1 
   2 /*
   3  * Copyright (c) 1998, 2012, Oracle and/or its affiliates. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "runtime/mutex.hpp"
  28 #include "runtime/osThread.hpp"
  29 #include "utilities/events.hpp"
  30 #ifdef TARGET_OS_FAMILY_linux
  31 # include "mutex_linux.inline.hpp"
  32 # include "thread_linux.inline.hpp"
  33 #endif
  34 #ifdef TARGET_OS_FAMILY_solaris
  35 # include "mutex_solaris.inline.hpp"
  36 # include "thread_solaris.inline.hpp"
  37 #endif
  38 #ifdef TARGET_OS_FAMILY_windows
  39 # include "mutex_windows.inline.hpp"
  40 # include "thread_windows.inline.hpp"
  41 #endif
  42 #ifdef TARGET_OS_FAMILY_bsd
  43 # include "mutex_bsd.inline.hpp"
  44 # include "thread_bsd.inline.hpp"
  45 #endif
  46 
  47 // 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
  48 //
  49 // Native Monitor-Mutex locking - theory of operations
  50 //
  51 // * Native Monitors are completely unrelated to Java-level monitors,
  52 //   although the "back-end" slow-path implementations share a common lineage.
  53 //   See objectMonitor:: in synchronizer.cpp.
  54 //   Native Monitors do *not* support nesting or recursion but otherwise
  55 //   they're basically Hoare-flavor monitors.
  56 //
  57 // * A thread acquires ownership of a Monitor/Mutex by CASing the LockByte
  58 //   in the _LockWord from zero to non-zero.  Note that the _Owner field
  59 //   is advisory and is used only to verify that the thread calling unlock()
  60 //   is indeed the last thread to have acquired the lock.
  61 //
  62 // * Contending threads "push" themselves onto the front of the contention
  63 //   queue -- called the cxq -- with CAS and then spin/park.
  64 //   The _LockWord contains the LockByte as well as the pointer to the head
  65 //   of the cxq.  Colocating the LockByte with the cxq precludes certain races.
  66 //
  67 // * Using a separately addressable LockByte allows for CAS:MEMBAR or CAS:0
  68 //   idioms.  We currently use MEMBAR in the uncontended unlock() path, as
  69 //   MEMBAR often has less latency than CAS.  If warranted, we could switch to
  70 //   a CAS:0 mode, using timers to close the resultant race, as is done
  71 //   with Java Monitors in synchronizer.cpp.
  72 //
  73 //   See the following for a discussion of the relative cost of atomics (CAS)
  74 //   MEMBAR, and ways to eliminate such instructions from the common-case paths:
  75 //   -- http://blogs.sun.com/dave/entry/biased_locking_in_hotspot
  76 //   -- http://blogs.sun.com/dave/resource/MustangSync.pdf
  77 //   -- http://blogs.sun.com/dave/resource/synchronization-public2.pdf
  78 //   -- synchronizer.cpp
  79 //
  80 // * Overall goals - desiderata
  81 //   1. Minimize context switching
  82 //   2. Minimize lock migration
  83 //   3. Minimize CPI -- affinity and locality
  84 //   4. Minimize the execution of high-latency instructions such as CAS or MEMBAR
  85 //   5. Minimize outer lock hold times
  86 //   6. Behave gracefully on a loaded system
  87 //
  88 // * Thread flow and list residency:
  89 //
  90 //   Contention queue --> EntryList --> OnDeck --> Owner --> !Owner
  91 //   [..resident on monitor list..]
  92 //   [...........contending..................]
  93 //
  94 //   -- The contention queue (cxq) contains recently-arrived threads (RATs).
  95 //      Threads on the cxq eventually drain into the EntryList.
  96 //   -- Invariant: a thread appears on at most one list -- cxq, EntryList
  97 //      or WaitSet -- at any one time.
  98 //   -- For a given monitor there can be at most one "OnDeck" thread at any
  99 //      given time but if needbe this particular invariant could be relaxed.
 100 //
 101 // * The WaitSet and EntryList linked lists are composed of ParkEvents.
 102 //   I use ParkEvent instead of threads as ParkEvents are immortal and
 103 //   type-stable, meaning we can safely unpark() a possibly stale
 104 //   list element in the unlock()-path.  (That's benign).
 105 //
 106 // * Succession policy - providing for progress:
 107 //
 108 //   As necessary, the unlock()ing thread identifies, unlinks, and unparks
 109 //   an "heir presumptive" tentative successor thread from the EntryList.
 110 //   This becomes the so-called "OnDeck" thread, of which there can be only
 111 //   one at any given time for a given monitor.  The wakee will recontend
 112 //   for ownership of monitor.
 113 //
 114 //   Succession is provided for by a policy of competitive handoff.
 115 //   The exiting thread does _not_ grant or pass ownership to the
 116 //   successor thread.  (This is also referred to as "handoff" succession").
 117 //   Instead the exiting thread releases ownership and possibly wakes
 118 //   a successor, so the successor can (re)compete for ownership of the lock.
 119 //
 120 //   Competitive handoff provides excellent overall throughput at the expense
 121 //   of short-term fairness.  If fairness is a concern then one remedy might
 122 //   be to add an AcquireCounter field to the monitor.  After a thread acquires
 123 //   the lock it will decrement the AcquireCounter field.  When the count
 124 //   reaches 0 the thread would reset the AcquireCounter variable, abdicate
 125 //   the lock directly to some thread on the EntryList, and then move itself to the
 126 //   tail of the EntryList.
 127 //
 128 //   But in practice most threads engage or otherwise participate in resource
 129 //   bounded producer-consumer relationships, so lock domination is not usually
 130 //   a practical concern.  Recall too, that in general it's easier to construct
 131 //   a fair lock from a fast lock, but not vice-versa.
 132 //
 133 // * The cxq can have multiple concurrent "pushers" but only one concurrent
 134 //   detaching thread.  This mechanism is immune from the ABA corruption.
 135 //   More precisely, the CAS-based "push" onto cxq is ABA-oblivious.
 136 //   We use OnDeck as a pseudo-lock to enforce the at-most-one detaching
 137 //   thread constraint.
 138 //
 139 // * Taken together, the cxq and the EntryList constitute or form a
 140 //   single logical queue of threads stalled trying to acquire the lock.
 141 //   We use two distinct lists to reduce heat on the list ends.
 142 //   Threads in lock() enqueue onto cxq while threads in unlock() will
 143 //   dequeue from the EntryList.  (c.f. Michael Scott's "2Q" algorithm).
 144 //   A key desideratum is to minimize queue & monitor metadata manipulation
 145 //   that occurs while holding the "outer" monitor lock -- that is, we want to
 146 //   minimize monitor lock holds times.
