Index: lib/lsan/lsan_interceptors.cc
===================================================================
--- lib/lsan/lsan_interceptors.cc
+++ lib/lsan/lsan_interceptors.cc
@@ -389,10 +389,10 @@
 
 INTERCEPTOR(int, pthread_join, void *th, void **ret) {
   ENSURE_LSAN_INITED;
-  int tid = ThreadTid((uptr)th);
+  ThreadContextBase *tctx = ThreadTctx((uptr)th);
   int res = REAL(pthread_join)(th, ret);
   if (res == 0)
-    ThreadJoin(tid);
+    ThreadJoin(tctx->tid);
   return res;
 }
 
Index: lib/lsan/lsan_thread.h
===================================================================
--- lib/lsan/lsan_thread.h
+++ lib/lsan/lsan_thread.h
@@ -49,7 +49,7 @@
 void ThreadFinish();
 u32 ThreadCreate(u32 tid, uptr uid, bool detached);
 void ThreadJoin(u32 tid);
-u32 ThreadTid(uptr uid);
+ThreadContextBase *ThreadTctx(uptr uid);
 
 u32 GetCurrentThread();
 void SetCurrentThread(u32 tid);
Index: lib/lsan/lsan_thread.cc
===================================================================
--- lib/lsan/lsan_thread.cc
+++ lib/lsan/lsan_thread.cc
@@ -111,7 +111,7 @@
   return false;
 }
 
-u32 ThreadTid(uptr uid) {
+ThreadContextBase *ThreadTctx(uptr uid) {
   return thread_registry->FindThread(FindThreadByUid, (void*)uid);
 }
 
Index: lib/sanitizer_common/sanitizer_thread_registry.h
===================================================================
--- lib/sanitizer_common/sanitizer_thread_registry.h
+++ lib/sanitizer_common/sanitizer_thread_registry.h
@@ -50,6 +50,8 @@
   u32 parent_tid;
   ThreadContextBase *next;  // For storing thread contexts in a list.
 
+  atomic_uint32_t thread_destroyed; // To address race of Joined vs Finished
+
   void SetName(const char *new_name);
 
   void SetDead();
@@ -60,6 +62,9 @@
                   u32 _parent_tid, void *arg);
   void Reset();
 
+  void SetDestroyed();
+  bool GetDestroyed();
+
   // The following methods may be overriden by subclasses.
   // Some of them take opaque arg that may be optionally be used
   // by subclasses.
@@ -102,9 +107,9 @@
   void RunCallbackForEachThreadLocked(ThreadCallback cb, void *arg);
 
   typedef bool (*FindThreadCallback)(ThreadContextBase *tctx, void *arg);
-  // Finds a thread using the provided callback. Returns kUnknownTid if no
+  // Finds a thread using the provided callback. Returns NULL if no
   // thread is found.
-  u32 FindThread(FindThreadCallback cb, void *arg);
+  ThreadContextBase *FindThread(FindThreadCallback cb, void *arg);
   // Should be guarded by ThreadRegistryLock. Return 0 if no thread
   // is found.
   ThreadContextBase *FindThreadContextLocked(FindThreadCallback cb,
Index: lib/sanitizer_common/sanitizer_thread_registry.cc
===================================================================
--- lib/sanitizer_common/sanitizer_thread_registry.cc
+++ lib/sanitizer_common/sanitizer_thread_registry.cc
@@ -21,6 +21,7 @@
       status(ThreadStatusInvalid),
       detached(false), workerthread(false), parent_tid(0), next(0) {
   name[0] = '\0';
+  atomic_store_relaxed(&thread_destroyed, 0);
 }
 
 ThreadContextBase::~ThreadContextBase() {
@@ -88,6 +89,14 @@
   OnReset();
 }
 
+void ThreadContextBase::SetDestroyed() {
+  atomic_store(&thread_destroyed, 1, memory_order_release);
+}
+
+bool ThreadContextBase::GetDestroyed() {
+  return !!atomic_load(&thread_destroyed, memory_order_acquire);
+}
+
 // ThreadRegistry implementation.
 
