< prev index next >

src/hotspot/share/runtime/synchronizer.cpp

Print this page
rev 56044 : imported patch 8230184.patch
rev 56046 : v2.00 -> v2.05 (CR5/v2.05/8-for-jdk13) patches combined into one; merge with 8229212.patch; merge with jdk-14+11; merge with 8230184.patch.

@@ -123,10 +123,13 @@
 ObjectMonitor* volatile ObjectSynchronizer::g_free_list = NULL;
 // Global ObjectMonitor in-use list. When a JavaThread is exiting,
 // ObjectMonitors on its per-thread in-use list are prepended here.
 ObjectMonitor* volatile ObjectSynchronizer::g_om_in_use_list = NULL;
 int ObjectSynchronizer::g_om_in_use_count = 0;  // # on g_om_in_use_list
+bool volatile ObjectSynchronizer::_is_async_deflation_requested = false;
+bool volatile ObjectSynchronizer::_is_special_deflation_requested = false;
+jlong ObjectSynchronizer::_last_async_deflation_time_ns = 0;
 
 static volatile intptr_t gListLock = 0;   // protects global monitor lists
 static volatile int g_om_free_count = 0;  // # on g_free_list
 static volatile int g_om_population = 0;  // # Extant -- in circulation
 

@@ -209,14 +212,22 @@
   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 markWord 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

@@ -242,10 +253,22 @@
 
     if (owner == NULL && Atomic::replace_if_null(self, &(m->_owner))) {
       assert(m->_recursions == 0, "invariant");
       return true;
     }
+
+      if (AsyncDeflateIdleMonitors &&
+          Atomic::cmpxchg(self, &m->_owner, DEFLATER_MARKER) == DEFLATER_MARKER) {
+        // The deflation protocol finished the first part (setting owner),
+        // but it failed the second part (making ref_count negative) and
+        // bailed. Or the ObjectMonitor was async deflated and reused.
+        // Acquired the monitor.
+        assert(m->_recursions == 0, "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:

@@ -325,11 +348,13 @@
       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

@@ -358,11 +383,13 @@
   // 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(markWord::unused_mark());
-  inflate(THREAD, obj(), inflate_cause_monitor_enter)->enter(THREAD);
+  ObjectMonitorHandle omh;
+  inflate(&omh, THREAD, obj(), inflate_cause_monitor_enter);
+  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 +414,26 @@
   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);
+  ObjectMonitorHandle omh;
+  inflate(&omh, THREAD, obj(), inflate_cause_vm_internal);
+  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 +441,13 @@
   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);
+  ObjectMonitorHandle omh;
+  inflate(&omh, THREAD, obj(), inflate_cause_jni_enter);
+  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 +456,13 @@
     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. We
   // intentionally do not use CHECK here because we must exit the
   // monitor even if an exception is pending.
   if (monitor->check_owner(THREAD)) {
     monitor->exit(true, THREAD);

@@ -467,31 +499,36 @@
     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::wait_uninterruptibly(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);

@@ -500,11 +537,13 @@
 
   markWord 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) {

@@ -514,11 +553,13 @@
 
   markWord 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
 //

@@ -708,10 +749,11 @@
   assert(Universe::verify_in_progress() || DumpSharedSpaces ||
          self->is_Java_thread() , "invariant");
   assert(Universe::verify_in_progress() || DumpSharedSpaces ||
          ((JavaThread *)self)->thread_state() != _thread_blocked, "invariant");
 
+  while (true) {
   ObjectMonitor* monitor = NULL;
   markWord temp, test;
   intptr_t hash;
   markWord mark = read_stable_mark(obj);
 

@@ -732,11 +774,17 @@
     }
     // 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");
+        continue;
+      }
+      monitor = omh.om_ptr();
     temp = monitor->header();
     assert(temp.is_neutral(), "invariant: header=" INTPTR_FORMAT, temp.value());
     hash = temp.hash();
     if (hash != 0) {
       return hash;

@@ -758,11 +806,13 @@
     // during an inflate() call so any change to that stack memory
     // 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, mark.value());
   hash = mark.hash();
   if (hash == 0) {

@@ -774,17 +824,22 @@
     if (test != mark) {
       // The only non-deflation update to the ObjectMonitor's
       // header/dmw field is to merge in the hash code. If someone
       // adds a new usage of the header/dmw field, please update
       // this code.
+        // ObjectMonitor::install_displaced_markword_in_object()
+        // does mark the header/dmw field as part of async deflation,
+        // but that protocol cannot happen now due to the
+        // ObjectMonitorHandle above.
       hash = test.hash();
       assert(test.is_neutral(), "invariant: header=" INTPTR_FORMAT, test.value());
       assert(hash != 0, "Trivial unexpected object/monitor header usage.");
     }
   }
   // We finally get the hash
   return hash;
+  }
 }
 
 // Deprecated -- use FastHashCode() instead.
 
 intptr_t ObjectSynchronizer::identity_hash_value_for(Handle obj) {

@@ -800,24 +855,32 @@
   }
 
   assert(thread == JavaThread::current(), "Can only be called on current thread");
   oop obj = h_obj();
 
+  while (true) {
   markWord mark = read_stable_mark(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

