jdk-sandbox: comparison hotspot/src/share/vm/runtime/objectMonitor.cpp

equal deleted inserted replaced

-:8f17e084029b
+:a02753d5a0b2
 #include "utilities/dtrace.hpp"
 #include "utilities/macros.hpp"
 #include "utilities/preserveException.hpp"
 #if defined(__GNUC__) && !defined(IA64) && !defined(PPC64)
 // Need to inhibit inlining for older versions of GCC to avoid build-time failures
 #define NOINLINE __attribute__((noinline))
 #else
 #define NOINLINE
 #endif
 // Enter support
 bool ObjectMonitor::try_enter(Thread* THREAD) {
 if (THREAD != _owner) {
 if (THREAD->is_lock_owned ((address)_owner)) {
 assert(_recursions == 0, "internal state error");
 _owner = THREAD;
 _recursions = 1;
 OwnerIsThread = 1;
 return true;
 }
 if (Atomic::cmpxchg_ptr (THREAD, &_owner, NULL) != NULL) {
 return false;
 }
 return true;
 // and to reduce RTS->RTO cache line upgrades on SPARC and IA32 processors.
 Thread * const Self = THREAD;
 void * cur = Atomic::cmpxchg_ptr (Self, &_owner, NULL);
 if (cur == NULL) {
 // Either ASSERT _recursions == 0 or explicitly set _recursions = 0.
 assert(_recursions == 0   , "invariant");
 assert(_owner      == Self, "invariant");
 // CONSIDER: set or assert OwnerIsThread == 1
 return;
 }
 if (cur == Self) {
 // TODO-FIXME: check for integer overflow!  BUGID 6557169.
 _recursions++;
 return;
 }
 if (Self->is_lock_owned ((address)cur)) {
 assert(_recursions == 0, "internal state error");
 _recursions = 1;
 // and before going through the awkward and expensive state
 // transitions.  The following spin is strictly optional ...
 // Note that if we acquire the monitor from an initial spin
 // we forgo posting JVMTI events and firing DTRACE probes.
 if (Knob_SpinEarly && TrySpin (Self) > 0) {
 assert(_owner == Self      , "invariant");
 assert(_recursions == 0    , "invariant");
 assert(((oop)(object()))->mark() == markOopDesc::encode(this), "invariant");
 Self->_Stalled = 0;
 return;
 }
 assert(_owner != Self          , "invariant");
 assert(_succ  != Self          , "invariant");
 assert(Self->is_Java_thread()  , "invariant");
 // We have acquired the contended monitor, but while we were
 // waiting another thread suspended us. We don't want to enter
 // the monitor while suspended because that would surprise the
 // thread that suspended us.
 //
 _recursions = 0;
 _succ = NULL;
 exit(false, Self);
 jt->java_suspend_self();
 }
 event.set_address((TYPE_ADDRESS)(uintptr_t)(this->object_addr()));
 event.commit();
 }
 if (ObjectMonitor::_sync_ContendedLockAttempts != NULL) {
 ObjectMonitor::_sync_ContendedLockAttempts->inc();
 }
 }
 // Caveat: TryLock() is not necessarily serializing if it returns failure.
 // Retry doesn't make as much sense because the lock was just acquired.
 return -1;
 }
 void NOINLINE ObjectMonitor::EnterI (TRAPS) {
 Thread * const Self = THREAD;
 assert(Self->is_Java_thread(), "invariant");
 assert(((JavaThread *) Self)->thread_state() == _thread_blocked   , "invariant");
 // Try the lock - TATAS
+if (TryLock (Self) > 0) {
+assert(_succ != Self              , "invariant");
+assert(_owner == Self             , "invariant");
+assert(_Responsible != Self       , "invariant");
+return;
+}
+DeferredInitialize();
+// We try one round of spinning *before* enqueueing Self.
+//
+// If the _owner is ready but OFFPROC we could use a YieldTo()
+// operation to donate the remainder of this thread's quantum
+// to the owner.  This has subtle but beneficial affinity
+// effects.
+if (TrySpin (Self) > 0) {
+assert(_owner == Self        , "invariant");
+assert(_succ != Self         , "invariant");
+assert(_Responsible != Self  , "invariant");
+return;
+}
+// The Spin failed -- Enqueue and park the thread ...
+assert(_succ  != Self            , "invariant");
+assert(_owner != Self            , "invariant");
+assert(_Responsible != Self      , "invariant");
+// Enqueue "Self" on ObjectMonitor's _cxq.
+//
+// Node acts as a proxy for Self.
+// As an aside, if were to ever rewrite the synchronization code mostly
+// in Java, WaitNodes, ObjectMonitors, and Events would become 1st-class
+// Java objects.  This would avoid awkward lifecycle and liveness issues,
+// as well as eliminate a subset of ABA issues.
+// TODO: eliminate ObjectWaiter and enqueue either Threads or Events.
+//
+ObjectWaiter node(Self);
+Self->_ParkEvent->reset();
+node._prev   = (ObjectWaiter *) 0xBAD;
+node.TState  = ObjectWaiter::TS_CXQ;
+// Push "Self" onto the front of the _cxq.
+// Once on cxq/EntryList, Self stays on-queue until it acquires the lock.
+// Note that spinning tends to reduce the rate at which threads
+// enqueue and dequeue on EntryList|cxq.
+ObjectWaiter * nxt;
+for (;;) {
+node._next = nxt = _cxq;
+if (Atomic::cmpxchg_ptr(&node, &_cxq, nxt) == nxt) break;
+// Interference - the CAS failed because _cxq changed.  Just retry.
+// As an optional optimization we retry the lock.
 if (TryLock (Self) > 0) {
-assert(_succ != Self              , "invariant");
+assert(_succ != Self         , "invariant");
-assert(_owner == Self             , "invariant");
+assert(_owner == Self        , "invariant");
-assert(_Responsible != Self       , "invariant");
+assert(_Responsible != Self  , "invariant");
 return;
 }
+}
-DeferredInitialize();
+// Check for cxq|EntryList edge transition to non-null.  This indicates
-// We try one round of spinning *before* enqueueing Self.
+// the onset of contention.  While contention persists exiting threads
+// will use a ST:MEMBAR:LD 1-1 exit protocol.  When contention abates exit
+// operations revert to the faster 1-0 mode.  This enter operation may interleave
+// (race) a concurrent 1-0 exit operation, resulting in stranding, so we
+// arrange for one of the contending thread to use a timed park() operations
+// to detect and recover from the race.  (Stranding is form of progress failure
+// where the monitor is unlocked but all the contending threads remain parked).
+// That is, at least one of the contended threads will periodically poll _owner.
+// One of the contending threads will become the designated "Responsible" thread.
+// The Responsible thread uses a timed park instead of a normal indefinite park
+// operation -- it periodically wakes and checks for and recovers from potential
+// strandings admitted by 1-0 exit operations.   We need at most one Responsible
+// thread per-monitor at any given moment.  Only threads on cxq|EntryList may
+// be responsible for a monitor.
+//
+// Currently, one of the contended threads takes on the added role of "Responsible".
+// A viable alternative would be to use a dedicated "stranding checker" thread
+// that periodically iterated over all the threads (or active monitors) and unparked
+// successors where there was risk of stranding.  This would help eliminate the
+// timer scalability issues we see on some platforms as we'd only have one thread
+// -- the checker -- parked on a timer.
+if ((SyncFlags & 16) == 0 && nxt == NULL && _EntryList == NULL) {
+// Try to assume the role of responsible thread for the monitor.
+// CONSIDER:  ST vs CAS vs { if (Responsible==null) Responsible=Self }
+Atomic::cmpxchg_ptr(Self, &_Responsible, NULL);
+}
+// The lock might have been released while this thread was occupied queueing
+// itself onto _cxq.  To close the race and avoid "stranding" and
+// progress-liveness failure we must resample-retry _owner before parking.
+// Note the Dekker/Lamport duality: ST cxq; MEMBAR; LD Owner.
+// In this case the ST-MEMBAR is accomplished with CAS().
+//
+// TODO: Defer all thread state transitions until park-time.
+// Since state transitions are heavy and inefficient we'd like
+// to defer the state transitions until absolutely necessary,
+// and in doing so avoid some transitions ...
+TEVENT(Inflated enter - Contention);
+int nWakeups = 0;
+int RecheckInterval = 1;
+for (;;) {
+if (TryLock(Self) > 0) break;
+assert(_owner != Self, "invariant");
+if ((SyncFlags & 2) && _Responsible == NULL) {
+Atomic::cmpxchg_ptr(Self, &_Responsible, NULL);
+}
+// park self
+if (_Responsible == Self || (SyncFlags & 1)) {
+TEVENT(Inflated enter - park TIMED);
+Self->_ParkEvent->park((jlong) RecheckInterval);
+// Increase the RecheckInterval, but clamp the value.
+RecheckInterval *= 8;
+if (RecheckInterval > 1000) RecheckInterval = 1000;
+} else {
+TEVENT(Inflated enter - park UNTIMED);
+Self->_ParkEvent->park();
+}
+if (TryLock(Self) > 0) break;
+// The lock is still contested.
+// Keep a tally of the # of futile wakeups.
+// Note that the counter is not protected by a lock or updated by atomics.
+// That is by design - we trade "lossy" counters which are exposed to
+// races during updates for a lower probe effect.
+TEVENT(Inflated enter - Futile wakeup);
+if (ObjectMonitor::_sync_FutileWakeups != NULL) {
+ObjectMonitor::_sync_FutileWakeups->inc();
+}
+++nWakeups;
+// Assuming this is not a spurious wakeup we'll normally find _succ == Self.
+// We can defer clearing _succ until after the spin completes
+// TrySpin() must tolerate being called with _succ == Self.
+// Try yet another round of adaptive spinning.
+if ((Knob_SpinAfterFutile & 1) && TrySpin(Self) > 0) break;
+// We can find that we were unpark()ed and redesignated _succ while
+// we were spinning.  That's harmless.  If we iterate and call park(),
+// park() will consume the event and return immediately and we'll
+// just spin again.  This pattern can repeat, leaving _succ to simply
+// spin on a CPU.  Enable Knob_ResetEvent to clear pending unparks().
+// Alternately, we can sample fired() here, and if set, forgo spinning
+// in the next iteration.
+if ((Knob_ResetEvent & 1) && Self->_ParkEvent->fired()) {
+Self->_ParkEvent->reset();
+OrderAccess::fence();
+}
+if (_succ == Self) _succ = NULL;
+// Invariant: after clearing _succ a thread *must* retry _owner before parking.
+OrderAccess::fence();
+}
+// Egress :
+// Self has acquired the lock -- Unlink Self from the cxq or EntryList.
+// Normally we'll find Self on the EntryList .
+// From the perspective of the lock owner (this thread), the
+// EntryList is stable and cxq is prepend-only.
+// The head of cxq is volatile but the interior is stable.
+// In addition, Self.TState is stable.
+assert(_owner == Self      , "invariant");
+assert(object() != NULL    , "invariant");
+// I'd like to write:
+//   guarantee (((oop)(object()))->mark() == markOopDesc::encode(this), "invariant") ;
+// but as we're at a safepoint that's not safe.
+UnlinkAfterAcquire(Self, &node);
+if (_succ == Self) _succ = NULL;
+assert(_succ != Self, "invariant");
+if (_Responsible == Self) {
+_Responsible = NULL;
+OrderAccess::fence(); // Dekker pivot-point
+// We may leave threads on cxq|EntryList without a designated
+// "Responsible" thread.  This is benign.  When this thread subsequently
+// exits the monitor it can "see" such preexisting "old" threads --
+// threads that arrived on the cxq|EntryList before the fence, above --
+// by LDing cxq|EntryList.  Newly arrived threads -- that is, threads
+// that arrive on cxq after the ST:MEMBAR, above -- will set Responsible
+// non-null and elect a new "Responsible" timer thread.
 //
-// If the _owner is ready but OFFPROC we could use a YieldTo()
+// This thread executes:
-// operation to donate the remainder of this thread's quantum
+//    ST Responsible=null; MEMBAR    (in enter epilogue - here)
-// to the owner.  This has subtle but beneficial affinity
+//    LD cxq|EntryList               (in subsequent exit)
-// effects.
-if (TrySpin (Self) > 0) {
-assert(_owner == Self        , "invariant");
-assert(_succ != Self         , "invariant");
-assert(_Responsible != Self  , "invariant");
-return;
-}
-// The Spin failed -- Enqueue and park the thread ...
