diff --git a/src/debugger.cc b/src/debugger.cc
index e0e02cf..672b660 100644
--- a/src/debugger.cc
+++ b/src/debugger.cc
@@ -1497,7 +1497,7 @@
         // query all threads, so it's easier if we just don't tell them about this thread.
         return;
       }
-      Object* peer = soa_.Decode<Object*>(t->GetPeer());
+      Object* peer = t->GetPeer();
       if (IsInDesiredThreadGroup(peer)) {
         thread_ids_.push_back(gRegistry->Add(peer));
       }
@@ -1632,7 +1632,7 @@
 
 JDWP::ObjectId Dbg::GetThreadSelfId() {
   ScopedObjectAccessUnchecked soa(Thread::Current());
-  return gRegistry->Add(soa.Decode<Object*>(Thread::Current()->GetPeer()));
+  return gRegistry->Add(soa.Self()->GetPeer());
 }
 
 void Dbg::SuspendVM() {
@@ -2755,7 +2755,7 @@
 void Dbg::PostThreadStartOrStop(Thread* t, uint32_t type) {
   if (IsDebuggerActive()) {
     ScopedObjectAccessUnchecked soa(Thread::Current());
-    JDWP::ObjectId id = gRegistry->Add(soa.Decode<Object*>(t->GetPeer()));
+    JDWP::ObjectId id = gRegistry->Add(t->GetPeer());
     gJdwpState->PostThreadChange(id, type == CHUNK_TYPE("THCR"));
     // If this thread's just joined the party while we're already debugging, make sure it knows
     // to give us updates when it's running.
diff --git a/src/native/dalvik_system_VMStack.cc b/src/native/dalvik_system_VMStack.cc
index 5ef512a..0e6e675 100644
--- a/src/native/dalvik_system_VMStack.cc
+++ b/src/native/dalvik_system_VMStack.cc
@@ -25,10 +25,11 @@
 
 static jobject GetThreadStack(JNIEnv* env, jobject peer) {
   bool timeout;
-  Thread* self = Thread::Current();
-  if (env->IsSameObject(peer, self->GetPeer())) {
+  {
     ScopedObjectAccess soa(env);
-    return self->CreateInternalStackTrace(soa);
+    if (soa.Decode<Object*>(peer) == soa.Self()->GetPeer()) {
+      return soa.Self()->CreateInternalStackTrace(soa);
+    }
   }
   // Suspend thread to build stack trace.
   Thread* thread = Thread::SuspendForDebugger(peer, true, &timeout);
diff --git a/src/native/java_lang_Thread.cc b/src/native/java_lang_Thread.cc
index cf475e2..f14c03b 100644
--- a/src/native/java_lang_Thread.cc
+++ b/src/native/java_lang_Thread.cc
@@ -25,11 +25,12 @@
 namespace art {
 
 static jobject Thread_currentThread(JNIEnv* env, jclass) {
-  return reinterpret_cast<JNIEnvExt*>(env)->self->GetPeer();
+  ScopedObjectAccess soa(env);
+  return soa.AddLocalReference<jobject>(soa.Self()->GetPeer());
 }
 
 static jboolean Thread_interrupted(JNIEnv* env, jclass) {
-  return reinterpret_cast<JNIEnvExt*>(env)->self->Interrupted() ? JNI_TRUE : JNI_FALSE;
+  return static_cast<JNIEnvExt*>(env)->self->Interrupted() ? JNI_TRUE : JNI_FALSE;
 }
 