 147 //
 148 //   The EntryList is ordered by the prevailing queue discipline and
 149 //   can be organized in any convenient fashion, such as a doubly-linked list or
 150 //   a circular doubly-linked list.  If we need a priority queue then something akin
 151 //   to Solaris' sleepq would work nicely.  Viz.,
 152 //   -- http://agg.eng/ws/on10_nightly/source/usr/src/uts/common/os/sleepq.c.
 153 //   -- http://cvs.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/uts/common/os/sleepq.c
 154 //   Queue discipline is enforced at ::unlock() time, when the unlocking thread
 155 //   drains the cxq into the EntryList, and orders or reorders the threads on the
 156 //   EntryList accordingly.
 157 //
 158 //   Barring "lock barging", this mechanism provides fair cyclic ordering,
 159 //   somewhat similar to an elevator-scan.
 160 //
 161 // * OnDeck
 162 //   --  For a given monitor there can be at most one OnDeck thread at any given
 163 //       instant.  The OnDeck thread is contending for the lock, but has been
 164 //       unlinked from the EntryList and cxq by some previous unlock() operations.
 165 //       Once a thread has been designated the OnDeck thread it will remain so
 166 //       until it manages to acquire the lock -- being OnDeck is a stable property.
 167 //   --  Threads on the EntryList or cxq are _not allowed to attempt lock acquisition.
 168 //   --  OnDeck also serves as an "inner lock" as follows.  Threads in unlock() will, after
 169 //       having cleared the LockByte and dropped the outer lock,  attempt to "trylock"
 170 //       OnDeck by CASing the field from null to non-null.  If successful, that thread
 171 //       is then responsible for progress and succession and can use CAS to detach and
 172 //       drain the cxq into the EntryList.  By convention, only this thread, the holder of
 173 //       the OnDeck inner lock, can manipulate the EntryList or detach and drain the
 174 //       RATs on the cxq into the EntryList.  This avoids ABA corruption on the cxq as
 175 //       we allow multiple concurrent "push" operations but restrict detach concurrency
 176 //       to at most one thread.  Having selected and detached a successor, the thread then
 177 //       changes the OnDeck to refer to that successor, and then unparks the successor.
 178 //       That successor will eventually acquire the lock and clear OnDeck.  Beware
 179 //       that the OnDeck usage as a lock is asymmetric.  A thread in unlock() transiently
 180 //       "acquires" OnDeck, performs queue manipulations, passes OnDeck to some successor,
 181 //       and then the successor eventually "drops" OnDeck.  Note that there's never
 182 //       any sense of contention on the inner lock, however.  Threads never contend
 183 //       or wait for the inner lock.
 184 //   --  OnDeck provides for futile wakeup throttling a described in section 3.3 of
 185 //       See http://www.usenix.org/events/jvm01/full_papers/dice/dice.pdf
 186 //       In a sense, OnDeck subsumes the ObjectMonitor _Succ and ObjectWaiter
 187 //       TState fields found in Java-level objectMonitors.  (See synchronizer.cpp).
 188 //
 189 // * Waiting threads reside on the WaitSet list -- wait() puts
 190 //   the caller onto the WaitSet.  Notify() or notifyAll() simply
 191 //   transfers threads from the WaitSet to either the EntryList or cxq.
 192 //   Subsequent unlock() operations will eventually unpark the notifyee.
 193 //   Unparking a notifee in notify() proper is inefficient - if we were to do so
 194 //   it's likely the notifyee would simply impale itself on the lock held
 195 //   by the notifier.
 196 //
 197 // * The mechanism is obstruction-free in that if the holder of the transient
 198 //   OnDeck lock in unlock() is preempted or otherwise stalls, other threads
 199 //   can still acquire and release the outer lock and continue to make progress.
 200 //   At worst, waking of already blocked contending threads may be delayed,
 201 //   but nothing worse.  (We only use "trylock" operations on the inner OnDeck
 202 //   lock).
 203 //
 204 // * Note that thread-local storage must be initialized before a thread
 205 //   uses Native monitors or mutexes.  The native monitor-mutex subsystem
 206 //   depends on Thread::current().
 207 //
 208 // * The monitor synchronization subsystem avoids the use of native
 209 //   synchronization primitives except for the narrow platform-specific
 210 //   park-unpark abstraction.  See the comments in os_solaris.cpp regarding
 211 //   the semantics of park-unpark.  Put another way, this monitor implementation
 212 //   depends only on atomic operations and park-unpark.  The monitor subsystem
 213 //   manages all RUNNING->BLOCKED and BLOCKED->READY transitions while the
 214 //   underlying OS manages the READY<->RUN transitions.
 215 //
 216 // * The memory consistency model provide by lock()-unlock() is at least as
 217 //   strong or stronger than the Java Memory model defined by JSR-133.
 218 //   That is, we guarantee at least entry consistency, if not stronger.
 219 //   See http://g.oswego.edu/dl/jmm/cookbook.html.
 220 //
 221 // * Thread:: currently contains a set of purpose-specific ParkEvents:
 222 //   _MutexEvent, _ParkEvent, etc.  A better approach might be to do away with
 223 //   the purpose-specific ParkEvents and instead implement a general per-thread
 224 //   stack of available ParkEvents which we could provision on-demand.  The
 225 //   stack acts as a local cache to avoid excessive calls to ParkEvent::Allocate()
 226 //   and ::Release().  A thread would simply pop an element from the local stack before it
 227 //   enqueued or park()ed.  When the contention was over the thread would
 228 //   push the no-longer-needed ParkEvent back onto its stack.
 229 //
 230 // * A slightly reduced form of ILock() and IUnlock() have been partially
 231 //   model-checked (Murphi) for safety and progress at T=1,2,3 and 4.
 232 //   It'd be interesting to see if TLA/TLC could be useful as well.
 233 //
 234 // * Mutex-Monitor is a low-level "leaf" subsystem.  That is, the monitor
 235 //   code should never call other code in the JVM that might itself need to
 236 //   acquire monitors or mutexes.  That's true *except* in the case of the
 237 //   ThreadBlockInVM state transition wrappers.  The ThreadBlockInVM DTOR handles
 238 //   mutator reentry (ingress) by checking for a pending safepoint in which case it will
 239 //   call SafepointSynchronize::block(), which in turn may call Safepoint_lock->lock(), etc.
 240 //   In that particular case a call to lock() for a given Monitor can end up recursively
 241 //   calling lock() on another monitor.   While distasteful, this is largely benign
 242 //   as the calls come from jacket that wraps lock(), and not from deep within lock() itself.
 243 //
 244 //   It's unfortunate that native mutexes and thread state transitions were convolved.
 245 //   They're really separate concerns and should have remained that way.  Melding
 246 //   them together was facile -- a bit too facile.   The current implementation badly
 247 //   conflates the two concerns.
 248 //
 249 // * TODO-FIXME:
 250 //
 251 //   -- Add DTRACE probes for contended acquire, contended acquired, contended unlock
 252 //      We should also add DTRACE probes in the ParkEvent subsystem for
 253 //      Park-entry, Park-exit, and Unpark.
