aoqi@0: aoqi@0: /* aoqi@0: * Copyright (c) 1998, 2014, Oracle and/or its affiliates. All rights reserved. aoqi@0: * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. aoqi@0: * aoqi@0: * This code is free software; you can redistribute it and/or modify it aoqi@0: * under the terms of the GNU General Public License version 2 only, as aoqi@0: * published by the Free Software Foundation. aoqi@0: * aoqi@0: * This code is distributed in the hope that it will be useful, but WITHOUT aoqi@0: * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or aoqi@0: * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License aoqi@0: * version 2 for more details (a copy is included in the LICENSE file that aoqi@0: * accompanied this code). aoqi@0: * aoqi@0: * You should have received a copy of the GNU General Public License version aoqi@0: * 2 along with this work; if not, write to the Free Software Foundation, aoqi@0: * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. aoqi@0: * aoqi@0: * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA aoqi@0: * or visit www.oracle.com if you need additional information or have any aoqi@0: * questions. aoqi@0: * aoqi@0: */ aoqi@0: aoqi@0: #include "precompiled.hpp" aoqi@0: #include "runtime/mutex.hpp" goetz@6911: #include "runtime/orderAccess.inline.hpp" aoqi@0: #include "runtime/osThread.hpp" aoqi@0: #include "runtime/thread.inline.hpp" aoqi@0: #include "utilities/events.hpp" aoqi@0: #ifdef TARGET_OS_FAMILY_linux aoqi@0: # include "mutex_linux.inline.hpp" aoqi@0: #endif aoqi@0: #ifdef TARGET_OS_FAMILY_solaris aoqi@0: # include "mutex_solaris.inline.hpp" aoqi@0: #endif aoqi@0: #ifdef TARGET_OS_FAMILY_windows aoqi@0: # include "mutex_windows.inline.hpp" aoqi@0: #endif aoqi@0: #ifdef TARGET_OS_FAMILY_bsd aoqi@0: # include "mutex_bsd.inline.hpp" aoqi@0: #endif aoqi@0: aoqi@0: PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC aoqi@0: aoqi@0: // o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o aoqi@0: // aoqi@0: // Native Monitor-Mutex locking - theory of operations aoqi@0: // aoqi@0: // * Native Monitors are completely unrelated to Java-level monitors, aoqi@0: // although the "back-end" slow-path implementations share a common lineage. aoqi@0: // See objectMonitor:: in synchronizer.cpp. aoqi@0: // Native Monitors do *not* support nesting or recursion but otherwise aoqi@0: // they're basically Hoare-flavor monitors. aoqi@0: // aoqi@0: // * A thread acquires ownership of a Monitor/Mutex by CASing the LockByte aoqi@0: // in the _LockWord from zero to non-zero. Note that the _Owner field aoqi@0: // is advisory and is used only to verify that the thread calling unlock() aoqi@0: // is indeed the last thread to have acquired the lock. aoqi@0: // aoqi@0: // * Contending threads "push" themselves onto the front of the contention aoqi@0: // queue -- called the cxq -- with CAS and then spin/park. aoqi@0: // The _LockWord contains the LockByte as well as the pointer to the head aoqi@0: // of the cxq. Colocating the LockByte with the cxq precludes certain races. aoqi@0: // aoqi@0: // * Using a separately addressable LockByte allows for CAS:MEMBAR or CAS:0 aoqi@0: // idioms. We currently use MEMBAR in the uncontended unlock() path, as aoqi@0: // MEMBAR often has less latency than CAS. If warranted, we could switch to aoqi@0: // a CAS:0 mode, using timers to close the resultant race, as is done aoqi@0: // with Java Monitors in synchronizer.cpp. aoqi@0: // aoqi@0: // See the following for a discussion of the relative cost of atomics (CAS) aoqi@0: // MEMBAR, and ways to eliminate such instructions from the common-case paths: aoqi@0: // -- http://blogs.sun.com/dave/entry/biased_locking_in_hotspot aoqi@0: // -- http://blogs.sun.com/dave/resource/MustangSync.pdf aoqi@0: // -- http://blogs.sun.com/dave/resource/synchronization-public2.pdf aoqi@0: // -- synchronizer.cpp aoqi@0: // aoqi@0: // * Overall goals - desiderata aoqi@0: // 1. Minimize context switching aoqi@0: // 2. Minimize lock migration aoqi@0: // 3. Minimize CPI -- affinity and locality aoqi@0: // 4. Minimize the execution of high-latency instructions such as CAS or MEMBAR aoqi@0: // 5. Minimize outer lock hold times aoqi@0: // 6. Behave gracefully on a loaded system aoqi@0: // aoqi@0: // * Thread flow and list residency: aoqi@0: // aoqi@0: // Contention queue --> EntryList --> OnDeck --> Owner --> !Owner aoqi@0: // [..resident on monitor list..] aoqi@0: // [...........contending..................] aoqi@0: // aoqi@0: // -- The contention queue (cxq) contains recently-arrived threads (RATs). aoqi@0: // Threads on the cxq eventually drain into the EntryList. aoqi@0: // -- Invariant: a thread appears on at most one list -- cxq, EntryList aoqi@0: // or WaitSet -- at any one time. aoqi@0: // -- For a given monitor there can be at most one "OnDeck" thread at any aoqi@0: // given time but if needbe this particular invariant could be relaxed. aoqi@0: // aoqi@0: // * The WaitSet and EntryList linked lists are composed of ParkEvents. aoqi@0: // I use ParkEvent instead of threads as ParkEvents are immortal and aoqi@0: // type-stable, meaning we can safely unpark() a possibly stale aoqi@0: // list element in the unlock()-path. (That's benign). aoqi@0: // aoqi@0: // * Succession policy - providing for progress: aoqi@0: // aoqi@0: // As necessary, the unlock()ing thread identifies, unlinks, and unparks aoqi@0: // an "heir presumptive" tentative successor thread from the EntryList. aoqi@0: // This becomes the so-called "OnDeck" thread, of which there can be only aoqi@0: // one at any given time for a given monitor. The wakee will recontend aoqi@0: // for ownership of monitor. aoqi@0: // aoqi@0: // Succession is provided for by a policy of competitive handoff. aoqi@0: // The exiting thread does _not_ grant or pass ownership to the aoqi@0: // successor thread. (This is also referred to as "handoff" succession"). aoqi@0: // Instead the exiting thread releases ownership and possibly wakes aoqi@0: // a successor, so the successor can (re)compete for ownership of the lock. aoqi@0: // aoqi@0: // Competitive handoff provides excellent overall throughput at the expense aoqi@0: // of short-term fairness. If fairness is a concern then one remedy might aoqi@0: // be to add an AcquireCounter field to the monitor. After a thread acquires aoqi@0: // the lock it will decrement the AcquireCounter field. When the count aoqi@0: // reaches 0 the thread would reset the AcquireCounter variable, abdicate aoqi@0: // the lock directly to some thread on the EntryList, and then move itself to the aoqi@0: // tail of the EntryList. aoqi@0: // aoqi@0: // But in practice most threads engage or otherwise participate in resource aoqi@0: // bounded producer-consumer relationships, so lock domination is not usually aoqi@0: // a practical concern. Recall too, that in general it's easier to construct aoqi@0: // a fair lock from a fast lock, but not vice-versa. aoqi@0: // aoqi@0: // * The cxq can have multiple concurrent "pushers" but only one concurrent aoqi@0: // detaching thread. This mechanism is immune from the ABA corruption. aoqi@0: // More precisely, the CAS-based "push" onto cxq is ABA-oblivious. aoqi@0: // We use OnDeck as a pseudo-lock to enforce the at-most-one detaching aoqi@0: // thread constraint. aoqi@0: // aoqi@0: // * Taken together, the cxq and the EntryList constitute or form a aoqi@0: // single logical queue of threads stalled trying to acquire the lock. aoqi@0: // We use two distinct lists to reduce heat on the list ends. aoqi@0: // Threads in lock() enqueue onto cxq while threads in unlock() will aoqi@0: // dequeue from the EntryList. (c.f. Michael Scott's "2Q" algorithm). aoqi@0: // A key desideratum is to minimize queue & monitor metadata manipulation aoqi@0: // that occurs while holding the "outer" monitor lock -- that is, we want to aoqi@0: // minimize monitor lock holds times. aoqi@0: // aoqi@0: // The EntryList is ordered by the prevailing queue discipline and aoqi@0: // can be organized in any convenient fashion, such as a doubly-linked list or aoqi@0: // a circular doubly-linked list. If we need a priority queue then something akin aoqi@0: // to Solaris' sleepq would work nicely. Viz., aoqi@0: // -- http://agg.eng/ws/on10_nightly/source/usr/src/uts/common/os/sleepq.c. aoqi@0: // -- http://cvs.