diff --git a/src/native/java_lang_Thread.cc b/src/native/java_lang_Thread.cc
index 86b3a20..626255e 100644
--- a/src/native/java_lang_Thread.cc
+++ b/src/native/java_lang_Thread.cc
@@ -42,8 +42,9 @@
   return (thread != NULL) ? thread->IsInterrupted() : JNI_FALSE;
 }
 
-static void Thread_nativeCreate(JNIEnv* env, jclass, jobject java_thread, jlong stack_size) {
-  Thread::CreateNativeThread(env, java_thread, stack_size);
+static void Thread_nativeCreate(JNIEnv* env, jclass, jobject java_thread, jlong stack_size,
+                                jboolean daemon) {
+  Thread::CreateNativeThread(env, java_thread, stack_size, daemon == JNI_TRUE);
 }
 
 static jint Thread_nativeGetStatus(JNIEnv* env, jobject java_thread, jboolean has_been_started) {
@@ -140,7 +141,7 @@
   NATIVE_METHOD(Thread, currentThread, "()Ljava/lang/Thread;"),
   NATIVE_METHOD(Thread, interrupted, "()Z"),
   NATIVE_METHOD(Thread, isInterrupted, "()Z"),
-  NATIVE_METHOD(Thread, nativeCreate, "(Ljava/lang/Thread;J)V"),
+  NATIVE_METHOD(Thread, nativeCreate, "(Ljava/lang/Thread;JZ)V"),
   NATIVE_METHOD(Thread, nativeGetStatus, "(Z)I"),
   NATIVE_METHOD(Thread, nativeHoldsLock, "(Ljava/lang/Object;)Z"),
   NATIVE_METHOD(Thread, nativeInterrupt, "()V"),
diff --git a/src/scoped_jni_thread_state.h b/src/scoped_jni_thread_state.h
index 42ed19c..1c9ab2c 100644
--- a/src/scoped_jni_thread_state.h
+++ b/src/scoped_jni_thread_state.h
@@ -34,22 +34,24 @@
 class ScopedJniThreadState {
  public:
   explicit ScopedJniThreadState(JNIEnv* env, ThreadState new_state = kRunnable)
-    : env_(reinterpret_cast<JNIEnvExt*>(env)), vm_(env_->vm), self_(ThreadForEnv(env)),
-      old_thread_state_(self_->SetState(new_state)), thread_state_(new_state) {
+      : env_(reinterpret_cast<JNIEnvExt*>(env)), vm_(env_->vm), self_(ThreadForEnv(env)),
+        old_thread_state_(self_->SetState(new_state)), thread_state_(new_state) {
     self_->VerifyStack();
   }
 
   explicit ScopedJniThreadState(Thread* self, ThreadState new_state = kRunnable)
-    : env_(reinterpret_cast<JNIEnvExt*>(self->GetJniEnv())), vm_(env_->vm), self_(self),
-      old_thread_state_(self_->SetState(new_state)), thread_state_(new_state) {
+      : env_(reinterpret_cast<JNIEnvExt*>(self->GetJniEnv())), vm_(env_->vm), self_(self),
+        old_thread_state_(self_->SetState(new_state)), thread_state_(new_state) {
+    if (!Vm()->work_around_app_jni_bugs && self != Thread::Current()) {
+      UnexpectedThreads(self, Thread::Current());
+    }
     self_->VerifyStack();
   }
 
-  // Used when we want a scoped jni thread state but have no thread/JNIEnv.
+  // Used when we want a scoped JNI thread state but have no thread/JNIEnv.
   explicit ScopedJniThreadState(JavaVM* vm)
-    : env_(NULL), vm_(reinterpret_cast<JavaVMExt*>(vm)), self_(NULL),
-      old_thread_state_(kTerminated), thread_state_(kTerminated) {
-  }
+      : env_(NULL), vm_(reinterpret_cast<JavaVMExt*>(vm)), self_(NULL),
+        old_thread_state_(kTerminated), thread_state_(kTerminated) {}
 
   ~ScopedJniThreadState() {
     if (self_ != NULL) {
@@ -164,13 +166,19 @@
     Thread* env_self = full_env->self;
     Thread* self = work_around_app_jni_bugs ? Thread::Current() : env_self;
     if (!work_around_app_jni_bugs && self != env_self) {
-      // TODO: pass through function name so we can use it here instead of NULL...
-      JniAbortF(NULL, "JNIEnv for %s used on %s",
-                ToStr<Thread>(*env_self).c_str(), ToStr<Thread>(*self).c_str());
+      UnexpectedThreads(env_self, self);
     }
     return self;
   }
 