-assert(_succ  != Self            , "invariant");
-assert(_owner != Self            , "invariant");
-assert(_Responsible != Self      , "invariant");
-// Enqueue "Self" on ObjectMonitor's _cxq.
 //
-// Node acts as a proxy for Self.
+// Entering threads in the slow/contended path execute:
-// As an aside, if were to ever rewrite the synchronization code mostly
+//    ST cxq=nonnull; MEMBAR; LD Responsible (in enter prolog)
-// in Java, WaitNodes, ObjectMonitors, and Events would become 1st-class
+//    The (ST cxq; MEMBAR) is accomplished with CAS().
-// Java objects.  This would avoid awkward lifecycle and liveness issues,
-// as well as eliminate a subset of ABA issues.
-// TODO: eliminate ObjectWaiter and enqueue either Threads or Events.
 //
+// The MEMBAR, above, prevents the LD of cxq|EntryList in the subsequent
-ObjectWaiter node(Self);
+// exit operation from floating above the ST Responsible=null.
-Self->_ParkEvent->reset();
+}
-node._prev   = (ObjectWaiter *) 0xBAD;
-node.TState  = ObjectWaiter::TS_CXQ;
+// We've acquired ownership with CAS().
+// CAS is serializing -- it has MEMBAR/FENCE-equivalent semantics.
-// Push "Self" onto the front of the _cxq.
+// But since the CAS() this thread may have also stored into _succ,
-// Once on cxq/EntryList, Self stays on-queue until it acquires the lock.
+// EntryList, cxq or Responsible.  These meta-data updates must be
-// Note that spinning tends to reduce the rate at which threads
+// visible __before this thread subsequently drops the lock.
-// enqueue and dequeue on EntryList|cxq.
+// Consider what could occur if we didn't enforce this constraint --
-ObjectWaiter * nxt;
+// STs to monitor meta-data and user-data could reorder with (become
-for (;;) {
+// visible after) the ST in exit that drops ownership of the lock.
-node._next = nxt = _cxq;
+// Some other thread could then acquire the lock, but observe inconsistent
-if (Atomic::cmpxchg_ptr(&node, &_cxq, nxt) == nxt) break;
+// or old monitor meta-data and heap data.  That violates the JMM.
+// To that end, the 1-0 exit() operation must have at least STST|LDST
-// Interference - the CAS failed because _cxq changed.  Just retry.
+// "release" barrier semantics.  Specifically, there must be at least a
-// As an optional optimization we retry the lock.
+// STST|LDST barrier in exit() before the ST of null into _owner that drops
-if (TryLock (Self) > 0) {
+// the lock.   The barrier ensures that changes to monitor meta-data and data
-assert(_succ != Self         , "invariant");
+// protected by the lock will be visible before we release the lock, and
-assert(_owner == Self        , "invariant");
+// therefore before some other thread (CPU) has a chance to acquire the lock.
-assert(_Responsible != Self  , "invariant");
+// See also: http://gee.cs.oswego.edu/dl/jmm/cookbook.html.
-return;
+//
-}
+// Critically, any prior STs to _succ or EntryList must be visible before
-}
+// the ST of null into _owner in the *subsequent* (following) corresponding
+// monitorexit.  Recall too, that in 1-0 mode monitorexit does not necessarily
-// Check for cxq|EntryList edge transition to non-null.  This indicates
+// execute a serializing instruction.
-// the onset of contention.  While contention persists exiting threads
-// will use a ST:MEMBAR:LD 1-1 exit protocol.  When contention abates exit
+if (SyncFlags & 8) {
-// operations revert to the faster 1-0 mode.  This enter operation may interleave
+OrderAccess::fence();
-// (race) a concurrent 1-0 exit operation, resulting in stranding, so we
+}
-// arrange for one of the contending thread to use a timed park() operations
+return;
-// to detect and recover from the race.  (Stranding is form of progress failure
-// where the monitor is unlocked but all the contending threads remain parked).
-// That is, at least one of the contended threads will periodically poll _owner.
-// One of the contending threads will become the designated "Responsible" thread.
-// The Responsible thread uses a timed park instead of a normal indefinite park
-// operation -- it periodically wakes and checks for and recovers from potential
-// strandings admitted by 1-0 exit operations.   We need at most one Responsible
-// thread per-monitor at any given moment.  Only threads on cxq|EntryList may
-// be responsible for a monitor.
-//
-// Currently, one of the contended threads takes on the added role of "Responsible".
-// A viable alternative would be to use a dedicated "stranding checker" thread
-// that periodically iterated over all the threads (or active monitors) and unparked
-// successors where there was risk of stranding.  This would help eliminate the
-// timer scalability issues we see on some platforms as we'd only have one thread
-// -- the checker -- parked on a timer.
-if ((SyncFlags & 16) == 0 && nxt == NULL && _EntryList == NULL) {
-// Try to assume the role of responsible thread for the monitor.
-// CONSIDER:  ST vs CAS vs { if (Responsible==null) Responsible=Self }
-Atomic::cmpxchg_ptr(Self, &_Responsible, NULL);
-}
-// The lock might have been released while this thread was occupied queueing
-// itself onto _cxq.  To close the race and avoid "stranding" and
-// progress-liveness failure we must resample-retry _owner before parking.
-// Note the Dekker/Lamport duality: ST cxq; MEMBAR; LD Owner.
-// In this case the ST-MEMBAR is accomplished with CAS().
-//
-// TODO: Defer all thread state transitions until park-time.
-// Since state transitions are heavy and inefficient we'd like
-// to defer the state transitions until absolutely necessary,
-// and in doing so avoid some transitions ...
-TEVENT(Inflated enter - Contention);
-int nWakeups = 0;
-int RecheckInterval = 1;
-for (;;) {
-if (TryLock(Self) > 0) break;
-assert(_owner != Self, "invariant");
-if ((SyncFlags & 2) && _Responsible == NULL) {
-Atomic::cmpxchg_ptr(Self, &_Responsible, NULL);
-}
-// park self
-if (_Responsible == Self || (SyncFlags & 1)) {
-TEVENT(Inflated enter - park TIMED);
-Self->_ParkEvent->park((jlong) RecheckInterval);
-// Increase the RecheckInterval, but clamp the value.
-RecheckInterval *= 8;
-if (RecheckInterval > 1000) RecheckInterval = 1000;
-} else {
-TEVENT(Inflated enter - park UNTIMED);
-Self->_ParkEvent->park();
-}
-if (TryLock(Self) > 0) break;
-// The lock is still contested.
-// Keep a tally of the # of futile wakeups.
-// Note that the counter is not protected by a lock or updated by atomics.
-// That is by design - we trade "lossy" counters which are exposed to
-// races during updates for a lower probe effect.
-TEVENT(Inflated enter - Futile wakeup);
-if (ObjectMonitor::_sync_FutileWakeups != NULL) {
-ObjectMonitor::_sync_FutileWakeups->inc();
-}
-++nWakeups;
-// Assuming this is not a spurious wakeup we'll normally find _succ == Self.
-// We can defer clearing _succ until after the spin completes
-// TrySpin() must tolerate being called with _succ == Self.
-// Try yet another round of adaptive spinning.
-if ((Knob_SpinAfterFutile & 1) && TrySpin(Self) > 0) break;
-// We can find that we were unpark()ed and redesignated _succ while
-// we were spinning.  That's harmless.  If we iterate and call park(),
-// park() will consume the event and return immediately and we'll
-// just spin again.  This pattern can repeat, leaving _succ to simply
-// spin on a CPU.  Enable Knob_ResetEvent to clear pending unparks().
-// Alternately, we can sample fired() here, and if set, forgo spinning
-// in the next iteration.
-if ((Knob_ResetEvent & 1) && Self->_ParkEvent->fired()) {
-Self->_ParkEvent->reset();
-OrderAccess::fence();
-}
-if (_succ == Self) _succ = NULL;
-// Invariant: after clearing _succ a thread *must* retry _owner before parking.
-OrderAccess::fence();
-}
-// Egress :
-// Self has acquired the lock -- Unlink Self from the cxq or EntryList.
-// Normally we'll find Self on the EntryList .
-// From the perspective of the lock owner (this thread), the
-// EntryList is stable and cxq is prepend-only.
-// The head of cxq is volatile but the interior is stable.
-// In addition, Self.TState is stable.
-assert(_owner == Self      , "invariant");
-assert(object() != NULL    , "invariant");
-// I'd like to write:
-//   guarantee (((oop)(object()))->mark() == markOopDesc::encode(this), "invariant") ;
-// but as we're at a safepoint that's not safe.
-UnlinkAfterAcquire(Self, &node);
-if (_succ == Self) _succ = NULL;
-assert(_succ != Self, "invariant");
-if (_Responsible == Self) {
-_Responsible = NULL;
-OrderAccess::fence(); // Dekker pivot-point
-// We may leave threads on cxq|EntryList without a designated
-// "Responsible" thread.  This is benign.  When this thread subsequently
-// exits the monitor it can "see" such preexisting "old" threads --
-// threads that arrived on the cxq|EntryList before the fence, above --
-// by LDing cxq|EntryList.  Newly arrived threads -- that is, threads
-// that arrive on cxq after the ST:MEMBAR, above -- will set Responsible
-// non-null and elect a new "Responsible" timer thread.
-//
-// This thread executes:
-//    ST Responsible=null; MEMBAR    (in enter epilogue - here)
-//    LD cxq|EntryList               (in subsequent exit)
-//
-// Entering threads in the slow/contended path execute:
-//    ST cxq=nonnull; MEMBAR; LD Responsible (in enter prolog)
-//    The (ST cxq; MEMBAR) is accomplished with CAS().
-//
-// The MEMBAR, above, prevents the LD of cxq|EntryList in the subsequent
-// exit operation from floating above the ST Responsible=null.
-}
-// We've acquired ownership with CAS().
-// CAS is serializing -- it has MEMBAR/FENCE-equivalent semantics.
-// But since the CAS() this thread may have also stored into _succ,
-// EntryList, cxq or Responsible.  These meta-data updates must be
-// visible __before this thread subsequently drops the lock.
-// Consider what could occur if we didn't enforce this constraint --
-// STs to monitor meta-data and user-data could reorder with (become
-// visible after) the ST in exit that drops ownership of the lock.
-// Some other thread could then acquire the lock, but observe inconsistent
-// or old monitor meta-data and heap data.  That violates the JMM.
-// To that end, the 1-0 exit() operation must have at least STST|LDST
-// "release" barrier semantics.  Specifically, there must be at least a
-// STST|LDST barrier in exit() before the ST of null into _owner that drops
-// the lock.   The barrier ensures that changes to monitor meta-data and data
-// protected by the lock will be visible before we release the lock, and
-// therefore before some other thread (CPU) has a chance to acquire the lock.
-// See also: http://gee.cs.oswego.edu/dl/jmm/cookbook.html.
-//
-// Critically, any prior STs to _succ or EntryList must be visible before
-// the ST of null into _owner in the *subsequent* (following) corresponding
-// monitorexit.  Recall too, that in 1-0 mode monitorexit does not necessarily
-// execute a serializing instruction.
-if (SyncFlags & 8) {
-OrderAccess::fence();
-}
-return;
 }
 // ReenterI() is a specialized inline form of the latter half of the
 // contended slow-path from EnterI().  We use ReenterI() only for
 // monitor reentry in wait().
 // In the future we should reconcile EnterI() and ReenterI(), adding
 // Knob_Reset and Knob_SpinAfterFutile support and restructuring the
 // loop accordingly.
 void NOINLINE ObjectMonitor::ReenterI (Thread * Self, ObjectWaiter * SelfNode) {
 assert(Self != NULL                , "invariant");
 assert(SelfNode != NULL            , "invariant");
 assert(SelfNode->_thread == Self   , "invariant");
 assert(_waiters > 0                , "invariant");
 assert(((oop)(object()))->mark() == markOopDesc::encode(this) , "invariant");
 assert(((JavaThread *)Self)->thread_state() != _thread_blocked, "invariant");
 JavaThread * jt = (JavaThread *) Self;
 int nWakeups = 0;
 for (;;) {
 ObjectWaiter::TStates v = SelfNode->TState;
 guarantee(v == ObjectWaiter::TS_ENTER || v == ObjectWaiter::TS_CXQ, "invariant");
 assert(_owner != Self, "invariant");
 if (TryLock(Self) > 0) break;
 if (TrySpin(Self) > 0) break;
 TEVENT(Wait Reentry - parking);
 // State transition wrappers around park() ...
 // ReenterI() wisely defers state transitions until
 // it's clear we must park the thread.