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/uts/common/os/sleepq.c aoqi@0: // Queue discipline is enforced at ::unlock() time, when the unlocking thread aoqi@0: // drains the cxq into the EntryList, and orders or reorders the threads on the aoqi@0: // EntryList accordingly. aoqi@0: // aoqi@0: // Barring "lock barging", this mechanism provides fair cyclic ordering, aoqi@0: // somewhat similar to an elevator-scan. aoqi@0: // aoqi@0: // * OnDeck aoqi@0: // -- For a given monitor there can be at most one OnDeck thread at any given aoqi@0: // instant. The OnDeck thread is contending for the lock, but has been aoqi@0: // unlinked from the EntryList and cxq by some previous unlock() operations. aoqi@0: // Once a thread has been designated the OnDeck thread it will remain so aoqi@0: // until it manages to acquire the lock -- being OnDeck is a stable property. aoqi@0: // -- Threads on the EntryList or cxq are _not allowed to attempt lock acquisition. aoqi@0: // -- OnDeck also serves as an "inner lock" as follows. Threads in unlock() will, after aoqi@0: // having cleared the LockByte and dropped the outer lock, attempt to "trylock" aoqi@0: // OnDeck by CASing the field from null to non-null. If successful, that thread aoqi@0: // is then responsible for progress and succession and can use CAS to detach and aoqi@0: // drain the cxq into the EntryList. By convention, only this thread, the holder of aoqi@0: // the OnDeck inner lock, can manipulate the EntryList or detach and drain the aoqi@0: // RATs on the cxq into the EntryList. This avoids ABA corruption on the cxq as aoqi@0: // we allow multiple concurrent "push" operations but restrict detach concurrency aoqi@0: // to at most one thread. Having selected and detached a successor, the thread then aoqi@0: // changes the OnDeck to refer to that successor, and then unparks the successor. aoqi@0: // That successor will eventually acquire the lock and clear OnDeck. Beware aoqi@0: // that the OnDeck usage as a lock is asymmetric. A thread in unlock() transiently aoqi@0: // "acquires" OnDeck, performs queue manipulations, passes OnDeck to some successor, aoqi@0: // and then the successor eventually "drops" OnDeck. Note that there's never aoqi@0: // any sense of contention on the inner lock, however. Threads never contend aoqi@0: // or wait for the inner lock. aoqi@0: // -- OnDeck provides for futile wakeup throttling a described in section 3.3 of aoqi@0: // See http://www.usenix.org/events/jvm01/full_papers/dice/dice.pdf aoqi@0: // In a sense, OnDeck subsumes the ObjectMonitor _Succ and ObjectWaiter aoqi@0: // TState fields found in Java-level objectMonitors. (See synchronizer.cpp). aoqi@0: // aoqi@0: // * Waiting threads reside on the WaitSet list -- wait() puts aoqi@0: // the caller onto the WaitSet. Notify() or notifyAll() simply aoqi@0: // transfers threads from the WaitSet to either the EntryList or cxq. aoqi@0: // Subsequent unlock() operations will eventually unpark the notifyee. aoqi@0: // Unparking a notifee in notify() proper is inefficient - if we were to do so aoqi@0: // it's likely the notifyee would simply impale itself on the lock held aoqi@0: // by the notifier. aoqi@0: // aoqi@0: // * The mechanism is obstruction-free in that if the holder of the transient aoqi@0: // OnDeck lock in unlock() is preempted or otherwise stalls, other threads aoqi@0: // can still acquire and release the outer lock and continue to make progress. aoqi@0: // At worst, waking of already blocked contending threads may be delayed, aoqi@0: // but nothing worse. (We only use "trylock" operations on the inner OnDeck aoqi@0: // lock). aoqi@0: // aoqi@0: // * Note that thread-local storage must be initialized before a thread aoqi@0: // uses Native monitors or mutexes. The native monitor-mutex subsystem aoqi@0: // depends on Thread::current(). aoqi@0: // aoqi@0: // * The monitor synchronization subsystem avoids the use of native aoqi@0: // synchronization primitives except for the narrow platform-specific aoqi@0: // park-unpark abstraction. See the comments in os_solaris.cpp regarding aoqi@0: // the semantics of park-unpark. Put another way, this monitor implementation aoqi@0: // depends only on atomic operations and park-unpark. The monitor subsystem aoqi@0: // manages all RUNNING->BLOCKED and BLOCKED->READY transitions while the aoqi@0: // underlying OS manages the READY<->RUN transitions. aoqi@0: // aoqi@0: // * The memory consistency model provide by lock()-unlock() is at least as aoqi@0: // strong or stronger than the Java Memory model defined by JSR-133. aoqi@0: // That is, we guarantee at least entry consistency, if not stronger. aoqi@0: // See http://g.oswego.edu/dl/jmm/cookbook.html. aoqi@0: // aoqi@0: // * Thread:: currently contains a set of purpose-specific ParkEvents: aoqi@0: // _MutexEvent, _ParkEvent, etc. A better approach might be to do away with aoqi@0: // the purpose-specific ParkEvents and instead implement a general per-thread aoqi@0: // stack of available ParkEvents which we could provision on-demand. The aoqi@0: // stack acts as a local cache to avoid excessive calls to ParkEvent::Allocate() aoqi@0: // and ::Release(). A thread would simply pop an element from the local stack before it aoqi@0: // enqueued or park()ed. When the contention was over the thread would aoqi@0: // push the no-longer-needed ParkEvent back onto its stack. aoqi@0: // aoqi@0: // * A slightly reduced form of ILock() and IUnlock() have been partially aoqi@0: // model-checked (Murphi) for safety and progress at T=1,2,3 and 4. aoqi@0: // It'd be interesting to see if TLA/TLC could be useful as well. aoqi@0: // aoqi@0: // * Mutex-Monitor is a low-level "leaf" subsystem. That is, the monitor aoqi@0: // code should never call other code in the JVM that might itself need to aoqi@0: // acquire monitors or mutexes. That's true *except* in the case of the aoqi@0: // ThreadBlockInVM state transition wrappers. The ThreadBlockInVM DTOR handles aoqi@0: // mutator reentry (ingress) by checking for a pending safepoint in which case it will aoqi@0: // call SafepointSynchronize::block(), which in turn may call Safepoint_lock->lock(), etc. aoqi@0: // In that particular case a call to lock() for a given Monitor can end up recursively aoqi@0: // calling lock() on another monitor. While distasteful, this is largely benign aoqi@0: // as the calls come from jacket that wraps lock(), and not from deep within lock() itself. aoqi@0: // aoqi@0: // It's unfortunate that native mutexes and thread state transitions were convolved. aoqi@0: // They're really separate concerns and should have remained that way. Melding aoqi@0: // them together was facile -- a bit too facile. The current implementation badly aoqi@0: // conflates the two concerns. aoqi@0: // aoqi@0: // * TODO-FIXME: aoqi@0: // aoqi@0: // -- Add DTRACE probes for contended acquire, contended acquired, contended unlock aoqi@0: // We should also add DTRACE probes in the ParkEvent subsystem for aoqi@0: // Park-entry, Park-exit, and Unpark. aoqi@0: // aoqi@0: // -- We have an excess of mutex-like constructs in the JVM, namely: aoqi@0: // 1. objectMonitors for Java-level synchronization (synchronizer.cpp) aoqi@0: // 2. low-level muxAcquire and muxRelease aoqi@0: // 3. low-level spinAcquire and spinRelease aoqi@0: // 4. native Mutex:: and Monitor:: aoqi@0: // 5. jvm_raw_lock() and _unlock() aoqi@0: // 6. JVMTI raw monitors -- distinct from (5) despite having a confusingly aoqi@0: // similar name. aoqi@0: // aoqi@0: // o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o-o aoqi@0: aoqi@0: aoqi@0: // CASPTR() uses the canonical argument order that dominates in the literature. aoqi@0: // Our internal cmpxchg_ptr() uses a bastardized ordering to accommodate Sun .il templates. aoqi@0: aoqi@0: #define CASPTR(a,c,s) intptr_t(Atomic::cmpxchg_ptr ((void *)(s),(void *)(a),(void *)(c))) aoqi@0: #define UNS(x) (uintptr_t(x)) aoqi@0: #define TRACE(m) { static volatile int ctr = 0 ; int x = ++ctr ; if ((x & (x-1))==0) { ::printf ("%d:%s\n", x, #m); ::fflush(stdout); }} aoqi@0: aoqi@0: // Simplistic