+  static void UnexpectedThreads(Thread* found_self, Thread* expected_self) {
+    // TODO: pass through function name so we can use it here instead of NULL...
+    JniAbortF(NULL, "JNIEnv for %s used on %s",
+             found_self != NULL ? ToStr<Thread>(*found_self).c_str() : "NULL",
+             expected_self != NULL ? ToStr<Thread>(*expected_self).c_str() : "NULL");
+
+  }
+
   // The full JNIEnv.
   JNIEnvExt* const env_;
   // The full JavaVM.
diff --git a/src/thread.cc b/src/thread.cc
index 8fe0160..104b9b9 100644
--- a/src/thread.cc
+++ b/src/thread.cc
@@ -207,8 +207,8 @@
   delete[] allocated_signal_stack;
 }
 
-void Thread::CreateNativeThread(JNIEnv* env, jobject java_peer, size_t stack_size) {
-  Thread* native_thread = new Thread;
+void Thread::CreateNativeThread(JNIEnv* env, jobject java_peer, size_t stack_size, bool daemon) {
+  Thread* native_thread = new Thread(daemon);
   {
     ScopedJniThreadState ts(env);
     Object* peer = ts.Decode<Object*>(java_peer);
@@ -277,7 +277,7 @@
 }
 
 Thread* Thread::Attach(const char* thread_name, bool as_daemon, jobject thread_group) {
-  Thread* self = new Thread;
+  Thread* self = new Thread(as_daemon);
   self->Init();
 
   self->SetState(kNative);
@@ -599,12 +599,20 @@
 }
 
 void Thread::SetStateWithoutSuspendCheck(ThreadState new_state) {
+  DCHECK_EQ(this, Thread::Current());
   volatile void* raw = reinterpret_cast<volatile void*>(&state_);
   volatile int32_t* addr = reinterpret_cast<volatile int32_t*>(raw);
   android_atomic_release_store(new_state, addr);
 }
 
 ThreadState Thread::SetState(ThreadState new_state) {
+  if (new_state != kVmWait && new_state != kTerminated) {
+    // TODO: kVmWait is set by the parent thread to a child thread to indicate it can go. Similarly
+    // kTerminated may be set by a parent thread to its child if pthread creation fails.  This
+    // overloaded use of the state variable means we cannot fully assert that only threads
+    // themselves modify their state.
+    DCHECK_EQ(this, Thread::Current());
+  }
   ThreadState old_state = state_;
   if (old_state == kRunnable) {
     // Non-runnable states are points where we expect thread suspension can occur.
@@ -764,7 +772,7 @@
   CHECK_PTHREAD_CALL(pthread_key_delete, (Thread::pthread_key_self_), "self key");
 }
 
-Thread::Thread()
+Thread::Thread(bool daemon)
     : suspend_count_(0),
       card_table_(NULL),
       exception_(NULL),
@@ -793,7 +801,9 @@
       debug_invoke_req_(new DebugInvokeReq),
       trace_stack_(new std::vector<TraceStackFrame>),
       name_(new std::string(kThreadNameDuringStartup)),
-      no_thread_suspension_(0) {
+      daemon_(daemon),
+      no_thread_suspension_(0),
+      last_no_thread_suspension_cause_(NULL) {
   CHECK_EQ((sizeof(Thread) % 4), 0U) << sizeof(Thread);
   memset(&held_mutexes_[0], 0, sizeof(held_mutexes_));
 }
@@ -1065,14 +1075,18 @@
     // object graph.
     method_trace->Set(depth, dex_pc_trace);
     // Set the Object*s and assert that no thread suspension is now possible.
-    ts.Self()->StartAssertNoThreadSuspension();
+    const char* last_no_suspend_cause =
+        ts.Self()->StartAssertNoThreadSuspension("Building internal stack trace");
+    CHECK(last_no_suspend_cause == NULL) << last_no_suspend_cause;
     method_trace_ = method_trace.get();
     dex_pc_trace_ = dex_pc_trace;
     return true;
   }
 
   virtual ~BuildInternalStackTraceVisitor() {
-    Thread::Current()->EndAssertNoThreadSuspension();
+    if (method_trace_ != NULL) {
+      Thread::Current()->EndAssertNoThreadSuspension(NULL);
+    }
   }
 
   bool VisitFrame() {
@@ -1436,7 +1450,12 @@
         throw_frame_id_(0), throw_dex_pc_(0), handler_quick_frame_(NULL),
         handler_quick_frame_pc_(0), handler_dex_pc_(0), native_method_count_(0),
         method_tracing_active_(Runtime::Current()->IsMethodTracingActive()) {
-    self->StartAssertNoThreadSuspension();  // Exception not in root sets, can't allow GC.
+    // Exception not in root sets, can't allow GC.
+    last_no_assert_suspension_cause_ = self->StartAssertNoThreadSuspension("Finding catch block");
+  }
+
+  ~CatchBlockStackVisitor() {
+    LOG(FATAL) << "UNREACHABLE";  // Expected to take long jump.
   }
 
