src/hotspot/share/runtime/synchronizer.cpp

rev 54109 : Add logging to ObjectSynchronizer::omFlush(), add global count logging at Info level to ObjectSynchronizer::finish_deflate_idle_monitors(); for monitorinflation logging, switch from cumulative "deflating per-thread idle monitors" mesgs to per-thread "deflating per-thread idle monitors" mesgs; fix timer bug in deflate_thread_local_monitors() where time to acquire gListLock wasn't counted; fix misc typos.
rev 54110 : Checkpoint latest preliminary review patches for full OpenJDK review.

@@ -122,10 +122,12 @@
 // global monitor in-use list, for moribund threads,
 // monitors they inflated need to be scanned for deflation
 ObjectMonitor * volatile ObjectSynchronizer::gOmInUseList  = NULL;
 // count of entries in gOmInUseList
 int ObjectSynchronizer::gOmInUseCount = 0;
+bool ObjectSynchronizer::_gOmShouldDeflateIdleMonitors = false;
+bool volatile ObjectSynchronizer::_is_cleanup_requested = false;
 
 static volatile intptr_t gListLock = 0;      // protects global monitor lists
 static volatile int gMonitorFreeCount  = 0;  // # on gFreeList
 static volatile int gMonitorPopulation = 0;  // # Extant -- in circulation
 

@@ -208,14 +210,22 @@
   assert(!SafepointSynchronize::is_at_safepoint(), "invariant");
   assert(Self->is_Java_thread(), "invariant");
   assert(((JavaThread *) Self)->thread_state() == _thread_in_Java, "invariant");
   NoSafepointVerifier nsv;
   if (obj == NULL) return false;       // Need to throw NPE
+
+  while (true) {
   const markOop mark = obj->mark();
 
   if (mark->has_monitor()) {
-    ObjectMonitor * const m = mark->monitor();
+      ObjectMonitorHandle omh;
+      if (!omh.save_om_ptr(obj, mark)) {
+        // Lost a race with async deflation so try again.
+        assert(AsyncDeflateIdleMonitors, "sanity check");
+        continue;
+      }
+      ObjectMonitor * const m = omh.om_ptr();
     assert(oopDesc::equals((oop) m->object(), obj), "invariant");
     Thread * const owner = (Thread *) m->_owner;
 
     // Lock contention and Transactional Lock Elision (TLE) diagnostics
     // and observability

@@ -243,10 +253,12 @@
       assert(m->_recursions == 0, "invariant");
       assert(m->_owner == Self, "invariant");
       return true;
     }
   }
+    break;
+  }
 
   // Note that we could inflate in quick_enter.
   // This is likely a useful optimization
   // Critically, in quick_enter() we must not:
   // -- perform bias revocation, or

@@ -325,19 +337,23 @@
       return;
     }
   }
 
   // We have to take the slow-path of possible inflation and then exit.
-  inflate(THREAD, object, inflate_cause_vm_internal)->exit(true, THREAD);
+  ObjectMonitorHandle omh;
+  inflate(&omh, THREAD, object, inflate_cause_vm_internal);
+  omh.om_ptr()->exit(true, THREAD);
 }
 
 // -----------------------------------------------------------------------------
 // Interpreter/Compiler Slow Case
 // This routine is used to handle interpreter/compiler slow case
 // We don't need to use fast path here, because it must have been
 // failed in the interpreter/compiler code.
 void ObjectSynchronizer::slow_enter(Handle obj, BasicLock* lock, TRAPS) {
+  bool do_loop = true;
+  while (do_loop) {
   markOop mark = obj->mark();
   assert(!mark->has_bias_pattern(), "should not see bias pattern here");
 
   if (mark->is_neutral()) {
     // Anticipate successful CAS -- the ST of the displaced mark must

@@ -358,11 +374,14 @@
   // The object header will never be displaced to this lock,
   // so it does not matter what the value is, except that it
   // must be non-zero to avoid looking like a re-entrant lock,
   // and must not look locked either.
   lock->set_displaced_header(markOopDesc::unused_mark());
-  inflate(THREAD, obj(), inflate_cause_monitor_enter)->enter(THREAD);
+    ObjectMonitorHandle omh;
+    inflate(&omh, THREAD, obj(), inflate_cause_monitor_enter);
+    do_loop = !omh.om_ptr()->enter(THREAD);
+  }
 }
 
 // This routine is used to handle interpreter/compiler slow case
 // We don't need to use fast path here, because it must have
 // failed in the interpreter/compiler code. Simply use the heavy

@@ -387,25 +406,29 @@
   if (UseBiasedLocking) {
     BiasedLocking::revoke_and_rebias(obj, false, THREAD);
     assert(!obj->mark()->has_bias_pattern(), "biases should be revoked by now");
   }
 
-  ObjectMonitor* monitor = inflate(THREAD, obj(), inflate_cause_vm_internal);
-
-  return monitor->complete_exit(THREAD);
+  ObjectMonitorHandle omh;
+  inflate(&omh, THREAD, obj(), inflate_cause_vm_internal);
+  intptr_t ret_code = omh.om_ptr()->complete_exit(THREAD);
+  return ret_code;
 }
 
 // NOTE: must use heavy weight monitor to handle complete_exit/reenter()
 void ObjectSynchronizer::reenter(Handle obj, intptr_t recursion, TRAPS) {
   if (UseBiasedLocking) {
     BiasedLocking::revoke_and_rebias(obj, false, THREAD);
     assert(!obj->mark()->has_bias_pattern(), "biases should be revoked by now");
   }
 
-  ObjectMonitor* monitor = inflate(THREAD, obj(), inflate_cause_vm_internal);
-
-  monitor->reenter(recursion, THREAD);
+  bool do_loop = true;
+  while (do_loop) {
+    ObjectMonitorHandle omh;
+    inflate(&omh, THREAD, obj(), inflate_cause_vm_internal);
+    do_loop = !omh.om_ptr()->reenter(recursion, THREAD);
+  }
 }
 // -----------------------------------------------------------------------------
 // JNI locks on java objects
 // NOTE: must use heavy weight monitor to handle jni monitor enter
 void ObjectSynchronizer::jni_enter(Handle obj, TRAPS) {

@@ -413,11 +436,16 @@
   if (UseBiasedLocking) {
     BiasedLocking::revoke_and_rebias(obj, false, THREAD);
     assert(!obj->mark()->has_bias_pattern(), "biases should be revoked by now");
   }
   THREAD->set_current_pending_monitor_is_from_java(false);
-  inflate(THREAD, obj(), inflate_cause_jni_enter)->enter(THREAD);
+  bool do_loop = true;
+  while (do_loop) {
+    ObjectMonitorHandle omh;
+    inflate(&omh, THREAD, obj(), inflate_cause_jni_enter);
+    do_loop = !omh.om_ptr()->enter(THREAD);
+  }
   THREAD->set_current_pending_monitor_is_from_java(true);
 }
 
 // NOTE: must use heavy weight monitor to handle jni monitor exit
 void ObjectSynchronizer::jni_exit(oop obj, Thread* THREAD) {