 {
 OSThreadContendState osts(Self->osthread());
 ThreadBlockInVM tbivm(jt);
 // cleared by handle_special_suspend_equivalent_condition()
 // or java_suspend_self()
 jt->set_suspend_equivalent();
 if (SyncFlags & 1) {
 Self->_ParkEvent->park((jlong)1000);
 } else {
 Self->_ParkEvent->park();
 }
 // were we externally suspended while we were waiting?
 for (;;) {
 if (!ExitSuspendEquivalent(jt)) break;
 if (_succ == Self) { _succ = NULL; OrderAccess::fence(); }
 jt->java_suspend_self();
 jt->set_suspend_equivalent();
 }
 }
 // Try again, but just so we distinguish between futile wakeups and
 // successful wakeups.  The following test isn't algorithmically
 // necessary, but it helps us maintain sensible statistics.
 if (TryLock(Self) > 0) break;
 // The lock is still contested.
 // Keep a tally of the # of futile wakeups.
 // Note that the counter is not protected by a lock or updated by atomics.
 // That is by design - we trade "lossy" counters which are exposed to
 // races during updates for a lower probe effect.
 TEVENT(Wait Reentry - futile wakeup);
 ++nWakeups;
 // Assuming this is not a spurious wakeup we'll normally
 // find that _succ == Self.
-if (_succ == Self) _succ = NULL;
-// Invariant: after clearing _succ a contending thread
-// *must* retry  _owner before parking.
-OrderAccess::fence();
-if (ObjectMonitor::_sync_FutileWakeups != NULL) {
-ObjectMonitor::_sync_FutileWakeups->inc();
-}
-}
-// Self has acquired the lock -- Unlink Self from the cxq or EntryList .
-// Normally we'll find Self on the EntryList.
-// Unlinking from the EntryList is constant-time and atomic-free.
-// From the perspective of the lock owner (this thread), the
-// EntryList is stable and cxq is prepend-only.
-// The head of cxq is volatile but the interior is stable.
-// In addition, Self.TState is stable.
-assert(_owner == Self, "invariant");
-assert(((oop)(object()))->mark() == markOopDesc::encode(this), "invariant");
-UnlinkAfterAcquire(Self, SelfNode);
 if (_succ == Self) _succ = NULL;
-assert(_succ != Self, "invariant");
-SelfNode->TState = ObjectWaiter::TS_RUN;
+// Invariant: after clearing _succ a contending thread
-OrderAccess::fence();      // see comments at the end of EnterI()
+// *must* retry  _owner before parking.
+OrderAccess::fence();
+if (ObjectMonitor::_sync_FutileWakeups != NULL) {
+ObjectMonitor::_sync_FutileWakeups->inc();
+}
+}
+// Self has acquired the lock -- Unlink Self from the cxq or EntryList .
+// Normally we'll find Self on the EntryList.
+// Unlinking from the EntryList is constant-time and atomic-free.
+// From the perspective of the lock owner (this thread), the
+// EntryList is stable and cxq is prepend-only.
+// The head of cxq is volatile but the interior is stable.
+// In addition, Self.TState is stable.
+assert(_owner == Self, "invariant");
+assert(((oop)(object()))->mark() == markOopDesc::encode(this), "invariant");
+UnlinkAfterAcquire(Self, SelfNode);
+if (_succ == Self) _succ = NULL;
+assert(_succ != Self, "invariant");
+SelfNode->TState = ObjectWaiter::TS_RUN;
+OrderAccess::fence();      // see comments at the end of EnterI()
 }
 // By convention we unlink a contending thread from EntryList|cxq immediately
 // after the thread acquires the lock in ::enter().  Equally, we could defer
 // unlinking the thread until ::exit()-time.
 void ObjectMonitor::UnlinkAfterAcquire (Thread * Self, ObjectWaiter * SelfNode)
 {
 assert(_owner == Self, "invariant");
 assert(SelfNode->_thread == Self, "invariant");
 if (SelfNode->TState == ObjectWaiter::TS_ENTER) {
 // Normal case: remove Self from the DLL EntryList .
 // This is a constant-time operation.
 ObjectWaiter * nxt = SelfNode->_next;
 ObjectWaiter * prv = SelfNode->_prev;
 if (nxt != NULL) nxt->_prev = prv;
 if (prv != NULL) prv->_next = nxt;
 if (SelfNode == _EntryList) _EntryList = nxt;
 assert(nxt == NULL || nxt->TState == ObjectWaiter::TS_ENTER, "invariant");
 assert(prv == NULL || prv->TState == ObjectWaiter::TS_ENTER, "invariant");
 TEVENT(Unlink from EntryList);
 } else {
 assert(SelfNode->TState == ObjectWaiter::TS_CXQ, "invariant");
 // Inopportune interleaving -- Self is still on the cxq.
 // This usually means the enqueue of self raced an exiting thread.
 // Normally we'll find Self near the front of the cxq, so
 // dequeueing is typically fast.  If needbe we can accelerate
 // this with some MCS/CHL-like bidirectional list hints and advisory
 // back-links so dequeueing from the interior will normally operate
 // in constant-time.
 // Dequeue Self from either the head (with CAS) or from the interior
 // with a linear-time scan and normal non-atomic memory operations.
 // CONSIDER: if Self is on the cxq then simply drain cxq into EntryList
 // and then unlink Self from EntryList.  We have to drain eventually,
 // so it might as well be now.
 ObjectWaiter * v = _cxq;
 assert(v != NULL, "invariant");
 if (v != SelfNode || Atomic::cmpxchg_ptr (SelfNode->_next, &_cxq, v) != v) {
 // The CAS above can fail from interference IFF a "RAT" arrived.
 // In that case Self must be in the interior and can no longer be
 // at the head of cxq.
 if (v == SelfNode) {
 assert(_cxq != v, "invariant");
 v = _cxq;          // CAS above failed - start scan at head of list
 }
 ObjectWaiter * p;
 ObjectWaiter * q = NULL;
 for (p = v; p != NULL && p != SelfNode; p = p->_next) {
 q = p;
 assert(p->TState == ObjectWaiter::TS_CXQ, "invariant");
 }
 assert(v != SelfNode, "invariant");
 assert(p == SelfNode, "Node not found on cxq");
 assert(p != _cxq, "invariant");
 assert(q != NULL, "invariant");
 assert(q->_next == p, "invariant");
 q->_next = p->_next;
 }
 TEVENT(Unlink from cxq);
 }
 #ifdef ASSERT
 // Diagnostic hygiene ...
 SelfNode->_prev  = (ObjectWaiter *) 0xBAD;
 SelfNode->_next  = (ObjectWaiter *) 0xBAD;
 SelfNode->TState = ObjectWaiter::TS_RUN;
 #endif
 }
 // -----------------------------------------------------------------------------
 // Exit support
 // then wake a thread unnecessarily. This is benign, and we've
 // structured the code so the windows are short and the frequency
 // of such futile wakups is low.
 void NOINLINE ObjectMonitor::exit(bool not_suspended, TRAPS) {
 Thread * const Self = THREAD;
 if (THREAD != _owner) {
 if (THREAD->is_lock_owned((address) _owner)) {
 // Transmute _owner from a BasicLock pointer to a Thread address.
 // We don't need to hold _mutex for this transition.
 // Non-null to Non-null is safe as long as all readers can
 // tolerate either flavor.
 assert(_recursions == 0, "invariant");
 _owner = THREAD;
 _recursions = 0;
 OwnerIsThread = 1;
 } else {
 // Apparent unbalanced locking ...
 // Naively we'd like to throw IllegalMonitorStateException.
 // As a practical matter we can neither allocate nor throw an
 // exception as ::exit() can be called from leaf routines.
 // see x86_32.ad Fast_Unlock() and the I1 and I2 properties.
 // Upon deeper reflection, however, in a properly run JVM the only
 // way we should encounter this situation is in the presence of
 // unbalanced JNI locking. TODO: CheckJNICalls.
 // See also: CR4414101
 TEVENT(Exit - Throw IMSX);
 assert(false, "Non-balanced monitor enter/exit! Likely JNI locking");
 return;
 }
 }
 if (_recursions != 0) {
 _recursions--;        // this is simple recursive enter
 TEVENT(Inflated exit - recursive);
 return;
 }
 // Invariant: after setting Responsible=null an thread must execute
 // a MEMBAR or other serializing instruction before fetching EntryList|cxq.
 if ((SyncFlags & 4) == 0) {
 _Responsible = NULL;
 }
 #if INCLUDE_TRACE
 // get the owner's thread id for the MonitorEnter event
 // if it is enabled and the thread isn't suspended
 if (not_suspended && Tracing::is_event_enabled(TraceJavaMonitorEnterEvent)) {
 _previous_owner_tid = SharedRuntime::get_java_tid(Self);
 }
 #endif
 for (;;) {
 assert(THREAD == _owner, "invariant");
 if (Knob_ExitPolicy == 0) {
 // release semantics: prior loads and stores from within the critical section
 // must not float (reorder) past the following store that drops the lock.
 // On SPARC that requires MEMBAR #loadstore|#storestore.
 // But of course in TSO #loadstore|#storestore is not required.
 // I'd like to write one of the following:
 // A.  OrderAccess::release() ; _owner = NULL
 // B.  OrderAccess::loadstore(); OrderAccess::storestore(); _owner = NULL;
 // Unfortunately OrderAccess::release() and OrderAccess::loadstore() both
 // store into a _dummy variable.  That store is not needed, but can result
 // in massive wasteful coherency traffic on classic SMP systems.
 // Instead, I use release_store(), which is implemented as just a simple
 // ST on x64, x86 and SPARC.
 OrderAccess::release_store_ptr(&_owner, NULL);   // drop the lock
 OrderAccess::storeload();                         // See if we need to wake a successor
 if ((intptr_t(_EntryList)|intptr_t(_cxq)) == 0 || _succ != NULL) {
 TEVENT(Inflated exit - simple egress);
 return;
 }
 TEVENT(Inflated exit - complex egress);
 // Other threads are blocked trying to acquire the lock.
 // Normally the exiting thread is responsible for ensuring succession,
 // but if other successors are ready or other entering threads are spinning
 // then this thread can simply store NULL into _owner and exit without
 // waking a successor.  The existence of spinners or ready successors
 // guarantees proper succession (liveness).  Responsibility passes to the
 // ready or running successors.  The exiting thread delegates the duty.
 // More precisely, if a successor already exists this thread is absolved
 // of the responsibility of waking (unparking) one.
 //
 // The _succ variable is critical to reducing futile wakeup frequency.
 // _succ identifies the "heir presumptive" thread that has been made
 // ready (unparked) but that has not yet run.  We need only one such
 // successor thread to guarantee progress.
 // See http://www.usenix.org/events/jvm01/full_papers/dice/dice.pdf
 // section 3.3 "Futile Wakeup Throttling" for details.
 //
 // Note that spinners in Enter() also set _succ non-null.
 // In the current implementation spinners opportunistically set
 // _succ so that exiting threads might avoid waking a successor.
 // Another less appealing alternative would be for the exiting thread
 // to drop the lock and then spin briefly to see if a spinner managed
 // to acquire the lock.  If so, the exiting thread could exit
 // immediately without waking a successor, otherwise the exiting
 // thread would need to dequeue and wake a successor.
 // (Note that we'd need to make the post-drop spin short, but no
 // shorter than the worst-case round-trip cache-line migration time.
 // The dropped lock needs to become visible to the spinner, and then
 // the acquisition of the lock by the spinner must become visible to
 // the exiting thread).
 //
 // It appears that an heir-presumptive (successor) must be made ready.
 // Only the current lock owner can manipulate the EntryList or
 // drain _cxq, so we need to reacquire the lock.  If we fail
 // to reacquire the lock the responsibility for ensuring succession
 // falls to the new owner.
 //
 if (Atomic::cmpxchg_ptr (THREAD, &_owner, NULL) != NULL) {
 return;
 }
 TEVENT(Exit - Reacquired);
+} else {
+if ((intptr_t(_EntryList)|intptr_t(_cxq)) == 0 || _succ != NULL) {
+OrderAccess::release_store_ptr(&_owner, NULL);   // drop the lock
+OrderAccess::storeload();
+// Ratify the previously observed values.
+if (_cxq == NULL || _succ != NULL) {
+TEVENT(Inflated exit - simple egress);
+return;
+}
+// inopportune interleaving -- the exiting thread (this thread)
+// in the fast-exit path raced an entering thread in the slow-enter
+// path.