 254 //
 255 //   -- We have an excess of mutex-like constructs in the JVM, namely:
 256 //      1. objectMonitors for Java-level synchronization (synchronizer.cpp)
 257 //      2. low-level muxAcquire and muxRelease
 258 //      3. low-level spinAcquire and spinRelease
 259 //      4. native Mutex:: and Monitor::
 260 //      5. jvm_raw_lock() and _unlock()
 261 //      6. JVMTI raw monitors -- distinct from (5) despite having a confusingly
 262 //         similar name.
 263 //
 264 // 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
 265 
 266 
 267 // CASPTR() uses the canonical argument order that dominates in the literature.
 268 // Our internal cmpxchg_ptr() uses a bastardized ordering to accommodate Sun .il templates.
 269 
 270 #define CASPTR(a,c,s) intptr_t(Atomic::cmpxchg_ptr ((void *)(s),(void *)(a),(void *)(c)))
 271 #define UNS(x) (uintptr_t(x))
 272 #define TRACE(m) { static volatile int ctr = 0 ; int x = ++ctr ; if ((x & (x-1))==0) { ::printf ("%d:%s\n", x, #m); ::fflush(stdout); }}
 273 
 274 // Simplistic low-quality Marsaglia SHIFT-XOR RNG.
 275 // Bijective except for the trailing mask operation.
 276 // Useful for spin loops as the compiler can't optimize it away.
 277 
 278 static inline jint MarsagliaXORV (jint x) {
 279   if (x == 0) x = 1|os::random() ;
 280   x ^= x << 6;
 281   x ^= ((unsigned)x) >> 21;
 282   x ^= x << 7 ;
 283   return x & 0x7FFFFFFF ;
 284 }
 285 
 286 static inline jint MarsagliaXOR (jint * const a) {
 287   jint x = *a ;
 288   if (x == 0) x = UNS(a)|1 ;
 289   x ^= x << 6;
 290   x ^= ((unsigned)x) >> 21;
 291   x ^= x << 7 ;
 292   *a = x ;
 293   return x & 0x7FFFFFFF ;
 294 }
 295 
 296 static int Stall (int its) {
 297   static volatile jint rv = 1 ;
 298   volatile int OnFrame = 0 ;
 299   jint v = rv ^ UNS(OnFrame) ;
 300   while (--its >= 0) {
 301     v = MarsagliaXORV (v) ;
 302   }
 303   // Make this impossible for the compiler to optimize away,
 304   // but (mostly) avoid W coherency sharing on MP systems.
 305   if (v == 0x12345) rv = v ;
 306   return v ;
 307 }
 308 
 309 int Monitor::TryLock () {
 310   intptr_t v = _LockWord.FullWord ;
 311   for (;;) {
 312     if ((v & _LBIT) != 0) return 0 ;
 313     const intptr_t u = CASPTR (&_LockWord, v, v|_LBIT) ;
 314     if (v == u) return 1 ;
 315     v = u ;
 316   }
 317 }
 318 
 319 int Monitor::TryFast () {
 320   // Optimistic fast-path form ...
 321   // Fast-path attempt for the common uncontended case.
 322   // Avoid RTS->RTO $ coherence upgrade on typical SMP systems.
 323   intptr_t v = CASPTR (&_LockWord, 0, _LBIT) ;  // agro ...
 324   if (v == 0) return 1 ;
 325 
 326   for (;;) {
 327     if ((v & _LBIT) != 0) return 0 ;
 328     const intptr_t u = CASPTR (&_LockWord, v, v|_LBIT) ;
 329     if (v == u) return 1 ;
 330     v = u ;
 331   }
 332 }
 333 
 334 int Monitor::ILocked () {
 335   const intptr_t w = _LockWord.FullWord & 0xFF ;
 336   assert (w == 0 || w == _LBIT, "invariant") ;
 337   return w == _LBIT ;
 338 }
 339 
 340 // Polite TATAS spinlock with exponential backoff - bounded spin.
 341 // Ideally we'd use processor cycles, time or vtime to control
 342 // the loop, but we currently use iterations.
 343 // All the constants within were derived empirically but work over
 344 // over the spectrum of J2SE reference platforms.
 345 // On Niagara-class systems the back-off is unnecessary but
 346 // is relatively harmless.  (At worst it'll slightly retard
 347 // acquisition times).  The back-off is critical for older SMP systems
 348 // where constant fetching of the LockWord would otherwise impair
 349 // scalability.
 350 //
 351 // Clamp spinning at approximately 1/2 of a context-switch round-trip.
 352 // See synchronizer.cpp for details and rationale.
 353 
 354 int Monitor::TrySpin (Thread * const Self) {
 355   if (TryLock())    return 1 ;
 356   if (!os::is_MP()) return 0 ;
 357 
 358   int Probes  = 0 ;
 359   int Delay   = 0 ;
 360   int Steps   = 0 ;
 361   int SpinMax = NativeMonitorSpinLimit ;
 362   int flgs    = NativeMonitorFlags ;
 363   for (;;) {
 364     intptr_t v = _LockWord.FullWord;
 365     if ((v & _LBIT) == 0) {
 366       if (CASPTR (&_LockWord, v, v|_LBIT) == v) {
 367         return 1 ;
 368       }
 369       continue ;
 370     }
 371 
 372     if ((flgs & 8) == 0) {
 373       SpinPause () ;
 374     }
 375 
 376     // Periodically increase Delay -- variable Delay form
 377     // conceptually: delay *= 1 + 1/Exponent
 378     ++ Probes;
 379     if (Probes > SpinMax) return 0 ;
 380 
 381     if ((Probes & 0x7) == 0) {
 382       Delay = ((Delay << 1)|1) & 0x7FF ;
 383       // CONSIDER: Delay += 1 + (Delay/4); Delay &= 0x7FF ;
 384     }
 385 
 386     if (flgs & 2) continue ;
 387 
 388     // Consider checking _owner's schedctl state, if OFFPROC abort spin.
 389     // If the owner is OFFPROC then it's unlike that the lock will be dropped
 390     // in a timely fashion, which suggests that spinning would not be fruitful
 391     // or profitable.
 392 
 393     // Stall for "Delay" time units - iterations in the current implementation.
 394     // Avoid generating coherency traffic while stalled.
 395     // Possible ways to delay:
 396     //   PAUSE, SLEEP, MEMBAR #sync, MEMBAR #halt,
 397     //   wr %g0,%asi, gethrtime, rdstick, rdtick, rdtsc, etc. ...
 398     // Note that on Niagara-class systems we want to minimize STs in the
 399     // spin loop.  N1 and brethren write-around the L1$ over the xbar into the L2$.
 400     // Furthermore, they don't have a W$ like traditional SPARC processors.
 401     // We currently use a Marsaglia Shift-Xor RNG loop.