 static jboolean Thread_isInterrupted(JNIEnv* env, jobject java_thread) {
@@ -108,7 +109,7 @@
   ScopedUtfChars name(env, java_name);
   {
     ScopedObjectAccess soa(env);
-    if (soa.Env()->IsSameObject(peer, soa.Self()->GetPeer())) {
+    if (soa.Decode<Object*>(peer) == soa.Self()->GetPeer()) {
       soa.Self()->SetThreadName(name.c_str());
       return;
     }
diff --git a/src/native/org_apache_harmony_dalvik_ddmc_DdmVmInternal.cc b/src/native/org_apache_harmony_dalvik_ddmc_DdmVmInternal.cc
index 0ed964b..9423795 100644
--- a/src/native/org_apache_harmony_dalvik_ddmc_DdmVmInternal.cc
+++ b/src/native/org_apache_harmony_dalvik_ddmc_DdmVmInternal.cc
@@ -61,7 +61,8 @@
     Runtime::Current()->GetThreadList()->ForEach(ThreadFinder::Callback, &finder);
   }
   if (finder.thread != NULL) {
-    return finder.thread->GetPeer();
+    ScopedObjectAccess soa(env);
+    return soa.AddLocalReference<jobject>(finder.thread->GetPeer());
   } else {
     return NULL;
   }
diff --git a/src/runtime.cc b/src/runtime.cc
index ec2f569..8d88270 100644
--- a/src/runtime.cc
+++ b/src/runtime.cc
@@ -644,7 +644,7 @@
                                                                       "Ljava/lang/ClassLoader;");
   CHECK(contextClassLoader != NULL);
 
-  contextClassLoader->SetObject(soa.Decode<Object*>(soa.Self()->GetPeer()), class_loader);
+  contextClassLoader->SetObject(soa.Self()->GetPeer(), class_loader);
 }
 
 void Runtime::Start() {
diff --git a/src/scoped_thread_state_change.h b/src/scoped_thread_state_change.h
index 39f5c3f..c0fb649 100644
--- a/src/scoped_thread_state_change.h
+++ b/src/scoped_thread_state_change.h
@@ -165,8 +165,7 @@
    * passed in), or NULL on failure.
    */
   template<typename T>
-  T AddLocalReference(Object* obj) const
-      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+  T AddLocalReference(Object* obj) const SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
     DCHECK_EQ(thread_state_, kRunnable);  // Don't work with raw objects in non-runnable states.
     if (obj == NULL) {
       return NULL;
diff --git a/src/thread.cc b/src/thread.cc
index c51d45f..72ceaf0 100644
--- a/src/thread.cc
+++ b/src/thread.cc
@@ -121,6 +121,13 @@
   }
   {
     ScopedObjectAccess soa(self);
+
+    // Copy peer into self, deleting global reference when done.
+    CHECK(self->jpeer_ != NULL);
+    self->opeer_ = soa.Decode<Object*>(self->jpeer_);
+    self->GetJniEnv()->DeleteGlobalRef(self->jpeer_);
+    self->jpeer_ = NULL;
+
     {
       SirtRef<String> thread_name(self, self->GetThreadName(soa));
       self->SetThreadName(thread_name->ToModifiedUtf8().c_str());
@@ -128,10 +135,10 @@
     Dbg::PostThreadStart(self);
 
     // Invoke the 'run' method of our java.lang.Thread.
-    CHECK(self->peer_ != NULL);
-    Object* receiver = soa.Decode<Object*>(self->peer_);
+    Object* receiver = self->opeer_;
     jmethodID mid = WellKnownClasses::java_lang_Thread_run;
-    AbstractMethod* m = receiver->GetClass()->FindVirtualMethodForVirtualOrInterface(soa.DecodeMethod(mid));
+    AbstractMethod* m =
+        receiver->GetClass()->FindVirtualMethodForVirtualOrInterface(soa.DecodeMethod(mid));
     m->Invoke(self, receiver, NULL, NULL);
   }
   // Detach and delete self.
@@ -244,7 +251,7 @@
 
   Thread* child_thread = new Thread(is_daemon);
   // Use global JNI ref to hold peer live while child thread starts.
-  child_thread->peer_ = env->NewGlobalRef(java_peer);
+  child_thread->jpeer_ = env->NewGlobalRef(java_peer);
   stack_size = FixStackSize(stack_size);
 