@@ -839,31 +902,41 @@
            "biases should be revoked by now");
   }
 
   assert(self == JavaThread::current(), "Can only be called on current thread");
   oop obj = h_obj();
+
+  while (true) {
   markWord mark = read_stable_mark(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) {

@@ -874,22 +947,29 @@
     }
     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;
   markWord mark = read_stable_mark(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) {

@@ -901,23 +981,33 @@
   // 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) {
   PaddedObjectMonitor* block = OrderAccess::load_acquire(&g_block_list);
   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 = (PaddedObjectMonitor*)block->_next_om;
   }

@@ -925,22 +1015,70 @@
 
 static bool monitors_used_above_threshold() {
   if (g_om_population == 0) {
     return false;
   }
+  if (MonitorUsedDeflationThreshold > 0) {
   int monitors_used = g_om_population - g_om_free_count;
   int monitor_usage = (monitors_used * 100LL) / g_om_population;
   return monitor_usage > MonitorUsedDeflationThreshold;
+  }
+  return false;
 }
 
-bool ObjectSynchronizer::is_cleanup_needed() {
-  if (MonitorUsedDeflationThreshold > 0) {
-    return monitors_used_above_threshold();
+// Returns true if MonitorBound is set (> 0) and if the specified
+// cnt is > MonitorBound. Otherwise returns false.
+static bool is_MonitorBound_exceeded(const int cnt) {
+  const int mx = MonitorBound;
+  return mx > 0 && cnt > mx;
+}
+
+bool ObjectSynchronizer::is_async_deflation_needed() {
+  if (!AsyncDeflateIdleMonitors) {
+    return false;
+  }
+  if (is_async_deflation_requested()) {
+    // Async deflation request.
+    return true;
+  }
+  if (AsyncDeflationInterval > 0 &&
+      time_since_last_async_deflation_ms() > AsyncDeflationInterval &&
+      monitors_used_above_threshold()) {
+    // It's been longer than our specified deflate interval and there
+    // are too many monitors in use. We don't deflate more frequently
+    // than AsyncDeflationInterval (unless is_async_deflation_requested)
+    // in order to not swamp the ServiceThread.
+    _last_async_deflation_time_ns = os::javaTimeNanos();
+    return true;
+  }
+  if (is_MonitorBound_exceeded(g_om_population - g_om_free_count)) {
+    // Not enough ObjectMonitors on the global free list.
+    return true;
   }
   return false;
 }
 
+bool ObjectSynchronizer::is_safepoint_deflation_needed() {
+  if (!AsyncDeflateIdleMonitors) {
+    if (monitors_used_above_threshold()) {
+      // Too many monitors in use.
+      return true;
+    }
+    return false;
+  }
+  if (is_special_deflation_requested()) {
+    // For AsyncDeflateIdleMonitors only do a safepoint deflation
+    // if there is a special deflation request.
+    return true;
+  }
+  return false;
+}
+
+jlong ObjectSynchronizer::time_since_last_async_deflation_ms() {
+  return (os::javaTimeNanos() - _last_async_deflation_time_ns) / (NANOUNITS / MILLIUNITS);
+}
+
 void ObjectSynchronizer::oops_do(OopClosure* f) {
   // We only scan the global used list here (for moribund threads), and
   // the thread-local monitors in Thread::oops_do().
   global_used_oops_do(f);
 }

@@ -988,10 +1126,11 @@
 
 // Constraining monitor pool growth via MonitorBound ...
 //
 // If MonitorBound is not set (<= 0), MonitorBound checks are disabled.
 //
+// When safepoint deflation is being used (!AsyncDeflateIdleMonitors):
 // The monitor pool is grow-only.  We scavenge at STW safepoint-time, but the
 // the rate of scavenging is driven primarily by GC.  As such,  we can find
 // an inordinate number of monitors in circulation.
 // To avoid that scenario we can artificially induce a STW safepoint
 // if the pool appears to be growing past some reasonable bound.

@@ -1002,17 +1141,30 @@
 // we'll incur more safepoints, which are harmful to performance.
 // See also: GuaranteedSafepointInterval
 //
 // The current implementation uses asynchronous VM operations.
 //
-// If MonitorBound is set, the boundry applies to
+// When safepoint deflation is being used and MonitorBound is set, the
+// boundry applies to
 //     (g_om_population - g_om_free_count)
 // i.e., if there are not enough ObjectMonitors on the global free list,
 // then a safepoint deflation is induced. Picking a good MonitorBound value
 // is non-trivial.
+//
+// When async deflation is being used:
+// The monitor pool is still grow-only. Async deflation is requested
+// by a safepoint's cleanup phase or by the ServiceThread at periodic
+// intervals when is_async_deflation_needed() returns true. In
+// addition to other policies that are checked, if there are not
+// enough ObjectMonitors on the global free list, then
+// is_async_deflation_needed() will return true. The ServiceThread
+// calls deflate_global_idle_monitors_using_JT() and also sets the
+// per-thread om_request_deflation flag as needed.
 
 static void InduceScavenge(Thread* self, const char * Whence) {
+  assert(!AsyncDeflateIdleMonitors, "is not used by async deflation");
+
   // Induce STW safepoint to trim monitors
   // Ultimately, this results in a call to deflate_idle_monitors() in the near future.
   // More precisely, trigger an asynchronous STW safepoint as the number
   // of active monitors passes the specified threshold.
   // TODO: assert thread state is reasonable