@@ -426,11 +454,13 @@
     BiasedLocking::revoke_and_rebias(h_obj, false, THREAD);
     obj = h_obj();
   }
   assert(!obj->mark()->has_bias_pattern(), "biases should be revoked by now");
 
-  ObjectMonitor* monitor = inflate(THREAD, obj, inflate_cause_jni_exit);
+  ObjectMonitorHandle omh;
+  inflate(&omh, THREAD, obj, inflate_cause_jni_exit);
+  ObjectMonitor * monitor = omh.om_ptr();
   // If this thread has locked the object, exit the monitor.  Note:  can't use
   // monitor->check(CHECK); must exit even if an exception is pending.
   if (monitor->check(THREAD)) {
     monitor->exit(true, THREAD);
   }

@@ -466,31 +496,36 @@
     assert(!obj->mark()->has_bias_pattern(), "biases should be revoked by now");
   }
   if (millis < 0) {
     THROW_MSG_0(vmSymbols::java_lang_IllegalArgumentException(), "timeout value is negative");
   }
-  ObjectMonitor* monitor = inflate(THREAD, obj(), inflate_cause_wait);
+  ObjectMonitorHandle omh;
+  inflate(&omh, THREAD, obj(), inflate_cause_wait);
+  ObjectMonitor * monitor = omh.om_ptr();
 
   DTRACE_MONITOR_WAIT_PROBE(monitor, obj(), THREAD, millis);
   monitor->wait(millis, true, THREAD);
 
   // This dummy call is in place to get around dtrace bug 6254741.  Once
   // that's fixed we can uncomment the following line, remove the call
   // and change this function back into a "void" func.
   // DTRACE_MONITOR_PROBE(waited, monitor, obj(), THREAD);
-  return dtrace_waited_probe(monitor, obj, THREAD);
+  int ret_code = dtrace_waited_probe(monitor, obj, THREAD);
+  return ret_code;
 }
 
 void ObjectSynchronizer::waitUninterruptibly(Handle obj, jlong millis, TRAPS) {
   if (UseBiasedLocking) {
     BiasedLocking::revoke_and_rebias(obj, false, THREAD);
     assert(!obj->mark()->has_bias_pattern(), "biases should be revoked by now");
   }
   if (millis < 0) {
     THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(), "timeout value is negative");
   }
-  inflate(THREAD, obj(), inflate_cause_wait)->wait(millis, false, THREAD);
+  ObjectMonitorHandle omh;
+  inflate(&omh, THREAD, obj(), inflate_cause_wait);
+  omh.om_ptr()->wait(millis, false, THREAD);
 }
 
 void ObjectSynchronizer::notify(Handle obj, TRAPS) {
   if (UseBiasedLocking) {
     BiasedLocking::revoke_and_rebias(obj, false, THREAD);

@@ -499,11 +534,13 @@
 
   markOop mark = obj->mark();
   if (mark->has_locker() && THREAD->is_lock_owned((address)mark->locker())) {
     return;
   }
-  inflate(THREAD, obj(), inflate_cause_notify)->notify(THREAD);
+  ObjectMonitorHandle omh;
+  inflate(&omh, THREAD, obj(), inflate_cause_notify);
+  omh.om_ptr()->notify(THREAD);
 }
 
 // NOTE: see comment of notify()
 void ObjectSynchronizer::notifyall(Handle obj, TRAPS) {
   if (UseBiasedLocking) {

@@ -513,11 +550,13 @@
 
   markOop mark = obj->mark();
   if (mark->has_locker() && THREAD->is_lock_owned((address)mark->locker())) {
     return;
   }
-  inflate(THREAD, obj(), inflate_cause_notify)->notifyAll(THREAD);
+  ObjectMonitorHandle omh;
+  inflate(&omh, THREAD, obj(), inflate_cause_notify);
+  omh.om_ptr()->notifyAll(THREAD);
 }
 
 // -----------------------------------------------------------------------------
 // Hash Code handling
 //

@@ -707,10 +746,11 @@
   assert(Universe::verify_in_progress() || DumpSharedSpaces ||
          Self->is_Java_thread() , "invariant");
   assert(Universe::verify_in_progress() || DumpSharedSpaces ||
          ((JavaThread *)Self)->thread_state() != _thread_blocked, "invariant");
 
+ Retry:
   ObjectMonitor* monitor = NULL;
   markOop temp, test;
   intptr_t hash;
   markOop mark = ReadStableMark(obj);
 

@@ -731,11 +771,17 @@
     }
     // If atomic operation failed, we must inflate the header
     // into heavy weight monitor. We could add more code here
     // for fast path, but it does not worth the complexity.
   } else if (mark->has_monitor()) {
-    monitor = mark->monitor();
+    ObjectMonitorHandle omh;
+    if (!omh.save_om_ptr(obj, mark)) {
+      // Lost a race with async deflation so try again.
+      assert(AsyncDeflateIdleMonitors, "sanity check");
+      goto Retry;
+    }
+    monitor = omh.om_ptr();
     temp = monitor->header();
     assert(temp->is_neutral(), "invariant: header=" INTPTR_FORMAT, p2i((address)temp));
     hash = temp->hash();
     if (hash != 0) {
       return hash;

@@ -758,11 +804,13 @@
     // Any change to stack may not propagate to other threads
     // correctly.
   }
 
   // Inflate the monitor to set hash code
-  monitor = inflate(Self, obj, inflate_cause_hash_code);
+  ObjectMonitorHandle omh;
+  inflate(&omh, Self, obj, inflate_cause_hash_code);
+  monitor = omh.om_ptr();
   // Load displaced header and check it has hash code
   mark = monitor->header();
   assert(mark->is_neutral(), "invariant: header=" INTPTR_FORMAT, p2i((address)mark));
   hash = mark->hash();
   if (hash == 0) {

@@ -798,24 +846,32 @@
   }
 
   assert(thread == JavaThread::current(), "Can only be called on current thread");
   oop obj = h_obj();
 
+  while (true) {
   markOop mark = ReadStableMark(obj);
 
   // Uncontended case, header points to stack
   if (mark->has_locker()) {
     return thread->is_lock_owned((address)mark->locker());
   }
   // Contended case, header points to ObjectMonitor (tagged pointer)
   if (mark->has_monitor()) {
-    ObjectMonitor* monitor = mark->monitor();
-    return monitor->is_entered(thread) != 0;
+      ObjectMonitorHandle omh;
+      if (!omh.save_om_ptr(obj, mark)) {
+        // Lost a race with async deflation so try again.
+        assert(AsyncDeflateIdleMonitors, "sanity check");
+        continue;
+      }
+      bool ret_code = omh.om_ptr()->is_entered(thread) != 0;
+      return ret_code;
   }
   // Unlocked case, header in place
   assert(mark->is_neutral(), "sanity check");
   return false;
+  }
 }
 
 // Be aware of this method could revoke bias of the lock object.
 // This method queries the ownership of the lock handle specified by 'h_obj'.
 // If the current thread owns the lock, it returns owner_self. If no

@@ -837,31 +893,41 @@
            "biases should be revoked by now");
   }
 
   assert(self == JavaThread::current(), "Can only be called on current thread");
   oop obj = h_obj();
+
+  while (true) {
   markOop mark = ReadStableMark(obj);
 
