diff --git a/classpath/java/lang/Runtime.java b/classpath/java/lang/Runtime.java
index 51f15006c7..efdedff039 100644
--- a/classpath/java/lang/Runtime.java
+++ b/classpath/java/lang/Runtime.java
@@ -59,6 +59,8 @@ public class Runtime {
     return new MyProcess(process[0], (int) process[1], (int) process[2], (int) process[3]);
   }
 
+  public native void addShutdownHook(Thread t);
+
   private static native void exec(String[] command, long[] process);
 
   private static native int exitValue(long pid);
diff --git a/classpath/java/lang/Thread.java b/classpath/java/lang/Thread.java
index e284c00df3..2e494c850c 100644
--- a/classpath/java/lang/Thread.java
+++ b/classpath/java/lang/Thread.java
@@ -252,10 +252,14 @@ public class Thread implements Runnable {
     return daemon;
   }
 
-  public void setDaemon(boolean v) {
-    daemon = v;
+  public synchronized void setDaemon(boolean v) {
+    if (v != daemon) {
+      setDaemon(this, v);
+    }
   }
 
+  private static native void setDaemon(Thread t, boolean increment);
+
   public static native void yield();
 
   public synchronized void join() throws InterruptedException {
diff --git a/src/builtin.cpp b/src/builtin.cpp
index 3a126a0312..a045d8d451 100644
--- a/src/builtin.cpp
+++ b/src/builtin.cpp
@@ -726,6 +726,8 @@ extern "C" JNIEXPORT void JNICALL
 Avian_java_lang_Runtime_exit
 (Thread* t, object, uintptr_t* arguments)
 {
+  shutDown(t);
+
   t->m->system->exit(*arguments);
 }
 
@@ -745,6 +747,18 @@ Avian_java_lang_Runtime_totalMemory
   return 0;
 }
 
+extern "C" JNIEXPORT void JNICALL
+Avian_java_lang_Runtime_addShutdownHook
+(Thread* t, object, uintptr_t* arguments)
+{
+  object hook = reinterpret_cast<object>(arguments[1]);
+  PROTECT(t, hook);
+
+  ACQUIRE(t, t->m->shutdownLock);
+
+  t->m->shutdownHooks = makePair(t, hook, t->m->shutdownHooks);
+}
+
 extern "C" JNIEXPORT int64_t JNICALL
 Avian_java_lang_Throwable_trace
 (Thread* t, object, uintptr_t* arguments)
@@ -840,16 +854,8 @@ extern "C" JNIEXPORT int64_t JNICALL
 Avian_java_lang_Thread_doStart
 (Thread* t, object, uintptr_t* arguments)
 {
-  object this_ = reinterpret_cast<object>(*arguments);
-
-  Thread* p = t->m->processor->makeThread(t->m, this_, t);
-
-  if (t->m->system->success(t->m->system->start(&(p->runnable)))) {
-    return reinterpret_cast<int64_t>(p);
-  } else {
-    p->exit();
-    return 0;
-  }
+  return reinterpret_cast<int64_t>
+    (startThread(t, reinterpret_cast<object>(*arguments)));
 }
 
 extern "C" JNIEXPORT void JNICALL
@@ -896,6 +902,27 @@ Avian_java_lang_Thread_enumerate
   return count;
 }
 
+extern "C" JNIEXPORT void JNICALL
+Avian_java_lang_Thread_setDaemon
+(Thread* t, object, uintptr_t* arguments)
+{
+  object thread = reinterpret_cast<object>(arguments[0]);
+  bool daemon = arguments[1] != 0;
+
+  ACQUIRE_RAW(t, t->m->stateLock);
+
+  threadDaemon(t, thread) = daemon;
+
+  if (daemon) {
+    ++ t->m->daemonCount;
+  } else {
+    expect(t, t->m->daemonCount);
+    -- t->m->daemonCount;
+  }
+
+  t->m->stateLock->notifyAll(t->systemThread);
+}
+
 extern "C" JNIEXPORT int64_t JNICALL
 Avian_avian_resource_Handler_00024ResourceInputStream_getContentLength
 (Thread* t, object, uintptr_t* arguments)
diff --git a/src/jnienv.cpp b/src/jnienv.cpp
index 5195057e4a..7e6316f025 100644
--- a/src/jnienv.cpp
+++ b/src/jnienv.cpp
@@ -25,37 +25,6 @@ const uintptr_t InterfaceMethodID
 const uintptr_t NonVirtualMethodID
 = (static_cast<uintptr_t>(1) << (BitsPerWord - 2));
 