@@ -1024,17 +1176,35 @@
     // The VMThread will delete the op when completed.
     VMThread::execute(new VM_ScavengeMonitors());
   }
 }
 
-ObjectMonitor* ObjectSynchronizer::om_alloc(Thread* self) {
+ObjectMonitor* ObjectSynchronizer::om_alloc(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->om_request_deflation && jt->om_in_use_count > 0 &&
+        cause != inflate_cause_vm_internal) {
+      // Deflate any per-thread idle monitors for this JavaThread if
+      // this is not an internal inflation; internal inflations can
+      // occur in places where it is not safe to pause for a safepoint.
+      // Clean up your own mess (Gibbs Rule 45). Otherwise, skip this
+      // deflation. deflate_global_idle_monitors_using_JT() is called
+      // by the ServiceThread. Per-thread async deflation is triggered
+      // by the ServiceThread via om_request_deflation.
+      debug_only(jt->check_for_valid_safepoint_state(false);)
+      ObjectSynchronizer::deflate_per_thread_idle_monitors_using_JT();
+    }
+  }
+
   stringStream ss;
   for (;;) {
     ObjectMonitor* m;
 
     // 1: try to allocate from the thread's local om_free_list.

@@ -1046,10 +1216,11 @@
     m = self->om_free_list;
     if (m != NULL) {
       self->om_free_list = m->_next_om;
       self->om_free_count--;
       guarantee(m->object() == NULL, "invariant");
+      m->set_allocation_state(ObjectMonitor::New);
       m->_next_om = self->om_in_use_list;
       self->om_in_use_list = m;
       self->om_in_use_count++;
       return m;
     }

@@ -1067,19 +1238,36 @@
       for (int i = self->om_free_provision; --i >= 0 && g_free_list != NULL;) {
         g_om_free_count--;
         ObjectMonitor* take = g_free_list;
         g_free_list = take->_next_om;
         guarantee(take->object() == NULL, "invariant");
+        if (AsyncDeflateIdleMonitors) {
+          // We allowed 3 field values to linger during async deflation.
+          // We clear header and restore ref_count here, but we leave
+          // owner == DEFLATER_MARKER so the simple C2 ObjectMonitor
+          // enter optimization can no longer race with async deflation
+          // and reuse.
+          take->set_header(markWord::zero());
+          if (take->ref_count() < 0) {
+            // Add back max_jint to restore the ref_count field to its
+            // proper value.
+            Atomic::add(max_jint, &take->_ref_count);
+
+            assert(take->ref_count() >= 0, "must not be negative: ref_count=%d",
+                   take->ref_count());
+          }
+        }
         take->Recycle();
+        assert(take->is_free(), "invariant");
         om_release(self, take, false);
       }
       Thread::muxRelease(&gListLock);
       self->om_free_provision += 1 + (self->om_free_provision/2);
       if (self->om_free_provision > MAXPRIVATE) self->om_free_provision = MAXPRIVATE;
 
-      const int mx = MonitorBound;
-      if (mx > 0 && (g_om_population-g_om_free_count) > mx) {
+      if (!AsyncDeflateIdleMonitors &&
+          is_MonitorBound_exceeded(g_om_population - g_om_free_count)) {
         // Not enough ObjectMonitors on the global free list.
         // We can't safely induce a STW safepoint from om_alloc() as our thread
         // state may not be appropriate for such activities and callers may hold
         // naked oops, so instead we defer the action.
         InduceScavenge(self, "om_alloc");

@@ -1120,10 +1308,11 @@
     // 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]._next_om = (ObjectMonitor *)&temp[i+1];
+      assert(temp[i].is_free(), "invariant");
     }
 
     // terminate the last monitor as the end of list
     temp[_BLOCKSIZE - 1]._next_om = NULL;
 

@@ -1162,21 +1351,22 @@
 // a CAS 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 while we
-// are trying to release them.
+// scavenger -- deflate_monitor_list() or deflate_monitor_list_using_JT()
+// -- from reclaiming them while we are trying to release them.
 
 void ObjectSynchronizer::om_release(Thread* self, ObjectMonitor* m,
                                     bool from_per_thread_alloc) {
   guarantee(m->header().value() == 0, "invariant");
   guarantee(m->object() == NULL, "invariant");
   stringStream ss;
   guarantee((m->is_busy() | m->_recursions) == 0, "freeing in-use monitor: "
             "%s, recursions=" INTPTR_FORMAT, m->is_busy_to_string(&ss),
             m->_recursions);
+  m->set_allocation_state(ObjectMonitor::Free);
   // _next_om is used for both per-thread in-use and free lists so
   // we have to remove 'm' from the in-use list first (as needed).
   if (from_per_thread_alloc) {
     // Need to remove 'm' from om_in_use_list.
     ObjectMonitor* cur_mid_in_use = NULL;

@@ -1196,10 +1386,11 @@
     }
     assert(extracted, "Should have extracted from in-use list");
   }
 
   m->_next_om = self->om_free_list;
+  guarantee(m->is_free(), "invariant");
   self->om_free_list = m;
   self->om_free_count++;
 }
 