 402     Steps += Delay ;
 403     if (Self != NULL) {
 404       jint rv = Self->rng[0] ;
 405       for (int k = Delay ; --k >= 0; ) {
 406         rv = MarsagliaXORV (rv) ;
 407         if ((flgs & 4) == 0 && SafepointSynchronize::do_call_back()) return 0 ;
 408       }
 409       Self->rng[0] = rv ;
 410     } else {
 411       Stall (Delay) ;
 412     }
 413   }
 414 }
 415 
 416 static int ParkCommon (ParkEvent * ev, jlong timo) {
 417   // Diagnostic support - periodically unwedge blocked threads
 418   intx nmt = NativeMonitorTimeout ;
 419   if (nmt > 0 && (nmt < timo || timo <= 0)) {
 420      timo = nmt ;
 421   }
 422   int err = OS_OK ;
 423   if (0 == timo) {
 424     ev->park() ;
 425   } else {
 426     err = ev->park(timo) ;
 427   }
 428   return err ;
 429 }
 430 
 431 inline int Monitor::AcquireOrPush (ParkEvent * ESelf) {
 432   intptr_t v = _LockWord.FullWord ;
 433   for (;;) {
 434     if ((v & _LBIT) == 0) {
 435       const intptr_t u = CASPTR (&_LockWord, v, v|_LBIT) ;
 436       if (u == v) return 1 ;        // indicate acquired
 437       v = u ;
 438     } else {
 439       // Anticipate success ...
 440       ESelf->ListNext = (ParkEvent *) (v & ~_LBIT) ;
 441       const intptr_t u = CASPTR (&_LockWord, v, intptr_t(ESelf)|_LBIT) ;
 442       if (u == v) return 0 ;        // indicate pushed onto cxq
 443       v = u ;
 444     }
 445     // Interference - LockWord change - just retry
 446   }
 447 }
 448 
 449 // ILock and IWait are the lowest level primitive internal blocking
 450 // synchronization functions.  The callers of IWait and ILock must have
 451 // performed any needed state transitions beforehand.
 452 // IWait and ILock may directly call park() without any concern for thread state.
 453 // Note that ILock and IWait do *not* access _owner.
 454 // _owner is a higher-level logical concept.
 455 
 456 void Monitor::ILock (Thread * Self) {
 457   assert (_OnDeck != Self->_MutexEvent, "invariant") ;
 458 
 459   if (TryFast()) {
 460  Exeunt:
 461     assert (ILocked(), "invariant") ;
 462     return ;
 463   }
 464 
 465   ParkEvent * const ESelf = Self->_MutexEvent ;
 466   assert (_OnDeck != ESelf, "invariant") ;
 467 
 468   // As an optimization, spinners could conditionally try to set ONDECK to _LBIT
 469   // Synchronizer.cpp uses a similar optimization.
 470   if (TrySpin (Self)) goto Exeunt ;
 471 
 472   // Slow-path - the lock is contended.
 473   // Either Enqueue Self on cxq or acquire the outer lock.
 474   // LockWord encoding = (cxq,LOCKBYTE)
 475   ESelf->reset() ;
 476   OrderAccess::fence() ;
 477 
 478   // Optional optimization ... try barging on the inner lock
 479   if ((NativeMonitorFlags & 32) && CASPTR (&_OnDeck, NULL, UNS(Self)) == 0) {
 480     goto OnDeck_LOOP ;
 481   }
 482 
 483   if (AcquireOrPush (ESelf)) goto Exeunt ;
 484 
 485   // At any given time there is at most one ondeck thread.
 486   // ondeck implies not resident on cxq and not resident on EntryList
 487   // Only the OnDeck thread can try to acquire -- contended for -- the lock.
 488   // CONSIDER: use Self->OnDeck instead of m->OnDeck.
 489   // Deschedule Self so that others may run.
 490   while (_OnDeck != ESelf) {
 491     ParkCommon (ESelf, 0) ;
 492   }
 493 
 494   // Self is now in the ONDECK position and will remain so until it
 495   // manages to acquire the lock.
 496  OnDeck_LOOP:
 497   for (;;) {
 498     assert (_OnDeck == ESelf, "invariant") ;
 499     if (TrySpin (Self)) break ;
 500     // CONSIDER: if ESelf->TryPark() && TryLock() break ...
 501     // It's probably wise to spin only if we *actually* blocked
 502     // CONSIDER: check the lockbyte, if it remains set then
 503     // preemptively drain the cxq into the EntryList.
 504     // The best place and time to perform queue operations -- lock metadata --
 505     // is _before having acquired the outer lock, while waiting for the lock to drop.
 506     ParkCommon (ESelf, 0) ;
 507   }
 508 
 509   assert (_OnDeck == ESelf, "invariant") ;
 510   _OnDeck = NULL ;
 511 
 512   // Note that we current drop the inner lock (clear OnDeck) in the slow-path
 513   // epilog immediately after having acquired the outer lock.
 514   // But instead we could consider the following optimizations:
 515   // A. Shift or defer dropping the inner lock until the subsequent IUnlock() operation.
 516   //    This might avoid potential reacquisition of the inner lock in IUlock().
 517   // B. While still holding the inner lock, attempt to opportunistically select
 518   //    and unlink the next ONDECK thread from the EntryList.
 519   //    If successful, set ONDECK to refer to that thread, otherwise clear ONDECK.
 520   //    It's critical that the select-and-unlink operation run in constant-time as
 521   //    it executes when holding the outer lock and may artificially increase the
 522   //    effective length of the critical section.
 523   // Note that (A) and (B) are tantamount to succession by direct handoff for
 524   // the inner lock.
 525   goto Exeunt ;
 526 }
 527 
 528 void Monitor::IUnlock (bool RelaxAssert) {
 529   assert (ILocked(), "invariant") ;
 530   // Conceptually we need a MEMBAR #storestore|#loadstore barrier or fence immediately
 531   // before the store that releases the lock.  Crucially, all the stores and loads in the
 532   // critical section must be globally visible before the store of 0 into the lock-word
 533   // that releases the lock becomes globally visible.  That is, memory accesses in the
 534   // critical section should not be allowed to bypass or overtake the following ST that
 535   // releases the lock.  As such, to prevent accesses within the critical section
 536   // from "leaking" out, we need a release fence between the critical section and the
 537   // store that releases the lock.  In practice that release barrier is elided on
 538   // platforms with strong memory models such as TSO.
 539   //
 540   // Note that the OrderAccess::storeload() fence that appears after unlock store
 541   // provides for progress conditions and succession and is _not related to exclusion
 542   // safety or lock release consistency.
 543   OrderAccess::release_store(&_LockWord.Bytes[_LSBINDEX], 0); // drop outer lock
 544 
 545   OrderAccess::storeload ();
 546   ParkEvent * const w = _OnDeck ;
 547   assert (RelaxAssert || w != Thread::current()->_MutexEvent, "invariant") ;
 548   if (w != NULL) {
 549     // Either we have a valid ondeck thread or ondeck is transiently "locked"
 550     // by some exiting thread as it arranges for succession.  The LSBit of
 551     // OnDeck allows us to discriminate two cases.  If the latter, the
 552     // responsibility for progress and succession lies with that other thread.