   // CASE: stack-locked.  Mark points to a BasicLock on the owner's stack.
   if (mark->has_locker()) {
     return self->is_lock_owned((address)mark->locker()) ?
       owner_self : owner_other;
   }
 
   // CASE: inflated. Mark (tagged pointer) points to an ObjectMonitor.
   // The Object:ObjectMonitor relationship is stable as long as we're
-  // not at a safepoint.
+    // not at a safepoint and AsyncDeflateIdleMonitors is false.
   if (mark->has_monitor()) {
-    void * owner = mark->monitor()->_owner;
+      ObjectMonitorHandle omh;
+      if (!omh.save_om_ptr(obj, mark)) {
+        // Lost a race with async deflation so try again.
+        assert(AsyncDeflateIdleMonitors, "sanity check");
+        continue;
+      }
+      ObjectMonitor * monitor = omh.om_ptr();
+      void * owner = monitor->_owner;
     if (owner == NULL) return owner_none;
     return (owner == self ||
             self->is_lock_owned((address)owner)) ? owner_self : owner_other;
   }
 
   // CASE: neutral
   assert(mark->is_neutral(), "sanity check");
   return owner_none;           // it's unlocked
+  }
 }
 
 // FIXME: jvmti should call this
 JavaThread* ObjectSynchronizer::get_lock_owner(ThreadsList * t_list, Handle h_obj) {
   if (UseBiasedLocking) {

@@ -872,22 +938,29 @@
     }
     assert(!h_obj->mark()->has_bias_pattern(), "biases should be revoked by now");
   }
 
   oop obj = h_obj();
-  address owner = NULL;
 
+  while (true) {
+    address owner = NULL;
   markOop mark = ReadStableMark(obj);
 
   // Uncontended case, header points to stack
   if (mark->has_locker()) {
     owner = (address) mark->locker();
   }
 
   // Contended case, header points to ObjectMonitor (tagged pointer)
   else if (mark->has_monitor()) {
-    ObjectMonitor* monitor = mark->monitor();
+      ObjectMonitorHandle omh;
+      if (!omh.save_om_ptr(obj, mark)) {
+        // Lost a race with async deflation so try again.
+        assert(AsyncDeflateIdleMonitors, "sanity check");
+        continue;
+      }
+      ObjectMonitor* monitor = omh.om_ptr();
     assert(monitor != NULL, "monitor should be non-null");
     owner = (address) monitor->owner();
   }
 
   if (owner != NULL) {

@@ -899,22 +972,33 @@
   // Cannot have assertion since this object may have been
   // locked by another thread when reaching here.
   // assert(mark->is_neutral(), "sanity check");
 
   return NULL;
+  }
 }
 
 // Visitors ...
 
 void ObjectSynchronizer::monitors_iterate(MonitorClosure* closure) {
   PaddedEnd<ObjectMonitor> * block = OrderAccess::load_acquire(&gBlockList);
   while (block != NULL) {
     assert(block->object() == CHAINMARKER, "must be a block header");
     for (int i = _BLOCKSIZE - 1; i > 0; i--) {
       ObjectMonitor* mid = (ObjectMonitor *)(block + i);
-      oop object = (oop)mid->object();
-      if (object != NULL) {
+      if (mid->is_active()) {
+        ObjectMonitorHandle omh(mid);
+
+        if (mid->object() == NULL ||
+            (AsyncDeflateIdleMonitors && mid->_owner == DEFLATER_MARKER)) {
+          // Only process with closure if the object is set.
+          // For async deflation, race here if monitor is not owned!
+          // The above ref_count bump (in ObjectMonitorHandle ctr)
+          // will cause subsequent async deflation to skip it.
+          // However, previous or concurrent async deflation is a race.
+          continue;
+        }
         closure->do_monitor(mid);
       }
     }
     block = (PaddedEnd<ObjectMonitor> *)block->FreeNext;
   }

@@ -1021,17 +1105,32 @@
     // The VMThread will delete the op when completed.
     VMThread::execute(new VM_ScavengeMonitors());
   }
 }
 
-ObjectMonitor* ObjectSynchronizer::omAlloc(Thread * Self) {
+ObjectMonitor* ObjectSynchronizer::omAlloc(Thread * Self,
+                                           const InflateCause cause) {
   // A large MAXPRIVATE value reduces both list lock contention
   // and list coherency traffic, but also tends to increase the
   // number of objectMonitors in circulation as well as the STW
   // scavenge costs.  As usual, we lean toward time in space-time
   // tradeoffs.
   const int MAXPRIVATE = 1024;
+
+  if (AsyncDeflateIdleMonitors) {
+    JavaThread * jt = (JavaThread *)Self;
+    if (jt->omShouldDeflateIdleMonitors && jt->omInUseCount > 0 &&
+        cause != inflate_cause_vm_internal) {
+      // Deflate any per-thread idle monitors for this JavaThread if
+      // this is not an internal inflation. Clean up your own mess.
+      // (Gibbs Rule 45) Otherwise, skip this cleanup.
+      // deflate_global_idle_monitors_using_JT() is called by the ServiceThread.
+      debug_only(jt->check_for_valid_safepoint_state(false);)
+      ObjectSynchronizer::deflate_per_thread_idle_monitors_using_JT();
+    }
+  }
+
   for (;;) {
     ObjectMonitor * m;
 
     // 1: try to allocate from the thread's local omFreeList.
     // Threads will attempt to allocate first from their local list, then

@@ -1042,10 +1141,11 @@
     m = Self->omFreeList;
     if (m != NULL) {
       Self->omFreeList = m->FreeNext;
       Self->omFreeCount--;
       guarantee(m->object() == NULL, "invariant");
+      m->set_allocation_state(ObjectMonitor::New);
       m->FreeNext = Self->omInUseList;
       Self->omInUseList = m;
       Self->omInUseCount++;
       return m;
     }

@@ -1063,12 +1163,17 @@
       for (int i = Self->omFreeProvision; --i >= 0 && gFreeList != NULL;) {
         gMonitorFreeCount--;
         ObjectMonitor * take = gFreeList;
         gFreeList = take->FreeNext;
         guarantee(take->object() == NULL, "invariant");
+        if (AsyncDeflateIdleMonitors) {
+          take->set_owner(NULL);
+          take->_count = 0;
+        }
         guarantee(!take->is_busy(), "invariant");
         take->Recycle();
+        assert(take->is_free(), "invariant");
         omRelease(Self, take, false);
       }
       Thread::muxRelease(&gListLock);
       Self->omFreeProvision += 1 + (Self->omFreeProvision/2);
       if (Self->omFreeProvision > MAXPRIVATE) Self->omFreeProvision = MAXPRIVATE;

@@ -1117,10 +1222,11 @@
     // linkage should be reconsidered.  A better implementation would
     // look like: class Block { Block * next; int N; ObjectMonitor Body [N] ; }
 
     for (int i = 1; i < _BLOCKSIZE; i++) {
       temp[i].FreeNext = (ObjectMonitor *)&temp[i+1];
+      assert(temp[i].is_free(), "invariant");
     }
 