   // Thread.start is synchronized, so we know that vmData is 0, and know that we're not racing to
@@ -267,8 +274,8 @@
       runtime->EndThreadBirth();
     }
     // Manually delete the global reference since Thread::Init will not have been run.
-    env->DeleteGlobalRef(child_thread->peer_);
-    child_thread->peer_ = NULL;
+    env->DeleteGlobalRef(child_thread->jpeer_);
+    child_thread->jpeer_ = NULL;
     delete child_thread;
     child_thread = NULL;
     // TODO: remove from thread group?
@@ -302,7 +309,7 @@
   CHECK_PTHREAD_CALL(pthread_setspecific, (Thread::pthread_key_self_, this), "attach self");
   DCHECK_EQ(Thread::Current(), this);
 
-  thin_lock_id_ = thread_list->AllocThreadId();
+  thin_lock_id_ = thread_list->AllocThreadId(this);
   InitStackHwm();
 
   jni_env_ = new JNIEnvExt(this, java_vm);
@@ -367,7 +374,10 @@
     CHECK(IsExceptionPending());
     return;
   }
-  peer_ = env->NewGlobalRef(peer.get());
+  {
+    ScopedObjectAccess soa(this);
+    opeer_ = soa.Decode<Object*>(peer.get());
+  }
   env->CallNonvirtualVoidMethod(peer.get(),
                                 WellKnownClasses::java_lang_Thread,
                                 WellKnownClasses::java_lang_Thread_init,
@@ -382,19 +392,18 @@
   ScopedObjectAccess soa(self);
   SirtRef<String> peer_thread_name(soa.Self(), GetThreadName(soa));
   if (peer_thread_name.get() == NULL) {
-    Object* native_peer = soa.Decode<Object*>(peer.get());
     // The Thread constructor should have set the Thread.name to a
     // non-null value. However, because we can run without code
     // available (in the compiler, in tests), we manually assign the
     // fields the constructor should have set.
     soa.DecodeField(WellKnownClasses::java_lang_Thread_daemon)->
-        SetBoolean(native_peer, thread_is_daemon);
+        SetBoolean(opeer_, thread_is_daemon);
     soa.DecodeField(WellKnownClasses::java_lang_Thread_group)->
-        SetObject(native_peer, soa.Decode<Object*>(thread_group));
+        SetObject(opeer_, soa.Decode<Object*>(thread_group));
     soa.DecodeField(WellKnownClasses::java_lang_Thread_name)->
-        SetObject(native_peer, soa.Decode<Object*>(thread_name.get()));
+        SetObject(opeer_, soa.Decode<Object*>(thread_name.get()));
     soa.DecodeField(WellKnownClasses::java_lang_Thread_priority)->
-        SetInt(native_peer, thread_priority);
+        SetInt(opeer_, thread_priority);
     peer_thread_name.reset(GetThreadName(soa));
   }
   // 'thread_name' may have been null, so don't trust 'peer_thread_name' to be non-null.
@@ -472,7 +481,7 @@
   }
   os << GetState()
            << ",Thread*=" << this
-           << ",peer=" << peer_
+           << ",peer=" << opeer_
            << ",\"" << *name_ << "\""
            << "]";
 }
@@ -484,8 +493,7 @@
 
 String* Thread::GetThreadName(const ScopedObjectAccessUnchecked& soa) const {
   Field* f = soa.DecodeField(WellKnownClasses::java_lang_Thread_name);
-  Object* native_peer = soa.Decode<Object*>(peer_);
-  return (peer_ != NULL) ? reinterpret_cast<String*>(f->GetObject(native_peer)) : NULL;
+  return (opeer_ != NULL) ? reinterpret_cast<String*>(f->GetObject(opeer_)) : NULL;
 }
 
 void Thread::GetThreadName(std::string& name) const {
@@ -536,7 +544,9 @@
       << delta << " " << debug_suspend_count_ << " " << this;
   DCHECK_GE(suspend_count_, debug_suspend_count_) << this;
   Locks::thread_suspend_count_lock_->AssertHeld(self);
-
+  if (this != self && !IsSuspended()) {
+    Locks::thread_list_lock_->AssertHeld(self);
+  }
   if (UNLIKELY(delta < 0 && suspend_count_ <= 0)) {
     UnsafeLogFatalForSuspendCount(self, this);
     return;
@@ -718,14 +728,13 @@
   bool is_daemon = false;
   Thread* self = Thread::Current();
 