 // Return ObjectMonitors on a moribund thread's free and in-use

@@ -1212,10 +1403,14 @@
 // a safepoint and interleave with deflate_idle_monitors(). In
 // particular, this ensures that the thread's in-use monitors are
 // scanned by a GC safepoint, either via Thread::oops_do() (before
 // om_flush() is called) or via ObjectSynchronizer::oops_do() (after
 // om_flush() is called).
+//
+// 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 om_flush() so we have to be careful.
 
 void ObjectSynchronizer::om_flush(Thread* self) {
   ObjectMonitor* free_list = self->om_free_list;
   ObjectMonitor* free_tail = NULL;
   int free_count = 0;

@@ -1229,11 +1424,11 @@
       free_tail = s;
       guarantee(s->object() == NULL, "invariant");
       guarantee(!s->is_busy(), "must be !is_busy: %s", s->is_busy_to_string(&ss));
     }
     guarantee(free_tail != NULL, "invariant");
-    assert(self->om_free_count == free_count, "free-count off");
+    ADIM_guarantee(self->om_free_count == free_count, "free-count off");
     self->om_free_list = NULL;
     self->om_free_count = 0;
   }
 
   ObjectMonitor* in_use_list = self->om_in_use_list;

@@ -1246,13 +1441,14 @@
     // in-use list g_om_in_use_list below, under the gListLock.
     ObjectMonitor *cur_om;
     for (cur_om = in_use_list; cur_om != NULL; cur_om = cur_om->_next_om) {
       in_use_tail = cur_om;
       in_use_count++;
+      ADIM_guarantee(cur_om->is_active(), "invariant");
     }
     guarantee(in_use_tail != NULL, "invariant");
-    assert(self->om_in_use_count == in_use_count, "in-use count off");
+    ADIM_guarantee(self->om_in_use_count == in_use_count, "in-use count off");
     self->om_in_use_list = NULL;
     self->om_in_use_count = 0;
   }
 
   Thread::muxAcquire(&gListLock, "om_flush");

@@ -1296,23 +1492,32 @@
   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) {
   markWord 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");
+      markWord dmw = monitor->header();
+      assert(dmw.is_neutral(), "sanity check: header=" INTPTR_FORMAT, dmw.value());
+      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");
 

@@ -1329,16 +1534,21 @@
     // *  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();
       markWord dmw = inf->header();
       assert(dmw.is_neutral(), "invariant: header=" INTPTR_FORMAT, dmw.value());
       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.

@@ -1370,11 +1580,22 @@
     // See the comments in om_alloc().
 
     LogStreamHandle(Trace, monitorinflation) lsh;
 
     if (mark.has_locker()) {
-      ObjectMonitor* m = om_alloc(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 om_alloc.
+        m = om_alloc(self, cause);
+      } else {
+        // If AsyncDeflateIdleMonitors and not an internal inflation, then
+        // we may stop for a safepoint in om_alloc() so protect object.
+        Handle h_obj(self, object);
+        m = om_alloc(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;

@@ -1417,11 +1638,11 @@
       // object is in the mark.  Furthermore the owner can't complete
       // an unlock on the object, either.
       markWord dmw = mark.displaced_mark_helper();
       // Catch if the object's header is not neutral (not locked and
       // not marked is what we care about here).
-      assert(dmw.is_neutral(), "invariant: header=" INTPTR_FORMAT, dmw.value());
+      ADIM_guarantee(dmw.is_neutral(), "invariant: header=" INTPTR_FORMAT, dmw.value());
 
       // Setup monitor fields to proper values -- prepare the monitor
       m->set_header(dmw);
 
       // Optimization: if the mark.locker stack address is associated

@@ -1431,10 +1652,14 @@
       // with CAS.  That is, we can avoid the xchg-NULL .... ST idiom.
       m->set_owner(mark.locker());
       m->set_object(object);
       // TODO-FIXME: assert BasicLock->dhw != 0.
 
+      omh_p->set_om_ptr(m);
+      assert(m->is_new(), "freshly allocated monitor must be new");
+      m->set_allocation_state(ObjectMonitor::Old);
+
       // Must preserve store ordering. The monitor state must
       // be stable at the time of publishing the monitor address.
       guarantee(object->mark() == markWord::INFLATING(), "invariant");
       object->release_set_mark(markWord::encode(m));
 

@@ -1448,11 +1673,12 @@
                      object->mark().value(), object->klass()->external_name());
       }
       if (event.should_commit()) {
         post_monitor_inflate_event(&event, object, cause);
       }
-      return m;
+      ADIM_guarantee(!m->is_free(), "inflated monitor to be returned cannot be free");
+      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

@@ -1463,23 +1689,43 @@
     // An inflateTry() method that we could call from fast_enter() and slow_enter()
     // would be useful.
 