     // terminate the last monitor as the end of list
     temp[_BLOCKSIZE - 1].FreeNext = NULL;
 

@@ -1159,17 +1265,19 @@
 // attempt failed.  This doesn't allow unbounded #s of monitors to
 // accumulate on a thread's free list.
 //
 // Key constraint: all ObjectMonitors on a thread's free list and the global
 // free list must have their object field set to null. This prevents the
-// scavenger -- deflate_monitor_list() -- from reclaiming them.
+// scavenger -- deflate_monitor_list() or deflate_monitor_list_using_JT()
+// -- from reclaiming them while we are trying to release them.
 
 void ObjectSynchronizer::omRelease(Thread * Self, ObjectMonitor * m,
                                    bool fromPerThreadAlloc) {
   guarantee(m->header() == NULL, "invariant");
   guarantee(m->object() == NULL, "invariant");
   guarantee(((m->is_busy()|m->_recursions) == 0), "freeing in-use monitor");
+  m->set_allocation_state(ObjectMonitor::Free);
   // Remove from omInUseList
   if (fromPerThreadAlloc) {
     ObjectMonitor* cur_mid_in_use = NULL;
     bool extracted = false;
     for (ObjectMonitor* mid = Self->omInUseList; mid != NULL; cur_mid_in_use = mid, mid = mid->FreeNext) {

@@ -1188,10 +1296,11 @@
     assert(extracted, "Should have extracted from in-use list");
   }
 
   // FreeNext is used for both omInUseList and omFreeList, so clear old before setting new
   m->FreeNext = Self->omFreeList;
+  guarantee(m->is_free(), "invariant");
   Self->omFreeList = m;
   Self->omFreeCount++;
 }
 
 // Return the monitors of a moribund thread's local free list to

@@ -1211,10 +1320,14 @@
 // interleave with the deflate_idle_monitors scavenge operator. In particular,
 // this ensures that the thread's monitors are scanned by a GC safepoint,
 // either via Thread::oops_do() (if safepoint happens before omFlush()) or via
 // ObjectSynchronizer::oops_do() (if it happens after omFlush() and the thread's
 // monitors have been transferred to the global in-use list).
+//
+// With AsyncDeflateIdleMonitors, deflate_global_idle_monitors_using_JT()
+// and deflate_per_thread_idle_monitors_using_JT() (in another thread) can
+// run at the same time as omFlush() so we have to be careful.
 
 void ObjectSynchronizer::omFlush(Thread * Self) {
   ObjectMonitor * list = Self->omFreeList;  // Null-terminated SLL
   ObjectMonitor * tail = NULL;
   int tally = 0;

@@ -1230,11 +1343,11 @@
       guarantee(s->object() == NULL, "invariant");
       guarantee(!s->is_busy(), "invariant");
       s->set_owner(NULL);   // redundant but good hygiene
     }
     guarantee(tail != NULL, "invariant");
-    assert(Self->omFreeCount == tally, "free-count off");
+    guarantee(Self->omFreeCount == tally, "free-count off");
     Self->omFreeList = NULL;
     Self->omFreeCount = 0;
   }
 
   ObjectMonitor * inUseList = Self->omInUseList;

@@ -1247,13 +1360,14 @@
     // Link them to inUseTail, which will be linked into the global in-use list
     // gOmInUseList below, under the gListLock
     for (cur_om = inUseList; cur_om != NULL; cur_om = cur_om->FreeNext) {
       inUseTail = cur_om;
       inUseTally++;
+      guarantee(cur_om->is_active(), "invariant");
     }
     guarantee(inUseTail != NULL, "invariant");
-    assert(Self->omInUseCount == inUseTally, "in-use count off");
+    guarantee(Self->omInUseCount == inUseTally, "in-use count off");
     Self->omInUseList = NULL;
     Self->omInUseCount = 0;
   }
 
   Thread::muxAcquire(&gListLock, "omFlush");

@@ -1297,23 +1411,32 @@
   event->set_cause((u1)cause);
   event->commit();
 }
 
 // Fast path code shared by multiple functions
-void ObjectSynchronizer::inflate_helper(oop obj) {
+void ObjectSynchronizer::inflate_helper(ObjectMonitorHandle * omh_p, oop obj) {
+  while (true) {
   markOop mark = obj->mark();
   if (mark->has_monitor()) {
-    assert(ObjectSynchronizer::verify_objmon_isinpool(mark->monitor()), "monitor is invalid");
-    assert(mark->monitor()->header()->is_neutral(), "monitor must record a good object header");
+      if (!omh_p->save_om_ptr(obj, mark)) {
+        // Lost a race with async deflation so try again.
+        assert(AsyncDeflateIdleMonitors, "sanity check");
+        continue;
+      }
+      ObjectMonitor * monitor = omh_p->om_ptr();
+      assert(ObjectSynchronizer::verify_objmon_isinpool(monitor), "monitor is invalid");
+      markOop dmw = monitor->header();
+      assert(dmw->is_neutral(), "sanity check: header=" INTPTR_FORMAT, p2i((address)dmw));
+      return;
+    }
+    inflate(omh_p, Thread::current(), obj, inflate_cause_vm_internal);
     return;
   }
-  inflate(Thread::current(), obj, inflate_cause_vm_internal);
 }
 
-ObjectMonitor* ObjectSynchronizer::inflate(Thread * Self,
-                                           oop object,
-                                           const InflateCause cause) {
+void ObjectSynchronizer::inflate(ObjectMonitorHandle * omh_p, Thread * Self,
+                                 oop object, const InflateCause cause) {
   // Inflate mutates the heap ...
   // Relaxing assertion for bug 6320749.
   assert(Universe::verify_in_progress() ||
          !SafepointSynchronize::is_at_safepoint(), "invariant");
 

@@ -1330,16 +1453,21 @@
     // *  Neutral      - aggressively inflate the object.
     // *  BIASED       - Illegal.  We should never see this
 
     // CASE: inflated
     if (mark->has_monitor()) {
-      ObjectMonitor * inf = mark->monitor();
+      if (!omh_p->save_om_ptr(object, mark)) {
+        // Lost a race with async deflation so try again.
+        assert(AsyncDeflateIdleMonitors, "sanity check");
+        continue;
+      }
+      ObjectMonitor * inf = omh_p->om_ptr();
       markOop dmw = inf->header();
       assert(dmw->is_neutral(), "invariant: header=" INTPTR_FORMAT, p2i((address)dmw));
       assert(oopDesc::equals((oop) inf->object(), object), "invariant");
       assert(ObjectSynchronizer::verify_objmon_isinpool(inf), "monitor is invalid");
-      return inf;
+      return;
     }
 
     // CASE: inflation in progress - inflating over a stack-lock.
     // Some other thread is converting from stack-locked to inflated.
     // Only that thread can complete inflation -- other threads must wait.

@@ -1371,11 +1499,22 @@
     // See the comments in omAlloc().
 