-jint JNICALL
-DestroyJavaVM(Machine* m)
-{
-  System* s = m->system;
-  Heap* h = m->heap;
-  Processor* p = m->processor;
-  Finder* f = m->finder;
-  Thread* t = m->rootThread;
-
-  // wait for other threads to exit
-  { ACQUIRE(t, m->stateLock);
-
-    while (m->liveCount > 1) {
-      t->m->stateLock->wait(t->systemThread, 0);
-    }
-  }
-
-  int exitCode = (t->exception ? -1 : 0);
-  enter(t, Thread::ActiveState);
-  t->exit();
-
-  m->dispose();
-  h->disposeFixies();
-  p->dispose();
-  h->dispose();
-  f->dispose();
-  s->dispose();
-
-  return exitCode;
-}
-
 jint JNICALL
 AttachCurrentThread(Machine* m, Thread** t, void*)
 {
@@ -94,6 +63,30 @@ DetachCurrentThread(Machine* m)
   }
 }
 
+jint JNICALL
+DestroyJavaVM(Machine* m)
+{
+  Thread* t; AttachCurrentThread(m, &t, 0);
+
+  // wait for other non-daemon threads to exit
+  { ACQUIRE(t, t->m->stateLock);
+    while (t->m->liveCount - t->m->daemonCount > 1) {
+      t->m->stateLock->wait(t->systemThread, 0);
+    }
+  }
+
+  { ENTER(t, Thread::ActiveState);
+
+    shutDown(t);
+  }
+
+  int exitCode = (t->exception ? -1 : 0);
+
+  t->exit();
+
+  return exitCode;
+}
+
 jint JNICALL
 GetEnv(Machine* m, Thread** t, jint version)
 {
@@ -431,7 +424,7 @@ CallBooleanMethodV(Thread* t, jobject o, jmethodID m, va_list a)
 
   object method = getMethod(t, m);
   object r = t->m->processor->invokeList(t, method, *o, true, a);
-  return (t->exception ? false : (intValue(t, r) != 0));
+  return (r ? (intValue(t, r) != 0) : false);
 }
 
 jboolean JNICALL
@@ -454,7 +447,7 @@ CallByteMethodV(Thread* t, jobject o, jmethodID m, va_list a)
 
   object method = getMethod(t, m);
   object r = t->m->processor->invokeList(t, method, *o, true, a);
-  return (t->exception ? 0 : intValue(t, r));
+  return (r ? intValue(t, r) : 0);
 }
 
 jbyte JNICALL
@@ -477,7 +470,7 @@ CallCharMethodV(Thread* t, jobject o, jmethodID m, va_list a)
 
   object method = getMethod(t, m);
   object r = t->m->processor->invokeList(t, method, *o, true, a);
-  return (t->exception ? 0 : intValue(t, r));
+  return (r ? intValue(t, r) : 0);
 }
 
 jchar JNICALL
@@ -500,7 +493,7 @@ CallShortMethodV(Thread* t, jobject o, jmethodID m, va_list a)
 
   object method = getMethod(t, m);
   object r = t->m->processor->invokeList(t, method, *o, true, a);
-  return (t->exception ? 0 : intValue(t, r));
+  return (r ? intValue(t, r) : 0);
 }
 
 jshort JNICALL
@@ -523,7 +516,7 @@ CallIntMethodV(Thread* t, jobject o, jmethodID m, va_list a)
 
   object method = getMethod(t, m);
   object r = t->m->processor->invokeList(t, method, *o, true, a);
-  return (t->exception ? 0 : intValue(t, r));
+  return (r ? intValue(t, r) : 0);
 }
 
 jint JNICALL
@@ -546,7 +539,7 @@ CallLongMethodV(Thread* t, jobject o, jmethodID m, va_list a)
 
   object method = getMethod(t, m);
   object r = t->m->processor->invokeList(t, method, *o, true, a);
-  return (t->exception ? 0 : longValue(t, r));
+  return (r ? longValue(t, r) : 0);
 }
 
 jlong JNICALL
@@ -569,7 +562,7 @@ CallFloatMethodV(Thread* t, jobject o, jmethodID m, va_list a)
 