     // Catch if the object's header is not neutral (not locked and
     // not marked is what we care about here).
-    assert(mark.is_neutral(), "invariant: header=" INTPTR_FORMAT, mark.value());
-    ObjectMonitor* m = om_alloc(self);
+    ADIM_guarantee(mark.is_neutral(), "invariant: header=" INTPTR_FORMAT, mark.value());
+    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 om_alloc.
+      m = om_alloc(self, cause);
+    } else {
+      // If AsyncDeflateIdleMonitors and not an internal inflation, then
+      // we may stop for a safepoint in om_alloc() so protect object.
+      Handle h_obj(self, object);
+      m = om_alloc(self, cause);
+      object = h_obj();  // Refresh object.
+    }
     // prepare m for installation - set monitor to initial state
     m->Recycle();
     m->set_header(mark);
+    // If we leave _owner == DEFLATER_MARKER here, then the simple C2
+    // ObjectMonitor enter optimization can no longer race with async
+    // deflation and reuse.
     m->set_object(object);
     m->_Responsible  = NULL;
     m->_SpinDuration = ObjectMonitor::Knob_SpinLimit;       // consider: keep metastats by type/class
 
+    omh_p->set_om_ptr(m);
+    assert(m->is_new(), "freshly allocated monitor must be new");
+    m->set_allocation_state(ObjectMonitor::Old);
+
     if (object->cas_set_mark(markWord::encode(m), mark) != mark) {
       m->set_header(markWord::zero());
       m->set_object(NULL);
       m->Recycle();
+      omh_p->set_om_ptr(NULL);
+      // om_release() will reset the allocation state
       om_release(self, m, true);
       m = NULL;
       continue;
       // interference - the markword changed - just retry.
       // The state-transitions are one-way, so there's no chance of

@@ -1496,17 +1742,19 @@
                    object->mark().value(), object->klass()->external_name());
     }
     if (event.should_commit()) {
       post_monitor_inflate_event(&event, object, cause);
     }
-    return m;
+    ADIM_guarantee(!m->is_free(), "inflated monitor to be returned cannot be free");
+    return;
   }
 }
 
 
 // We maintain a list of in-use monitors for each thread.
 //
+// For safepoint based deflation:
 // deflate_thread_local_monitors() scans a single thread's in-use list, while
 // deflate_idle_monitors() scans only a global list of in-use monitors which
 // is populated only as a thread dies (see om_flush()).
 //
 // These operations are called at all safepoints, immediately after mutators

@@ -1521,10 +1769,35 @@
 //
 // 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.
+//
+// For async deflation:
+// If a special deflation request is made, then the safepoint based
+// deflation mechanism is used. Otherwise, an async deflation request
+// is registered with the ServiceThread and it is notified.
+
+void ObjectSynchronizer::do_safepoint_work(DeflateMonitorCounters* _counters) {
+  assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
+
+  // The per-thread in-use lists are handled in
+  // ParallelSPCleanupThreadClosure::do_thread().
+
+  if (!AsyncDeflateIdleMonitors || is_special_deflation_requested()) {
+    // Use the older mechanism for the global in-use list or if a
+    // special deflation has been requested before the safepoint.
+    ObjectSynchronizer::deflate_idle_monitors(_counters);
+    return;
+  }
+
+  log_debug(monitorinflation)("requesting async deflation of idle monitors.");
+  // Request deflation of idle monitors by the ServiceThread:
+  set_is_async_deflation_requested(true);
+  MonitorLocker ml(Service_lock, Mutex::_no_safepoint_check_flag);
+  ml.notify_all();
+}
 
 // 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** free_head_p,

@@ -1539,11 +1812,13 @@
   guarantee(mark.monitor() == mid, "should match: monitor()=" INTPTR_FORMAT
             ", mid=" INTPTR_FORMAT, p2i(mark.monitor()), p2i(mid));
   const markWord dmw = mid->header();
   guarantee(dmw.is_neutral(), "invariant: header=" INTPTR_FORMAT, dmw.value());
 
-  if (mid->is_busy()) {
+  if (mid->is_busy() || mid->ref_count() != 0) {
+    // Easy checks are first - the ObjectMonitor is busy or ObjectMonitor*
+    // is in use so no deflation.
     deflated = false;
   } else {
     // Deflate the monitor if it is no longer being used
     // It's idle - scavenge and return to the global free list
     // plain old deflation ...

@@ -1555,14 +1830,21 @@
                                   mark.value(), obj->klass()->external_name());
     }
 
     // Restore the header back to obj
     obj->release_set_mark(dmw);
+    if (AsyncDeflateIdleMonitors) {
+      // clear() expects the owner field to be NULL and we won't race
+      // with the simple C2 ObjectMonitor enter optimization since
+      // we're at a safepoint.
+      mid->set_owner(NULL);
+    }
     mid->clear();
 
     assert(mid->object() == NULL, "invariant: object=" INTPTR_FORMAT,
            p2i(mid->object()));
+    assert(mid->is_free(), "invariant");
 
     // Move the deflated ObjectMonitor to the working free list
     // defined by free_head_p and free_tail_p.
     if (*free_head_p == NULL) *free_head_p = mid;
     if (*free_tail_p != NULL) {