 const u32 ThreadRegistry::kUnknownTid = ~0U;
@@ -170,14 +179,15 @@
   }
 }
 
-u32 ThreadRegistry::FindThread(FindThreadCallback cb, void *arg) {
+ThreadContextBase *
+ThreadRegistry::FindThread(FindThreadCallback cb, void *arg) {
   BlockingMutexLock l(&mtx_);
   for (u32 tid = 0; tid < n_contexts_; tid++) {
     ThreadContextBase *tctx = threads_[tid];
     if (tctx != 0 && cb(tctx, arg))
-      return tctx->tid;
+      return tctx;
   }
-  return kUnknownTid;
+  return 0;
 }
 
 ThreadContextBase *
Index: lib/tsan/rtl/tsan_interceptors.cc
===================================================================
--- lib/tsan/rtl/tsan_interceptors.cc
+++ lib/tsan/rtl/tsan_interceptors.cc
@@ -868,6 +868,7 @@
   }
   DTLS_Destroy();
   cur_thread_finalize();
+  thr->tctx->SetDestroyed();
 }
 }  // namespace __tsan
 
@@ -982,12 +983,14 @@
 
 TSAN_INTERCEPTOR(int, pthread_join, void *th, void **ret) {
   SCOPED_INTERCEPTOR_RAW(pthread_join, th, ret);
-  int tid = ThreadTid(thr, pc, (uptr)th);
+  ThreadContextBase *tctx = ThreadTctx(thr, pc, (uptr)th);
   ThreadIgnoreBegin(thr, pc);
   int res = BLOCK_REAL(pthread_join)(th, ret);
   ThreadIgnoreEnd(thr, pc);
   if (res == 0) {
-    ThreadJoin(thr, pc, tid);
+    while (!tctx->GetDestroyed())
+       internal_sched_yield();
+    ThreadJoin(thr, pc, tctx->tid);
   }
   return res;
 }
@@ -996,10 +999,10 @@
 
 TSAN_INTERCEPTOR(int, pthread_detach, void *th) {
   SCOPED_TSAN_INTERCEPTOR(pthread_detach, th);
-  int tid = ThreadTid(thr, pc, (uptr)th);
+  ThreadContextBase *tctx = ThreadTctx(thr, pc, (uptr)th);
   int res = REAL(pthread_detach)(th);
   if (res == 0) {
-    ThreadDetach(thr, pc, tid);
+    ThreadDetach(thr, pc, tctx->tid);
   }
   return res;
 }
Index: lib/tsan/rtl/tsan_rtl.h
===================================================================
--- lib/tsan/rtl/tsan_rtl.h
+++ lib/tsan/rtl/tsan_rtl.h
@@ -757,7 +757,7 @@
 int ThreadCreate(ThreadState *thr, uptr pc, uptr uid, bool detached);
 void ThreadStart(ThreadState *thr, int tid, tid_t os_id, bool workerthread);
 void ThreadFinish(ThreadState *thr);
-int ThreadTid(ThreadState *thr, uptr pc, uptr uid);
+ThreadContextBase *ThreadTctx(ThreadState *thr, uptr pc, uptr uid);
 void ThreadJoin(ThreadState *thr, uptr pc, int tid);
 void ThreadDetach(ThreadState *thr, uptr pc, int tid);
 void ThreadFinalize(ThreadState *thr);
Index: lib/tsan/rtl/tsan_rtl_thread.cc
===================================================================
--- lib/tsan/rtl/tsan_rtl_thread.cc
+++ lib/tsan/rtl/tsan_rtl_thread.cc
@@ -293,9 +293,10 @@
   return false;
 }
 
-int ThreadTid(ThreadState *thr, uptr pc, uptr uid) {
-  int res = ctx->thread_registry->FindThread(FindThreadByUid, (void*)uid);
-  DPrintf("#%d: ThreadTid uid=%zu tid=%d\n", thr->tid, uid, res);
+ThreadContextBase *ThreadTctx(ThreadState *thr, uptr pc, uptr uid) {
+  ThreadContextBase *res;
+  res = ctx->thread_registry->FindThread(FindThreadByUid, (void*)uid);
+  DPrintf("#%d: ThreadTid uid=%zu tid=%d\n", thr->tid, uid, res->tid);
   return res;
 }