 553     // For good performance, we also depend on the fact that redundant unpark()
 554     // operations are cheap.  That is, repeated Unpark()ing of the ONDECK thread
 555     // is inexpensive.  This approach provides implicit futile wakeup throttling.
 556     // Note that the referent "w" might be stale with respect to the lock.
 557     // In that case the following unpark() is harmless and the worst that'll happen
 558     // is a spurious return from a park() operation.  Critically, if "w" _is stale,
 559     // then progress is known to have occurred as that means the thread associated
 560     // with "w" acquired the lock.  In that case this thread need take no further
 561     // action to guarantee progress.
 562     if ((UNS(w) & _LBIT) == 0) w->unpark() ;
 563     return ;
 564   }
 565 
 566   intptr_t cxq = _LockWord.FullWord ;
 567   if (((cxq & ~_LBIT)|UNS(_EntryList)) == 0) {
 568     return ;      // normal fast-path exit - cxq and EntryList both empty
 569   }
 570   if (cxq & _LBIT) {
 571     // Optional optimization ...
 572     // Some other thread acquired the lock in the window since this
 573     // thread released it.  Succession is now that thread's responsibility.
 574     return ;
 575   }
 576 
 577  Succession:
 578   // Slow-path exit - this thread must ensure succession and progress.
 579   // OnDeck serves as lock to protect cxq and EntryList.
 580   // Only the holder of OnDeck can manipulate EntryList or detach the RATs from cxq.
 581   // Avoid ABA - allow multiple concurrent producers (enqueue via push-CAS)
 582   // but only one concurrent consumer (detacher of RATs).
 583   // Consider protecting this critical section with schedctl on Solaris.
 584   // Unlike a normal lock, however, the exiting thread "locks" OnDeck,
 585   // picks a successor and marks that thread as OnDeck.  That successor
 586   // thread will then clear OnDeck once it eventually acquires the outer lock.
 587   if (CASPTR (&_OnDeck, NULL, _LBIT) != UNS(NULL)) {
 588     return ;
 589   }
 590 
 591   ParkEvent * List = _EntryList ;
 592   if (List != NULL) {
 593     // Transfer the head of the EntryList to the OnDeck position.
 594     // Once OnDeck, a thread stays OnDeck until it acquires the lock.
 595     // For a given lock there is at most OnDeck thread at any one instant.
 596    WakeOne:
 597     assert (List == _EntryList, "invariant") ;
 598     ParkEvent * const w = List ;
 599     assert (RelaxAssert || w != Thread::current()->_MutexEvent, "invariant") ;
 600     _EntryList = w->ListNext ;
 601     // as a diagnostic measure consider setting w->_ListNext = BAD
 602     assert (UNS(_OnDeck) == _LBIT, "invariant") ;
 603     _OnDeck = w ;           // pass OnDeck to w.
 604                             // w will clear OnDeck once it acquires the outer lock
 605 
 606     // Another optional optimization ...
 607     // For heavily contended locks it's not uncommon that some other
 608     // thread acquired the lock while this thread was arranging succession.
 609     // Try to defer the unpark() operation - Delegate the responsibility
 610     // for unpark()ing the OnDeck thread to the current or subsequent owners
 611     // That is, the new owner is responsible for unparking the OnDeck thread.
 612     OrderAccess::storeload() ;
 613     cxq = _LockWord.FullWord ;
 614     if (cxq & _LBIT) return ;
 615 
 616     w->unpark() ;
 617     return ;
 618   }
 619 
 620   cxq = _LockWord.FullWord ;
 621   if ((cxq & ~_LBIT) != 0) {
 622     // The EntryList is empty but the cxq is populated.
 623     // drain RATs from cxq into EntryList
 624     // Detach RATs segment with CAS and then merge into EntryList
 625     for (;;) {
 626       // optional optimization - if locked, the owner is responsible for succession
 627       if (cxq & _LBIT) goto Punt ;
 628       const intptr_t vfy = CASPTR (&_LockWord, cxq, cxq & _LBIT) ;
 629       if (vfy == cxq) break ;
 630       cxq = vfy ;
 631       // Interference - LockWord changed - Just retry
 632       // We can see concurrent interference from contending threads
 633       // pushing themselves onto the cxq or from lock-unlock operations.
 634       // From the perspective of this thread, EntryList is stable and
 635       // the cxq is prepend-only -- the head is volatile but the interior
 636       // of the cxq is stable.  In theory if we encounter interference from threads
 637       // pushing onto cxq we could simply break off the original cxq suffix and
 638       // move that segment to the EntryList, avoiding a 2nd or multiple CAS attempts
 639       // on the high-traffic LockWord variable.   For instance lets say the cxq is "ABCD"
 640       // when we first fetch cxq above.  Between the fetch -- where we observed "A"
 641       // -- and CAS -- where we attempt to CAS null over A -- "PQR" arrive,
 642       // yielding cxq = "PQRABCD".  In this case we could simply set A.ListNext
 643       // null, leaving cxq = "PQRA" and transfer the "BCD" segment to the EntryList.
 644       // Note too, that it's safe for this thread to traverse the cxq
 645       // without taking any special concurrency precautions.
 646     }
 647 
 648     // We don't currently reorder the cxq segment as we move it onto
 649     // the EntryList, but it might make sense to reverse the order
 650     // or perhaps sort by thread priority.  See the comments in
 651     // synchronizer.cpp objectMonitor::exit().
 652     assert (_EntryList == NULL, "invariant") ;
 653     _EntryList = List = (ParkEvent *)(cxq & ~_LBIT) ;
 654     assert (List != NULL, "invariant") ;
 655     goto WakeOne ;
 656   }
 657 
 658   // cxq|EntryList is empty.
 659   // w == NULL implies that cxq|EntryList == NULL in the past.
 660   // Possible race - rare inopportune interleaving.
 661   // A thread could have added itself to cxq since this thread previously checked.
 662   // Detect and recover by refetching cxq.
 663  Punt:
 664   assert (UNS(_OnDeck) == _LBIT, "invariant") ;
 665   _OnDeck = NULL ;            // Release inner lock.
 666   OrderAccess::storeload();   // Dekker duality - pivot point
 667 
 668   // Resample LockWord/cxq to recover from possible race.
 669   // For instance, while this thread T1 held OnDeck, some other thread T2 might
 670   // acquire the outer lock.  Another thread T3 might try to acquire the outer
 671   // lock, but encounter contention and enqueue itself on cxq.  T2 then drops the
 672   // outer lock, but skips succession as this thread T1 still holds OnDeck.
 673   // T1 is and remains responsible for ensuring succession of T3.
 674   //
 675   // Note that we don't need to recheck EntryList, just cxq.