-  if (thread != NULL && thread->peer_ != NULL) {
-    ScopedObjectAccess soa(self);
-    Object* native_peer = soa.Decode<Object*>(thread->peer_);
-    priority = soa.DecodeField(WellKnownClasses::java_lang_Thread_priority)->GetInt(native_peer);
-    is_daemon = soa.DecodeField(WellKnownClasses::java_lang_Thread_daemon)->GetBoolean(native_peer);
+  if (thread != NULL && thread->opeer_ != NULL) {
+    ScopedObjectAccessUnchecked soa(self);
+    priority = soa.DecodeField(WellKnownClasses::java_lang_Thread_priority)->GetInt(thread->opeer_);
+    is_daemon = soa.DecodeField(WellKnownClasses::java_lang_Thread_daemon)->GetBoolean(thread->opeer_);
 
     Object* thread_group =
-        soa.DecodeField(WellKnownClasses::java_lang_Thread_group)->GetObject(native_peer);
+        soa.DecodeField(WellKnownClasses::java_lang_Thread_group)->GetObject(thread->opeer_);
 
     if (thread_group != NULL) {
       Field* group_name_field = soa.DecodeField(WellKnownClasses::java_lang_ThreadGroup_name);
@@ -760,7 +769,7 @@
     os << "  | group=\"" << group_name << "\""
        << " sCount=" << thread->suspend_count_
        << " dsCount=" << thread->debug_suspend_count_
-       << " obj=" << reinterpret_cast<void*>(thread->peer_)
+       << " obj=" << reinterpret_cast<void*>(thread->opeer_)
        << " self=" << reinterpret_cast<const void*>(thread) << "\n";
   }
 
@@ -936,7 +945,8 @@
       managed_stack_(),
       jni_env_(NULL),
       self_(NULL),
-      peer_(NULL),
+      opeer_(NULL),
+      jpeer_(NULL),
       stack_begin_(NULL),
       stack_size_(0),
       thin_lock_id_(0),
@@ -1002,29 +1012,24 @@
   Thread* self = this;
   DCHECK_EQ(self, Thread::Current());
 
-  if (peer_ != NULL) {
+  if (opeer_ != NULL) {
+    ScopedObjectAccess soa(self);
     // We may need to call user-supplied managed code, do this before final clean-up.
-    HandleUncaughtExceptions();
-    RemoveFromThreadGroup();
+    HandleUncaughtExceptions(soa);
+    RemoveFromThreadGroup(soa);
 
     // this.vmData = 0;
-    jni_env_->SetIntField(peer_, WellKnownClasses::java_lang_Thread_vmData, 0);
+    soa.DecodeField(WellKnownClasses::java_lang_Thread_vmData)->SetInt(opeer_, 0);
+    Dbg::PostThreadDeath(self);
 
-    {
-      ScopedObjectAccess soa(self);
-      Dbg::PostThreadDeath(self);
-    }
-
-    // Thread.join() is implemented as an Object.wait() on the Thread.lock
-    // object. Signal anyone who is waiting.
-    ScopedLocalRef<jobject> lock(jni_env_,
-                                 jni_env_->GetObjectField(peer_,
-                                                          WellKnownClasses::java_lang_Thread_lock));
+    // Thread.join() is implemented as an Object.wait() on the Thread.lock object. Signal anyone
+    // who is waiting.
+    Object* lock = soa.DecodeField(WellKnownClasses::java_lang_Thread_lock)->GetObject(opeer_);
     // (This conditional is only needed for tests, where Thread.lock won't have been set.)
-    if (lock.get() != NULL) {
-      jni_env_->MonitorEnter(lock.get());
-      jni_env_->CallVoidMethod(lock.get(), WellKnownClasses::java_lang_Object_notify);
-      jni_env_->MonitorExit(lock.get());
+    if (lock != NULL) {
+      lock->MonitorEnter(self);
+      lock->Notify();
+      lock->MonitorExit(self);
     }
   }
 