     LogStreamHandle(Trace, monitorinflation) lsh;
 
     if (mark->has_locker()) {
-      ObjectMonitor * m = omAlloc(Self);
+      ObjectMonitor * m;
+      if (!AsyncDeflateIdleMonitors || cause == inflate_cause_vm_internal) {
+        // If !AsyncDeflateIdleMonitors or if an internal inflation, then
+        // we won't stop for a potential safepoint in omAlloc.
+        m = omAlloc(Self, cause);
+      } else {
+        // If AsyncDeflateIdleMonitors and not an internal inflation, then
+        // we may stop for a safepoint in omAlloc() so protect object.
+        Handle h_obj(Self, object);
+        m = omAlloc(Self, cause);
+        object = h_obj();  // Refresh object.
+      }
       // Optimistically prepare the objectmonitor - anticipate successful CAS
       // We do this before the CAS in order to minimize the length of time
       // in which INFLATING appears in the mark.
       m->Recycle();
       m->_Responsible  = NULL;

@@ -1448,11 +1587,13 @@
                      p2i(object->mark()), object->klass()->external_name());
       }
       if (event.should_commit()) {
         post_monitor_inflate_event(&event, object, cause);
       }
-      return m;
+      assert(!m->is_free(), "post-condition");
+      omh_p->set_om_ptr(m);
+      return;
     }
 
     // CASE: neutral
     // TODO-FIXME: for entry we currently inflate and then try to CAS _owner.
     // If we know we're inflating for entry it's better to inflate by swinging a

@@ -1462,11 +1603,22 @@
     // to inflate and then CAS() again to try to swing _owner from NULL to Self.
     // An inflateTry() method that we could call from fast_enter() and slow_enter()
     // would be useful.
 
     assert(mark->is_neutral(), "invariant");
-    ObjectMonitor * m = omAlloc(Self);
+    ObjectMonitor * m;
+    if (!AsyncDeflateIdleMonitors || cause == inflate_cause_vm_internal) {
+      // If !AsyncDeflateIdleMonitors or if an internal inflation, then
+      // we won't stop for a potential safepoint in omAlloc.
+      m = omAlloc(Self, cause);
+    } else {
+      // If AsyncDeflateIdleMonitors and not an internal inflation, then
+      // we may stop for a safepoint in omAlloc() so protect object.
+      Handle h_obj(Self, object);
+      m = omAlloc(Self, cause);
+      object = h_obj();  // Refresh object.
+    }
     // prepare m for installation - set monitor to initial state
     m->Recycle();
     m->set_header(mark);
     m->set_owner(NULL);
     m->set_object(object);

@@ -1496,11 +1648,12 @@
                    p2i(object->mark()), object->klass()->external_name());
     }
     if (event.should_commit()) {
       post_monitor_inflate_event(&event, object, cause);
     }
-    return m;
+    omh_p->set_om_ptr(m);
+    return;
   }
 }
 
 
 // We create a list of in-use monitors for each thread.

@@ -1522,10 +1675,34 @@
 // Perversely, the heap size -- and thus the STW safepoint rate --
 // typically drives the scavenge rate.  Large heaps can mean infrequent GC,
 // which in turn can mean large(r) numbers of objectmonitors in circulation.
 // This is an unfortunate aspect of this design.
 
+void ObjectSynchronizer::do_safepoint_work(DeflateMonitorCounters* _counters) {
+  if (!AsyncDeflateIdleMonitors) {
+    // Use the older mechanism for the global in-use list.
+    ObjectSynchronizer::deflate_idle_monitors(_counters);
+    return;
+  }
+
+  assert(_counters == NULL, "not used with AsyncDeflateIdleMonitors");
+
+  log_debug(monitorinflation)("requesting deflation of idle monitors.");
+  // Request deflation of global idle monitors by the ServiceThread:
+  _gOmShouldDeflateIdleMonitors = true;
+  MutexLockerEx ml(Service_lock, Mutex::_no_safepoint_check_flag);
+  Service_lock->notify_all();
+
+  // Request deflation of per-thread idle monitors by each JavaThread:
+  for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
+    if (jt->omInUseCount > 0) {
+      // This JavaThread is using monitors so check it.
+      jt->omShouldDeflateIdleMonitors = true;
+    }
+  }
+}
+
 // Deflate a single monitor if not in-use
 // Return true if deflated, false if in-use
 bool ObjectSynchronizer::deflate_monitor(ObjectMonitor* mid, oop obj,
                                          ObjectMonitor** freeHeadp,
                                          ObjectMonitor** freeTailp) {

@@ -1552,10 +1729,11 @@
     // Restore the header back to obj
     obj->release_set_mark(mid->header());
     mid->clear();
 
     assert(mid->object() == NULL, "invariant");
+    assert(mid->is_free(), "invariant");
 
     // Move the object to the working free list defined by freeHeadp, freeTailp
     if (*freeHeadp == NULL) *freeHeadp = mid;
     if (*freeTailp != NULL) {
       ObjectMonitor * prevtail = *freeTailp;