 676   // If threads moved onto EntryList since we dropped OnDeck
 677   // that implies some other thread forced succession.
 678   cxq = _LockWord.FullWord ;
 679   if ((cxq & ~_LBIT) != 0 && (cxq & _LBIT) == 0) {
 680     goto Succession ;         // potential race -- re-run succession
 681   }
 682   return ;
 683 }
 684 
 685 bool Monitor::notify() {
 686   assert (_owner == Thread::current(), "invariant") ;
 687   assert (ILocked(), "invariant") ;
 688   if (_WaitSet == NULL) return true ;
 689   NotifyCount ++ ;
 690 
 691   // Transfer one thread from the WaitSet to the EntryList or cxq.
 692   // Currently we just unlink the head of the WaitSet and prepend to the cxq.
 693   // And of course we could just unlink it and unpark it, too, but
 694   // in that case it'd likely impale itself on the reentry.
 695   Thread::muxAcquire (_WaitLock, "notify:WaitLock") ;
 696   ParkEvent * nfy = _WaitSet ;
 697   if (nfy != NULL) {                  // DCL idiom
 698     _WaitSet = nfy->ListNext ;
 699     assert (nfy->Notified == 0, "invariant") ;
 700     // push nfy onto the cxq
 701     for (;;) {
 702       const intptr_t v = _LockWord.FullWord ;
 703       assert ((v & 0xFF) == _LBIT, "invariant") ;
 704       nfy->ListNext = (ParkEvent *)(v & ~_LBIT);
 705       if (CASPTR (&_LockWord, v, UNS(nfy)|_LBIT) == v) break;
 706       // interference - _LockWord changed -- just retry
 707     }
 708     // Note that setting Notified before pushing nfy onto the cxq is
 709     // also legal and safe, but the safety properties are much more
 710     // subtle, so for the sake of code stewardship ...
 711     OrderAccess::fence() ;
 712     nfy->Notified = 1;
 713   }
 714   Thread::muxRelease (_WaitLock) ;
 715   if (nfy != NULL && (NativeMonitorFlags & 16)) {
 716     // Experimental code ... light up the wakee in the hope that this thread (the owner)
 717     // will drop the lock just about the time the wakee comes ONPROC.
 718     nfy->unpark() ;
 719   }
 720   assert (ILocked(), "invariant") ;
 721   return true ;
 722 }
 723 
 724 // Currently notifyAll() transfers the waiters one-at-a-time from the waitset
 725 // to the cxq.  This could be done more efficiently with a single bulk en-mass transfer,
 726 // but in practice notifyAll() for large #s of threads is rare and not time-critical.
 727 // Beware too, that we invert the order of the waiters.  Lets say that the
 728 // waitset is "ABCD" and the cxq is "XYZ".  After a notifyAll() the waitset
 729 // will be empty and the cxq will be "DCBAXYZ".  This is benign, of course.
 730 
 731 bool Monitor::notify_all() {
 732   assert (_owner == Thread::current(), "invariant") ;
 733   assert (ILocked(), "invariant") ;
 734   while (_WaitSet != NULL) notify() ;
 735   return true ;
 736 }
 737 
 738 int Monitor::IWait (Thread * Self, jlong timo) {
 739   assert (ILocked(), "invariant") ;
 740 
 741   // Phases:
 742   // 1. Enqueue Self on WaitSet - currently prepend
 743   // 2. unlock - drop the outer lock
 744   // 3. wait for either notification or timeout
 745   // 4. lock - reentry - reacquire the outer lock
 746 
 747   ParkEvent * const ESelf = Self->_MutexEvent ;
 748   ESelf->Notified = 0 ;
 749   ESelf->reset() ;
 750   OrderAccess::fence() ;
 751 
 752   // Add Self to WaitSet
 753   // Ideally only the holder of the outer lock would manipulate the WaitSet -
 754   // That is, the outer lock would implicitly protect the WaitSet.
 755   // But if a thread in wait() encounters a timeout it will need to dequeue itself
 756   // from the WaitSet _before it becomes the owner of the lock.  We need to dequeue
 757   // as the ParkEvent -- which serves as a proxy for the thread -- can't reside
 758   // on both the WaitSet and the EntryList|cxq at the same time..  That is, a thread
 759   // on the WaitSet can't be allowed to compete for the lock until it has managed to
 760   // unlink its ParkEvent from WaitSet.  Thus the need for WaitLock.
 761   // Contention on the WaitLock is minimal.
 762   //
 763   // Another viable approach would be add another ParkEvent, "WaitEvent" to the
 764   // thread class.  The WaitSet would be composed of WaitEvents.  Only the
 765   // owner of the outer lock would manipulate the WaitSet.  A thread in wait()
 766   // could then compete for the outer lock, and then, if necessary, unlink itself
 767   // from the WaitSet only after having acquired the outer lock.  More precisely,
 768   // there would be no WaitLock.  A thread in in wait() would enqueue its WaitEvent
 769   // on the WaitSet; release the outer lock; wait for either notification or timeout;
 770   // reacquire the inner lock; and then, if needed, unlink itself from the WaitSet.
 771   //
 772   // Alternatively, a 2nd set of list link fields in the ParkEvent might suffice.
 773   // One set would be for the WaitSet and one for the EntryList.
 774   // We could also deconstruct the ParkEvent into a "pure" event and add a
 775   // new immortal/TSM "ListElement" class that referred to ParkEvents.
 776   // In that case we could have one ListElement on the WaitSet and another
 777   // on the EntryList, with both referring to the same pure Event.
 778 
 779   Thread::muxAcquire (_WaitLock, "wait:WaitLock:Add") ;
 780   ESelf->ListNext = _WaitSet ;
 781   _WaitSet = ESelf ;
 782   Thread::muxRelease (_WaitLock) ;
 783 
 784   // Release the outer lock
 785   // We call IUnlock (RelaxAssert=true) as a thread T1 might
 786   // enqueue itself on the WaitSet, call IUnlock(), drop the lock,
 787   // and then stall before it can attempt to wake a successor.
 788   // Some other thread T2 acquires the lock, and calls notify(), moving
 789   // T1 from the WaitSet to the cxq.  T2 then drops the lock.  T1 resumes,
 790   // and then finds *itself* on the cxq.  During the course of a normal
 791   // IUnlock() call a thread should _never find itself on the EntryList
 792   // or cxq, but in the case of wait() it's possible.
 793   // See synchronizer.cpp objectMonitor::wait().
 794   IUnlock (true) ;
 795 
 796   // Wait for either notification or timeout
 797   // Beware that in some circumstances we might propagate
 798   // spurious wakeups back to the caller.
 799 
 800   for (;;) {
 801     if (ESelf->Notified) break ;
 802     int err = ParkCommon (ESelf, timo) ;
 803     if (err == OS_TIMEOUT || (NativeMonitorFlags & 1)) break ;
 804   }
 805 
 806   // Prepare for reentry - if necessary, remove ESelf from WaitSet
 807   // ESelf can be:
 808   // 1. Still on the WaitSet.  This can happen if we exited the loop by timeout.