@@ -1035,11 +1040,12 @@
 }
 
 Thread::~Thread() {
-  if (jni_env_ != NULL && peer_ != NULL) {
+  if (jni_env_ != NULL && jpeer_ != NULL) {
     // If pthread_create fails we don't have a jni env here.
-    jni_env_->DeleteGlobalRef(peer_);
+    jni_env_->DeleteGlobalRef(jpeer_);
+    jpeer_ = NULL;
   }
-  peer_ = NULL;
+  opeer_ = NULL;
 
   delete jni_env_;
   jni_env_ = NULL;
@@ -1062,10 +1068,12 @@
   TearDownAlternateSignalStack();
 }
 
-void Thread::HandleUncaughtExceptions() {
+void Thread::HandleUncaughtExceptions(ScopedObjectAccess& soa) {
   if (!IsExceptionPending()) {
     return;
   }
+  ScopedLocalRef<jobject> peer(jni_env_, soa.AddLocalReference<jobject>(opeer_));
+  ScopedThreadStateChange tsc(this, kNative);
 
   // Get and clear the exception.
   ScopedLocalRef<jthrowable> exception(jni_env_, jni_env_->ExceptionOccurred());
@@ -1073,31 +1081,32 @@
 
   // If the thread has its own handler, use that.
   ScopedLocalRef<jobject> handler(jni_env_,
-                                  jni_env_->GetObjectField(peer_,
+                                  jni_env_->GetObjectField(peer.get(),
                                                            WellKnownClasses::java_lang_Thread_uncaughtHandler));
   if (handler.get() == NULL) {
     // Otherwise use the thread group's default handler.
-    handler.reset(jni_env_->GetObjectField(peer_, WellKnownClasses::java_lang_Thread_group));
+    handler.reset(jni_env_->GetObjectField(peer.get(), WellKnownClasses::java_lang_Thread_group));
   }
 
   // Call the handler.
   jni_env_->CallVoidMethod(handler.get(),
                            WellKnownClasses::java_lang_Thread$UncaughtExceptionHandler_uncaughtException,
-                           peer_, exception.get());
+                           peer.get(), exception.get());
 
   // If the handler threw, clear that exception too.
   jni_env_->ExceptionClear();
 }
 
-void Thread::RemoveFromThreadGroup() {
+void Thread::RemoveFromThreadGroup(ScopedObjectAccess& soa) {
   // this.group.removeThread(this);
   // group can be null if we're in the compiler or a test.
-  ScopedLocalRef<jobject> group(jni_env_,
-                                jni_env_->GetObjectField(peer_,
-                                                         WellKnownClasses::java_lang_Thread_group));
-  if (group.get() != NULL) {
+  Object* ogroup = soa.DecodeField(WellKnownClasses::java_lang_Thread_group)->GetObject(opeer_);
+  if (ogroup != NULL) {
+    ScopedLocalRef<jobject> group(soa.Env(), soa.AddLocalReference<jobject>(ogroup));
+    ScopedLocalRef<jobject> peer(soa.Env(), soa.AddLocalReference<jobject>(opeer_));
+    ScopedThreadStateChange tsc(soa.Self(), kNative);
     jni_env_->CallVoidMethod(group.get(), WellKnownClasses::java_lang_ThreadGroup_removeThread,
-                             peer_);
+                             peer.get());
   }
 }
 