@@ -1581,10 +1863,150 @@
     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 ref_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 or to a thread that just
+// wants to use the ObjectMonitor*.
+//
+// The ObjectMonitor has been successfully async deflated when:
+// (owner == DEFLATER_MARKER && ref_count < 0)
+// Contending threads or ObjectMonitor* using threads that see those
+// values know to retry their operation.
+//
+bool ObjectSynchronizer::deflate_monitor_using_JT(ObjectMonitor* mid,
+                                                  ObjectMonitor** free_head_p,
+                                                  ObjectMonitor** free_tail_p) {
+  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(), "must be old: allocation_state=%d",
+         (int) mid->allocation_state());
+
+  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::replace_if_null(DEFLATER_MARKER, &(mid->_owner))) {
+    // 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->_contentions != 0 || mid->_waiters != 0) {
+      // Another thread has raced to enter the ObjectMonitor after
+      // mid->is_busy() above or has already entered and waited on
+      // it which makes it busy so no deflation. 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->_ref_count, (jint)0) == 0) {
+      // Make ref_count negative to force any contending threads or
+      // ObjectMonitor* using threads to retry. This is the second
+      // part of the async deflation dance.
+
+      if (mid->owner_is_DEFLATER_MARKER()) {
+        // If owner is still DEFLATER_MARKER, then we have successfully
+        // signaled any contending threads to retry. If it is not, then we
+        // have lost the race to an entering 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 ref_count
+        // where another thread acquired the ObjectMonitor, finished
+        // using it and restored the ref_count to zero.
+
+        // Sanity checks for the races:
+        guarantee(mid->_contentions == 0, "must be 0: contentions=%d",
+                  mid->_contentions);
+        guarantee(mid->_waiters == 0, "must be 0: waiters=%d", mid->_waiters);
+        guarantee(mid->_cxq == NULL, "must be no contending threads: cxq="
+                  INTPTR_FORMAT, p2i(mid->_cxq));
+        guarantee(mid->_EntryList == NULL,
+                  "must be no entering threads: EntryList=" INTPTR_FORMAT,
+                  p2i(mid->_EntryList));
+
+        const oop obj = (oop) mid->object();
+        if (log_is_enabled(Trace, monitorinflation)) {
+          ResourceMark rm;
+          log_trace(monitorinflation)("deflate_monitor_using_JT: "
+                                      "object=" INTPTR_FORMAT ", mark="
+                                      INTPTR_FORMAT ", type='%s'",
+                                      p2i(obj), obj->mark().value(),
+                                      obj->klass()->external_name());
+        }
+
+        // Install the old mark word if nobody else has already done it.
+        mid->install_displaced_markword_in_object(obj);
+        mid->clear_using_JT();
+
+        assert(mid->object() == NULL, "must be NULL: object=" INTPTR_FORMAT,
+               p2i(mid->object()));
+        assert(mid->is_free(), "must be free: allocation_state=%d",
+               (int) mid->allocation_state());
+
+        // Move the deflated ObjectMonitor to the working free list
+        // defined by free_head_p and free_tail_p.
+        if (*free_head_p == NULL) {
+          // First one on the list.
+          *free_head_p = mid;
+        }
+        if (*free_tail_p != NULL) {
+          // We append to the list so the caller can use mid->_next_om
+          // to fix the linkages in its context.
+          ObjectMonitor* prevtail = *free_tail_p;
+          // Should have been cleaned up by the caller:
+          assert(prevtail->_next_om == NULL, "must be NULL: _next_om="
+                 INTPTR_FORMAT, p2i(prevtail->_next_om));
+          prevtail->_next_om = mid;
+        }
+        *free_tail_p = mid;
+
+        // At this point, mid->_next_om still refers to its current
+        // value and another ObjectMonitor's _next_om 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 ref_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 ref_count field to its
+      // proper value (which may not be what we saw above):
+      Atomic::add(max_jint, &mid->_ref_count);
+
+      assert(mid->ref_count() >= 0, "must not be negative: ref_count=%d",
+             mid->ref_count());
+      return false;
+    }
+
+    // The ref_count was no longer 0 so we lost the race since the
+    // ObjectMonitor is now busy or the ObjectMonitor* is now is use.
+    // Restore owner to NULL if it is still DEFLATER_MARKER:
+    Atomic::cmpxchg((void*)NULL, &mid->_owner, DEFLATER_MARKER);
+  }
+
+  // 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,