+// We have two choices:
+// A.  Try to reacquire the lock.
+//     If the CAS() fails return immediately, otherwise
+//     we either restart/rerun the exit operation, or simply
+//     fall-through into the code below which wakes a successor.
+// B.  If the elements forming the EntryList|cxq are TSM
+//     we could simply unpark() the lead thread and return
+//     without having set _succ.
+if (Atomic::cmpxchg_ptr (THREAD, &_owner, NULL) != NULL) {
+TEVENT(Inflated exit - reacquired succeeded);
+return;
+}
+TEVENT(Inflated exit - reacquired failed);
 } else {
-if ((intptr_t(_EntryList)|intptr_t(_cxq)) == 0 || _succ != NULL) {
+TEVENT(Inflated exit - complex egress);
-OrderAccess::release_store_ptr(&_owner, NULL);   // drop the lock
+}
-OrderAccess::storeload();
+}
-// Ratify the previously observed values.
-if (_cxq == NULL || _succ != NULL) {
+guarantee(_owner == THREAD, "invariant");
-TEVENT(Inflated exit - simple egress);
-return;
+ObjectWaiter * w = NULL;
-}
+int QMode = Knob_QMode;
-// inopportune interleaving -- the exiting thread (this thread)
+if (QMode == 2 && _cxq != NULL) {
-// in the fast-exit path raced an entering thread in the slow-enter
+// QMode == 2 : cxq has precedence over EntryList.
-// path.
+// Try to directly wake a successor from the cxq.
-// We have two choices:
+// If successful, the successor will need to unlink itself from cxq.
-// A.  Try to reacquire the lock.
-//     If the CAS() fails return immediately, otherwise
-//     we either restart/rerun the exit operation, or simply
-//     fall-through into the code below which wakes a successor.
-// B.  If the elements forming the EntryList|cxq are TSM
-//     we could simply unpark() the lead thread and return
-//     without having set _succ.
-if (Atomic::cmpxchg_ptr (THREAD, &_owner, NULL) != NULL) {
-TEVENT(Inflated exit - reacquired succeeded);
-return;
-}
-TEVENT(Inflated exit - reacquired failed);
-} else {
-TEVENT(Inflated exit - complex egress);
-}
-}
-guarantee(_owner == THREAD, "invariant");
-ObjectWaiter * w = NULL;
-int QMode = Knob_QMode;
-if (QMode == 2 && _cxq != NULL) {
-// QMode == 2 : cxq has precedence over EntryList.
-// Try to directly wake a successor from the cxq.
-// If successful, the successor will need to unlink itself from cxq.
-w = _cxq;
-assert(w != NULL, "invariant");
-assert(w->TState == ObjectWaiter::TS_CXQ, "Invariant");
-ExitEpilog(Self, w);
-return;
-}
-if (QMode == 3 && _cxq != NULL) {
-// Aggressively drain cxq into EntryList at the first opportunity.
-// This policy ensure that recently-run threads live at the head of EntryList.
-// Drain _cxq into EntryList - bulk transfer.
-// First, detach _cxq.
-// The following loop is tantamount to: w = swap (&cxq, NULL)
-w = _cxq;
-for (;;) {
-assert(w != NULL, "Invariant");
-ObjectWaiter * u = (ObjectWaiter *) Atomic::cmpxchg_ptr(NULL, &_cxq, w);
-if (u == w) break;
-w = u;
-}
-assert(w != NULL              , "invariant");
-ObjectWaiter * q = NULL;
-ObjectWaiter * p;
-for (p = w; p != NULL; p = p->_next) {
-guarantee(p->TState == ObjectWaiter::TS_CXQ, "Invariant");
-p->TState = ObjectWaiter::TS_ENTER;
-p->_prev = q;
-q = p;
-}
-// Append the RATs to the EntryList
-// TODO: organize EntryList as a CDLL so we can locate the tail in constant-time.
-ObjectWaiter * Tail;
-for (Tail = _EntryList; Tail != NULL && Tail->_next != NULL; Tail = Tail->_next);
-if (Tail == NULL) {
-_EntryList = w;
-} else {
-Tail->_next = w;
-w->_prev = Tail;
-}
-// Fall thru into code that tries to wake a successor from EntryList
-}
-if (QMode == 4 && _cxq != NULL) {
-// Aggressively drain cxq into EntryList at the first opportunity.
-// This policy ensure that recently-run threads live at the head of EntryList.
-// Drain _cxq into EntryList - bulk transfer.
-// First, detach _cxq.
-// The following loop is tantamount to: w = swap (&cxq, NULL)
-w = _cxq;
-for (;;) {
-assert(w != NULL, "Invariant");
-ObjectWaiter * u = (ObjectWaiter *) Atomic::cmpxchg_ptr(NULL, &_cxq, w);
-if (u == w) break;
-w = u;
-}
-assert(w != NULL              , "invariant");
-ObjectWaiter * q = NULL;
-ObjectWaiter * p;
-for (p = w; p != NULL; p = p->_next) {
-guarantee(p->TState == ObjectWaiter::TS_CXQ, "Invariant");
-p->TState = ObjectWaiter::TS_ENTER;
-p->_prev = q;
-q = p;
-}
-// Prepend the RATs to the EntryList
-if (_EntryList != NULL) {
-q->_next = _EntryList;
-_EntryList->_prev = q;
-}
-_EntryList = w;
-// Fall thru into code that tries to wake a successor from EntryList
-}
-w = _EntryList;
-if (w != NULL) {
-// I'd like to write: guarantee (w->_thread != Self).
-// But in practice an exiting thread may find itself on the EntryList.
-// Let's say thread T1 calls O.wait().  Wait() enqueues T1 on O's waitset and
-// then calls exit().  Exit release the lock by setting O._owner to NULL.
-// Let's say T1 then stalls.  T2 acquires O and calls O.notify().  The
-// notify() operation moves T1 from O's waitset to O's EntryList. T2 then
-// release the lock "O".  T2 resumes immediately after the ST of null into
-// _owner, above.  T2 notices that the EntryList is populated, so it
-// reacquires the lock and then finds itself on the EntryList.
-// Given all that, we have to tolerate the circumstance where "w" is
-// associated with Self.
-assert(w->TState == ObjectWaiter::TS_ENTER, "invariant");
-ExitEpilog(Self, w);
-return;
-}
-// If we find that both _cxq and EntryList are null then just
-// re-run the exit protocol from the top.
 w = _cxq;
-if (w == NULL) continue;
+assert(w != NULL, "invariant");
+assert(w->TState == ObjectWaiter::TS_CXQ, "Invariant");
+ExitEpilog(Self, w);
+return;
+}
+if (QMode == 3 && _cxq != NULL) {
+// Aggressively drain cxq into EntryList at the first opportunity.
+// This policy ensure that recently-run threads live at the head of EntryList.
 // Drain _cxq into EntryList - bulk transfer.
 // First, detach _cxq.
 // The following loop is tantamount to: w = swap (&cxq, NULL)
+w = _cxq;
 for (;;) {
 assert(w != NULL, "Invariant");
 ObjectWaiter * u = (ObjectWaiter *) Atomic::cmpxchg_ptr(NULL, &_cxq, w);
 if (u == w) break;
 w = u;
 }
-TEVENT(Inflated exit - drain cxq into EntryList);
 assert(w != NULL              , "invariant");
-assert(_EntryList  == NULL    , "invariant");
+ObjectWaiter * q = NULL;
-// Convert the LIFO SLL anchored by _cxq into a DLL.
+ObjectWaiter * p;
-// The list reorganization step operates in O(LENGTH(w)) time.
+for (p = w; p != NULL; p = p->_next) {
-// It's critical that this step operate quickly as
+guarantee(p->TState == ObjectWaiter::TS_CXQ, "Invariant");
-// "Self" still holds the outer-lock, restricting parallelism
+p->TState = ObjectWaiter::TS_ENTER;
-// and effectively lengthening the critical section.
+p->_prev = q;
-// Invariant: s chases t chases u.
+q = p;
-// TODO-FIXME: consider changing EntryList from a DLL to a CDLL so
+}
-// we have faster access to the tail.
+// Append the RATs to the EntryList
-if (QMode == 1) {
+// TODO: organize EntryList as a CDLL so we can locate the tail in constant-time.
-// QMode == 1 : drain cxq to EntryList, reversing order
+ObjectWaiter * Tail;
-// We also reverse the order of the list.
+for (Tail = _EntryList; Tail != NULL && Tail->_next != NULL; Tail = Tail->_next);
-ObjectWaiter * s = NULL;
+if (Tail == NULL) {
-ObjectWaiter * t = w;
+_EntryList = w;
-ObjectWaiter * u = NULL;
-while (t != NULL) {
-guarantee(t->TState == ObjectWaiter::TS_CXQ, "invariant");
-t->TState = ObjectWaiter::TS_ENTER;
-u = t->_next;
-t->_prev = u;
-t->_next = s;
-s = t;
-t = u;
-}
-_EntryList  = s;
-assert(s != NULL, "invariant");
 } else {
-// QMode == 0 or QMode == 2
+Tail->_next = w;
-_EntryList = w;
+w->_prev = Tail;
-ObjectWaiter * q = NULL;
+}
-ObjectWaiter * p;
-for (p = w; p != NULL; p = p->_next) {
+// Fall thru into code that tries to wake a successor from EntryList
-guarantee(p->TState == ObjectWaiter::TS_CXQ, "Invariant");
+}
-p->TState = ObjectWaiter::TS_ENTER;
-p->_prev = q;
+if (QMode == 4 && _cxq != NULL) {
-q = p;
+// Aggressively drain cxq into EntryList at the first opportunity.
-}
+// This policy ensure that recently-run threads live at the head of EntryList.
-}
+// Drain _cxq into EntryList - bulk transfer.
-// In 1-0 mode we need: ST EntryList; MEMBAR #storestore; ST _owner = NULL
+// First, detach _cxq.
-// The MEMBAR is satisfied by the release_store() operation in ExitEpilog().
+// The following loop is tantamount to: w = swap (&cxq, NULL)
+w = _cxq;
-// See if we can abdicate to a spinner instead of waking a thread.
+for (;;) {
-// A primary goal of the implementation is to reduce the
+assert(w != NULL, "Invariant");
-// context-switch rate.
+ObjectWaiter * u = (ObjectWaiter *) Atomic::cmpxchg_ptr(NULL, &_cxq, w);
-if (_succ != NULL) continue;
+if (u == w) break;
+w = u;
-w = _EntryList;
+}
-if (w != NULL) {
+assert(w != NULL              , "invariant");
-guarantee(w->TState == ObjectWaiter::TS_ENTER, "invariant");
-ExitEpilog(Self, w);
+ObjectWaiter * q = NULL;
-return;
+ObjectWaiter * p;
-}
+for (p = w; p != NULL; p = p->_next) {
-}
+guarantee(p->TState == ObjectWaiter::TS_CXQ, "Invariant");
+p->TState = ObjectWaiter::TS_ENTER;
+p->_prev = q;
+q = p;
+}
+// Prepend the RATs to the EntryList
+if (_EntryList != NULL) {
+q->_next = _EntryList;
+_EntryList->_prev = q;
+}
+_EntryList = w;
+// Fall thru into code that tries to wake a successor from EntryList
+}
+w = _EntryList;
+if (w != NULL) {
+// I'd like to write: guarantee (w->_thread != Self).
+// But in practice an exiting thread may find itself on the EntryList.
+// Let's say thread T1 calls O.wait().  Wait() enqueues T1 on O's waitset and
+// then calls exit().  Exit release the lock by setting O._owner to NULL.
+// Let's say T1 then stalls.  T2 acquires O and calls O.notify().  The
+// notify() operation moves T1 from O's waitset to O's EntryList. T2 then
+// release the lock "O".  T2 resumes immediately after the ST of null into
+// _owner, above.  T2 notices that the EntryList is populated, so it
+// reacquires the lock and then finds itself on the EntryList.
+// Given all that, we have to tolerate the circumstance where "w" is
+// associated with Self.
+assert(w->TState == ObjectWaiter::TS_ENTER, "invariant");
+ExitEpilog(Self, w);
+return;
+}
+// If we find that both _cxq and EntryList are null then just
+// re-run the exit protocol from the top.
+w = _cxq;
+if (w == NULL) continue;
+// Drain _cxq into EntryList - bulk transfer.
+// First, detach _cxq.