low-quality Marsaglia SHIFT-XOR RNG. aoqi@0: // Bijective except for the trailing mask operation. aoqi@0: // Useful for spin loops as the compiler can't optimize it away. aoqi@0: aoqi@0: static inline jint MarsagliaXORV (jint x) { aoqi@0: if (x == 0) x = 1|os::random() ; aoqi@0: x ^= x << 6; aoqi@0: x ^= ((unsigned)x) >> 21; aoqi@0: x ^= x << 7 ; aoqi@0: return x & 0x7FFFFFFF ; aoqi@0: } aoqi@0: aoqi@0: static inline jint MarsagliaXOR (jint * const a) { aoqi@0: jint x = *a ; aoqi@0: if (x == 0) x = UNS(a)|1 ; aoqi@0: x ^= x << 6; aoqi@0: x ^= ((unsigned)x) >> 21; aoqi@0: x ^= x << 7 ; aoqi@0: *a = x ; aoqi@0: return x & 0x7FFFFFFF ; aoqi@0: } aoqi@0: aoqi@0: static int Stall (int its) { aoqi@0: static volatile jint rv = 1 ; aoqi@0: volatile int OnFrame = 0 ; aoqi@0: jint v = rv ^ UNS(OnFrame) ; aoqi@0: while (--its >= 0) { aoqi@0: v = MarsagliaXORV (v) ; aoqi@0: } aoqi@0: // Make this impossible for the compiler to optimize away, aoqi@0: // but (mostly) avoid W coherency sharing on MP systems. aoqi@0: if (v == 0x12345) rv = v ; aoqi@0: return v ; aoqi@0: } aoqi@0: aoqi@0: int Monitor::TryLock () { aoqi@0: intptr_t v = _LockWord.FullWord ; aoqi@0: for (;;) { aoqi@0: if ((v & _LBIT) != 0) return 0 ; aoqi@0: const intptr_t u = CASPTR (&_LockWord, v, v|_LBIT) ; aoqi@0: if (v == u) return 1 ; aoqi@0: v = u ; aoqi@0: } aoqi@0: } aoqi@0: aoqi@0: int Monitor::TryFast () { aoqi@0: // Optimistic fast-path form ... aoqi@0: // Fast-path attempt for the common uncontended case. aoqi@0: // Avoid RTS->RTO $ coherence upgrade on typical SMP systems. aoqi@0: intptr_t v = CASPTR (&_LockWord, 0, _LBIT) ; // agro ... aoqi@0: if (v == 0) return 1 ; aoqi@0: aoqi@0: for (;;) { aoqi@0: if ((v & _LBIT) != 0) return 0 ; aoqi@0: const intptr_t u = CASPTR (&_LockWord, v, v|_LBIT) ; aoqi@0: if (v == u) return 1 ; aoqi@0: v = u ; aoqi@0: } aoqi@0: } aoqi@0: aoqi@0: int Monitor::ILocked () { aoqi@0: const intptr_t w = _LockWord.FullWord & 0xFF ; aoqi@0: assert (w == 0 || w == _LBIT, "invariant") ; aoqi@0: return w == _LBIT ; aoqi@0: } aoqi@0: aoqi@0: // Polite TATAS spinlock with exponential backoff - bounded spin. aoqi@0: // Ideally we'd use processor cycles, time or vtime to control aoqi@0: // the loop, but we currently use iterations. aoqi@0: // All the constants within were derived empirically but work over aoqi@0: // over the spectrum of J2SE reference platforms. aoqi@0: // On Niagara-class systems the back-off is unnecessary but aoqi@0: // is relatively harmless. (At worst it'll slightly retard aoqi@0: // acquisition times). The back-off is critical for older SMP systems aoqi@0: // where constant fetching of the LockWord would otherwise impair aoqi@0: // scalability. aoqi@0: // aoqi@0: // Clamp spinning at approximately 1/2 of a context-switch round-trip. aoqi@0: // See synchronizer.cpp for details and rationale. aoqi@0: aoqi@0: int Monitor::TrySpin (Thread * const Self) { aoqi@0: if (TryLock()) return 1 ; aoqi@0: if (!os::is_MP()) return 0 ; aoqi@0: aoqi@0: int Probes = 0 ; aoqi@0: int Delay = 0 ; aoqi@0: int Steps = 0 ; aoqi@0: int SpinMax = NativeMonitorSpinLimit ; aoqi@0: int flgs = NativeMonitorFlags ; aoqi@0: for (;;) { aoqi@0: intptr_t v = _LockWord.FullWord; aoqi@0: if ((v & _LBIT) == 0) { aoqi@0: if (CASPTR (&_LockWord, v, v|_LBIT) == v) { aoqi@0: return 1 ; aoqi@0: } aoqi@0: continue ; aoqi@0: } aoqi@0: aoqi@0: if ((flgs & 8) == 0) { aoqi@0: SpinPause () ; aoqi@0: } aoqi@0: aoqi@0: // Periodically increase Delay -- variable Delay form aoqi@0: // conceptually: delay *= 1 + 1/Exponent aoqi@0: ++ Probes; aoqi@0: if (Probes > SpinMax) return 0 ; aoqi@0: aoqi@0: if ((Probes & 0x7) == 0) { aoqi@0: Delay = ((Delay << 1)|1) & 0x7FF ; aoqi@0: // CONSIDER: Delay += 1 + (Delay/4); Delay &= 0x7FF ; aoqi@0: } aoqi@0: aoqi@0: if (flgs & 2) continue ; aoqi@0: aoqi@0: // Consider checking _owner's schedctl state, if OFFPROC abort spin. aoqi@0: // If the owner is OFFPROC then it's unlike that the lock will be dropped aoqi@0: // in a timely fashion, which suggests that spinning would not be fruitful aoqi@0: // or profitable. aoqi@0: aoqi@0: // Stall for "Delay" time units - iterations in the current implementation. aoqi@0: // Avoid generating coherency traffic while stalled. aoqi@0: // Possible ways to delay: aoqi@0: // PAUSE, SLEEP, MEMBAR #sync, MEMBAR #halt, aoqi@0: // wr %g0,%asi, gethrtime, rdstick, rdtick, rdtsc, etc. ... aoqi@0: // Note that on Niagara-class systems we want to minimize STs in the aoqi@0: // spin loop. N1 and brethren write-around the L1$ over the xbar into the L2$. aoqi@0: // Furthermore, they don't have a W$ like traditional SPARC processors. aoqi@0: // We currently use a Marsaglia Shift-Xor RNG loop. aoqi@0: Steps += Delay ; aoqi@0: if (Self != NULL) { aoqi@0: jint rv = Self->rng[0] ; aoqi@0: for (int k = Delay ; --k >= 0; ) { aoqi@0: rv = MarsagliaXORV (rv) ; aoqi@0: if ((flgs & 4) == 0 && SafepointSynchronize::do_call_back()) return 0 ; aoqi@0: } aoqi@0: Self->rng[0] = rv ; aoqi@0: } else { aoqi@0: Stall (Delay) ; aoqi@0: } aoqi@0: } aoqi@0: } aoqi@0: aoqi@0: static int ParkCommon (ParkEvent * ev, jlong timo) { aoqi@0: // Diagnostic support - periodically unwedge blocked threads aoqi@0: intx nmt = NativeMonitorTimeout ; aoqi@0: if (nmt > 0 && (nmt < timo || timo <= 0)) { aoqi@0: timo = nmt ; aoqi@0: } aoqi@0: int err = OS_OK ; aoqi@0: if (0 == timo) { aoqi@0: ev->park() ; aoqi@0: } else { aoqi@0: err = ev->park(timo) ; aoqi@0: } aoqi@0: return err ; aoqi@0: } aoqi@0: aoqi@0: inline int Monitor::AcquireOrPush (ParkEvent * ESelf) { aoqi@0: intptr_t v = _LockWord.FullWord ; aoqi@0: for (;;) { aoqi@0: if ((v & _LBIT) == 0) { aoqi@0: const intptr_t u = CASPTR (&_LockWord, v, v|_LBIT) ; aoqi@0: if (u == v) return 1 ; // indicate acquired aoqi@0: v = u ; aoqi@0: } else { aoqi@0: // Anticipate success ... aoqi@0: ESelf->ListNext = (ParkEvent *) (v & ~_LBIT) ; aoqi@0: const intptr_t u = CASPTR (&_LockWord, v, intptr_t(ESelf)|_LBIT) ; aoqi@0: if (u == v) return 0 ; // indicate pushed onto cxq aoqi@0: v = u ; aoqi@0: } aoqi@0: // Interference - LockWord change - just retry aoqi@0: } aoqi@0: } aoqi@0: aoqi@0: // ILock and IWait are the lowest level primitive internal blocking aoqi@0: // synchronization functions. The callers of IWait and ILock must have aoqi@0: // performed any needed state transitions beforehand. aoqi@0: // IWait and ILock may directly call park() without any concern for thread state. aoqi@0: // Note that ILock and IWait do *not* access _owner. aoqi@0: // _owner is a higher-level logical concept. aoqi@0: aoqi@0: void Monitor::ILock (Thread * Self) { aoqi@0: assert (_OnDeck != Self->_MutexEvent, "invariant") ; aoqi@0: aoqi@0: if (TryFast()) { aoqi@0: Exeunt: aoqi@0: assert (ILocked(), "invariant") ; aoqi@0: return ; aoqi@0: } aoqi@0: aoqi@0: ParkEvent * const ESelf = Self->_MutexEvent ; aoqi@0: assert (_OnDeck != ESelf, "invariant") ; aoqi@0: aoqi@0: // As an optimization, spinners could conditionally try to set ONDECK to _LBIT aoqi@0: // Synchronizer.cpp uses a similar optimization. aoqi@0: if (TrySpin (Self)) goto Exeunt ; aoqi@0: aoqi@0: // Slow-path - the lock is contended. aoqi@0: // Either Enqueue Self on cxq or acquire the outer lock. aoqi@0: // LockWord encoding = (cxq,LOCKBYTE) aoqi@0: ESelf->reset() ; aoqi@0: OrderAccess::fence() ; aoqi@0: aoqi@0: // Optional optimization ... try barging on the inner lock aoqi@0: if ((NativeMonitorFlags & 32) && CASPTR (&_OnDeck, NULL, UNS(Self)) == 0) { aoqi@0: goto OnDeck_LOOP ; aoqi@0: } aoqi@0: aoqi@0: if (AcquireOrPush (ESelf)) goto Exeunt ; aoqi@0: aoqi@0: // At any given time there is at most one ondeck thread. aoqi@0: // ondeck implies not resident on cxq and not resident on EntryList aoqi@0: // Only the OnDeck thread can try to acquire -- contended for -- the lock. aoqi@0: // CONSIDER: use Self->OnDeck instead of m->OnDeck. aoqi@0: // Deschedule Self so that others may run. aoqi@0: while (_OnDeck != ESelf) { aoqi@0: ParkCommon (ESelf, 0) ; aoqi@0: } aoqi@0: aoqi@0: // Self is now in the ONDECK position and will remain so until it aoqi@0: // manages to acquire the lock. aoqi@0: OnDeck_LOOP: aoqi@0: for (;;) { aoqi@0: assert (_OnDeck == ESelf, "invariant") ; aoqi@0: if (TrySpin (Self)) break ; aoqi@0: // CONSIDER: if ESelf->TryPark() && TryLock() break ... aoqi@0: // It's probably wise to spin only if we *actually* blocked aoqi@0: // CONSIDER: check the lockbyte, if it remains set then aoqi@0: // preemptively drain the cxq into the EntryList. aoqi@0: // The best place and time to perform queue operations -- lock metadata -- aoqi@0: // is _before having acquired the outer lock, while waiting for the lock to drop. aoqi@0: ParkCommon (ESelf, 0) ; aoqi@0: } aoqi@0: aoqi@0: assert (_OnDeck == ESelf, "invariant") ; aoqi@0: _OnDeck = NULL ; aoqi@0: aoqi@0: // Note that we current drop the inner lock (clear OnDeck) in the slow-path aoqi@0: // epilog immediately after having acquired the outer lock. aoqi@0: // But instead we could consider the following optimizations: aoqi@0: // A. Shift or defer dropping the inner lock until the subsequent IUnlock() operation. aoqi@0: // This might avoid potential reacquisition of the inner lock in IUlock(). aoqi@0: // B. While still holding the inner lock, attempt to opportunistically select aoqi@0: // and unlink the next ONDECK thread from the EntryList. aoqi@0: // If successful, set ONDECK to refer to that thread, otherwise clear ONDECK. aoqi@0: // It's critical that the select-and-unlink operation run in constant-time as aoqi@0: // it executes when holding the outer lock and may artificially increase the aoqi@0: // effective length of the critical section. aoqi@0: // Note that (A) and (B) are tantamount to succession by direct handoff for aoqi@0: // the inner lock. aoqi@0: goto Exeunt ; aoqi@0: } aoqi@0: aoqi@0: void Monitor::IUnlock (bool RelaxAssert) { aoqi@0: assert (ILocked(), "invariant") ; aoqi@0: // Conceptually we need a MEMBAR #storestore|#loadstore barrier or fence immediately aoqi@0: // before the store that releases the lock. Crucially, all the stores and loads in the aoqi@0: // critical section must be globally visible before the store of 0 into the lock-word aoqi@0: // that releases the lock becomes globally visible. That is, memory accesses in the aoqi@0: // critical section should not be allowed to bypass or overtake the following ST that aoqi@0: // releases the lock. As such, to prevent accesses within the critical section aoqi@0: // from "leaking" out, we need a release fence between the critical section and the aoqi@0: // store that releases the lock. In practice that release barrier is elided on aoqi@0: // platforms with strong memory models such as TSO. aoqi@0: // aoqi@0: // Note that the OrderAccess::storeload() fence that appears after unlock store aoqi@0: // provides for progress conditions and succession and is _not related to exclusion aoqi@0: // safety or lock release consistency. aoqi@0: OrderAccess::release_store(&_LockWord.Bytes[_LSBINDEX], 0); // drop outer lock aoqi@0: aoqi@0: OrderAccess::storeload (); aoqi@0: ParkEvent * const w = _OnDeck ; aoqi@0: assert (RelaxAssert || w != Thread::current()->_MutexEvent, "invariant") ; aoqi@0: if (w != NULL) { aoqi@0: // Either we have a valid ondeck thread or ondeck is transiently "locked" aoqi@0: // by some exiting thread as it arranges for succession. The LSBit of aoqi@0: // OnDeck allows us to discriminate two cases. If the latter, the aoqi@0: // responsibility for progress and succession lies with that other thread. aoqi@0: // For good performance, we also depend on the fact that redundant unpark() aoqi@0: // operations are cheap. That is, repeated Unpark()ing of the ONDECK thread aoqi@0: // is inexpensive. This approach provides implicit futile wakeup throttling. aoqi@0: // Note that the referent "w" might be stale with respect to the lock. aoqi@0: // In that case the following unpark() is harmless and the worst that'll happen aoqi@0: // is a spurious return from a park() operation. Critically, if "w" _is stale, aoqi@0: // then progress is known to have occurred as that means the thread associated aoqi@0: // with "w" acquired the lock. In that case this thread need take no further aoqi@0: // action to guarantee progress. aoqi@0: if ((UNS(w) & _LBIT) == 0) w->unpark() ; aoqi@0: return ; aoqi@0: } aoqi@0: aoqi@0: intptr_t cxq = _LockWord.FullWord ; aoqi@0: if (((cxq & ~_LBIT)|UNS(_EntryList)) == 0) { aoqi@0: return ; // normal fast-path exit - cxq and EntryList both empty aoqi@0: } aoqi@0: if (cxq & _LBIT) { aoqi@0: // Optional optimization ... aoqi@0: // Some other thread acquired the lock in the window since this aoqi@0: // thread released it. Succession is now that thread's responsibility. aoqi@0: return ; aoqi@0: } aoqi@0: aoqi@0: Succession: aoqi@0: // Slow-path exit - this thread must ensure succession and progress. aoqi@0: // OnDeck serves as lock to protect cxq and EntryList. aoqi@0: // Only the holder of OnDeck can manipulate EntryList or detach the RATs from cxq. aoqi@0: // Avoid ABA - allow multiple concurrent producers (enqueue via push-CAS) aoqi@0: // but only one concurrent consumer (detacher of RATs). aoqi@0: // Consider protecting this critical section with schedctl on Solaris. aoqi@0: // Unlike a normal lock, however, the exiting thread "locks" OnDeck, aoqi@0: // picks a successor and marks that thread as OnDeck. That successor aoqi@0: // thread will then clear OnDeck once it eventually acquires the outer lock. aoqi@0: if (CASPTR (&_OnDeck, NULL, _LBIT) != UNS(NULL)) { aoqi@0: return ; aoqi@0: } aoqi@0: aoqi@0: ParkEvent * List = _EntryList ; aoqi@0: if (List != NULL) { aoqi@0: // Transfer the head of the EntryList to the OnDeck position. aoqi@0: // Once OnDeck, a thread stays OnDeck until it acquires the lock. aoqi@0: // For a given lock there is at most OnDeck thread at any one instant. aoqi@0: WakeOne: aoqi@0: assert (List == _EntryList, "invariant") ; aoqi@0: ParkEvent * const w = List ; aoqi@0: assert (RelaxAssert || w != Thread::current()->_MutexEvent, "invariant") ; aoqi@0: _EntryList = w->ListNext ; aoqi@0: // as a diagnostic measure consider setting w->_ListNext = BAD aoqi@0: assert (UNS(_OnDeck) == _LBIT, "invariant") ; aoqi@0: _OnDeck = w ; // pass OnDeck to w. aoqi@0: // w will clear OnDeck once it acquires the outer lock aoqi@0: aoqi@0: // Another optional optimization ... aoqi@0: // For heavily contended locks it's not uncommon that some other aoqi@0: // thread acquired the lock while this thread was arranging succession. aoqi@0: // Try to defer the unpark() operation - Delegate the responsibility aoqi@0: // for unpark()ing the OnDeck thread to the current or subsequent owners aoqi@0: // That is, the new owner is responsible for unparking the OnDeck thread. aoqi@0: OrderAccess::storeload() ; aoqi@0: cxq = _LockWord.FullWord ; aoqi@0: if (cxq & _LBIT) return ; aoqi@0: aoqi@0: w->unpark() ; aoqi@0: return ; aoqi@0: } aoqi@0: aoqi@0: cxq = _LockWord.FullWord ; aoqi@0: if ((cxq & ~_LBIT) != 0) { aoqi@0: // The EntryList is empty but the cxq is populated. aoqi@0: // drain RATs from cxq into EntryList aoqi@0: // Detach RATs segment with CAS and then merge into EntryList aoqi@0: for (;;) { aoqi@0: // optional optimization - if locked, the owner is responsible for succession aoqi@0: if (cxq & _LBIT) goto Punt ; aoqi@0: const intptr_t vfy = CASPTR (&_LockWord, cxq, cxq & _LBIT) ; aoqi@0: if (vfy == cxq) break ; aoqi@0: cxq = vfy ; aoqi@0: // Interference - LockWord changed - Just retry aoqi@0: // We can see concurrent interference from contending threads aoqi@0: // pushing themselves onto the cxq or from lock-unlock operations. aoqi@0: // From the perspective of this thread, EntryList is stable and aoqi@0: // the cxq is prepend-only -- the head is volatile but the interior aoqi@0: // of the cxq is stable. In theory if we encounter interference from threads aoqi@0: // pushing onto cxq we could simply break off the original cxq suffix and aoqi@0: // move that segment to the EntryList, avoiding a 2nd or multiple CAS attempts aoqi@0: // on the high-traffic LockWord variable. For instance lets say the cxq is "ABCD" aoqi@0: // when we first fetch cxq above. Between the fetch -- where we observed "A" aoqi@0: // -- and CAS -- where we attempt to CAS null over A -- "PQR" arrive, aoqi@0: // yielding cxq = "PQRABCD". In this case we could simply set A.ListNext aoqi@0: // null, leaving cxq = "PQRA" and transfer the "BCD" segment to the EntryList. aoqi@0: // Note too, that it's safe for this thread to traverse the cxq aoqi@0: // without taking any special concurrency precautions. aoqi@0: } aoqi@0: aoqi@0: // We don't currently reorder the cxq segment as we move it onto aoqi@0: // the EntryList, but it might make sense to reverse the order aoqi@0: // or perhaps sort by thread priority. See the comments in aoqi@0: // synchronizer.cpp objectMonitor::exit(). aoqi@0: assert (_EntryList == NULL, "invariant") ; aoqi@0: _EntryList = List = (ParkEvent *)(cxq & ~_LBIT) ; aoqi@0: assert (List != NULL, "invariant") ; aoqi@0: goto WakeOne ; aoqi@0: } aoqi@0: aoqi@0: // cxq|EntryList is empty. aoqi@0: // w == NULL implies that cxq|EntryList == NULL in the past. aoqi@0: // Possible race - rare inopportune interleaving. aoqi@0: // A thread could have added itself to cxq since this thread previously checked. aoqi@0: // Detect and recover by refetching cxq. aoqi@0: Punt: aoqi@0: assert (UNS(_OnDeck) == _LBIT, "invariant") ; aoqi@0: _OnDeck = NULL ; // Release inner lock. aoqi@0: OrderAccess::storeload(); // Dekker duality - pivot point aoqi@0: aoqi@0: // Resample LockWord/cxq to recover from possible race. aoqi@0: // For instance, while this thread T1 held OnDeck, some other thread T2 might aoqi@0: // acquire the outer lock. Another thread T3 might try to acquire the outer aoqi@0: // lock, but encounter contention and enqueue itself on cxq. T2 then drops the aoqi@0: // outer lock, but skips succession as this thread T1 still holds OnDeck. aoqi@0: // T1 is and remains responsible for ensuring succession of T3. aoqi@0: // aoqi@0: // Note that we don't need to recheck EntryList, just cxq. aoqi@0: // If threads moved onto EntryList since we dropped OnDeck aoqi@0: // that implies some other thread forced succession. aoqi@0: cxq = _LockWord.FullWord ; aoqi@0: if ((cxq & ~_LBIT) != 0 && (cxq & _LBIT) == 0) { aoqi@0: goto Succession ; // potential race -- re-run succession aoqi@0: } aoqi@0: return ; aoqi@0: } aoqi@0: aoqi@0: bool Monitor::notify() { aoqi@0: assert (_owner == Thread::current(), "invariant") ; aoqi@0: assert (ILocked(), "invariant") ; aoqi@0: if (_WaitSet == NULL) return true ; aoqi@0: NotifyCount ++ ; aoqi@0: aoqi@0: // Transfer one thread from the WaitSet to the EntryList or cxq. aoqi@0: // Currently we just unlink the head of the WaitSet and prepend to the cxq. aoqi@0: // And of course we could just unlink it and unpark it, too, but aoqi@0: // in that case it'd likely impale itself on the reentry. aoqi@0: Thread::muxAcquire (_WaitLock, "notify:WaitLock") ; aoqi@0: ParkEvent * nfy = _WaitSet ; aoqi@0: if (nfy != NULL) { // DCL idiom aoqi@0: _WaitSet = nfy->ListNext ; aoqi@0: assert (nfy->Notified == 0, "invariant") ; aoqi@0: // push nfy onto the cxq aoqi@0: for (;;) { aoqi@0: const intptr_t v = _LockWord.FullWord ; aoqi@0: assert ((v & 0xFF) == _LBIT, "invariant") ; aoqi@0: nfy->ListNext = (ParkEvent *)(v & ~_LBIT); aoqi@0: if (CASPTR (&_LockWord, v, UNS(nfy)|_LBIT) == v) break; aoqi@0: // interference - _LockWord changed -- just retry aoqi@0: } aoqi@0: // Note that setting Notified before pushing nfy onto the cxq is aoqi@0: // also legal and safe, but the safety properties are much more aoqi@0: // subtle, so for the sake of code stewardship ... aoqi@0: OrderAccess::fence() ; aoqi@0: nfy->Notified = 1; aoqi@0: } aoqi@0: Thread::muxRelease (_WaitLock) ; aoqi@0: if (nfy != NULL && (NativeMonitorFlags & 16)) { aoqi@0: // Experimental code ... light up the wakee in the hope that this thread (the owner) aoqi@0: // will drop the lock just about the time the wakee comes ONPROC. aoqi@0: nfy->unpark() ; aoqi@0: } aoqi@0: assert (ILocked(), "invariant") ; aoqi@0: return true ; aoqi@0: } aoqi@0: aoqi@0: // Currently notifyAll() transfers the waiters one-at-a-time from the waitset aoqi@0: // to the cxq. This could be done more efficiently with a single bulk en-mass transfer, aoqi@0: // but in practice notifyAll() for large #s of threads is rare and not time-critical. aoqi@0: // Beware too, that we invert the order of the waiters. Lets say that the aoqi@0: // waitset is "ABCD" and the cxq is "XYZ". After a notifyAll() the waitset aoqi@0: // will be empty and the cxq will be "DCBAXYZ". This is benign, of course. aoqi@0: aoqi@0: bool Monitor::notify_all() { aoqi@0: assert (_owner == Thread::current(), "invariant") ; aoqi@0: assert (ILocked(), "invariant") ; aoqi@0: while (_WaitSet != NULL) notify() ; aoqi@0: return true ; aoqi@0: } aoqi@0: aoqi@0: int Monitor::IWait (Thread * Self, jlong timo) { aoqi@0: assert (ILocked(), "invariant") ; aoqi@0: aoqi@0: // Phases: aoqi@0: // 1. Enqueue Self on WaitSet - currently prepend aoqi@0: // 2. unlock - drop the outer lock aoqi@0: // 3. wait for either notification or timeout aoqi@0: // 4. lock - reentry - reacquire the outer lock aoqi@0: aoqi@0: ParkEvent * const ESelf = Self->_MutexEvent ; aoqi@0: ESelf->Notified = 0 ; aoqi@0: ESelf->reset() ; aoqi@0: OrderAccess::fence() ; aoqi@0: aoqi@0: // Add Self to WaitSet aoqi@0: // Ideally only the holder of the outer lock would manipulate the WaitSet - aoqi@0: // That is, the outer lock would implicitly protect the WaitSet. aoqi@0: // But if a thread in wait() encounters a timeout it will need to dequeue itself aoqi@0: // from the WaitSet _before it becomes the owner of the lock. We need to dequeue aoqi@0: // as the ParkEvent -- which serves as a proxy for the thread -- can't reside aoqi@0: // on both the WaitSet and the EntryList|cxq at the same time.. That is, a thread aoqi@0: // on the WaitSet can't be allowed to compete for the lock until it has managed to aoqi@0: // unlink its ParkEvent from WaitSet. Thus the need for WaitLock. aoqi@0: // Contention on the WaitLock is minimal. aoqi@0: // aoqi@0: // Another viable approach would be add another ParkEvent, "WaitEvent" to the aoqi@0: // thread class. The WaitSet would be composed of WaitEvents. Only the aoqi@0: // owner of the outer lock would manipulate the WaitSet. A thread in wait() aoqi@0: // could then compete for the outer lock, and then, if necessary, unlink itself aoqi@0: // from the WaitSet only after having acquired the outer lock. More precisely, aoqi@0: // there would be no WaitLock. A thread in in wait() would enqueue its WaitEvent aoqi@0: // on the WaitSet; release the outer lock; wait for either notification or timeout; aoqi@0: // reacquire the inner lock; and then, if needed, unlink itself from the WaitSet. aoqi@0: // aoqi@0: // Alternatively, a 2nd set of list link fields in the ParkEvent might suffice. aoqi@0: // One set would be for the WaitSet and one for the EntryList. aoqi@0: // We could also deconstruct the ParkEvent into a "pure" event and add a aoqi@0: // new immortal/TSM "ListElement" class that referred to ParkEvents. aoqi@0: // In that case we could have one ListElement on the WaitSet and another aoqi@0: // on the EntryList, with both referring to the same pure Event. aoqi@0: aoqi@0: Thread::muxAcquire (_WaitLock, "wait:WaitLock:Add") ; aoqi@0: ESelf->ListNext = _WaitSet ; aoqi@0: _WaitSet = ESelf ; aoqi@0: Thread::muxRelease (_WaitLock) ; aoqi@0: aoqi@0: // Release the outer lock aoqi@0: // We call IUnlock (RelaxAssert=true) as a thread T1 might aoqi@0: // enqueue itself on the WaitSet, call IUnlock(), drop the lock, aoqi@0: // and then stall before it can attempt to wake a successor. aoqi@0: // Some other thread T2 acquires the lock, and calls notify(), moving aoqi@0: // T1 from the WaitSet to the cxq. T2 then drops the lock. T1 resumes, aoqi@0: // and then finds *itself* on the cxq. During the course of a normal aoqi@0: // IUnlock() call a thread should _never find itself on the EntryList aoqi@0: // or cxq, but in the case of wait() it's possible. aoqi@0: // See synchronizer.cpp objectMonitor::wait(). aoqi@0: IUnlock (true) ; aoqi@0: aoqi@0: // Wait for either notification or timeout aoqi@0: // Beware that in some circumstances we might propagate aoqi@0: // spurious wakeups back to the caller. aoqi@0: aoqi@0: for (;;) { aoqi@0: if (ESelf->Notified) break ; aoqi@0: int err = ParkCommon (ESelf, timo) ; aoqi@0: if (err == OS_TIMEOUT || (NativeMonitorFlags & 1)) break ; aoqi@0: } aoqi@0: aoqi@0: // Prepare for reentry - if necessary, remove ESelf from WaitSet aoqi@0: // ESelf can be: aoqi@0: // 1. Still on the WaitSet. This can happen if we exited the loop by timeout. aoqi@0: // 2. On the cxq or EntryList aoqi@0: // 3. Not resident on cxq, EntryList or WaitSet, but in the OnDeck position. aoqi@0: aoqi@0: OrderAccess::fence() ; aoqi@0: int WasOnWaitSet = 0 ; aoqi@0: if (ESelf->Notified == 0) { aoqi@0: Thread::muxAcquire (_WaitLock, "wait:WaitLock:remove") ; aoqi@0: if (ESelf->Notified == 0) { // DCL idiom aoqi@0: assert (_OnDeck != ESelf, "invariant") ; // can't be both OnDeck and on WaitSet aoqi@0: // ESelf is resident on the WaitSet -- unlink it. aoqi@0: // A doubly-linked list would be better here so we can unlink in constant-time. aoqi@0: // We have to unlink before we potentially recontend as ESelf might otherwise aoqi@0: // end up on the cxq|EntryList -- it can't be on two lists at once. aoqi@0: ParkEvent * p = _WaitSet ; aoqi@0: ParkEvent * q = NULL ; // classic q chases p aoqi@0: while (p != NULL && p != ESelf) { aoqi@0: q = p ; aoqi@0: p = p->ListNext ; aoqi@0: } aoqi@0: assert (p == ESelf, "invariant") ; aoqi@0: if (p == _WaitSet) { // found at head aoqi@0: assert (q == NULL, "invariant") ; aoqi@0: _WaitSet = p->ListNext ; aoqi@0: } else { // found in interior aoqi@0: assert (q->ListNext == p, "invariant") ; aoqi@0: q->ListNext = p->ListNext ; aoqi@0: } aoqi@0: WasOnWaitSet = 1 ; // We were *not* notified but instead encountered timeout aoqi@0: } aoqi@0: Thread::muxRelease (_WaitLock) ; aoqi@0: } aoqi@0: aoqi@0: // Reentry phase - reacquire the lock aoqi@0: if (WasOnWaitSet) { aoqi@0: // ESelf was previously on the WaitSet but we just unlinked it above aoqi@0: // because of a timeout. ESelf is not resident on any list and is not OnDeck aoqi@0: assert (_OnDeck != ESelf, "invariant") ; aoqi@0: ILock (Self) ; aoqi@0: } else { aoqi@0: // A prior notify() operation moved ESelf from the WaitSet to the cxq. aoqi@0: // ESelf is now on the cxq, EntryList or at the OnDeck position. aoqi@0: // The following fragment is extracted from Monitor::ILock() aoqi@0: for (;;) { aoqi@0: if (_OnDeck == ESelf && TrySpin(Self)) break ; aoqi@0: ParkCommon (ESelf, 0) ; aoqi@0: } aoqi@0: assert (_OnDeck == ESelf, "invariant") ; aoqi@0: _OnDeck = NULL ; aoqi@0: } aoqi@0: aoqi@0: assert (ILocked(), "invariant") ; aoqi@0: return WasOnWaitSet != 0 ; // return true IFF timeout aoqi@0: } aoqi@0: aoqi@0: aoqi@0: // ON THE VMTHREAD SNEAKING PAST HELD LOCKS: aoqi@0: // In particular, there are certain types of global lock that may be held aoqi@0: // by a Java thread while it is blocked at a safepoint but before it has aoqi@0: // written the _owner field. These locks may be sneakily acquired by the aoqi@0: // VM thread during a safepoint to avoid deadlocks. Alternatively, one should aoqi@0: // identify all such locks, and ensure that Java threads never block at aoqi@0: // safepoints while holding them (_no_safepoint_check_flag). While it aoqi@0: // seems as though this could increase the time to reach a safepoint aoqi@0: // (or at least increase the mean, if not the variance), the latter aoqi@0: // approach might make for a cleaner, more maintainable JVM design. aoqi@0: // aoqi@0: // Sneaking is vile and reprehensible and should be excised at the 1st aoqi@0: // opportunity. It's possible that the need for sneaking could be obviated aoqi@0: // as follows. Currently, a thread might (a) while TBIVM, call pthread_mutex_lock aoqi@0: // or ILock() thus acquiring the "physical" lock underlying Monitor/Mutex. aoqi@0: // (b) stall at the TBIVM exit point as a safepoint is in effect. Critically, aoqi@0: // it'll stall at the TBIVM reentry state transition after having acquired the aoqi@0: // underlying lock, but before having set _owner and having entered the actual aoqi@0: // critical section. The lock-sneaking facility leverages that fact and allowed the aoqi@0: // VM thread to logically acquire locks that had already be physically locked by mutators aoqi@0: // but where mutators were known blocked by the reentry thread state transition. aoqi@0: // aoqi@0: // If we were to modify the Monitor-Mutex so that TBIVM state transitions tightly aoqi@0: // wrapped calls to park(), then we could likely do away with sneaking. We'd aoqi@0: // decouple lock acquisition and parking. The critical invariant to eliminating aoqi@0: // sneaking is to ensure that we never "physically" acquire the lock while TBIVM. aoqi@0: // An easy way to accomplish this is to wrap the park calls in a narrow TBIVM jacket. aoqi@0: // One difficulty with this approach is that the TBIVM wrapper could recurse and aoqi@0: // call lock() deep from within a lock() call, while the MutexEvent was already enqueued. aoqi@0: // Using a stack (N=2 at minimum) of ParkEvents would take care of that problem. aoqi@0: // aoqi@0: // But of course the proper ultimate approach is to avoid schemes that require explicit aoqi@0: // sneaking or dependence on any any clever invariants or subtle implementation properties aoqi@0: // of Mutex-Monitor and instead directly address the underlying design flaw. aoqi@0: aoqi@0: void Monitor::lock (Thread * Self) { aoqi@0: #ifdef CHECK_UNHANDLED_OOPS aoqi@0: // Clear unhandled oops so we get a crash right away. Only clear for non-vm aoqi@0: // or GC threads. aoqi@0: if (Self->is_Java_thread()) { aoqi@0: Self->clear_unhandled_oops(); aoqi@0: } aoqi@0: #endif // CHECK_UNHANDLED_OOPS aoqi@0: aoqi@0: debug_only(check_prelock_state(Self)); aoqi@0: assert (_owner != Self , "invariant") ; aoqi@0: assert (_OnDeck != Self->_MutexEvent, "invariant") ; aoqi@0: aoqi@0: if (TryFast()) { aoqi@0: Exeunt: aoqi@0: assert (ILocked(), "invariant") ; aoqi@0: assert (owner() == NULL, "invariant"); aoqi@0: set_owner (Self); aoqi@0: return ; aoqi@0: } aoqi@0: aoqi@0: // The lock is contended ... aoqi@0: aoqi@0: bool can_sneak = Self->is_VM_thread() && SafepointSynchronize::is_at_safepoint(); aoqi@0: if (can_sneak && _owner == NULL) { aoqi@0: // a java thread has locked the lock but has not entered the aoqi@0: // critical region -- let's just pretend we've locked the lock aoqi@0: // and go on. we note this with _snuck so we can also aoqi@0: // pretend to unlock when the time comes. aoqi@0: _snuck = true; aoqi@0: goto Exeunt ; aoqi@0: } aoqi@0: aoqi@0: // Try a brief spin to avoid passing thru thread state transition ... aoqi@0: if (TrySpin (Self)) goto Exeunt ; aoqi@0: aoqi@0: check_block_state(Self); aoqi@0: if (Self->is_Java_thread()) { aoqi@0: // Horribile dictu - we suffer through a state transition aoqi@0: assert(rank() > Mutex::special, "Potential deadlock with special or lesser rank mutex"); aoqi@0: ThreadBlockInVM tbivm ((JavaThread *) Self) ; aoqi@0: ILock (Self) ; aoqi@0: } else { aoqi@0: // Mirabile dictu aoqi@0: ILock (Self) ; aoqi@0: } aoqi@0: goto Exeunt ; aoqi@0: } aoqi@0: aoqi@0: void Monitor::lock() { aoqi@0: this->lock(Thread::current()); aoqi@0: } aoqi@0: aoqi@0: // Lock without safepoint check - a degenerate variant of lock(). aoqi@0: // Should ONLY be used by safepoint code and other code aoqi@0: // that is guaranteed not to block while running inside the VM. If this is called with aoqi@0: // thread state set to be in VM, the safepoint synchronization code will deadlock! aoqi@0: aoqi@0: void Monitor::lock_without_safepoint_check (Thread * Self) { aoqi@0: assert (_owner != Self, "invariant") ; aoqi@0: ILock (Self) ; aoqi@0: assert (_owner == NULL, "invariant"); aoqi@0: set_owner (Self); aoqi@0: } aoqi@0: aoqi@0: void Monitor::lock_without_safepoint_check () { aoqi@0: lock_without_safepoint_check (Thread::current()) ; aoqi@0: } aoqi@0: aoqi@0: aoqi@0: // Returns true if thread succeceed [sic] in grabbing the lock, otherwise false. aoqi@0: aoqi@0: bool Monitor::try_lock() { aoqi@0: Thread * const Self = Thread::current(); aoqi@0: debug_only(check_prelock_state(Self)); aoqi@0: // assert(!thread->is_inside_signal_handler(), "don't lock inside signal handler"); aoqi@0: aoqi@0: // Special case, where all Java threads are stopped. aoqi@0: // The lock may have been acquired but _owner is not yet set. aoqi@0: // In that case the VM thread can safely grab the lock. aoqi@0: // It strikes me this should appear _after the