)           ";
   name[StatClockReleaseFull]             = "  full (slow)                     ";
   name[StatClockReleaseAcquired]         = "  was acquired                    ";
diff --git a/lib/tsan/rtl/tsan_stat.h b/lib/tsan/rtl/tsan_stat.h
index 7d2791ebbfcc..42d6a2b63532 100644
--- a/lib/tsan/rtl/tsan_stat.h
+++ b/lib/tsan/rtl/tsan_stat.h
@@ -74,15 +74,12 @@ enum StatType {
   StatClockAcquire,
   StatClockAcquireEmpty,
   StatClockAcquireFastRelease,
-  StatClockAcquireLarge,
-  StatClockAcquireRepeat,
   StatClockAcquireFull,
   StatClockAcquiredSomething,
   // Clocks - release.
   StatClockRelease,
   StatClockReleaseResize,
-  StatClockReleaseFast1,
-  StatClockReleaseFast2,
+  StatClockReleaseFast,
   StatClockReleaseSlow,
   StatClockReleaseFull,
   StatClockReleaseAcquired,
diff --git a/lib/tsan/rtl/tsan_sync.cc b/lib/tsan/rtl/tsan_sync.cc
index 4cc3cb89c34f..44ae558fa1b2 100644
--- a/lib/tsan/rtl/tsan_sync.cc
+++ b/lib/tsan/rtl/tsan_sync.cc
@@ -53,7 +53,9 @@ void SyncVar::Reset(Processor *proc) {
   }
 }
 
-MetaMap::MetaMap() {
+MetaMap::MetaMap()
+    : block_alloc_("heap block allocator")
+    , sync_alloc_("sync allocator") {
   atomic_store(&uid_gen_, 0, memory_order_relaxed);
 }