+// The following loop is tantamount to: w = swap (&cxq, NULL)
+for (;;) {
+assert(w != NULL, "Invariant");
+ObjectWaiter * u = (ObjectWaiter *) Atomic::cmpxchg_ptr(NULL, &_cxq, w);
+if (u == w) break;
+w = u;
+}
+TEVENT(Inflated exit - drain cxq into EntryList);
+assert(w != NULL              , "invariant");
+assert(_EntryList  == NULL    , "invariant");
+// Convert the LIFO SLL anchored by _cxq into a DLL.
+// The list reorganization step operates in O(LENGTH(w)) time.
+// It's critical that this step operate quickly as
+// "Self" still holds the outer-lock, restricting parallelism
+// and effectively lengthening the critical section.
+// Invariant: s chases t chases u.
+// TODO-FIXME: consider changing EntryList from a DLL to a CDLL so
+// we have faster access to the tail.
+if (QMode == 1) {
+// QMode == 1 : drain cxq to EntryList, reversing order
+// We also reverse the order of the list.
+ObjectWaiter * s = NULL;
+ObjectWaiter * t = w;
+ObjectWaiter * u = NULL;
+while (t != NULL) {
+guarantee(t->TState == ObjectWaiter::TS_CXQ, "invariant");
+t->TState = ObjectWaiter::TS_ENTER;
+u = t->_next;
+t->_prev = u;
+t->_next = s;
+s = t;
+t = u;
+}
+_EntryList  = s;
+assert(s != NULL, "invariant");
+} else {
+// QMode == 0 or QMode == 2
+_EntryList = w;
+ObjectWaiter * q = NULL;
+ObjectWaiter * p;
+for (p = w; p != NULL; p = p->_next) {
+guarantee(p->TState == ObjectWaiter::TS_CXQ, "Invariant");
+p->TState = ObjectWaiter::TS_ENTER;
+p->_prev = q;
+q = p;
+}
+}
+// In 1-0 mode we need: ST EntryList; MEMBAR #storestore; ST _owner = NULL
+// The MEMBAR is satisfied by the release_store() operation in ExitEpilog().
+// See if we can abdicate to a spinner instead of waking a thread.
+// A primary goal of the implementation is to reduce the
+// context-switch rate.
+if (_succ != NULL) continue;
+w = _EntryList;
+if (w != NULL) {
+guarantee(w->TState == ObjectWaiter::TS_ENTER, "invariant");
+ExitEpilog(Self, w);
+return;
+}
+}
 }
 // ExitSuspendEquivalent:
 // A faster alternate to handle_special_suspend_equivalent_condition()
 //
 // with a more efficient implementation, the need to use "FastHSSEC" has
 // decreased. - Dave
 bool ObjectMonitor::ExitSuspendEquivalent (JavaThread * jSelf) {
 const int Mode = Knob_FastHSSEC;
 if (Mode && !jSelf->is_external_suspend()) {
 assert(jSelf->is_suspend_equivalent(), "invariant");
 jSelf->clear_suspend_equivalent();
 if (2 == Mode) OrderAccess::storeload();
 if (!jSelf->is_external_suspend()) return false;
 // We raced a suspension -- fall thru into the slow path
 TEVENT(ExitSuspendEquivalent - raced);
 jSelf->set_suspend_equivalent();
 }
 return jSelf->handle_special_suspend_equivalent_condition();
 }
 void ObjectMonitor::ExitEpilog (Thread * Self, ObjectWaiter * Wakee) {
 assert(_owner == Self, "invariant");
 // Exit protocol:
 // 1. ST _succ = wakee
 // 2. membar #loadstore|#storestore;
 // 2. ST _owner = NULL
 // 3. unpark(wakee)
 _succ = Knob_SuccEnabled ? Wakee->_thread : NULL;
 ParkEvent * Trigger = Wakee->_event;
 // Hygiene -- once we've set _owner = NULL we can't safely dereference Wakee again.
 // The thread associated with Wakee may have grabbed the lock and "Wakee" may be
 // out-of-scope (non-extant).
 Wakee  = NULL;
 // Drop the lock
 OrderAccess::release_store_ptr(&_owner, NULL);
 OrderAccess::fence();                               // ST _owner vs LD in unpark()
 if (SafepointSynchronize::do_call_back()) {
 TEVENT(unpark before SAFEPOINT);
 }
 DTRACE_MONITOR_PROBE(contended__exit, this, object(), Self);
 Trigger->unpark();
 // Maintain stats and report events to JVMTI
 if (ObjectMonitor::_sync_Parks != NULL) {
 ObjectMonitor::_sync_Parks->inc();
 }
 }
 // -----------------------------------------------------------------------------
 // Class Loader deadlock handling.
 // complete_exit requires an inflated monitor
 // The _owner field is not always the Thread addr even with an
 // inflated monitor, e.g. the monitor can be inflated by a non-owning
 // thread due to contention.
 intptr_t ObjectMonitor::complete_exit(TRAPS) {
 Thread * const Self = THREAD;
 assert(Self->is_Java_thread(), "Must be Java thread!");
 JavaThread *jt = (JavaThread *)THREAD;
 DeferredInitialize();
 if (THREAD != _owner) {
 if (THREAD->is_lock_owned ((address)_owner)) {
 assert(_recursions == 0, "internal state error");
 _owner = THREAD;   /* Convert from basiclock addr to Thread addr */
 _recursions = 0;
 OwnerIsThread = 1;
 }
 }
 guarantee(Self == _owner, "complete_exit not owner");
 intptr_t save = _recursions; // record the old recursion count
 _recursions = 0;        // set the recursion level to be 0
 exit(true, Self);           // exit the monitor
 guarantee(_owner != Self, "invariant");
 return save;
 }
 // reenter() enters a lock and sets recursion count
 // complete_exit/reenter operate as a wait without waiting
 void ObjectMonitor::reenter(intptr_t recursions, TRAPS) {
 Thread * const Self = THREAD;
 assert(Self->is_Java_thread(), "Must be Java thread!");
 JavaThread *jt = (JavaThread *)THREAD;
 guarantee(_owner != Self, "reenter already owner");
 enter(THREAD);       // enter the monitor
 guarantee(_recursions == 0, "reenter recursion");
 _recursions = recursions;
 return;
 }
 // -----------------------------------------------------------------------------
 // A macro is used below because there may already be a pending
 return v;
 }
 // helper method for posting a monitor wait event
 void ObjectMonitor::post_monitor_wait_event(EventJavaMonitorWait* event,
 jlong notifier_tid,
 jlong timeout,
 bool timedout) {
 event->set_klass(((oop)this->object())->klass());
 event->set_timeout((TYPE_ULONG)timeout);
 event->set_address((TYPE_ADDRESS)(uintptr_t)(this->object_addr()));
 event->set_notifier((TYPE_OSTHREAD)notifier_tid);
 event->set_timedOut((TYPE_BOOLEAN)timedout);
 // Wait/Notify/NotifyAll
 //
 // Note: a subset of changes to ObjectMonitor::wait()
 // will need to be replicated in complete_exit
 void ObjectMonitor::wait(jlong millis, bool interruptible, TRAPS) {
 Thread * const Self = THREAD;
 assert(Self->is_Java_thread(), "Must be Java thread!");
 JavaThread *jt = (JavaThread *)THREAD;
 DeferredInitialize();
 // Throw IMSX or IEX.
 CHECK_OWNER();
 EventJavaMonitorWait event;
 // check for a pending interrupt
 if (interruptible && Thread::is_interrupted(Self, true) && !HAS_PENDING_EXCEPTION) {
 // post monitor waited event.  Note that this is past-tense, we are done waiting.
 if (JvmtiExport::should_post_monitor_waited()) {
 // Note: 'false' parameter is passed here because the
 // wait was not timed out due to thread interrupt.
 JvmtiExport::post_monitor_waited(jt, this, false);
 // In this short circuit of the monitor wait protocol, the
 // current thread never drops ownership of the monitor and
 // never gets added to the wait queue so the current thread
 // cannot be made the successor. This means that the
 // JVMTI_EVENT_MONITOR_WAITED event handler cannot accidentally
 // consume an unpark() meant for the ParkEvent associated with
 // this ObjectMonitor.
 }
 if (event.should_commit()) {
 post_monitor_wait_event(&event, 0, millis, false);
 }
 TEVENT(Wait - Throw IEX);
 THROW(vmSymbols::java_lang_InterruptedException());
 return;
 }
 TEVENT(Wait);
 assert(Self->_Stalled == 0, "invariant");
 Self->_Stalled = intptr_t(this);
 jt->set_current_waiting_monitor(this);
 // create a node to be put into the queue
 // Critically, after we reset() the event but prior to park(), we must check
 // for a pending interrupt.
 ObjectWaiter node(Self);
 node.TState = ObjectWaiter::TS_WAIT;
 Self->_ParkEvent->reset();
 OrderAccess::fence();          // ST into Event; membar ; LD interrupted-flag
 // Enter the waiting queue, which is a circular doubly linked list in this case
 // but it could be a priority queue or any data structure.
 // _WaitSetLock protects the wait queue.  Normally the wait queue is accessed only
 // by the the owner of the monitor *except* in the case where park()
 // returns because of a timeout of interrupt.  Contention is exceptionally rare
 // so we use a simple spin-lock instead of a heavier-weight blocking lock.
 Thread::SpinAcquire(&_WaitSetLock, "WaitSet - add");
 AddWaiter(&node);
 Thread::SpinRelease(&_WaitSetLock);
 if ((SyncFlags & 4) == 0) {
 _Responsible = NULL;
 }
 intptr_t save = _recursions; // record the old recursion count
 _waiters++;                  // increment the number of waiters
 _recursions = 0;             // set the recursion level to be 1
 exit(true, Self);                    // exit the monitor
 guarantee(_owner != Self, "invariant");
 // The thread is on the WaitSet list - now park() it.
 // On MP systems it's conceivable that a brief spin before we park
 // could be profitable.
 //
 // TODO-FIXME: change the following logic to a loop of the form
 //   while (!timeout && !interrupted && _notified == 0) park()
 int ret = OS_OK;
 int WasNotified = 0;
 { // State transition wrappers
 OSThread* osthread = Self->osthread();
 OSThreadWaitState osts(osthread, true);
 {
 ThreadBlockInVM tbivm(jt);
 // Thread is in thread_blocked state and oop access is unsafe.
 jt->set_suspend_equivalent();
 if (interruptible && (Thread::is_interrupted(THREAD, false) || HAS_PENDING_EXCEPTION)) {
 // Intentionally empty
 } else
 if (node._notified == 0) {
 if (millis <= 0) {
 Self->_ParkEvent->park();
 } else {
 ret = Self->_ParkEvent->park(millis);
 }
 }
 // were we externally suspended while we were waiting?
 if (ExitSuspendEquivalent (jt)) {
 // TODO-FIXME: add -- if succ == Self then succ = null.
 jt->java_suspend_self();
 }
 } // Exit thread safepoint: transition _thread_blocked -> _thread_in_vm
 // Node may be on the WaitSet, the EntryList (or cxq), or in transition
 // from the WaitSet to the EntryList.
 // See if we need to remove Node from the WaitSet.
 // We use double-checked locking to avoid grabbing _WaitSetLock
 // if the thread is not on the wait queue.
 //
 // Note that we don't need a fence before the fetch of TState.
 // In the worst case we'll fetch a old-stale value of TS_WAIT previously
 // written by the is thread. (perhaps the fetch might even be satisfied
 // by a look-aside into the processor's own store buffer, although given
 // the length of the code path between the prior ST and this load that's
 // highly unlikely).  If the following LD fetches a stale TS_WAIT value
 // then we'll acquire the lock and then re-fetch a fresh TState value.
 // That is, we fail toward safety.
 if (node.TState == ObjectWaiter::TS_WAIT) {
 Thread::SpinAcquire(&_WaitSetLock, "WaitSet - unlink");
 if (node.TState == ObjectWaiter::TS_WAIT) {
 DequeueSpecificWaiter(&node);       // unlink from WaitSet
 assert(node._notified == 0, "invariant");
 node.TState = ObjectWaiter::TS_RUN;
 }
 Thread::SpinRelease(&_WaitSetLock);
 }
 // The thread is now either on off-list (TS_RUN),
 // on the EntryList (TS_ENTER), or on the cxq (TS_CXQ).
 // The Node's TState variable is stable from the perspective of this thread.
 // No other threads will asynchronously modify TState.
 guarantee(node.TState != ObjectWaiter::TS_WAIT, "invariant");
 OrderAccess::loadload();
 if (_succ == Self) _succ = NULL;
 WasNotified = node._notified;
 // Reentry phase -- reacquire the monitor.