 809   // 2. On the cxq or EntryList
 810   // 3. Not resident on cxq, EntryList or WaitSet, but in the OnDeck position.
 811 
 812   OrderAccess::fence() ;
 813   int WasOnWaitSet = 0 ;
 814   if (ESelf->Notified == 0) {
 815     Thread::muxAcquire (_WaitLock, "wait:WaitLock:remove") ;
 816     if (ESelf->Notified == 0) {     // DCL idiom
 817       assert (_OnDeck != ESelf, "invariant") ;   // can't be both OnDeck and on WaitSet
 818       // ESelf is resident on the WaitSet -- unlink it.
 819       // A doubly-linked list would be better here so we can unlink in constant-time.
 820       // We have to unlink before we potentially recontend as ESelf might otherwise
 821       // end up on the cxq|EntryList -- it can't be on two lists at once.
 822       ParkEvent * p = _WaitSet ;
 823       ParkEvent * q = NULL ;            // classic q chases p
 824       while (p != NULL && p != ESelf) {
 825         q = p ;
 826         p = p->ListNext ;
 827       }
 828       assert (p == ESelf, "invariant") ;
 829       if (p == _WaitSet) {      // found at head
 830         assert (q == NULL, "invariant") ;
 831         _WaitSet = p->ListNext ;
 832       } else {                  // found in interior
 833         assert (q->ListNext == p, "invariant") ;
 834         q->ListNext = p->ListNext ;
 835       }
 836       WasOnWaitSet = 1 ;        // We were *not* notified but instead encountered timeout
 837     }
 838     Thread::muxRelease (_WaitLock) ;
 839   }
 840 
 841   // Reentry phase - reacquire the lock
 842   if (WasOnWaitSet) {
 843     // ESelf was previously on the WaitSet but we just unlinked it above
 844     // because of a timeout.  ESelf is not resident on any list and is not OnDeck
 845     assert (_OnDeck != ESelf, "invariant") ;
 846     ILock (Self) ;
 847   } else {
 848     // A prior notify() operation moved ESelf from the WaitSet to the cxq.
 849     // ESelf is now on the cxq, EntryList or at the OnDeck position.
 850     // The following fragment is extracted from Monitor::ILock()
 851     for (;;) {
 852       if (_OnDeck == ESelf && TrySpin(Self)) break ;
 853       ParkCommon (ESelf, 0) ;
 854     }
 855     assert (_OnDeck == ESelf, "invariant") ;
 856     _OnDeck = NULL ;
 857   }
 858 
 859   assert (ILocked(), "invariant") ;
 860   return WasOnWaitSet != 0 ;        // return true IFF timeout
 861 }
 862 
 863 
 864 // ON THE VMTHREAD SNEAKING PAST HELD LOCKS:
 865 // In particular, there are certain types of global lock that may be held
 866 // by a Java thread while it is blocked at a safepoint but before it has
 867 // written the _owner field. These locks may be sneakily acquired by the
 868 // VM thread during a safepoint to avoid deadlocks. Alternatively, one should
 869 // identify all such locks, and ensure that Java threads never block at
 870 // safepoints while holding them (_no_safepoint_check_flag). While it
 871 // seems as though this could increase the time to reach a safepoint
 872 // (or at least increase the mean, if not the variance), the latter
 873 // approach might make for a cleaner, more maintainable JVM design.
 874 //
 875 // Sneaking is vile and reprehensible and should be excised at the 1st
 876 // opportunity.  It's possible that the need for sneaking could be obviated
 877 // as follows.  Currently, a thread might (a) while TBIVM, call pthread_mutex_lock
 878 // or ILock() thus acquiring the "physical" lock underlying Monitor/Mutex.
 879 // (b) stall at the TBIVM exit point as a safepoint is in effect.  Critically,
 880 // it'll stall at the TBIVM reentry state transition after having acquired the
 881 // underlying lock, but before having set _owner and having entered the actual
 882 // critical section.  The lock-sneaking facility leverages that fact and allowed the
 883 // VM thread to logically acquire locks that had already be physically locked by mutators
 884 // but where mutators were known blocked by the reentry thread state transition.
 885 //
 886 // If we were to modify the Monitor-Mutex so that TBIVM state transitions tightly
 887 // wrapped calls to park(), then we could likely do away with sneaking.  We'd
 888 // decouple lock acquisition and parking.  The critical invariant  to eliminating
 889 // sneaking is to ensure that we never "physically" acquire the lock while TBIVM.
 890 // An easy way to accomplish this is to wrap the park calls in a narrow TBIVM jacket.
 891 // One difficulty with this approach is that the TBIVM wrapper could recurse and
 892 // call lock() deep from within a lock() call, while the MutexEvent was already enqueued.
 893 // Using a stack (N=2 at minimum) of ParkEvents would take care of that problem.
 894 //
 895 // But of course the proper ultimate approach is to avoid schemes that require explicit
 896 // sneaking or dependence on any any clever invariants or subtle implementation properties
 897 // of Mutex-Monitor and instead directly address the underlying design flaw.
 898 
 899 void Monitor::lock (Thread * Self) {
 900 #ifdef CHECK_UNHANDLED_OOPS
 901   // Clear unhandled oops so we get a crash right away.  Only clear for non-vm
 902   // or GC threads.
 903   if (Self->is_Java_thread()) {
 904     Self->clear_unhandled_oops();
 905   }
 906 #endif // CHECK_UNHANDLED_OOPS
 907 
 908   debug_only(check_prelock_state(Self));
 909   assert (_owner != Self              , "invariant") ;
 910   assert (_OnDeck != Self->_MutexEvent, "invariant") ;
 911 
 912   if (TryFast()) {
 913  Exeunt:
 914     assert (ILocked(), "invariant") ;
 915     assert (owner() == NULL, "invariant");
 916     set_owner (Self);
 917     return ;
 918   }
 919 
 920   // The lock is contended ...
 921 
 922   bool can_sneak = Self->is_VM_thread() && SafepointSynchronize::is_at_safepoint();
 923   if (can_sneak && _owner == NULL) {
 924     // a java thread has locked the lock but has not entered the
 925     // critical region -- let's just pretend we've locked the lock
 926     // and go on.  we note this with _snuck so we can also
 927     // pretend to unlock when the time comes.
 928     _snuck = true;
 929     goto Exeunt ;
 930   }
 931 
 932   // Try a brief spin to avoid passing thru thread state transition ...