@@ -1566,10 +1744,136 @@
     deflated = true;
   }
   return deflated;
 }
 
+// Deflate the specified ObjectMonitor if not in-use using a JavaThread.
+// Returns true if it was deflated and false otherwise.
+//
+// The async deflation protocol sets _owner to DEFLATER_MARKER and
+// makes _count negative as signals to contending threads that an
+// async deflation is in progress. There are a number of checks as
+// part of the protocol to make sure that the calling thread has
+// not lost the race to a contending thread.
+//
+// The ObjectMonitor has been successfully async deflated when:
+// (_owner == DEFLATER_MARKER && _count < 0). Contending threads that
+// see those values know to retry their operation.
+//
+bool ObjectSynchronizer::deflate_monitor_using_JT(ObjectMonitor* mid,
+                                                  ObjectMonitor** freeHeadp,
+                                                  ObjectMonitor** freeTailp) {
+  assert(AsyncDeflateIdleMonitors, "sanity check");
+  assert(Thread::current()->is_Java_thread(), "precondition");
+  // A newly allocated ObjectMonitor should not be seen here so we
+  // avoid an endless inflate/deflate cycle.
+  assert(mid->is_old(), "precondition");
+
+  if (mid->is_busy() || mid->ref_count() != 0) {
+    // Easy checks are first - the ObjectMonitor is busy or ObjectMonitor*
+    // is in use so no deflation.
+    return false;
+  }
+
+  if (Atomic::cmpxchg(DEFLATER_MARKER, &mid->_owner, (void*)NULL) == NULL) {
+    // ObjectMonitor is not owned by another thread. Our setting
+    // _owner to DEFLATER_MARKER forces any contending thread through
+    // the slow path. This is just the first part of the async
+    // deflation dance.
+
+    if (mid->_waiters != 0 || mid->ref_count() != 0) {
+      // Another thread has raced to enter the ObjectMonitor after
+      // mid->is_busy() above and has already waited on it which
+      // makes it busy so no deflation. Or the ObjectMonitor* is
+      // in use for some other operation like inflate(). Restore
+      // _owner to NULL if it is still DEFLATER_MARKER.
+      Atomic::cmpxchg((void*)NULL, &mid->_owner, DEFLATER_MARKER);
+      return false;
+    }
+
+    if (Atomic::cmpxchg(-max_jint, &mid->_count, (jint)0) == 0) {
+      // Make _count negative to force racing threads to retry.
+      // This is the second part of the async deflation dance.
+
+      if (mid->_owner == DEFLATER_MARKER) {
+        // If _owner is still DEFLATER_MARKER, then we have successfully
+        // signaled any racing threads to retry. If it is not, then we
+        // have lost the race to another thread and the ObjectMonitor is
+        // now busy. This is the third and final part of the async
+        // deflation dance.
+        // Note: This _owner check solves the ABA problem with _count
+        // where another thread acquired the ObjectMonitor, finished
+        // using it and restored the _count to zero.
+
+        // Sanity checks for the races:
+        guarantee(mid->_waiters == 0, "should be no waiters");
+        guarantee(mid->_cxq == NULL, "should be no contending threads");
+        guarantee(mid->_EntryList == NULL, "should be no entering threads");
+
+        if (log_is_enabled(Trace, monitorinflation)) {
+          oop obj = (oop) mid->object();
+          assert(obj != NULL, "sanity check");
+          if (obj->is_instance()) {
+            ResourceMark rm;
+            log_trace(monitorinflation)("deflate_monitor_using_JT: "
+                                        "object=" INTPTR_FORMAT ", mark=" INTPTR_FORMAT ", type='%s'",
+                                        p2i(obj), p2i(obj->mark()),
+                                        obj->klass()->external_name());
+          }
+        }
+
+        // Install the old mark word if nobody else has already done it.
+        mid->install_displaced_markword_in_object();
+        mid->clear_using_JT();
+
+        assert(mid->object() == NULL, "invariant");
+        assert(mid->is_free(), "invariant");
+
+        // Move the deflated ObjectMonitor to the working free list
+        // defined by freeHeadp and freeTailp.
+        if (*freeHeadp == NULL) {
+          // First one on the list.
+          *freeHeadp = mid;
+        }
+        if (*freeTailp != NULL) {
+          // We append to the list so the caller can use mid->FreeNext
+          // to fix the linkages in its context.
+          ObjectMonitor * prevtail = *freeTailp;
+          assert(prevtail->FreeNext == NULL, "not cleaned up by the caller");
+          prevtail->FreeNext = mid;
+        }
+        *freeTailp = mid;
+
+        // At this point, mid->FreeNext still refers to its current
+        // value and another ObjectMonitor's FreeNext field still
+        // refers to this ObjectMonitor. Those linkages have to be
+        // cleaned up by the caller who has the complete context.
+
+        // We leave _owner == DEFLATER_MARKER and _count < 0 to
+        // force any racing threads to retry.
+        return true;  // Success, ObjectMonitor has been deflated.
+      }
+
+      // The _owner was changed from DEFLATER_MARKER so we lost the
+      // race since the ObjectMonitor is now busy. Add back max_jint
+      // to restore the _count field to its proper value (which may
+      // not be what we saw above).
+      Atomic::add(max_jint, &mid->_count);
+
+      assert(mid->_count >= 0, "_count should not be negative");
+    }
+
+    // The _count was no longer 0 so we lost the race since the
+    // ObjectMonitor is now busy.
+    assert(mid->_owner != DEFLATER_MARKER, "should no longer be set");
+  }
+
+  // The _owner field is no longer NULL so we lost the race since the
+  // ObjectMonitor is now busy.
+  return false;
+}
+
 // Walk a given monitor list, and deflate idle monitors
 // The given list could be a per-thread list or a global list
 // Caller acquires gListLock as needed.
 //
 // In the case of parallel processing of thread local monitor lists,

@@ -1609,19 +1913,101 @@
     }
   }
   return deflated_count;
 }
 
+// Walk a given ObjectMonitor list and deflate idle ObjectMonitors using
+// a JavaThread. Returns the number of deflated ObjectMonitors. The given
+// list could be a per-thread in-use list or the global in-use list.
+// Caller acquires gListLock as appropriate. If a safepoint has started,
+// then we save state via savedMidInUsep and return to the caller to
+// honor the safepoint.
+//
+int ObjectSynchronizer::deflate_monitor_list_using_JT(ObjectMonitor** listHeadp,
+                                                      ObjectMonitor** freeHeadp,
+                                                      ObjectMonitor** freeTailp,
+                                                      ObjectMonitor** savedMidInUsep) {
+  assert(AsyncDeflateIdleMonitors, "sanity check");
+  assert(Thread::current()->is_Java_thread(), "precondition");
+
+  ObjectMonitor* mid;
+  ObjectMonitor* next;
+  ObjectMonitor* cur_mid_in_use = NULL;
+  int deflated_count = 0;
+
+  if (*savedMidInUsep == NULL) {
+    // No saved state so start at the beginning.
+    mid = *listHeadp;
+  } else {
+    // We're restarting after a safepoint so restore the necessary state
+    // before we resume.
+    cur_mid_in_use = *savedMidInUsep;
+    mid = cur_mid_in_use->FreeNext;
+  }
+  while (mid != NULL) {
+    // Only try to deflate if there is an associated Java object and if
+    // mid is old (is not newly allocated and is not newly freed).
+    if (mid->object() != NULL && mid->is_old() &&
+        deflate_monitor_using_JT(mid, freeHeadp, freeTailp)) {
+      // Deflation succeeded so update the in-use list.
+      if (mid == *listHeadp) {
+        *listHeadp = mid->FreeNext;
+      } else if (cur_mid_in_use != NULL) {
+        // Maintain the current in-use list.
+        cur_mid_in_use->FreeNext = mid->FreeNext;
+      }
+      next = mid->FreeNext;
+      mid->FreeNext = NULL;
+      // At this point mid is disconnected from the in-use list
+      // and is the current tail in the freeHeadp list.
+      mid = next;
+      deflated_count++;
+    } else {
+      // mid is considered in-use if it does not have an associated
+      // Java object or mid is not old or deflation did not succeed.
+      // A mid->is_new() node can be seen here when it is freshly returned
+      // by omAlloc() (and skips the deflation code path).
+      // A mid->is_old() node can be seen here when deflation failed.
+      // A mid->is_free() node can be seen here when a fresh node from
+      // omAlloc() is released by omRelease() due to losing the race
+      // in inflate().
+
+      if (mid->object() != NULL && mid->is_new()) {
+        // mid has an associated Java object and has now been seen
+        // as newly allocated so mark it as "old".
+        mid->set_allocation_state(ObjectMonitor::Old);
+      }
+      cur_mid_in_use = mid;
+      mid = mid->FreeNext;
+
+      if (SafepointSynchronize::is_synchronizing() &&
+          cur_mid_in_use != *listHeadp && cur_mid_in_use->is_old()) {
+        // If a safepoint has started and cur_mid_in_use is not the list
+        // head and is old, then it is safe to use as saved state. Return
+        // to the caller so gListLock can be dropped as appropriate
+        // before blocking.
+        *savedMidInUsep = cur_mid_in_use;
+        return deflated_count;
+      }
+    }
+  }
+  // We finished the list without a safepoint starting so there's
+  // no need to save state.
+  *savedMidInUsep = NULL;
+  return deflated_count;
+}
+
 void ObjectSynchronizer::prepare_deflate_idle_monitors(DeflateMonitorCounters* counters) {
   counters->nInuse = 0;              // currently associated with objects
   counters->nInCirculation = 0;      // extant
   counters->nScavenged = 0;          // reclaimed (global and per-thread)
   counters->perThreadScavenged = 0;  // per-thread scavenge total
   counters->perThreadTimes = 0.0;    // per-thread scavenge times
 }
 
 void ObjectSynchronizer::deflate_idle_monitors(DeflateMonitorCounters* counters) {
+  assert(!AsyncDeflateIdleMonitors, "sanity check");
   assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
   bool deflated = false;
 