   object method = getMethod(t, m);
   object r = t->m->processor->invokeList(t, method, *o, true, a);
-  return (t->exception ? 0 : bitsToFloat(intValue(t, r)));
+  return (r ? bitsToFloat(intValue(t, r)) : 0);
 }
 
 jfloat JNICALL
@@ -592,7 +585,7 @@ CallDoubleMethodV(Thread* t, jobject o, jmethodID m, va_list a)
 
   object method = getMethod(t, m);
   object r = t->m->processor->invokeList(t, method, *o, true, a);
-  return (t->exception ? 0 : bitsToDouble(longValue(t, r)));
+  return (r ? bitsToDouble(longValue(t, r)) : 0);
 }
 
 jdouble JNICALL
@@ -666,7 +659,7 @@ CallStaticBooleanMethodV(Thread* t, jclass, jmethodID m, va_list a)
   ENTER(t, Thread::ActiveState);
 
   object r = t->m->processor->invokeList(t, getStaticMethod(t, m), 0, true, a);
-  return (t->exception ? 0 : (intValue(t, r) != 0));
+  return (r ? (intValue(t, r) != 0) : false);
 }
 
 jboolean JNICALL
@@ -688,7 +681,7 @@ CallStaticByteMethodV(Thread* t, jclass, jmethodID m, va_list a)
   ENTER(t, Thread::ActiveState);
 
   object r = t->m->processor->invokeList(t, getStaticMethod(t, m), 0, true, a);
-  return (t->exception ? 0 : intValue(t, r));
+  return (r ? intValue(t, r) : 0);
 }
 
 jbyte JNICALL
@@ -710,7 +703,7 @@ CallStaticCharMethodV(Thread* t, jclass, jmethodID m, va_list a)
   ENTER(t, Thread::ActiveState);
 
   object r = t->m->processor->invokeList(t, getStaticMethod(t, m), 0, true, a);
-  return (t->exception ? 0 : intValue(t, r));
+  return (r ? intValue(t, r) : 0);
 }
 
 jchar JNICALL
@@ -732,7 +725,7 @@ CallStaticShortMethodV(Thread* t, jclass, jmethodID m, va_list a)
   ENTER(t, Thread::ActiveState);
 
   object r = t->m->processor->invokeList(t, getStaticMethod(t, m), 0, true, a);
-  return (t->exception ? 0 : intValue(t, r));
+  return (r ? intValue(t, r) : 0);
 }
 
 jshort JNICALL
@@ -754,7 +747,7 @@ CallStaticIntMethodV(Thread* t, jclass, jmethodID m, va_list a)
   ENTER(t, Thread::ActiveState);
 
   object r = t->m->processor->invokeList(t, getStaticMethod(t, m), 0, true, a);
-  return (t->exception ? 0 : intValue(t, r));
+  return (r ? intValue(t, r) : 0);
 }
 
 jint JNICALL
@@ -776,7 +769,7 @@ CallStaticLongMethodV(Thread* t, jclass, jmethodID m, va_list a)
   ENTER(t, Thread::ActiveState);
 
   object r = t->m->processor->invokeList(t, getStaticMethod(t, m), 0, true, a);
-  return (t->exception ? 0 : longValue(t, r));
+  return (r ? longValue(t, r) : 0);
 }
 
 jlong JNICALL
@@ -798,7 +791,7 @@ CallStaticFloatMethodV(Thread* t, jclass, jmethodID m, va_list a)
   ENTER(t, Thread::ActiveState);
 
   object r = t->m->processor->invokeList(t, getStaticMethod(t, m), 0, true, a);
-  return (t->exception ? 0 : bitsToFloat(intValue(t, r)));
+  return (r ? bitsToFloat(intValue(t, r)) : 0);
 }
 
 jfloat JNICALL
@@ -820,7 +813,7 @@ CallStaticDoubleMethodV(Thread* t, jclass, jmethodID m, va_list a)
   ENTER(t, Thread::ActiveState);
 
   object r = t->m->processor->invokeList(t, getStaticMethod(t, m), 0, true, a);
-  return (t->exception ? 0 : bitsToDouble(longValue(t, r)));
+  return (r ? bitsToDouble(longValue(t, r)) : 0);
 }
 
 jdouble JNICALL
diff --git a/src/machine.cpp b/src/machine.cpp
index 1fa9791397..6e046843d0 100644
--- a/src/machine.cpp
+++ b/src/machine.cpp
@@ -146,6 +146,50 @@ disposeAll(Thread* m, Thread* o)
   dispose(m, o, false);
 }
 