@@ -1109,7 +1118,7 @@
   return count;
 }
 
-bool Thread::SirtContains(jobject obj) {
+bool Thread::SirtContains(jobject obj) const {
   Object** sirt_entry = reinterpret_cast<Object**>(obj);
   for (StackIndirectReferenceTable* cur = top_sirt_; cur; cur = cur->GetLink()) {
     if (cur->Contains(sirt_entry)) {
@@ -1132,7 +1141,7 @@
   }
 }
 
-Object* Thread::DecodeJObject(jobject obj) {
+Object* Thread::DecodeJObject(jobject obj) const {
   Locks::mutator_lock_->AssertSharedHeld(this);
   if (obj == NULL) {
     return NULL;
@@ -1151,7 +1160,7 @@
     {
       JavaVMExt* vm = Runtime::Current()->GetJavaVM();
       IndirectReferenceTable& globals = vm->globals;
-      MutexLock mu(this, vm->globals_lock);
+      MutexLock mu(const_cast<Thread*>(this), vm->globals_lock);
       result = const_cast<Object*>(globals.Get(ref));
       break;
     }
@@ -1159,7 +1168,7 @@
     {
       JavaVMExt* vm = Runtime::Current()->GetJavaVM();
       IndirectReferenceTable& weak_globals = vm->weak_globals;
-      MutexLock mu(this, vm->weak_globals_lock);
+      MutexLock mu(const_cast<Thread*>(this), vm->weak_globals_lock);
       result = const_cast<Object*>(weak_globals.Get(ref));
       if (result == kClearedJniWeakGlobal) {
         // This is a special case where it's okay to return NULL.
@@ -1967,6 +1976,9 @@
   wrapperArg.arg = arg;
   wrapperArg.visitor = visitor;
 
+  if (opeer_ != NULL) {
+    VerifyRootWrapperCallback(opeer_, &wrapperArg);
+  }
   if (exception_ != NULL) {
     VerifyRootWrapperCallback(exception_, &wrapperArg);
   }
@@ -1988,6 +2000,9 @@
 }
 
 void Thread::VisitRoots(Heap::RootVisitor* visitor, void* arg) {
+  if (opeer_ != NULL) {
+    visitor(opeer_, arg);
+  }
   if (exception_ != NULL) {
     visitor(exception_, arg);
   }
diff --git a/src/thread.h b/src/thread.h
index b56bcf1..4c065c5 100644
--- a/src/thread.h
+++ b/src/thread.h
@@ -148,7 +148,8 @@
   // Dumps the SIGQUIT per-thread header. 'thread' can be NULL for a non-attached thread, in which
   // case we use 'tid' to identify the thread, and we'll include as much information as we can.
   static void DumpState(std::ostream& os, const Thread* thread, pid_t tid)
-      LOCKS_EXCLUDED(Locks::thread_suspend_count_lock_);
+      LOCKS_EXCLUDED(Locks::thread_suspend_count_lock_)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   ThreadState GetState() const {
     return static_cast<ThreadState>(state_and_flags_.as_struct.state);
@@ -279,12 +280,14 @@
   // Sets the thread's name.
   void SetThreadName(const char* name) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
-  jobject GetPeer() const {
-    return peer_;
+  Object* GetPeer() const SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+    CHECK(jpeer_ == NULL);
+    return opeer_;
   }
 
   bool HasPeer() const {
-    return peer_ != NULL;
+    CHECK(jpeer_ == NULL);
+    return opeer_ != NULL;
   }
 
   RuntimeStats* GetStats() {
@@ -386,7 +389,7 @@
   }
 
   // Convert a jobject into a Object*
-  Object* DecodeJObject(jobject obj) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  Object* DecodeJObject(jobject obj) const SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // Implements java.lang.Thread.interrupted.
   bool Interrupted();
@@ -531,7 +534,7 @@
   };
 
   // Is the given obj in this thread's stack indirect reference table?
-  bool SirtContains(jobject obj);
+  bool SirtContains(jobject obj) const;
 
   void SirtVisitRoots(Heap::RootVisitor* visitor, void* arg);
 
@@ -618,7 +621,7 @@
   }
   friend class SignalCatcher;  // For SetStateUnsafe.
 