   ObjectMonitor * freeHeadp = NULL;  // Local SLL of scavenged monitors
   ObjectMonitor * freeTailp = NULL;

@@ -1671,19 +2057,175 @@
   if (ls != NULL) {
     ls->print_cr("deflating global idle monitors, %3.7f secs, %d monitors", timer.seconds(), deflated_count);
   }
 }
 
+// Deflate global idle ObjectMonitors using a JavaThread.
+//
+void ObjectSynchronizer::deflate_global_idle_monitors_using_JT() {
+  assert(AsyncDeflateIdleMonitors, "sanity check");
+  assert(Thread::current()->is_Java_thread(), "precondition");
+  JavaThread * cur_jt = JavaThread::current();
+
+  _gOmShouldDeflateIdleMonitors = false;
+
+  int deflated_count = 0;
+  ObjectMonitor * freeHeadp = NULL;  // Local SLL of scavenged ObjectMonitors
+  ObjectMonitor * freeTailp = NULL;
+  ObjectMonitor * savedMidInUsep = NULL;
+  elapsedTimer timer;
+
+  if (log_is_enabled(Info, monitorinflation)) {
+    timer.start();
+  }
+  Thread::muxAcquire(&gListLock, "deflate_global_idle_monitors_using_JT(1)");
+  OM_PERFDATA_OP(MonExtant, set_value(gOmInUseCount));
+
+  do {
+    int local_deflated_count = deflate_monitor_list_using_JT((ObjectMonitor **)&gOmInUseList, &freeHeadp, &freeTailp, &savedMidInUsep);
+    gOmInUseCount -= local_deflated_count;
+    deflated_count += local_deflated_count;
+
+    if (freeHeadp != NULL) {
+      // Move the scavenged ObjectMonitors to the global free list.
+      guarantee(freeTailp != NULL && local_deflated_count > 0, "freeTailp=" INTPTR_FORMAT ", local_deflated_count=%d", p2i(freeTailp), local_deflated_count);
+      assert(freeTailp->FreeNext == NULL, "invariant");
+
+      // Constant-time list splice - prepend scavenged segment to gFreeList.
+      freeTailp->FreeNext = gFreeList;
+      gFreeList = freeHeadp;
+
+      gMonitorFreeCount += local_deflated_count;
+      OM_PERFDATA_OP(Deflations, inc(local_deflated_count));
+    }
+
+    if (savedMidInUsep != NULL) {
+      // deflate_monitor_list_using_JT() detected a safepoint starting.
+      Thread::muxRelease(&gListLock);
+      timer.stop();
+      {
+        log_debug(monitorinflation)("pausing deflation of global idle monitors for a safepoint.");
+        assert(SafepointSynchronize::is_synchronizing(), "sanity check");
+        ThreadBlockInVM blocker(cur_jt);
+      }
+      // Prepare for another loop after the safepoint.
+      freeHeadp = NULL;
+      freeTailp = NULL;
+      if (log_is_enabled(Info, monitorinflation)) {
+        timer.start();
+      }
+      Thread::muxAcquire(&gListLock, "deflate_global_idle_monitors_using_JT(2)");
+    }
+  } while (savedMidInUsep != NULL);
+  Thread::muxRelease(&gListLock);
+  timer.stop();
+
+  LogStreamHandle(Debug, monitorinflation) lsh_debug;
+  LogStreamHandle(Info, monitorinflation) lsh_info;
+  LogStream * ls = NULL;
+  if (log_is_enabled(Debug, monitorinflation)) {
+    ls = &lsh_debug;
+  } else if (deflated_count != 0 && log_is_enabled(Info, monitorinflation)) {
+    ls = &lsh_info;
+  }
+  if (ls != NULL) {
+    ls->print_cr("async-deflating global idle monitors, %3.7f secs, %d monitors", timer.seconds(), deflated_count);
+  }
+}
+
+// Deflate per-thread idle ObjectMonitors using a JavaThread.
+//
+void ObjectSynchronizer::deflate_per_thread_idle_monitors_using_JT() {
+  assert(AsyncDeflateIdleMonitors, "sanity check");
+  assert(Thread::current()->is_Java_thread(), "precondition");
+  JavaThread * cur_jt = JavaThread::current();
+
+  cur_jt->omShouldDeflateIdleMonitors = false;
+
+  int deflated_count = 0;
+  ObjectMonitor * freeHeadp = NULL;  // Local SLL of scavenged ObjectMonitors
+  ObjectMonitor * freeTailp = NULL;
+  ObjectMonitor * savedMidInUsep = NULL;
+  elapsedTimer timer;
+
+  if (log_is_enabled(Info, monitorinflation)) {
+    timer.start();
+  }
+
+  OM_PERFDATA_OP(MonExtant, inc(cur_jt->omInUseCount));
+  do {
+    int local_deflated_count = deflate_monitor_list_using_JT(cur_jt->omInUseList_addr(), &freeHeadp, &freeTailp, &savedMidInUsep);
+    cur_jt->omInUseCount -= local_deflated_count;
+    deflated_count += local_deflated_count;
+
+    if (freeHeadp != NULL) {
+      // Move the scavenged ObjectMonitors to the global free list.
+      Thread::muxAcquire(&gListLock, "deflate_per_thread_idle_monitors_using_JT");
+      guarantee(freeTailp != NULL && local_deflated_count > 0, "freeTailp=" INTPTR_FORMAT ", local_deflated_count=%d", p2i(freeTailp), local_deflated_count);
+      assert(freeTailp->FreeNext == NULL, "invariant");
+
+      // Constant-time list splice - prepend scavenged segment to gFreeList.
+      freeTailp->FreeNext = gFreeList;
+      gFreeList = freeHeadp;
+
+      gMonitorFreeCount += local_deflated_count;
+      OM_PERFDATA_OP(Deflations, inc(local_deflated_count));
+      Thread::muxRelease(&gListLock);
+      // Prepare for another loop on the current JavaThread.
+      freeHeadp = NULL;
+      freeTailp = NULL;
+    }
+    timer.stop();
+
+    if (savedMidInUsep != NULL) {
+      // deflate_monitor_list_using_JT() detected a safepoint starting.
+      {
+        log_debug(monitorinflation)("jt=" INTPTR_FORMAT ": pausing deflation of per-thread idle monitors for a safepoint.", p2i(cur_jt));
+        assert(SafepointSynchronize::is_synchronizing(), "sanity check");
+        ThreadBlockInVM blocker(cur_jt);
+      }
+      // Prepare for another loop on the current JavaThread after
+      // the safepoint.
+      if (log_is_enabled(Info, monitorinflation)) {
+        timer.start();
+      }
+    }
+  } while (savedMidInUsep != NULL);
+
+  LogStreamHandle(Debug, monitorinflation) lsh_debug;
+  LogStreamHandle(Info, monitorinflation) lsh_info;
+  LogStream * ls = NULL;
+  if (log_is_enabled(Debug, monitorinflation)) {
+    ls = &lsh_debug;
+  } else if (deflated_count != 0 && log_is_enabled(Info, monitorinflation)) {
+    ls = &lsh_info;
+  }
+  if (ls != NULL) {
+    ls->print_cr("jt=" INTPTR_FORMAT ": async-deflating per-thread idle monitors, %3.7f secs, %d monitors", p2i(cur_jt), timer.seconds(), deflated_count);
+  }
+}
+
 void ObjectSynchronizer::finish_deflate_idle_monitors(DeflateMonitorCounters* counters) {
   // Report the cumulative time for deflating each thread's idle
   // monitors. Note: if the work is split among more than one
   // worker thread, then the reported time will likely be more
   // than a beginning to end measurement of the phase.
+  // Note: AsyncDeflateIdleMonitors only deflates per-thread idle
+  // monitors at a safepoint when a special cleanup has been requested.
   log_info(safepoint, cleanup)("deflating per-thread idle monitors, %3.7f secs, monitors=%d", counters->perThreadTimes, counters->perThreadScavenged);
 