@@ -1625,20 +2047,105 @@
     }
   }
   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 saved_mid_in_use_p and return to the caller to
+// honor the safepoint.
+//
+int ObjectSynchronizer::deflate_monitor_list_using_JT(ObjectMonitor** list_p,
+                                                      ObjectMonitor** free_head_p,
+                                                      ObjectMonitor** free_tail_p,
+                                                      ObjectMonitor** saved_mid_in_use_p) {
+  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 (*saved_mid_in_use_p == NULL) {
+    // No saved state so start at the beginning.
+    mid = *list_p;
+  } else {
+    // We're restarting after a safepoint so restore the necessary state
+    // before we resume.
+    cur_mid_in_use = *saved_mid_in_use_p;
+    mid = cur_mid_in_use->_next_om;
+  }
+  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, free_head_p, free_tail_p)) {
+      // Deflation succeeded so update the in-use list.
+      if (mid == *list_p) {
+        *list_p = mid->_next_om;
+      } else if (cur_mid_in_use != NULL) {
+        // Maintain the current in-use list.
+        cur_mid_in_use->_next_om = mid->_next_om;
+      }
+      next = mid->_next_om;
+      mid->_next_om = NULL;
+      // At this point mid is disconnected from the in-use list
+      // and is the current tail in the free_head_p 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 om_alloc() (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
+      // om_alloc() is released by om_release() due to losing the race
+      // in inflate().
+
+      cur_mid_in_use = mid;
+      mid = mid->_next_om;
+
+      if (SafepointSynchronize::is_synchronizing() &&
+          cur_mid_in_use != *list_p && 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.
+        *saved_mid_in_use_p = cur_mid_in_use;
+        return deflated_count;
+      }
+    }
+  }
+  // We finished the list without a safepoint starting so there's
+  // no need to save state.
+  *saved_mid_in_use_p = NULL;
+  return deflated_count;
+}
+
 void ObjectSynchronizer::prepare_deflate_idle_monitors(DeflateMonitorCounters* counters) {
   counters->n_in_use = 0;              // currently associated with objects
   counters->n_in_circulation = 0;      // extant
   counters->n_scavenged = 0;           // reclaimed (global and per-thread)
   counters->per_thread_scavenged = 0;  // per-thread scavenge total
   counters->per_thread_times = 0.0;    // per-thread scavenge times
 }
 
 void ObjectSynchronizer::deflate_idle_monitors(DeflateMonitorCounters* counters) {
   assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
+
+  if (AsyncDeflateIdleMonitors) {
+    // Nothing to do when global idle ObjectMonitors are deflated using
+    // a JavaThread unless a special deflation has been requested.
+    if (!is_special_deflation_requested()) {
+      return;
+    }
+  }
+
   bool deflated = false;
 
   ObjectMonitor* free_head_p = NULL;  // Local SLL of scavenged monitors
   ObjectMonitor* free_tail_p = NULL;
   elapsedTimer timer;

@@ -1687,19 +2194,150 @@
   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* self = JavaThread::current();
+
+  deflate_common_idle_monitors_using_JT(true /* is_global */, self);
+}
+
+// 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* self = JavaThread::current();
+
+  self->om_request_deflation = false;
+
+  deflate_common_idle_monitors_using_JT(false /* !is_global */, self);
+}
+
+// Deflate global or per-thread idle ObjectMonitors using a JavaThread.
+//
+void ObjectSynchronizer::deflate_common_idle_monitors_using_JT(bool is_global, JavaThread* self) {
+  int deflated_count = 0;
+  ObjectMonitor* free_head_p = NULL;  // Local SLL of scavenged ObjectMonitors
+  ObjectMonitor* free_tail_p = NULL;
+  ObjectMonitor* saved_mid_in_use_p = NULL;
+  elapsedTimer timer;
+
+  if (log_is_enabled(Info, monitorinflation)) {
+    timer.start();
+  }
+
+  if (is_global) {
+    Thread::muxAcquire(&gListLock, "deflate_global_idle_monitors_using_JT(1)");
+    OM_PERFDATA_OP(MonExtant, set_value(g_om_in_use_count));
+  } else {
+    OM_PERFDATA_OP(MonExtant, inc(self->om_in_use_count));
+  }
+
+  do {
+    int local_deflated_count;
+    if (is_global) {
+      local_deflated_count = deflate_monitor_list_using_JT((ObjectMonitor **)&g_om_in_use_list, &free_head_p, &free_tail_p, &saved_mid_in_use_p);
+      g_om_in_use_count -= local_deflated_count;
+    } else {
+      local_deflated_count = deflate_monitor_list_using_JT(self->om_in_use_list_addr(), &free_head_p, &free_tail_p, &saved_mid_in_use_p);
+      self->om_in_use_count -= local_deflated_count;
+    }
+    deflated_count += local_deflated_count;
+
+    if (free_head_p != NULL) {
+      // Move the scavenged ObjectMonitors to the global free list.
+      guarantee(free_tail_p != NULL && local_deflated_count > 0, "free_tail_p=" INTPTR_FORMAT ", local_deflated_count=%d", p2i(free_tail_p), local_deflated_count);
+      assert(free_tail_p->_next_om == NULL, "invariant");
+
+      if (!is_global) {
+        Thread::muxAcquire(&gListLock, "deflate_per_thread_idle_monitors_using_JT(2)");
+      }
+      // Constant-time list splice - prepend scavenged segment to g_free_list.
+      free_tail_p->_next_om = g_free_list;
+      g_free_list = free_head_p;
+
+      g_om_free_count += local_deflated_count;
+      OM_PERFDATA_OP(Deflations, inc(local_deflated_count));
+      if (!is_global) {
+        Thread::muxRelease(&gListLock);
+      }
+    }
+
+    if (saved_mid_in_use_p != NULL) {
+      // deflate_monitor_list_using_JT() detected a safepoint starting.
+      if (is_global) {
+        Thread::muxRelease(&gListLock);
+      }
+      timer.stop();
+      {
+        if (is_global) {
+          log_debug(monitorinflation)("pausing deflation of global idle monitors for a safepoint.");
+        } else {
+          log_debug(monitorinflation)("jt=" INTPTR_FORMAT ": pausing deflation of per-thread idle monitors for a safepoint.", p2i(self));
+        }
+        assert(SafepointSynchronize::is_synchronizing(), "sanity check");
+        ThreadBlockInVM blocker(self);
+      }
+      // Prepare for another loop after the safepoint.
+      free_head_p = NULL;
+      free_tail_p = NULL;
+      if (log_is_enabled(Info, monitorinflation)) {
+        timer.start();
+      }
+      if (is_global) {
+        Thread::muxAcquire(&gListLock, "deflate_global_idle_monitors_using_JT(3)");
+      }
+    }
+  } while (saved_mid_in_use_p != NULL);
+  if (is_global) {
+    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) {
+    if (is_global) {
+      ls->print_cr("async-deflating global idle monitors, %3.7f secs, %d monitors", timer.seconds(), deflated_count);
+    } else {
+      ls->print_cr("jt=" INTPTR_FORMAT ": async-deflating per-thread idle monitors, %3.7f secs, %d monitors", p2i(self), 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 deflation has been requested.
   log_info(safepoint, cleanup)("deflating per-thread idle monitors, %3.7f secs, monitors=%d", counters->per_thread_times, counters->per_thread_scavenged);
 