 // re-enter contended monitor after object.wait().
 // retain OBJECT_WAIT state until re-enter successfully completes
 // Thread state is thread_in_vm and oop access is again safe,
 // although the raw address of the object may have changed.
 // (Don't cache naked oops over safepoints, of course).
 // post monitor waited event. Note that this is past-tense, we are done waiting.
 if (JvmtiExport::should_post_monitor_waited()) {
 JvmtiExport::post_monitor_waited(jt, this, ret == OS_TIMEOUT);
 if (node._notified != 0 && _succ == Self) {
 // In this part of the monitor wait-notify-reenter protocol it
 // is possible (and normal) for another thread to do a fastpath
 // monitor enter-exit while this thread is still trying to get
 // to the reenter portion of the protocol.
 //
 // The ObjectMonitor was notified and the current thread is
 // the successor which also means that an unpark() has already
 // been done. The JVMTI_EVENT_MONITOR_WAITED event handler can
 // consume the unpark() that was done when the successor was
 // set because the same ParkEvent is shared between Java
 // monitors and JVM/TI RawMonitors (for now).
 //
 // We redo the unpark() to ensure forward progress, i.e., we
 // don't want all pending threads hanging (parked) with none
 // entering the unlocked monitor.
 node._event->unpark();
 }
 }
 if (event.should_commit()) {
 post_monitor_wait_event(&event, node._notifier_tid, millis, ret == OS_TIMEOUT);
 }
 OrderAccess::fence();
 assert(Self->_Stalled != 0, "invariant");
 Self->_Stalled = 0;
 assert(_owner != Self, "invariant");
 ObjectWaiter::TStates v = node.TState;
 if (v == ObjectWaiter::TS_RUN) {
 enter(Self);
 } else {
 guarantee(v == ObjectWaiter::TS_ENTER || v == ObjectWaiter::TS_CXQ, "invariant");
 ReenterI(Self, &node);
 node.wait_reenter_end(this);
 }
 // Self has reacquired the lock.
 // Lifecycle - the node representing Self must not appear on any queues.
 // Node is about to go out-of-scope, but even if it were immortal we wouldn't
 // want residual elements associated with this thread left on any lists.
 guarantee(node.TState == ObjectWaiter::TS_RUN, "invariant");
 assert(_owner == Self, "invariant");
 assert(_succ != Self , "invariant");
 } // OSThreadWaitState()
 jt->set_current_waiting_monitor(NULL);
 guarantee(_recursions == 0, "invariant");
 _recursions = save;     // restore the old recursion count
 _waiters--;             // decrement the number of waiters
 // Verify a few postconditions
 assert(_owner == Self       , "invariant");
 assert(_succ  != Self       , "invariant");
 assert(((oop)(object()))->mark() == markOopDesc::encode(this), "invariant");
 if (SyncFlags & 32) {
 OrderAccess::fence();
 }
 // check if the notification happened
 if (!WasNotified) {
 // no, it could be timeout or Thread.interrupt() or both
 // check for interrupt event, otherwise it is timeout
 if (interruptible && Thread::is_interrupted(Self, true) && !HAS_PENDING_EXCEPTION) {
 TEVENT(Wait - throw IEX from epilog);
 THROW(vmSymbols::java_lang_InterruptedException());
 }
 }
 // NOTE: Spurious wake up will be consider as timeout.
 // Monitor notify has precedence over thread interrupt.
 }
 // Consider:
 // If the lock is cool (cxq == null && succ == null) and we're on an MP system
 // we might just dequeue a thread from the WaitSet and directly unpark() it.
 void ObjectMonitor::notify(TRAPS) {
 CHECK_OWNER();
 if (_WaitSet == NULL) {
 TEVENT(Empty-Notify);
 return;
 }
 DTRACE_MONITOR_PROBE(notify, this, object(), THREAD);
 int Policy = Knob_MoveNotifyee;
 Thread::SpinAcquire(&_WaitSetLock, "WaitSet - notify");
 ObjectWaiter * iterator = DequeueWaiter();
 if (iterator != NULL) {
 TEVENT(Notify1 - Transfer);
 guarantee(iterator->TState == ObjectWaiter::TS_WAIT, "invariant");
 guarantee(iterator->_notified == 0, "invariant");
 if (Policy != 4) {
 iterator->TState = ObjectWaiter::TS_ENTER;
 }
 iterator->_notified = 1;
 Thread * Self = THREAD;
 iterator->_notifier_tid = Self->osthread()->thread_id();
 ObjectWaiter * List = _EntryList;
 if (List != NULL) {
 assert(List->_prev == NULL, "invariant");
 assert(List->TState == ObjectWaiter::TS_ENTER, "invariant");
 assert(List != iterator, "invariant");
 }
 if (Policy == 0) {       // prepend to EntryList
 if (List == NULL) {
 iterator->_next = iterator->_prev = NULL;
 _EntryList = iterator;
 } else {
 List->_prev = iterator;
 iterator->_next = List;
 iterator->_prev = NULL;
 _EntryList = iterator;
 }
 } else
 if (Policy == 1) {      // append to EntryList
 if (List == NULL) {
 iterator->_next = iterator->_prev = NULL;
 _EntryList = iterator;
 } else {
 // CONSIDER:  finding the tail currently requires a linear-time walk of
 // the EntryList.  We can make tail access constant-time by converting to
 // a CDLL instead of using our current DLL.
 ObjectWaiter * Tail;
 for (Tail = List; Tail->_next != NULL; Tail = Tail->_next);
 assert(Tail != NULL && Tail->_next == NULL, "invariant");
 Tail->_next = iterator;
 iterator->_prev = Tail;
 iterator->_next = NULL;
 }
 } else
 if (Policy == 2) {      // prepend to cxq
 // prepend to cxq
 if (List == NULL) {
 iterator->_next = iterator->_prev = NULL;
 _EntryList = iterator;
 } else {
-iterator->TState = ObjectWaiter::TS_CXQ;
-for (;;) {
-ObjectWaiter * Front = _cxq;
-iterator->_next = Front;
-if (Atomic::cmpxchg_ptr (iterator, &_cxq, Front) == Front) {
-break;
-}
-}
-}
-} else
-if (Policy == 3) {      // append to cxq
 iterator->TState = ObjectWaiter::TS_CXQ;
 for (;;) {
-ObjectWaiter * Tail;
+ObjectWaiter * Front = _cxq;
-Tail = _cxq;
+iterator->_next = Front;
-if (Tail == NULL) {
+if (Atomic::cmpxchg_ptr (iterator, &_cxq, Front) == Front) {
-iterator->_next = NULL;
+break;
-if (Atomic::cmpxchg_ptr (iterator, &_cxq, NULL) == NULL) {
+}
-break;
-}
-} else {
-while (Tail->_next != NULL) Tail = Tail->_next;
-Tail->_next = iterator;
-iterator->_prev = Tail;
-iterator->_next = NULL;
-break;
-}
 }
-} else {
+}
-ParkEvent * ev = iterator->_event;
+} else
-iterator->TState = ObjectWaiter::TS_RUN;
+if (Policy == 3) {      // append to cxq
-OrderAccess::fence();
+iterator->TState = ObjectWaiter::TS_CXQ;
-ev->unpark();
+for (;;) {
-}
+ObjectWaiter * Tail;
+Tail = _cxq;
-if (Policy < 4) {
+if (Tail == NULL) {
-iterator->wait_reenter_begin(this);
+iterator->_next = NULL;
-}
+if (Atomic::cmpxchg_ptr (iterator, &_cxq, NULL) == NULL) {
+break;
-// _WaitSetLock protects the wait queue, not the EntryList.  We could
+}
-// move the add-to-EntryList operation, above, outside the critical section
+} else {
-// protected by _WaitSetLock.  In practice that's not useful.  With the
+while (Tail->_next != NULL) Tail = Tail->_next;
-// exception of  wait() timeouts and interrupts the monitor owner
+Tail->_next = iterator;
-// is the only thread that grabs _WaitSetLock.  There's almost no contention
+iterator->_prev = Tail;
-// on _WaitSetLock so it's not profitable to reduce the length of the
+iterator->_next = NULL;
-// critical section.
+break;
+}
+}
+} else {
+ParkEvent * ev = iterator->_event;
+iterator->TState = ObjectWaiter::TS_RUN;
+OrderAccess::fence();
+ev->unpark();
+}
+if (Policy < 4) {
+iterator->wait_reenter_begin(this);
+}
+// _WaitSetLock protects the wait queue, not the EntryList.  We could
+// move the add-to-EntryList operation, above, outside the critical section
+// protected by _WaitSetLock.  In practice that's not useful.  With the
+// exception of  wait() timeouts and interrupts the monitor owner
+// is the only thread that grabs _WaitSetLock.  There's almost no contention
+// on _WaitSetLock so it's not profitable to reduce the length of the
+// critical section.
 }
 Thread::SpinRelease(&_WaitSetLock);
 if (iterator != NULL && ObjectMonitor::_sync_Notifications != NULL) {
 ObjectMonitor::_sync_Notifications->inc();
 }
 }
 void ObjectMonitor::notifyAll(TRAPS) {
 CHECK_OWNER();
 ObjectWaiter* iterator;
 if (_WaitSet == NULL) {
 TEVENT(Empty-NotifyAll);
 return;
 }
 DTRACE_MONITOR_PROBE(notifyAll, this, object(), THREAD);
 int Policy = Knob_MoveNotifyee;
 int Tally = 0;
 Thread::SpinAcquire(&_WaitSetLock, "WaitSet - notifyall");
 for (;;) {
 iterator = DequeueWaiter();
 if (iterator == NULL) break;
 TEVENT(NotifyAll - Transfer1);
 ++Tally;
 // Disposition - what might we do with iterator ?
 // a.  add it directly to the EntryList - either tail or head.
 // b.  push it onto the front of the _cxq.
 // For now we use (a).
 guarantee(iterator->TState == ObjectWaiter::TS_WAIT, "invariant");
 guarantee(iterator->_notified == 0, "invariant");
 iterator->_notified = 1;
 Thread * Self = THREAD;
 iterator->_notifier_tid = Self->osthread()->thread_id();
 if (Policy != 4) {
 iterator->TState = ObjectWaiter::TS_ENTER;
 }
 ObjectWaiter * List = _EntryList;
 if (List != NULL) {
 assert(List->_prev == NULL, "invariant");
 assert(List->TState == ObjectWaiter::TS_ENTER, "invariant");
 assert(List != iterator, "invariant");
 }
 if (Policy == 0) {       // prepend to EntryList
 if (List == NULL) {
 iterator->_next = iterator->_prev = NULL;
 _EntryList = iterator;
 } else {
 List->_prev = iterator;
 iterator->_next = List;
 iterator->_prev = NULL;
 _EntryList = iterator;
+}
+} else
+if (Policy == 1) {      // append to EntryList
+if (List == NULL) {
+iterator->_next = iterator->_prev = NULL;
+_EntryList = iterator;
+} else {
+// CONSIDER:  finding the tail currently requires a linear-time walk of
+// the EntryList.  We can make tail access constant-time by converting to
+// a CDLL instead of using our current DLL.
+ObjectWaiter * Tail;
+for (Tail = List; Tail->_next != NULL; Tail = Tail->_next);
+assert(Tail != NULL && Tail->_next == NULL, "invariant");
+Tail->_next = iterator;
+iterator->_prev = Tail;
+iterator->_next = NULL;
+}
+} else
+if (Policy == 2) {      // prepend to cxq
+// prepend to cxq
+iterator->TState = ObjectWaiter::TS_CXQ;
+for (;;) {
+ObjectWaiter * Front = _cxq;
+iterator->_next = Front;
+if (Atomic::cmpxchg_ptr (iterator, &_cxq, Front) == Front) {
+break;
 }
-} else
+}
-if (Policy == 1) {      // append to EntryList
+} else
-if (List == NULL) {
+if (Policy == 3) {      // append to cxq
-iterator->_next = iterator->_prev = NULL;
+iterator->TState = ObjectWaiter::TS_CXQ;
-_EntryList = iterator;
+for (;;) {
-} else {
+ObjectWaiter * Tail;
-// CONSIDER:  finding the tail currently requires a linear-time walk of
+Tail = _cxq;
-// the EntryList.  We can make tail access constant-time by converting to
+if (Tail == NULL) {
-// a CDLL instead of using our current DLL.