+  bool needs_special_cleanup = is_cleanup_requested();
+  if (!AsyncDeflateIdleMonitors || needs_special_cleanup) {
+    // AsyncDeflateIdleMonitors does not use these counters unless
+    // there is a special cleanup request.
+
   gMonitorFreeCount += counters->nScavenged;
 
+    OM_PERFDATA_OP(Deflations, inc(counters->nScavenged));
+    OM_PERFDATA_OP(MonExtant, set_value(counters->nInCirculation));
+  }
+
   if (log_is_enabled(Debug, monitorinflation)) {
     // exit_globals()'s call to audit_and_print_stats() is done
     // at the Info level.
     ObjectSynchronizer::audit_and_print_stats(false /* on_exit */);
   } else if (log_is_enabled(Info, monitorinflation)) {

@@ -1693,21 +2235,28 @@
                                gOmInUseCount, gMonitorFreeCount);
     Thread::muxRelease(&gListLock);
   }
 
   ForceMonitorScavenge = 0;    // Reset
-
-  OM_PERFDATA_OP(Deflations, inc(counters->nScavenged));
-  OM_PERFDATA_OP(MonExtant, set_value(counters->nInCirculation));
-
   GVars.stwRandom = os::random();
   GVars.stwCycle++;
+  if (needs_special_cleanup) {
+    set_is_cleanup_requested(false);  // special clean up is done
+  }
 }
 
 void ObjectSynchronizer::deflate_thread_local_monitors(Thread* thread, DeflateMonitorCounters* counters) {
   assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
 
+  if (AsyncDeflateIdleMonitors) {
+    // Nothing to do when idle ObjectMonitors are deflated using a
+    // JavaThread unless a special cleanup has been requested.
+    if (!is_cleanup_requested()) {
+      return;
+    }
+  }
+
   ObjectMonitor * freeHeadp = NULL;  // Local SLL of scavenged monitors
   ObjectMonitor * freeTailp = NULL;
   elapsedTimer timer;
 
   if (log_is_enabled(Info, safepoint, cleanup) ||

@@ -1915,11 +2464,12 @@
 }
 
 // Check a free monitor entry; log any errors.
 void ObjectSynchronizer::chk_free_entry(JavaThread * jt, ObjectMonitor * n,
                                         outputStream * out, int *error_cnt_p) {
-  if (n->is_busy()) {
+  if ((!AsyncDeflateIdleMonitors && n->is_busy()) ||
+      (AsyncDeflateIdleMonitors && n->is_busy_async())) {
     if (jt != NULL) {
       out->print_cr("ERROR: jt=" INTPTR_FORMAT ", monitor=" INTPTR_FORMAT
                     ": free per-thread monitor must not be busy.", p2i(jt),
                     p2i(n));
     } else {

@@ -2107,41 +2657,42 @@
   }
 
   if (gOmInUseCount > 0) {
     out->print_cr("In-use global monitor info:");
     out->print_cr("(B -> is_busy, H -> has hashcode, L -> lock status)");
-    out->print_cr("%18s  %s  %18s  %18s",
-                  "monitor", "BHL", "object", "object type");
-    out->print_cr("==================  ===  ==================  ==================");
+    out->print_cr("%18s  %s  %7s  %18s  %18s",
+                  "monitor", "BHL", "ref_cnt", "object", "object type");
+    out->print_cr("==================  ===  =======  ==================  ==================");
     for (ObjectMonitor * n = gOmInUseList; n != NULL; n = n->FreeNext) {
       const oop obj = (oop) n->object();
       const markOop mark = n->header();
       ResourceMark rm;
-      out->print_cr(INTPTR_FORMAT "  %d%d%d  " INTPTR_FORMAT "  %s", p2i(n),
-                    n->is_busy() != 0, mark->hash() != 0, n->owner() != NULL,
-                    p2i(obj), obj->klass()->external_name());
+      out->print_cr(INTPTR_FORMAT "  %d%d%d  %7d  " INTPTR_FORMAT "  %s",
+                    p2i(n), n->is_busy() != 0, mark->hash() != 0,
+                    n->owner() != NULL, (int)n->ref_count(), p2i(obj),
+                    obj->klass()->external_name());
     }
   }
 
   if (!on_exit) {
     Thread::muxRelease(&gListLock);
   }
 
   out->print_cr("In-use per-thread monitor info:");
   out->print_cr("(B -> is_busy, H -> has hashcode, L -> lock status)");
-  out->print_cr("%18s  %18s  %s  %18s  %18s",
-                "jt", "monitor", "BHL", "object", "object type");
-  out->print_cr("==================  ==================  ===  ==================  ==================");
+  out->print_cr("%18s  %18s  %s  %7s  %18s  %18s",
+                "jt", "monitor", "BHL", "ref_cnt", "object", "object type");
+  out->print_cr("==================  ==================  ===  =======  ==================  ==================");
   for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
     for (ObjectMonitor * n = jt->omInUseList; n != NULL; n = n->FreeNext) {
       const oop obj = (oop) n->object();
       const markOop mark = n->header();
       ResourceMark rm;
-      out->print_cr(INTPTR_FORMAT "  " INTPTR_FORMAT "  %d%d%d  " INTPTR_FORMAT
-                    "  %s", p2i(jt), p2i(n), n->is_busy() != 0,
-                    mark->hash() != 0, n->owner() != NULL, p2i(obj),
-                    obj->klass()->external_name());
+      out->print_cr(INTPTR_FORMAT "  " INTPTR_FORMAT "  %d%d%d  %7d  "
+                    INTPTR_FORMAT "  %s", p2i(jt), p2i(n), n->is_busy() != 0,
+                    mark->hash() != 0, n->owner() != NULL, (int)n->ref_count(),
+                    p2i(obj), obj->klass()->external_name());
     }
   }
 
   out->flush();
 }