-  void DumpState(std::ostream& os) const;
+  void DumpState(std::ostream& os) const SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   void DumpStack(std::ostream& os) const
       LOCKS_EXCLUDED(Locks::thread_suspend_count_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
@@ -630,8 +633,9 @@
 
   static void* CreateCallback(void* arg);
 
-  void HandleUncaughtExceptions();
-  void RemoveFromThreadGroup();
+  void HandleUncaughtExceptions(ScopedObjectAccess& soa)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  void RemoveFromThreadGroup(ScopedObjectAccess& soa) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   void Init(ThreadList*, JavaVMExt*) EXCLUSIVE_LOCKS_REQUIRED(Locks::runtime_shutdown_lock_);
   void InitCardTable();
@@ -698,8 +702,10 @@
   // is hard. This field can be read off of Thread::Current to give the address.
   Thread* self_;
 
-  // Our managed peer (an instance of java.lang.Thread).
-  jobject peer_;
+  // Our managed peer (an instance of java.lang.Thread). The jobject version is used during thread
+  // start up, until the thread is registered and the local opeer_ is used.
+  Object* opeer_;
+  jobject jpeer_;
 
   // The "lowest addressable byte" of the stack
   byte* stack_begin_;
diff --git a/src/thread_list.cc b/src/thread_list.cc
index a2a8fe8..d39d424 100644
--- a/src/thread_list.cc
+++ b/src/thread_list.cc
@@ -72,7 +72,9 @@
   DumpUnattachedThreads(os);
 }
 
-static void DumpUnattachedThread(std::ostream& os, pid_t tid) {
+static void DumpUnattachedThread(std::ostream& os, pid_t tid) NO_THREAD_SAFETY_ANALYSIS {
+  // TODO: No thread safety analysis as DumpState with a NULL thread won't access fields, should
+  // refactor DumpState to avoid skipping analysis.
   Thread::DumpState(os, NULL, tid);
   DumpKernelStack(os, tid, "  kernel: ", false);
   // TODO: Reenable this when the native code in system_server can handle it.
@@ -540,18 +542,23 @@
   // suspend and so on, must happen at this point, and not in ~Thread.
   self->Destroy();
 
-  {
-    // Remove this thread from the list.
+  uint32_t thin_lock_id = self->thin_lock_id_;
+  self->thin_lock_id_ = 0;
+  ReleaseThreadId(self, thin_lock_id);
+  while (self != NULL) {
+    // Remove and delete the Thread* while holding the thread_list_lock_ and
+    // thread_suspend_count_lock_ so that the unregistering thread cannot be suspended.
     MutexLock mu(self, *Locks::thread_list_lock_);
     CHECK(Contains(self));
-    list_.remove(self);
+    // Note: we don't take the thread_suspend_count_lock_ here as to be suspending a thread other
+    // than yourself you need to hold the thread_list_lock_ (see Thread::ModifySuspendCount).
+    if (!self->IsSuspended()) {
+      list_.remove(self);
+      delete self;
+      self = NULL;
+    }
   }
 
-  // Delete the Thread* and release the thin lock id.
-  uint32_t thin_lock_id = self->thin_lock_id_;
-  ReleaseThreadId(thin_lock_id);
-  delete self;
-
   // Clear the TLS data, so that the underlying native thread is recognizably detached.
   // (It may wish to reattach later.)
   CHECK_PTHREAD_CALL(pthread_setspecific, (Thread::pthread_key_self_, NULL), "detach self");
@@ -581,8 +588,8 @@
   }
 }
 
-uint32_t ThreadList::AllocThreadId() {
-  MutexLock mu(Thread::Current(), allocated_ids_lock_);
+uint32_t ThreadList::AllocThreadId(Thread* self) {
+  MutexLock mu(self, allocated_ids_lock_);
   for (size_t i = 0; i < allocated_ids_.size(); ++i) {
     if (!allocated_ids_[i]) {
       allocated_ids_.set(i);
@@ -593,8 +600,8 @@
   return 0;
 }
 