+  bool needs_special_deflation = is_special_deflation_requested();
+  if (!AsyncDeflateIdleMonitors || needs_special_deflation) {
+    // AsyncDeflateIdleMonitors does not use these counters unless
+    // there is a special deflation request.
+
   g_om_free_count += counters->n_scavenged;
 
+    OM_PERFDATA_OP(Deflations, inc(counters->n_scavenged));
+    OM_PERFDATA_OP(MonExtant, set_value(counters->n_in_circulation));
+  }
+
   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)) {

@@ -1709,21 +2347,32 @@
                                g_om_in_use_count, g_om_free_count);
     Thread::muxRelease(&gListLock);
   }
 
   ForceMonitorScavenge = 0;    // Reset
-
-  OM_PERFDATA_OP(Deflations, inc(counters->n_scavenged));
-  OM_PERFDATA_OP(MonExtant, set_value(counters->n_in_circulation));
-
   GVars.stw_random = os::random();
   GVars.stw_cycle++;
+  if (needs_special_deflation) {
+    set_is_special_deflation_requested(false);  // special deflation is done
+  }
 }
 
 void ObjectSynchronizer::deflate_thread_local_monitors(Thread* thread, DeflateMonitorCounters* counters) {
   assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
 
+  if (AsyncDeflateIdleMonitors) {
+    if (!is_special_deflation_requested()) {
+      // Mark the JavaThread for idle monitor deflation if a special
+      // deflation has NOT been requested.
+      if (thread->om_in_use_count > 0) {
+        // This JavaThread is using monitors so mark it.
+        thread->om_request_deflation = true;
+      }
+      return;
+    }
+  }
+
   ObjectMonitor* free_head_p = NULL;  // Local SLL of scavenged monitors
   ObjectMonitor* free_tail_p = NULL;
   elapsedTimer timer;
 
   if (log_is_enabled(Info, safepoint, cleanup) ||

@@ -1946,17 +2595,18 @@
     if (jt != NULL) {
       out->print_cr("ERROR: jt=" INTPTR_FORMAT ", monitor=" INTPTR_FORMAT
                     ": free per-thread monitor must have NULL _header "
                     "field: _header=" INTPTR_FORMAT, p2i(jt), p2i(n),
                     n->header().value());
-    } else {
+      *error_cnt_p = *error_cnt_p + 1;
+    } else if (!AsyncDeflateIdleMonitors) {
       out->print_cr("ERROR: monitor=" INTPTR_FORMAT ": free global monitor "
                     "must have NULL _header field: _header=" INTPTR_FORMAT,
                     p2i(n), n->header().value());
-    }
     *error_cnt_p = *error_cnt_p + 1;
   }
+  }
   if (n->object() != NULL) {
     if (jt != NULL) {
       out->print_cr("ERROR: jt=" INTPTR_FORMAT ", monitor=" INTPTR_FORMAT
                     ": free per-thread monitor must have NULL _object "
                     "field: _object=" INTPTR_FORMAT, p2i(jt), p2i(n),

@@ -2119,20 +2769,21 @@
 
   stringStream ss;
   if (g_om_in_use_count > 0) {
     out->print_cr("In-use global monitor info:");
     out->print_cr("(B -> is_busy, H -> has hash code, 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 = g_om_in_use_list; n != NULL; n = n->_next_om) {
       const oop obj = (oop) n->object();
       const markWord mark = n->header();
       ResourceMark rm;
-      out->print(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(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 (n->is_busy() != 0) {
         out->print(" (%s)", n->is_busy_to_string(&ss));
         ss.reset();
       }
       out->cr();

@@ -2143,22 +2794,22 @@
     Thread::muxRelease(&gListLock);
   }
 
   out->print_cr("In-use per-thread monitor info:");
   out->print_cr("(B -> is_busy, H -> has hash code, 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->om_in_use_list; n != NULL; n = n->_next_om) {
       const oop obj = (oop) n->object();
       const markWord mark = n->header();
       ResourceMark rm;
-      out->print(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(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());
       if (n->is_busy() != 0) {
         out->print(" (%s)", n->is_busy_to_string(&ss));
         ss.reset();
       }
       out->cr();
< prev index next >