TryLock() fails, below. aoqi@0: bool can_sneak = Self->is_VM_thread() && SafepointSynchronize::is_at_safepoint(); aoqi@0: if (can_sneak && _owner == NULL) { aoqi@0: set_owner(Self); // Do not need to be atomic, since we are at a safepoint aoqi@0: _snuck = true; aoqi@0: return true; aoqi@0: } aoqi@0: aoqi@0: if (TryLock()) { aoqi@0: // We got the lock aoqi@0: assert (_owner == NULL, "invariant"); aoqi@0: set_owner (Self); aoqi@0: return true; aoqi@0: } aoqi@0: return false; aoqi@0: } aoqi@0: aoqi@0: void Monitor::unlock() { aoqi@0: assert (_owner == Thread::current(), "invariant") ; aoqi@0: assert (_OnDeck != Thread::current()->_MutexEvent , "invariant") ; aoqi@0: set_owner (NULL) ; aoqi@0: if (_snuck) { aoqi@0: assert(SafepointSynchronize::is_at_safepoint() && Thread::current()->is_VM_thread(), "sneak"); aoqi@0: _snuck = false; aoqi@0: return ; aoqi@0: } aoqi@0: IUnlock (false) ; aoqi@0: } aoqi@0: aoqi@0: // Yet another degenerate version of Monitor::lock() or lock_without_safepoint_check() aoqi@0: // jvm_raw_lock() and _unlock() can be called by non-Java threads via JVM_RawMonitorEnter. aoqi@0: // aoqi@0: // There's no expectation that JVM_RawMonitors will interoperate properly with the native aoqi@0: // Mutex-Monitor constructs. We happen to implement JVM_RawMonitors in terms of aoqi@0: // native Mutex-Monitors simply as a matter of convenience. A simple abstraction layer aoqi@0: // over a pthread_mutex_t would work equally as well, but require more platform-specific aoqi@0: // code -- a "PlatformMutex". Alternatively, a simply layer over muxAcquire-muxRelease aoqi@0: // would work too. aoqi@0: // aoqi@0: // Since the caller might be a foreign thread, we don't necessarily have a Thread.MutexEvent aoqi@0: // instance available. Instead, we transiently allocate a ParkEvent on-demand if aoqi@0: // we encounter contention. That ParkEvent remains associated with the thread aoqi@0: // until it manages to acquire the lock, at which time we return the ParkEvent aoqi@0: // to the global ParkEvent free list. This is correct and suffices for our purposes. aoqi@0: // aoqi@0: // Beware that the original jvm_raw_unlock() had a "_snuck" test but that aoqi@0: // jvm_raw_lock() didn't have the corresponding test. I suspect that's an aoqi@0: // oversight, but I've replicated the original suspect logic in the new code ... aoqi@0: aoqi@0: void Monitor::jvm_raw_lock() { aoqi@0: assert(rank() == native, "invariant"); aoqi@0: aoqi@0: if (TryLock()) { aoqi@0: Exeunt: aoqi@0: assert (ILocked(), "invariant") ; aoqi@0: assert (_owner == NULL, "invariant"); aoqi@0: // This can potentially be called by non-java Threads. Thus, the ThreadLocalStorage aoqi@0: // might return NULL. Don't call set_owner since it will break on an NULL owner aoqi@0: // Consider installing a non-null "ANON" distinguished value instead of just NULL. aoqi@0: _owner = ThreadLocalStorage::thread(); aoqi@0: return ; aoqi@0: } aoqi@0: aoqi@0: if (TrySpin(NULL)) goto Exeunt ; aoqi@0: aoqi@0: // slow-path - apparent contention aoqi@0: // Allocate a ParkEvent for transient use. aoqi@0: // The ParkEvent remains associated with this thread until aoqi@0: // the time the thread manages to acquire the lock. aoqi@0: ParkEvent * const ESelf = ParkEvent::Allocate(NULL) ; aoqi@0: ESelf->reset() ; aoqi@0: OrderAccess::storeload() ; aoqi@0: aoqi@0: // Either Enqueue Self on cxq or acquire the outer lock. aoqi@0: if (AcquireOrPush (ESelf)) { aoqi@0: ParkEvent::Release (ESelf) ; // surrender the ParkEvent aoqi@0: goto Exeunt ; aoqi@0: } aoqi@0: aoqi@0: // At any given time there is at most one ondeck thread. aoqi@0: // ondeck implies not resident on cxq and not resident on EntryList aoqi@0: // Only the OnDeck thread can try to acquire -- contended for -- the lock. aoqi@0: // CONSIDER: use Self->OnDeck instead of m->OnDeck. aoqi@0: for (;;) { aoqi@0: if (_OnDeck == ESelf && TrySpin(NULL)) break ; aoqi@0: ParkCommon (ESelf, 0) ; aoqi@0: } aoqi@0: aoqi@0: assert (_OnDeck == ESelf, "invariant") ; aoqi@0: _OnDeck = NULL ; aoqi@0: ParkEvent::Release (ESelf) ; // surrender the ParkEvent aoqi@0: goto Exeunt ; aoqi@0: } aoqi@0: aoqi@0: void Monitor::jvm_raw_unlock() { aoqi@0: // Nearly the same as Monitor::unlock() ... aoqi@0: // directly set _owner instead of using set_owner(null) aoqi@0: _owner = NULL ; aoqi@0: if (_snuck) { // ??? aoqi@0: assert(SafepointSynchronize::is_at_safepoint() && Thread::current()->is_VM_thread(), "sneak"); aoqi@0: _snuck = false; aoqi@0: return ; aoqi@0: } aoqi@0: IUnlock(false) ; aoqi@0: } aoqi@0: aoqi@0: bool Monitor::wait(bool no_safepoint_check, long timeout, bool as_suspend_equivalent) { aoqi@0: Thread * const Self = Thread::current() ; aoqi@0: assert (_owner == Self, "invariant") ; aoqi@0: assert (ILocked(), "invariant") ; aoqi@0: aoqi@0: // as_suspend_equivalent logically implies !no_safepoint_check aoqi@0: guarantee (!as_suspend_equivalent || !no_safepoint_check, "invariant") ; aoqi@0: // !no_safepoint_check logically implies java_thread aoqi@0: guarantee (no_safepoint_check || Self->is_Java_thread(), "invariant") ; aoqi@0: aoqi@0: #ifdef ASSERT aoqi@0: Monitor * least = get_least_ranked_lock_besides_this(Self->owned_locks()); aoqi@0: assert(least != this, "Specification of get_least_... call above"); aoqi@0: if (least != NULL && least->rank() <= special) { aoqi@0: tty->print("Attempting to wait on monitor %s/%d while holding" aoqi@0: " lock %s/%d -- possible deadlock", aoqi@0: name(), rank(), least->name(), least->rank()); aoqi@0: assert(false, "Shouldn't block(wait) while holding a lock of rank special"); aoqi@0: } aoqi@0: #endif // ASSERT aoqi@0: aoqi@0: int wait_status ; aoqi@0: // conceptually set the owner to NULL in anticipation of aoqi@0: // abdicating the lock in wait aoqi@0: set_owner(NULL); aoqi@0: if (no_safepoint_check) { aoqi@0: wait_status = IWait (Self, timeout) ; aoqi@0: } else { aoqi@0: assert (Self->is_Java_thread(), "invariant") ; aoqi@0: JavaThread *jt = (JavaThread *)Self; aoqi@0: aoqi@0: // Enter safepoint region - ornate and Rococo ... aoqi@0: ThreadBlockInVM tbivm(jt); aoqi@0: OSThreadWaitState osts(Self->osthread(), false /* not Object.wait() */); aoqi@0: aoqi@0: if (as_suspend_equivalent) { aoqi@0: jt->set_suspend_equivalent(); aoqi@0: // cleared by handle_special_suspend_equivalent_condition() or aoqi@0: // java_suspend_self() aoqi@0: } aoqi@0: aoqi@0: wait_status = IWait (Self, timeout) ; aoqi@0: aoqi@0: // were we externally suspended while we were waiting? aoqi@0: if (as_suspend_equivalent && jt->handle_special_suspend_equivalent_condition()) { aoqi@0: // Our event wait has finished and we own the lock, but aoqi@0: // while we were waiting another thread suspended us. We don't aoqi@0: // want to hold the lock while suspended because that aoqi@0: // would surprise the thread that suspended us. aoqi@0: assert (ILocked(), "invariant") ; aoqi@0: IUnlock (true) ; aoqi@0: jt->java_suspend_self(); aoqi@0: ILock (Self) ; aoqi@0: assert (ILocked(), "invariant") ; aoqi@0: } aoqi@0: } aoqi@0: aoqi@0: // Conceptually reestablish ownership of the lock. aoqi@0: // The "real" lock -- the LockByte -- was reacquired by IWait(). aoqi@0: assert (ILocked(), "invariant") ; aoqi@0: assert (_owner == NULL, "invariant") ; aoqi@0: set_owner (Self) ; aoqi@0: return wait_status != 0 ; // return true IFF timeout aoqi@0: } aoqi@0: aoqi@0: Monitor::~Monitor() { aoqi@0: assert ((UNS(_owner)|UNS(_LockWord.FullWord)|UNS(_EntryList)|UNS(_WaitSet)|UNS(_OnDeck)) == 0, "") ; aoqi@0: } aoqi@0: aoqi@0: void Monitor::ClearMonitor (Monitor * m, const char *name) { aoqi@0: m->_owner = NULL ; aoqi@0: m->_snuck = false ; aoqi@0: if (name == NULL) { aoqi@0: strcpy(m->_name, "UNKNOWN") ; aoqi@0: } else { aoqi@0: strncpy(m->_name, name, MONITOR_NAME_LEN - 1); aoqi@0: m->_name[MONITOR_NAME_LEN - 1] = '\0'; aoqi@0: } aoqi@0: m->_LockWord.FullWord = 0 ; aoqi@0: m->_EntryList = NULL ; aoqi@0: m->_OnDeck = NULL ; aoqi@0: m->_WaitSet = NULL ; aoqi@0: m->_WaitLock[0] = 0 ; aoqi@0: } aoqi@0: aoqi@0: Monitor::Monitor() { ClearMonitor(this); } aoqi@0: aoqi@0: Monitor::Monitor (int Rank, const char * name, bool allow_vm_block) { aoqi@0: ClearMonitor (this, name) ; aoqi@0: #ifdef ASSERT aoqi@0: _allow_vm_block = allow_vm_block; aoqi@0: _rank = Rank ; aoqi@0: #endif aoqi@0: } aoqi@0: aoqi@0: Mutex::~Mutex() { aoqi@0: assert ((UNS(_owner)|UNS(_LockWord.FullWord)|UNS(_EntryList)|UNS(_WaitSet)|UNS(_OnDeck)) == 0, "") ; aoqi@0: } aoqi@0: aoqi@0: Mutex::Mutex (int Rank, const char * name, bool allow_vm_block) { aoqi@0: ClearMonitor ((Monitor *) this, name) ; aoqi@0: #ifdef ASSERT aoqi@0: _allow_vm_block = allow_vm_block; aoqi@0: _rank = Rank ; aoqi@0: #endif aoqi@0: } aoqi@0: aoqi@0: bool Monitor::owned_by_self() const { aoqi@0: bool ret = _owner == Thread::current(); aoqi@0: assert (!ret || _LockWord.Bytes[_LSBINDEX] != 0, "invariant") ; aoqi@0: return ret; aoqi@0: } aoqi@0: aoqi@0: void Monitor::print_on_error(outputStream* st) const { aoqi@0: st->print("[" PTR_FORMAT, this); aoqi@0: st->print("] %s", _name); aoqi@0: st->print(" - owner thread: " PTR_FORMAT, _owner); aoqi@0: } aoqi@0: aoqi@0: aoqi@0: aoqi@0: aoqi@0: // ---------------------------------------------------------------------------------- aoqi@0: // Non-product code aoqi@0: aoqi@0: #ifndef PRODUCT aoqi@0: void Monitor::print_on(outputStream* st) const { aoqi@0: st->print_cr("Mutex: [0x%lx/0x%lx] %s - owner: 0x%lx", this, _LockWord.FullWord, _name, _owner); aoqi@0: } aoqi@0: #endif aoqi@0: aoqi@0: #ifndef PRODUCT aoqi@0: #ifdef ASSERT aoqi@0: Monitor * Monitor::get_least_ranked_lock(Monitor * locks) { aoqi@0: Monitor *res, *tmp; aoqi@0: for (res = tmp = locks; tmp != NULL; tmp = tmp->next()) { aoqi@0: if (tmp->rank() < res->rank()) { aoqi@0: res = tmp; aoqi@0: } aoqi@0: } aoqi@0: if (!SafepointSynchronize::is_at_safepoint()) { aoqi@0: // In this case, we expect the held locks to be aoqi@0: // in increasing rank order (modulo any native ranks) aoqi@0: for (tmp = locks; tmp != NULL; tmp = tmp->next()) { aoqi@0: if (tmp->next() != NULL) { aoqi@0: assert(tmp->rank() == Mutex::native || aoqi@0: tmp->rank() <= tmp->next()->rank(), "mutex rank anomaly?"); aoqi@0: } aoqi@0: } aoqi@0: } aoqi@0: return res; aoqi@0: } aoqi@0: aoqi@0: Monitor* Monitor::get_least_ranked_lock_besides_this(Monitor* locks) { aoqi@0: Monitor *res, *tmp; aoqi@0: for (res = NULL, tmp = locks; tmp != NULL; tmp = tmp->next()) { aoqi@0: if (tmp != this && (res == NULL || tmp->rank() < res->rank())) { aoqi@0: res = tmp; aoqi@0: } aoqi@0: } aoqi@0: if (!SafepointSynchronize::is_at_safepoint()) { aoqi@0: // In this case, we expect the held locks to be aoqi@0: // in increasing rank order (modulo any native ranks) aoqi@0: for (tmp = locks; tmp != NULL; tmp = tmp->next()) { aoqi@0: if (tmp->next() != NULL) { aoqi@0: assert(tmp->rank() == Mutex::native || aoqi@0: tmp->rank() <= tmp->next()->rank(), "mutex rank anomaly?"); aoqi@0: } aoqi@0: } aoqi@0: } aoqi@0: return res; aoqi@0: } aoqi@0: aoqi@0: aoqi@0: bool Monitor::contains(Monitor* locks, Monitor * lock) { aoqi@0: for (; locks != NULL; locks = locks->next()) { aoqi@0: if (locks == lock) aoqi@0: return true; aoqi@0: } aoqi@0: return false; aoqi@0: } aoqi@0: #endif aoqi@0: aoqi@0: // Called immediately after lock acquisition or release as a diagnostic aoqi@0: // to track the lock-set of the thread and test for rank violations that aoqi@0: // might indicate exposure to deadlock. aoqi@0: // Rather like an EventListener for _owner (:>). aoqi@0: aoqi@0: void Monitor::set_owner_implementation(Thread *new_owner) { aoqi@0: // This function is solely responsible for maintaining aoqi@0: // and checking the invariant that threads and locks aoqi@0: // are in a 1/N relation, with some some locks unowned. aoqi@0: // It uses the Mutex::_owner, Mutex::_next, and aoqi@0: // Thread::_owned_locks fields, and no other function aoqi@0: // changes those fields. aoqi@0: // It is illegal to set the mutex from one non-NULL aoqi@0: // owner to another--it must be owned by NULL as an aoqi@0: // intermediate state. aoqi@0: aoqi@0: if (new_owner != NULL) { aoqi@0: // the thread is acquiring this lock aoqi@0: aoqi@0: assert(new_owner == Thread::current(), "Should I be doing this?"); aoqi@0: assert(_owner == NULL, "setting the owner thread of an already owned mutex"); aoqi@0: _owner = new_owner; // set the owner aoqi@0: aoqi@0: // link "this" into the owned locks list aoqi@0: aoqi@0: #ifdef ASSERT // Thread::_owned_locks is under the same ifdef aoqi@0: Monitor* locks = get_least_ranked_lock(new_owner->owned_locks()); aoqi@0: // Mutex::set_owner_implementation is a friend of Thread aoqi@0: aoqi@0: assert(this->rank() >= 0, "bad lock rank"); aoqi@0: aoqi@0: // Deadlock avoidance rules require us to acquire Mutexes only in aoqi@0: // a global total order. For example m1 is the lowest ranked mutex aoqi@0: // that the thread holds and m2 is the mutex the thread is trying aoqi@0: // to acquire, then deadlock avoidance rules require that the rank aoqi@0: // of m2 be less than the rank of m1. aoqi@0: // The rank Mutex::native is an exception in that it is not subject aoqi@0: // to the verification rules. aoqi@0: // Here are some further notes relating to mutex acquisition anomalies: aoqi@0: // . under Solaris, the interrupt lock gets acquired when doing aoqi@0: // profiling, so any lock could be held. aoqi@0: // . it is also ok to acquire Safepoint_lock at the very end while we aoqi@0: // already hold Terminator_lock - may happen because of periodic safepoints aoqi@0: if (this->rank() != Mutex::native && aoqi@0: this->rank() != Mutex::suspend_resume && aoqi@0: locks != NULL && locks->rank() <= this->rank() && aoqi@0: !SafepointSynchronize::is_at_safepoint() && aoqi@0: this != Interrupt_lock && this != ProfileVM_lock && aoqi@0: !(this == Safepoint_lock && contains(locks, Terminator_lock) && aoqi@0: SafepointSynchronize::is_synchronizing())) { aoqi@0: new_owner->print_owned_locks(); aoqi@0: fatal(err_msg("acquiring lock %s/%d out of order with lock %s/%d -- " aoqi@0: "possible deadlock", this->name(), this->rank(), aoqi@0: locks->name(), locks->rank())); aoqi@0: } aoqi@0: aoqi@0: this->_next = new_owner->_owned_locks; aoqi@0: new_owner->_owned_locks = this; aoqi@0: #endif aoqi@0: aoqi@0: } else { aoqi@0: // the thread is releasing this lock aoqi@0: aoqi@0: Thread* old_owner = _owner; aoqi@0: debug_only(_last_owner = old_owner); aoqi@0: aoqi@0: assert(old_owner != NULL, "removing the owner thread of an unowned mutex"); aoqi@0: assert(old_owner == Thread::current(), "removing the owner thread of an unowned mutex"); aoqi@0: aoqi@0: _owner = NULL; // set the owner aoqi@0: aoqi@0: #ifdef ASSERT aoqi@0: Monitor *locks = old_owner->owned_locks(); aoqi@0: aoqi@0: // remove "this" from the owned locks list aoqi@0: aoqi@0: Monitor *prev = NULL; aoqi@0: bool found = false; aoqi@0: for (; locks != NULL; prev = locks, locks = locks->next()) { aoqi@0: if (locks == this) { aoqi@0: found = true; aoqi@0: break; aoqi@0: } aoqi@0: } aoqi@0: assert(found, "Removing a lock not owned"); aoqi@0: if (prev == NULL) { aoqi@0: old_owner->_owned_locks = _next; aoqi@0: } else { aoqi@0: prev->_next = _next; aoqi@0: } aoqi@0: _next = NULL; aoqi@0: #endif aoqi@0: } aoqi@0: } aoqi@0: aoqi@0: aoqi@0: // Factored out common sanity checks for locking mutex'es. Used by lock() and try_lock() aoqi@0: void Monitor::check_prelock_state(Thread *thread) { aoqi@0: assert((!thread->is_Java_thread() || ((JavaThread *)thread)->thread_state() == _thread_in_vm) aoqi@0: || rank() == Mutex::special, "wrong thread state for using locks"); aoqi@0: if (StrictSafepointChecks) { aoqi@0: if (thread->is_VM_thread() && !allow_vm_block()) { aoqi@0: fatal(err_msg("VM thread using lock %s (not allowed to block on)", aoqi@0: name())); aoqi@0: } aoqi@0: debug_only(if (rank() != Mutex::special) \ aoqi@0: thread->check_for_valid_safepoint_state(false);) aoqi@0: } aoqi@0: if (thread->is_Watcher_thread()) { aoqi@0: assert(!WatcherThread::watcher_thread()->has_crash_protection(), aoqi@0: "locking not allowed when crash protection is set"); aoqi@0: } aoqi@0: } aoqi@0: aoqi@0: void Monitor::check_block_state(Thread *thread) { aoqi@0: if (!_allow_vm_block && thread->is_VM_thread()) { aoqi@0: warning("VM thread blocked on lock"); aoqi@0: print(); aoqi@0: BREAKPOINT; aoqi@0: } aoqi@0: assert(_owner != thread, "deadlock: blocking on monitor owned by current thread"); aoqi@0: } aoqi@0: aoqi@0: #endif // PRODUCT