   bool VisitFrame() {
@@ -1496,8 +1515,8 @@
         LOG(INFO) << "Handler: " << PrettyMethod(catch_method) << " (line: " << line_number << ")";
       }
     }
-    self_->SetException(exception_);
-    self_->EndAssertNoThreadSuspension();  // Exception back in root set.
+    self_->SetException(exception_);  // Exception back in root set.
+    self_->EndAssertNoThreadSuspension(last_no_assert_suspension_cause_);
     // Place context back on thread so it will be available when we continue.
     self_->ReleaseLongJumpContext(context_);
     context_->SetSP(reinterpret_cast<uintptr_t>(handler_quick_frame_));
@@ -1525,6 +1544,8 @@
   uint32_t native_method_count_;
   // Is method tracing active?
   const bool method_tracing_active_;
+  // Support for nesting no thread suspension checks.
+  const char* last_no_assert_suspension_cause_;
 };
 
 void Thread::DeliverException() {
@@ -1595,11 +1616,6 @@
   return object->GetThinLockId() == thin_lock_id_;
 }
 
-bool Thread::IsDaemon() {
-  ScopedJniThreadState ts(this);
-  return ts.DecodeField(WellKnownClasses::java_lang_Thread_daemon)->GetBoolean(peer_);
-}
-
 class ReferenceMapVisitor : public StackVisitor {
  public:
   ReferenceMapVisitor(const ManagedStack* stack, const std::vector<TraceStackFrame>* trace_stack,
diff --git a/src/thread.h b/src/thread.h
index 7cd55a3..5e60637 100644
--- a/src/thread.h
+++ b/src/thread.h
@@ -94,7 +94,7 @@
 
   // Creates a new native thread corresponding to the given managed peer.
   // Used to implement Thread.start.
-  static void CreateNativeThread(JNIEnv* env, jobject peer, size_t stack_size);
+  static void CreateNativeThread(JNIEnv* env, jobject peer, size_t stack_size, bool daemon);
 
   // Attaches the calling native thread to the runtime, returning the new native peer.
   // Used to implement JNI AttachCurrentThread and AttachCurrentThreadAsDaemon calls.
@@ -131,27 +131,45 @@
   ThreadState SetState(ThreadState new_state);
   void SetStateWithoutSuspendCheck(ThreadState new_state);
 
-  bool IsDaemon();
+  bool IsDaemon() const {
+    return daemon_;
+  }
+
   bool IsSuspended();
 
   void WaitUntilSuspended();
 
   // Once called thread suspension will cause an assertion failure.
-  void StartAssertNoThreadSuspension() {
 #ifndef NDEBUG
+  const char* StartAssertNoThreadSuspension(const char* cause) {
+    CHECK(cause != NULL);
+    const char* previous_cause = last_no_thread_suspension_cause_;
     no_thread_suspension_++;
-#endif
+    last_no_thread_suspension_cause_ = cause;
+    return previous_cause;
   }
+#else
+  const char* StartAssertNoThreadSuspension(const char* cause) {
+    CHECK(cause != NULL);
+    return NULL;
+  }
+#endif
+
   // End region where no thread suspension is expected.
-  void EndAssertNoThreadSuspension() {
 #ifndef NDEBUG
-    DCHECK_GT(no_thread_suspension_, 0U);
+  void EndAssertNoThreadSuspension(const char* old_cause) {
+    CHECK(old_cause != NULL || no_thread_suspension_ == 1);
+    CHECK_GT(no_thread_suspension_, 0U);
     no_thread_suspension_--;
-#endif
+    last_no_thread_suspension_cause_ = old_cause;
   }
+#else
+  void EndAssertNoThreadSuspension(const char*) {
+  }
+#endif
 
   void AssertThreadSuspensionIsAllowable() const {
-    DCHECK_EQ(0u, no_thread_suspension_);
+    DCHECK_EQ(0u, no_thread_suspension_) << last_no_thread_suspension_cause_;
   }
 
   bool CanAccessDirectReferences() const {
@@ -494,7 +512,7 @@
   void CheckSafeToWait(MutexRank rank);
 
  private:
-  Thread();
+  explicit Thread(bool daemon);
   ~Thread();
   void Destroy();
   friend class ThreadList;  // For ~Thread and Destroy.
@@ -626,6 +644,12 @@
   // A cached copy of the java.lang.Thread's name.
   std::string* name_;
 
+  // Is the thread a daemon?
+  const bool daemon_;
+
+  // Keep data fields within Thread 4 byte aligned.
+  byte pad_[3];
+
   // A cached pthread_t for the pthread underlying this Thread*.
   pthread_t pthread_self_;
 
@@ -634,6 +658,9 @@
 
   // A positive value implies we're in a region where thread suspension isn't expected.
   uint32_t no_thread_suspension_;
+
+  // Cause for last suspension.
+  const char* last_no_thread_suspension_cause_;
  public:
   // Runtime support function pointers
   EntryPoints entrypoints_;