+void
+turnOffTheLights(Thread* t)
+{
+  expect(t, t->m->liveCount == 1);
+
+  joinAll(t, t->m->rootThread);
+
+  enter(t, Thread::ExitState);
+
+  for (object* p = &(t->m->finalizers); *p;) {
+    object f = *p;
+    *p = finalizerNext(t, *p);
+
+    void (*function)(Thread*, object);
+    memcpy(&function, &finalizerFinalize(t, f), BytesPerWord);
+    function(t, finalizerTarget(t, f));
+  }
+
+  for (object* p = &(t->m->tenuredFinalizers); *p;) {
+    object f = *p;
+    *p = finalizerNext(t, *p);
+
+    void (*function)(Thread*, object);
+    memcpy(&function, &finalizerFinalize(t, f), BytesPerWord);
+    function(t, finalizerTarget(t, f));
+  }
+
+  Machine* m = t->m;
+
+  disposeAll(t, t->m->rootThread);
+
+  System* s = m->system;
+  Heap* h = m->heap;
+  Processor* p = m->processor;
+  Finder* f = m->finder;
+
+  m->dispose();
+  h->disposeFixies();
+  p->dispose();
+  h->dispose();
+  f->dispose();
+  s->dispose();
+}
+
 void
 killZombies(Thread* t, Thread* o)
 {
@@ -1976,12 +2020,14 @@ Machine::Machine(System* system, Heap* heap, Finder* finder,
   propertyCount(propertyCount),
   activeCount(0),
   liveCount(0),
+  daemonCount(0),
   fixedFootprint(0),
   localThread(0),
   stateLock(0),
   heapLock(0),
   classLock(0),
   referenceLock(0),
+  shutdownLock(0),
   libraries(0),
   loader(0),
   loadClassMethod(0),
@@ -1996,6 +2042,7 @@ Machine::Machine(System* system, Heap* heap, Finder* finder,
   finalizeQueue(0),
   weakReferences(0),
   tenuredWeakReferences(0),
+  shutdownHooks(0),
   unsafe(false),
   triedBuiltinOnLoad(false),
   heapPoolIndex(0)
@@ -2009,6 +2056,7 @@ Machine::Machine(System* system, Heap* heap, Finder* finder,
       not system->success(system->make(&heapLock)) or
       not system->success(system->make(&classLock)) or
       not system->success(system->make(&referenceLock)) or
+      not system->success(system->make(&shutdownLock)) or
       not system->success
       (system->load(&libraries, findProperty(this, "avian.bootstrap"), false)))
   {
@@ -2024,6 +2072,7 @@ Machine::dispose()
   heapLock->dispose();
   classLock->dispose();
   referenceLock->dispose();
+  shutdownLock->dispose();
 
   if (libraries) {
     libraries->disposeAll();
@@ -2150,7 +2199,7 @@ Thread::exit()
     enter(this, Thread::ExclusiveState);
 
     if (m->liveCount == 1) {
-      vm::exit(this);
+      turnOffTheLights(this);
     } else {
       enter(this, Thread::ZombieState);
     }
@@ -2160,6 +2209,8 @@ Thread::exit()
 void
 Thread::dispose()
 {
+  threadPeer(this, javaThread) = 0;
+
   if (systemThread) {
     systemThread->dispose();
   }
@@ -2170,31 +2221,41 @@ Thread::dispose()
 }
 
 void
-exit(Thread* t)
+shutDown(Thread* t)
 {
-  enter(t, Thread::ExitState);
+  ACQUIRE(t, t->m->shutdownLock);
 
-  joinAll(t, t->m->rootThread);
+  object hooks = t->m->shutdownHooks;
+  PROTECT(t, hooks);
 
-  for (object* p = &(t->m->finalizers); *p;) {
-    object f = *p;
-    *p = finalizerNext(t, *p);
+  t->m->shutdownHooks = 0;
 