+iterator->_next = NULL;
-ObjectWaiter * Tail;
+if (Atomic::cmpxchg_ptr (iterator, &_cxq, NULL) == NULL) {
-for (Tail = List; Tail->_next != NULL; Tail = Tail->_next);
+break;
-assert(Tail != NULL && Tail->_next == NULL, "invariant");
+}
-Tail->_next = iterator;
+} else {
-iterator->_prev = Tail;
+while (Tail->_next != NULL) Tail = Tail->_next;
-iterator->_next = NULL;
+Tail->_next = iterator;
+iterator->_prev = Tail;
+iterator->_next = NULL;
+break;
 }
-} else
+}
-if (Policy == 2) {      // prepend to cxq
+} else {
-// prepend to cxq
+ParkEvent * ev = iterator->_event;
-iterator->TState = ObjectWaiter::TS_CXQ;
+iterator->TState = ObjectWaiter::TS_RUN;
-for (;;) {
+OrderAccess::fence();
-ObjectWaiter * Front = _cxq;
+ev->unpark();
-iterator->_next = Front;
+}
-if (Atomic::cmpxchg_ptr (iterator, &_cxq, Front) == Front) {
-break;
+if (Policy < 4) {
-}
+iterator->wait_reenter_begin(this);
 }
-} else
-if (Policy == 3) {      // append to cxq
+// _WaitSetLock protects the wait queue, not the EntryList.  We could
-iterator->TState = ObjectWaiter::TS_CXQ;
+// move the add-to-EntryList operation, above, outside the critical section
-for (;;) {
+// protected by _WaitSetLock.  In practice that's not useful.  With the
-ObjectWaiter * Tail;
+// exception of  wait() timeouts and interrupts the monitor owner
-Tail = _cxq;
+// is the only thread that grabs _WaitSetLock.  There's almost no contention
-if (Tail == NULL) {
+// on _WaitSetLock so it's not profitable to reduce the length of the
-iterator->_next = NULL;
+// critical section.
-if (Atomic::cmpxchg_ptr (iterator, &_cxq, NULL) == NULL) {
-break;
-}
-} else {
-while (Tail->_next != NULL) Tail = Tail->_next;
-Tail->_next = iterator;
-iterator->_prev = Tail;
-iterator->_next = NULL;
-break;
-}
-}
-} else {
-ParkEvent * ev = iterator->_event;
-iterator->TState = ObjectWaiter::TS_RUN;
-OrderAccess::fence();
-ev->unpark();
-}
-if (Policy < 4) {
-iterator->wait_reenter_begin(this);
-}
-// _WaitSetLock protects the wait queue, not the EntryList.  We could
-// move the add-to-EntryList operation, above, outside the critical section
-// protected by _WaitSetLock.  In practice that's not useful.  With the
-// exception of  wait() timeouts and interrupts the monitor owner
-// is the only thread that grabs _WaitSetLock.  There's almost no contention
-// on _WaitSetLock so it's not profitable to reduce the length of the
-// critical section.
 }
 Thread::SpinRelease(&_WaitSetLock);
 if (Tally != 0 && ObjectMonitor::_sync_Notifications != NULL) {
 ObjectMonitor::_sync_Notifications->inc(Tally);
 }
 }
 // -----------------------------------------------------------------------------
 // Adaptive Spinning Support
 // Spinning: Fixed frequency (100%), vary duration
 int ObjectMonitor::TrySpin_VaryDuration (Thread * Self) {
 // Dumb, brutal spin.  Good for comparative measurements against adaptive spinning.
 int ctr = Knob_FixedSpin;
 if (ctr != 0) {
 while (--ctr >= 0) {
 if (TryLock(Self) > 0) return 1;
 SpinPause();
+}
+return 0;
+}
+for (ctr = Knob_PreSpin + 1; --ctr >= 0;) {
+if (TryLock(Self) > 0) {
+// Increase _SpinDuration ...
+// Note that we don't clamp SpinDuration precisely at SpinLimit.
+// Raising _SpurDuration to the poverty line is key.
+int x = _SpinDuration;
+if (x < Knob_SpinLimit) {
+if (x < Knob_Poverty) x = Knob_Poverty;
+_SpinDuration = x + Knob_BonusB;
+}
+return 1;
+}
+SpinPause();
+}
+// Admission control - verify preconditions for spinning
+//
+// We always spin a little bit, just to prevent _SpinDuration == 0 from
+// becoming an absorbing state.  Put another way, we spin briefly to
+// sample, just in case the system load, parallelism, contention, or lock
+// modality changed.
+//
+// Consider the following alternative:
+// Periodically set _SpinDuration = _SpinLimit and try a long/full
+// spin attempt.  "Periodically" might mean after a tally of
+// the # of failed spin attempts (or iterations) reaches some threshold.
+// This takes us into the realm of 1-out-of-N spinning, where we
+// hold the duration constant but vary the frequency.
+ctr = _SpinDuration;
+if (ctr < Knob_SpinBase) ctr = Knob_SpinBase;
+if (ctr <= 0) return 0;
+if (Knob_SuccRestrict && _succ != NULL) return 0;
+if (Knob_OState && NotRunnable (Self, (Thread *) _owner)) {
+TEVENT(Spin abort - notrunnable [TOP]);
+return 0;
+}
+int MaxSpin = Knob_MaxSpinners;
+if (MaxSpin >= 0) {
+if (_Spinner > MaxSpin) {
+TEVENT(Spin abort -- too many spinners);
+return 0;
+}
+// Slightly racy, but benign ...
+Adjust(&_Spinner, 1);
+}
+// We're good to spin ... spin ingress.
+// CONSIDER: use Prefetch::write() to avoid RTS->RTO upgrades
+// when preparing to LD...CAS _owner, etc and the CAS is likely
+// to succeed.
+int hits    = 0;
+int msk     = 0;
+int caspty  = Knob_CASPenalty;
+int oxpty   = Knob_OXPenalty;
+int sss     = Knob_SpinSetSucc;
+if (sss && _succ == NULL) _succ = Self;
+Thread * prv = NULL;
+// There are three ways to exit the following loop:
+// 1.  A successful spin where this thread has acquired the lock.
+// 2.  Spin failure with prejudice
+// 3.  Spin failure without prejudice
+while (--ctr >= 0) {
+// Periodic polling -- Check for pending GC
+// Threads may spin while they're unsafe.
+// We don't want spinning threads to delay the JVM from reaching
+// a stop-the-world safepoint or to steal cycles from GC.
+// If we detect a pending safepoint we abort in order that
+// (a) this thread, if unsafe, doesn't delay the safepoint, and (b)
+// this thread, if safe, doesn't steal cycles from GC.
+// This is in keeping with the "no loitering in runtime" rule.
+// We periodically check to see if there's a safepoint pending.
+if ((ctr & 0xFF) == 0) {
+if (SafepointSynchronize::do_call_back()) {
+TEVENT(Spin: safepoint);
+goto Abort;           // abrupt spin egress
+}
+if (Knob_UsePause & 1) SpinPause();
+int (*scb)(intptr_t,int) = SpinCallbackFunction;
+if (hits > 50 && scb != NULL) {
+int abend = (*scb)(SpinCallbackArgument, 0);
+}
+}
+if (Knob_UsePause & 2) SpinPause();
+// Exponential back-off ...  Stay off the bus to reduce coherency traffic.
+// This is useful on classic SMP systems, but is of less utility on
+// N1-style CMT platforms.
+//
+// Trade-off: lock acquisition latency vs coherency bandwidth.
+// Lock hold times are typically short.  A histogram
+// of successful spin attempts shows that we usually acquire
+// the lock early in the spin.  That suggests we want to
+// sample _owner frequently in the early phase of the spin,
+// but then back-off and sample less frequently as the spin
+// progresses.  The back-off makes a good citizen on SMP big
+// SMP systems.  Oversampling _owner can consume excessive
+// coherency bandwidth.  Relatedly, if we _oversample _owner we
+// can inadvertently interfere with the the ST m->owner=null.
+// executed by the lock owner.
+if (ctr & msk) continue;
+++hits;
+if ((hits & 0xF) == 0) {
+// The 0xF, above, corresponds to the exponent.
+// Consider: (msk+1)|msk
+msk = ((msk << 2)|3) & BackOffMask;
+}
+// Probe _owner with TATAS
+// If this thread observes the monitor transition or flicker
+// from locked to unlocked to locked, then the odds that this
+// thread will acquire the lock in this spin attempt go down
+// considerably.  The same argument applies if the CAS fails
+// or if we observe _owner change from one non-null value to
+// another non-null value.   In such cases we might abort
+// the spin without prejudice or apply a "penalty" to the
+// spin count-down variable "ctr", reducing it by 100, say.
+Thread * ox = (Thread *) _owner;
+if (ox == NULL) {
+ox = (Thread *) Atomic::cmpxchg_ptr(Self, &_owner, NULL);
+if (ox == NULL) {
+// The CAS succeeded -- this thread acquired ownership
+// Take care of some bookkeeping to exit spin state.
+if (sss && _succ == Self) {
+_succ = NULL;
 }
-return 0;
+if (MaxSpin > 0) Adjust(&_Spinner, -1);
-}
+// Increase _SpinDuration :
-for (ctr = Knob_PreSpin + 1; --ctr >= 0;) {
+// The spin was successful (profitable) so we tend toward
-if (TryLock(Self) > 0) {
+// longer spin attempts in the future.
-// Increase _SpinDuration ...
+// CONSIDER: factor "ctr" into the _SpinDuration adjustment.
+// If we acquired the lock early in the spin cycle it
+// makes sense to increase _SpinDuration proportionally.
 // Note that we don't clamp SpinDuration precisely at SpinLimit.
-// Raising _SpurDuration to the poverty line is key.
 int x = _SpinDuration;
 if (x < Knob_SpinLimit) {
 if (x < Knob_Poverty) x = Knob_Poverty;
-_SpinDuration = x + Knob_BonusB;
+_SpinDuration = x + Knob_Bonus;
 }
 return 1;
 }
-SpinPause();
-}
+// The CAS failed ... we can take any of the following actions:
+// * penalize: ctr -= Knob_CASPenalty
-// Admission control - verify preconditions for spinning
+// * exit spin with prejudice -- goto Abort;
-//
+// * exit spin without prejudice.
-// We always spin a little bit, just to prevent _SpinDuration == 0 from
+// * Since CAS is high-latency, retry again immediately.
-// becoming an absorbing state.  Put another way, we spin briefly to
+prv = ox;
-// sample, just in case the system load, parallelism, contention, or lock
+TEVENT(Spin: cas failed);
-// modality changed.
+if (caspty == -2) break;
-//
+if (caspty == -1) goto Abort;
-// Consider the following alternative:
+ctr -= caspty;
-// Periodically set _SpinDuration = _SpinLimit and try a long/full
+continue;
-// spin attempt.  "Periodically" might mean after a tally of
+}
-// the # of failed spin attempts (or iterations) reaches some threshold.
-// This takes us into the realm of 1-out-of-N spinning, where we
+// Did lock ownership change hands ?
-// hold the duration constant but vary the frequency.
+if (ox != prv && prv != NULL) {
+TEVENT(spin: Owner changed)
-ctr = _SpinDuration;
+if (oxpty == -2) break;
-if (ctr < Knob_SpinBase) ctr = Knob_SpinBase;
+if (oxpty == -1) goto Abort;
-if (ctr <= 0) return 0;
+ctr -= oxpty;
+}
-if (Knob_SuccRestrict && _succ != NULL) return 0;
+prv = ox;
-if (Knob_OState && NotRunnable (Self, (Thread *) _owner)) {
-TEVENT(Spin abort - notrunnable [TOP]);
+// Abort the spin if the owner is not executing.
-return 0;
+// The owner must be executing in order to drop the lock.
-}
+// Spinning while the owner is OFFPROC is idiocy.
+// Consider: ctr -= RunnablePenalty ;
-int MaxSpin = Knob_MaxSpinners;
+if (Knob_OState && NotRunnable (Self, ox)) {
-if (MaxSpin >= 0) {
+TEVENT(Spin abort - notrunnable);
-if (_Spinner > MaxSpin) {
+goto Abort;
-TEVENT(Spin abort -- too many spinners);
+}
-return 0;
-}
-// Slightly racy, but benign ...
-Adjust(&_Spinner, 1);
-}
-// We're good to spin ... spin ingress.
-// CONSIDER: use Prefetch::write() to avoid RTS->RTO upgrades
-// when preparing to LD...CAS _owner, etc and the CAS is likely
-// to succeed.