-void ThreadList::ReleaseThreadId(uint32_t id) {
-  MutexLock mu(Thread::Current(), allocated_ids_lock_);
+void ThreadList::ReleaseThreadId(Thread* self, uint32_t id) {
+  MutexLock mu(self, allocated_ids_lock_);
   --id; // Zero is reserved to mean "invalid".
   DCHECK(allocated_ids_[id]) << id;
   allocated_ids_.reset(id);
diff --git a/src/thread_list.h b/src/thread_list.h
index d64183b..fb989ab 100644
--- a/src/thread_list.h
+++ b/src/thread_list.h
@@ -98,8 +98,8 @@
  private:
   typedef std::list<Thread*>::const_iterator It; // TODO: C++0x auto
 
-  uint32_t AllocThreadId();
-  void ReleaseThreadId(uint32_t id) LOCKS_EXCLUDED(allocated_ids_lock_);
+  uint32_t AllocThreadId(Thread* self);
+  void ReleaseThreadId(Thread* self, uint32_t id) LOCKS_EXCLUDED(allocated_ids_lock_);
 
   bool Contains(Thread* thread) EXCLUSIVE_LOCKS_REQUIRED(Locks::thread_list_lock_);
   bool Contains(pid_t tid) EXCLUSIVE_LOCKS_REQUIRED(Locks::thread_list_lock_);
diff --git a/src/well_known_classes.cc b/src/well_known_classes.cc
index 03b9cb2..cb7b70c 100644
--- a/src/well_known_classes.cc
+++ b/src/well_known_classes.cc
@@ -57,7 +57,6 @@
 jmethodID WellKnownClasses::java_lang_Float_valueOf;
 jmethodID WellKnownClasses::java_lang_Integer_valueOf;
 jmethodID WellKnownClasses::java_lang_Long_valueOf;
-jmethodID WellKnownClasses::java_lang_Object_notify;
 jmethodID WellKnownClasses::java_lang_ref_FinalizerReference_add;
 jmethodID WellKnownClasses::java_lang_ref_ReferenceQueue_add;
 jmethodID WellKnownClasses::java_lang_reflect_InvocationHandler_invoke;
@@ -152,9 +151,6 @@
   java_lang_Daemons_requestHeapTrim = CacheMethod(env, java_lang_Daemons, true, "requestHeapTrim", "()V");
   java_lang_Daemons_start = CacheMethod(env, java_lang_Daemons, true, "start", "()V");
 
-  ScopedLocalRef<jclass> java_lang_Object(env, env->FindClass("java/lang/Object"));
-  java_lang_Object_notify = CacheMethod(env, java_lang_Object.get(), false, "notify", "()V");
-
   ScopedLocalRef<jclass> java_lang_ref_FinalizerReference(env, env->FindClass("java/lang/ref/FinalizerReference"));
   java_lang_ref_FinalizerReference_add = CacheMethod(env, java_lang_ref_FinalizerReference.get(), true, "add", "(Ljava/lang/Object;)V");
   ScopedLocalRef<jclass> java_lang_ref_ReferenceQueue(env, env->FindClass("java/lang/ref/ReferenceQueue"));
diff --git a/src/well_known_classes.h b/src/well_known_classes.h
index 1f4217d..90b33c1 100644
--- a/src/well_known_classes.h
+++ b/src/well_known_classes.h
@@ -68,7 +68,6 @@
   static jmethodID java_lang_Float_valueOf;
   static jmethodID java_lang_Integer_valueOf;
   static jmethodID java_lang_Long_valueOf;
-  static jmethodID java_lang_Object_notify;
   static jmethodID java_lang_ref_FinalizerReference_add;
   static jmethodID java_lang_ref_ReferenceQueue_add;
   static jmethodID java_lang_reflect_InvocationHandler_invoke;