-    void (*function)(Thread*, object);
-    memcpy(&function, &finalizerFinalize(t, f), BytesPerWord);
-    function(t, finalizerTarget(t, f));
+  object h = hooks;
+  PROTECT(t, h);
+  for (; h; h = pairSecond(t, h)) {
+    startThread(t, pairFirst(t, h));
   }
 
-  for (object* p = &(t->m->tenuredFinalizers); *p;) {
-    object f = *p;
-    *p = finalizerNext(t, *p);
+  // wait for hooks to exit
+  h = hooks;
+  for (; h; h = pairSecond(t, h)) {
+    while (true) {
+      Thread* ht = reinterpret_cast<Thread*>(threadPeer(t, pairFirst(t, h)));
 
-    void (*function)(Thread*, object);
-    memcpy(&function, &finalizerFinalize(t, f), BytesPerWord);
-    function(t, finalizerTarget(t, f));
+      { ACQUIRE(t, t->m->stateLock);
+
+        if (ht == 0
+            or ht->state == Thread::ZombieState
+            or ht->state == Thread::JoinedState)
+        {
+          break;
+        } else {
+          ENTER(t, Thread::IdleState);
+          t->m->stateLock->wait(t->systemThread, 0);
+        }
+      }
+    }
   }
-
-  disposeAll(t, t->m->rootThread);
 }
 
 void
@@ -2255,6 +2316,10 @@ enter(Thread* t, Thread::State s)
     if (s == Thread::ZombieState) {
       assert(t, t->m->liveCount > 0);
       -- t->m->liveCount;
+
+      if (threadDaemon(t, t->javaThread)) {
+        -- t->m->daemonCount;
+      }
     }
     t->state = s;
 
@@ -2308,7 +2373,7 @@ enter(Thread* t, Thread::State s)
 
     t->state = s;
 
-    while (t->m->liveCount > 1) {
+    while (t->m->liveCount - t->m->daemonCount > 1) {
       t->m->stateLock->wait(t->systemThread, 0);
     }
   } break;
@@ -3436,6 +3501,7 @@ visitRoots(Machine* m, Heap::Visitor* v)
   v->visit(&(m->byteArrayMap));
   v->visit(&(m->types));
   v->visit(&(m->jniMethodTable));
+  v->visit(&(m->shutdownHooks));
 
   for (Thread* t = m->rootThread; t; t = t->peer) {
     ::visitRoots(t, v);
diff --git a/src/machine.h b/src/machine.h
index 1539b80af0..078355dbe9 100644
--- a/src/machine.h
+++ b/src/machine.h
@@ -1178,12 +1178,14 @@ class Machine {
   unsigned propertyCount;
   unsigned activeCount;
   unsigned liveCount;
+  unsigned daemonCount;
   unsigned fixedFootprint;
   System::Local* localThread;
   System::Monitor* stateLock;
   System::Monitor* heapLock;
   System::Monitor* classLock;
   System::Monitor* referenceLock;
+  System::Monitor* shutdownLock;
   System::Library* libraries;
   object loader;
   object loadClassMethod;
@@ -1198,6 +1200,7 @@ class Machine {
   object finalizeQueue;
   object weakReferences;
   object tenuredWeakReferences;
+  object shutdownHooks;
   bool unsafe;
   bool triedBuiltinOnLoad;
   JavaVMVTable javaVMVTable;
@@ -1394,6 +1397,9 @@ dispose(Thread* t, Reference* r)
 void
 collect(Thread* t, Heap::CollectionType type);
 
+void
+shutDown(Thread* t);
+
 #ifdef VM_STRESS
 
 inline void
@@ -1599,6 +1605,19 @@ setObjectClass(Thread*, object o, object value)
      | (reinterpret_cast<uintptr_t>(cast<object>(o, 0)) & (~PointerMask)));
 }
 
+inline Thread*
+startThread(Thread* t, object javaThread)
+{
+  Thread* p = t->m->processor->makeThread(t->m, javaThread, t);
+
+  if (t->m->system->success(t->m->system->start(&(p->runnable)))) {
+    return p;
+  } else {
+    p->exit();
+    return 0;
+  }
+}
+
 inline const char*
 findProperty(Machine* m, const char* name)
 {
@@ -2341,9 +2360,6 @@ interrupt(Thread*, Thread* target)
 object
 intern(Thread* t, object s);
 
-void
-exit(Thread* t);
-
 void
 walk(Thread* t, Heap::Walker* w, object o, unsigned start);