-int hits    = 0;
-int msk     = 0;
-int caspty  = Knob_CASPenalty;
-int oxpty   = Knob_OXPenalty;
-int sss     = Knob_SpinSetSucc;
 if (sss && _succ == NULL) _succ = Self;
-Thread * prv = NULL;
+}
-// There are three ways to exit the following loop:
+// Spin failed with prejudice -- reduce _SpinDuration.
-// 1.  A successful spin where this thread has acquired the lock.
+// TODO: Use an AIMD-like policy to adjust _SpinDuration.
-// 2.  Spin failure with prejudice
+// AIMD is globally stable.
-// 3.  Spin failure without prejudice
+TEVENT(Spin failure);
+{
-while (--ctr >= 0) {
+int x = _SpinDuration;
+if (x > 0) {
-// Periodic polling -- Check for pending GC
+// Consider an AIMD scheme like: x -= (x >> 3) + 100
-// Threads may spin while they're unsafe.
+// This is globally sample and tends to damp the response.
-// We don't want spinning threads to delay the JVM from reaching
+x -= Knob_Penalty;
-// a stop-the-world safepoint or to steal cycles from GC.
+if (x < 0) x = 0;
-// If we detect a pending safepoint we abort in order that
+_SpinDuration = x;
-// (a) this thread, if unsafe, doesn't delay the safepoint, and (b)
+}
-// this thread, if safe, doesn't steal cycles from GC.
+}
-// This is in keeping with the "no loitering in runtime" rule.
-// We periodically check to see if there's a safepoint pending.
-if ((ctr & 0xFF) == 0) {
-if (SafepointSynchronize::do_call_back()) {
-TEVENT(Spin: safepoint);
-goto Abort;           // abrupt spin egress
-}
-if (Knob_UsePause & 1) SpinPause();
-int (*scb)(intptr_t,int) = SpinCallbackFunction;
-if (hits > 50 && scb != NULL) {
-int abend = (*scb)(SpinCallbackArgument, 0);
-}
-}
-if (Knob_UsePause & 2) SpinPause();
-// Exponential back-off ...  Stay off the bus to reduce coherency traffic.
-// This is useful on classic SMP systems, but is of less utility on
-// N1-style CMT platforms.
-//
-// Trade-off: lock acquisition latency vs coherency bandwidth.
-// Lock hold times are typically short.  A histogram
-// of successful spin attempts shows that we usually acquire
-// the lock early in the spin.  That suggests we want to
-// sample _owner frequently in the early phase of the spin,
-// but then back-off and sample less frequently as the spin
-// progresses.  The back-off makes a good citizen on SMP big
-// SMP systems.  Oversampling _owner can consume excessive
-// coherency bandwidth.  Relatedly, if we _oversample _owner we
-// can inadvertently interfere with the the ST m->owner=null.
-// executed by the lock owner.
-if (ctr & msk) continue;
-++hits;
-if ((hits & 0xF) == 0) {
-// The 0xF, above, corresponds to the exponent.
-// Consider: (msk+1)|msk
-msk = ((msk << 2)|3) & BackOffMask;
-}
-// Probe _owner with TATAS
-// If this thread observes the monitor transition or flicker
-// from locked to unlocked to locked, then the odds that this
-// thread will acquire the lock in this spin attempt go down
-// considerably.  The same argument applies if the CAS fails
-// or if we observe _owner change from one non-null value to
-// another non-null value.   In such cases we might abort
-// the spin without prejudice or apply a "penalty" to the
-// spin count-down variable "ctr", reducing it by 100, say.
-Thread * ox = (Thread *) _owner;
-if (ox == NULL) {
-ox = (Thread *) Atomic::cmpxchg_ptr(Self, &_owner, NULL);
-if (ox == NULL) {
-// The CAS succeeded -- this thread acquired ownership
-// Take care of some bookkeeping to exit spin state.
-if (sss && _succ == Self) {
-_succ = NULL;
-}
-if (MaxSpin > 0) Adjust(&_Spinner, -1);
-// Increase _SpinDuration :
-// The spin was successful (profitable) so we tend toward
-// longer spin attempts in the future.
-// CONSIDER: factor "ctr" into the _SpinDuration adjustment.
-// If we acquired the lock early in the spin cycle it
-// makes sense to increase _SpinDuration proportionally.
-// Note that we don't clamp SpinDuration precisely at SpinLimit.
-int x = _SpinDuration;
-if (x < Knob_SpinLimit) {
-if (x < Knob_Poverty) x = Knob_Poverty;
-_SpinDuration = x + Knob_Bonus;
-}
-return 1;
-}
-// The CAS failed ... we can take any of the following actions:
-// * penalize: ctr -= Knob_CASPenalty
-// * exit spin with prejudice -- goto Abort;
-// * exit spin without prejudice.
-// * Since CAS is high-latency, retry again immediately.
-prv = ox;
-TEVENT(Spin: cas failed);
-if (caspty == -2) break;
-if (caspty == -1) goto Abort;
-ctr -= caspty;
-continue;
-}
-// Did lock ownership change hands ?
-if (ox != prv && prv != NULL) {
-TEVENT(spin: Owner changed)
-if (oxpty == -2) break;
-if (oxpty == -1) goto Abort;
-ctr -= oxpty;
-}
-prv = ox;
-// Abort the spin if the owner is not executing.
-// The owner must be executing in order to drop the lock.
-// Spinning while the owner is OFFPROC is idiocy.
-// Consider: ctr -= RunnablePenalty ;
-if (Knob_OState && NotRunnable (Self, ox)) {
-TEVENT(Spin abort - notrunnable);
-goto Abort;
-}
-if (sss && _succ == NULL) _succ = Self;
-}
-// Spin failed with prejudice -- reduce _SpinDuration.
-// TODO: Use an AIMD-like policy to adjust _SpinDuration.
-// AIMD is globally stable.
-TEVENT(Spin failure);
-{
-int x = _SpinDuration;
-if (x > 0) {
-// Consider an AIMD scheme like: x -= (x >> 3) + 100
-// This is globally sample and tends to damp the response.
-x -= Knob_Penalty;
-if (x < 0) x = 0;
-_SpinDuration = x;
-}
-}
 Abort:
 if (MaxSpin >= 0) Adjust(&_Spinner, -1);
 if (sss && _succ == Self) {
 _succ = NULL;
 // Invariant: after setting succ=null a contending thread
 // must recheck-retry _owner before parking.  This usually happens
 // in the normal usage of TrySpin(), but it's safest
 // to make TrySpin() as foolproof as possible.
 OrderAccess::fence();
 if (TryLock(Self) > 0) return 1;
 }
 return 0;
 }
 // NotRunnable() -- informed spinning
 //
 // Don't bother spinning if the owner is not eligible to drop the lock.
 // The caller must tolerate false-negative and false-positive errors.
 // Spinning, in general, is probabilistic anyway.
 int ObjectMonitor::NotRunnable (Thread * Self, Thread * ox) {
 // Check either OwnerIsThread or ox->TypeTag == 2BAD.
 if (!OwnerIsThread) return 0;
 if (ox == NULL) return 0;
 // Avoid transitive spinning ...
 // Say T1 spins or blocks trying to acquire L.  T1._Stalled is set to L.
 // Immediately after T1 acquires L it's possible that T2, also
 // spinning on L, will see L.Owner=T1 and T1._Stalled=L.
 // This occurs transiently after T1 acquired L but before
 // T1 managed to clear T1.Stalled.  T2 does not need to abort
 // its spin in this circumstance.
 intptr_t BlockedOn = SafeFetchN((intptr_t *) &ox->_Stalled, intptr_t(1));
 if (BlockedOn == 1) return 1;
 if (BlockedOn != 0) {
 return BlockedOn != intptr_t(this) && _owner == ox;
 }
 assert(sizeof(((JavaThread *)ox)->_thread_state == sizeof(int)), "invariant");
 int jst = SafeFetch32((int *) &((JavaThread *) ox)->_thread_state, -1);;
 // consider also: jst != _thread_in_Java -- but that's overspecific.
 return jst == _thread_blocked || jst == _thread_in_native;
 }
 // -----------------------------------------------------------------------------
 // WaitSet management ...
 void ObjectMonitor::Initialize() {
 static int InitializationCompleted = 0;
 assert(InitializationCompleted == 0, "invariant");
 InitializationCompleted = 1;
 if (UsePerfData) {
 EXCEPTION_MARK;
 #define NEWPERFCOUNTER(n)   {n = PerfDataManager::create_counter(SUN_RT, #n, PerfData::U_Events,CHECK); }
 #define NEWPERFVARIABLE(n)  {n = PerfDataManager::create_variable(SUN_RT, #n, PerfData::U_Events,CHECK); }
 NEWPERFCOUNTER(_sync_Inflations);
 NEWPERFCOUNTER(_sync_Deflations);
 NEWPERFCOUNTER(_sync_ContendedLockAttempts);
 NEWPERFCOUNTER(_sync_FutileWakeups);
 NEWPERFCOUNTER(_sync_Parks);
 NEWPERFCOUNTER(_sync_EmptyNotifications);
 NEWPERFCOUNTER(_sync_Notifications);
 NEWPERFCOUNTER(_sync_SlowEnter);
 NEWPERFCOUNTER(_sync_SlowExit);
 NEWPERFCOUNTER(_sync_SlowNotify);
 NEWPERFCOUNTER(_sync_SlowNotifyAll);
 NEWPERFCOUNTER(_sync_FailedSpins);
 NEWPERFCOUNTER(_sync_SuccessfulSpins);
 NEWPERFCOUNTER(_sync_PrivateA);
 NEWPERFCOUNTER(_sync_PrivateB);
 NEWPERFCOUNTER(_sync_MonInCirculation);
 NEWPERFCOUNTER(_sync_MonScavenged);
 NEWPERFVARIABLE(_sync_MonExtant);
 #undef NEWPERFCOUNTER
 }
 }
 CTASSERT(offset_of (ObjectMonitor, _header) == 0);
 }
 static char * kvGet (char * kvList, const char * Key) {
 if (kvList == NULL) return NULL;
 size_t n = strlen(Key);
 char * Search;
 for (Search = kvList; *Search; Search += strlen(Search) + 1) {
 if (strncmp (Search, Key, n) == 0) {
 if (Search[n] == '=') return Search + n + 1;
 if (Search[n] == 0)   return(char *) "1";
 }
 }
 return NULL;
 }
 static int kvGetInt (char * kvList, const char * Key, int Default) {
 char * v = kvGet(kvList, Key);
 int rslt = v ? ::strtol(v, NULL, 0) : Default;
 if (Knob_ReportSettings && v != NULL) {
 ::printf ("  SyncKnob: %s %d(%d)\n", Key, rslt, Default) ;
 ::fflush(stdout);
 }
 return rslt;
 }
 void ObjectMonitor::DeferredInitialize() {
 if (InitDone > 0) return;
 if (Atomic::cmpxchg (-1, &InitDone, 0) != 0) {
 while (InitDone != 1);
 return;
 }
 // One-shot global initialization ...
 // The initialization is idempotent, so we don't need locks.
 // In the future consider doing this via os::init_2().
 if (SyncKnobs == NULL) SyncKnobs = "";
 size_t sz = strlen(SyncKnobs);
 char * knobs = (char *) malloc(sz + 2);
 if (knobs == NULL) {
 vm_exit_out_of_memory(sz + 2, OOM_MALLOC_ERROR, "Parse SyncKnobs");
 guarantee(0, "invariant");
 }
 strcpy(knobs, SyncKnobs);
 knobs[sz+1] = 0;
 for (char * p = knobs; *p; p++) {
 if (*p == ':') *p = 0;
 }
 #define SETKNOB(x) { Knob_##x = kvGetInt (knobs, #x, Knob_##x); }
 SETKNOB(ReportSettings);
 SETKNOB(Verbose);
 if (Knob_Verbose) {
 sanity_checks();
 }
 if (os::is_MP()) {
 BackOffMask = (1 << Knob_SpinBackOff) - 1;
 if (Knob_ReportSettings) ::printf("BackOffMask=%X\n", BackOffMask);
 // CONSIDER: BackOffMask = ROUNDUP_NEXT_POWER2 (ncpus-1)
 } else {
 Knob_SpinLimit = 0;
 Knob_SpinBase  = 0;
 Knob_PreSpin   = 0;
 Knob_FixedSpin = -1;
 }
 if (Knob_LogSpins == 0) {
 ObjectMonitor::_sync_FailedSpins = NULL;
 }
 free(knobs);
 OrderAccess::fence();
 InitDone = 1;

changeset 26683	a02753d5a0b2
parent 25633	4cd9c4622c8c
child 26684	d1221849ea3d