 933   if (TrySpin (Self)) goto Exeunt ;
 934 
 935   check_block_state(Self);
 936   if (Self->is_Java_thread()) {
 937     // Horribile dictu - we suffer through a state transition
 938     assert(rank() > Mutex::special, "Potential deadlock with special or lesser rank mutex");
 939     ThreadBlockInVM tbivm ((JavaThread *) Self) ;
 940     ILock (Self) ;
 941   } else {
 942     // Mirabile dictu
 943     ILock (Self) ;
 944   }
 945   goto Exeunt ;
 946 }
 947 
 948 void Monitor::lock() {
 949   this->lock(Thread::current());
 950 }
 951 
 952 // Lock without safepoint check - a degenerate variant of lock().
 953 // Should ONLY be used by safepoint code and other code
 954 // that is guaranteed not to block while running inside the VM. If this is called with
 955 // thread state set to be in VM, the safepoint synchronization code will deadlock!
 956 
 957 void Monitor::lock_without_safepoint_check (Thread * Self) {
 958   assert (_owner != Self, "invariant") ;
 959   ILock (Self) ;
 960   assert (_owner == NULL, "invariant");
 961   set_owner (Self);
 962 }
 963 
 964 void Monitor::lock_without_safepoint_check () {
 965   lock_without_safepoint_check (Thread::current()) ;
 966 }
 967 
 968 
 969 // Returns true if thread succeceed [sic] in grabbing the lock, otherwise false.
 970 
 971 bool Monitor::try_lock() {
 972   Thread * const Self = Thread::current();
 973   debug_only(check_prelock_state(Self));
 974   // assert(!thread->is_inside_signal_handler(), "don't lock inside signal handler");
 975 
 976   // Special case, where all Java threads are stopped.
 977   // The lock may have been acquired but _owner is not yet set.
 978   // In that case the VM thread can safely grab the lock.
 979   // It strikes me this should appear _after the TryLock() fails, below.
 980   bool can_sneak = Self->is_VM_thread() && SafepointSynchronize::is_at_safepoint();
 981   if (can_sneak && _owner == NULL) {
 982     set_owner(Self); // Do not need to be atomic, since we are at a safepoint
 983     _snuck = true;
 984     return true;
 985   }
 986 
 987   if (TryLock()) {
 988     // We got the lock
 989     assert (_owner == NULL, "invariant");
 990     set_owner (Self);
 991     return true;
 992   }
 993   return false;
 994 }
 995 
 996 void Monitor::unlock() {
 997   assert (_owner  == Thread::current(), "invariant") ;
 998   assert (_OnDeck != Thread::current()->_MutexEvent , "invariant") ;
 999   set_owner (NULL) ;
1000   if (_snuck) {
1001     assert(SafepointSynchronize::is_at_safepoint() && Thread::current()->is_VM_thread(), "sneak");
1002     _snuck = false;
1003     return ;
1004   }
1005   IUnlock (false) ;
1006 }
1007 
1008 // Yet another degenerate version of Monitor::lock() or lock_without_safepoint_check()
1009 // jvm_raw_lock() and _unlock() can be called by non-Java threads via JVM_RawMonitorEnter.
1010 //
1011 // There's no expectation that JVM_RawMonitors will interoperate properly with the native
1012 // Mutex-Monitor constructs.  We happen to implement JVM_RawMonitors in terms of
1013 // native Mutex-Monitors simply as a matter of convenience.  A simple abstraction layer
1014 // over a pthread_mutex_t would work equally as well, but require more platform-specific
1015 // code -- a "PlatformMutex".  Alternatively, a simply layer over muxAcquire-muxRelease
1016 // would work too.
1017 //
1018 // Since the caller might be a foreign thread, we don't necessarily have a Thread.MutexEvent
1019 // instance available.  Instead, we transiently allocate a ParkEvent on-demand if
1020 // we encounter contention.  That ParkEvent remains associated with the thread
1021 // until it manages to acquire the lock, at which time we return the ParkEvent
1022 // to the global ParkEvent free list.  This is correct and suffices for our purposes.
1023 //
1024 // Beware that the original jvm_raw_unlock() had a "_snuck" test but that
1025 // jvm_raw_lock() didn't have the corresponding test.  I suspect that's an
1026 // oversight, but I've replicated the original suspect logic in the new code ...
1027 
1028 void Monitor::jvm_raw_lock() {
1029   assert(rank() == native, "invariant");
1030 
1031   if (TryLock()) {
1032  Exeunt:
1033     assert (ILocked(), "invariant") ;
1034     assert (_owner == NULL, "invariant");
1035     // This can potentially be called by non-java Threads. Thus, the ThreadLocalStorage
1036     // might return NULL. Don't call set_owner since it will break on an NULL owner
1037     // Consider installing a non-null "ANON" distinguished value instead of just NULL.
1038     _owner = ThreadLocalStorage::thread();
1039     return ;
1040   }
1041 
1042   if (TrySpin(NULL)) goto Exeunt ;
1043 
1044   // slow-path - apparent contention
1045   // Allocate a ParkEvent for transient use.
1046   // The ParkEvent remains associated with this thread until
1047   // the time the thread manages to acquire the lock.
1048   ParkEvent * const ESelf = ParkEvent::Allocate(NULL) ;
1049   ESelf->reset() ;
1050   OrderAccess::storeload() ;
1051 
1052   // Either Enqueue Self on cxq or acquire the outer lock.
1053   if (AcquireOrPush (ESelf)) {
1054     ParkEvent::Release (ESelf) ;      // surrender the ParkEvent
1055     goto Exeunt ;
1056   }
1057 
1058   // At any given time there is at most one ondeck thread.
1059   // ondeck implies not resident on cxq and not resident on EntryList
1060   // Only the OnDeck thread can try to acquire -- contended for -- the lock.
1061   // CONSIDER: use Self->OnDeck instead of m->OnDeck.
1062   for (;;) {
1063     if (_OnDeck == ESelf && TrySpin(NULL)) break ;
1064     ParkCommon (ESelf, 0) ;
1065   }
1066 
1067   assert (_OnDeck == ESelf, "invariant") ;
1068   _OnDeck = NULL ;
1069   ParkEvent::Release (ESelf) ;      // surrender the ParkEvent
1070   goto Exeunt ;
1071 }
1072 
1073 void Monitor::jvm_raw_unlock() {
1074   // Nearly the same as Monitor::unlock() ...
1075   // directly set _owner instead of using set_owner(null)
1076   _owner = NULL ;
1077   if (_snuck) {         // ???
1078     assert(SafepointSynchronize::is_at_safepoint() && Thread::current()->is_VM_thread(), "sneak");
1079     _snuck = false;
1080     return ;
1081   }
1082   IUnlock(false) ;
1083 }
1084 
1085 bool Monitor::wait(bool no_safepoint_check, long timeout, bool as_suspend_equivalent) {
1086   Thread * const Self = Thread::current() ;
1087   assert (_owner == Self, "invariant") ;
1088   assert (ILocked(), "invariant") ;
1089 
1090   // as_suspend_equivalent logically implies !no_safepoint_check
1091   guarantee (!as_suspend_equivalent || !no_safepoint_check, "invariant") ;
1092   // !no_safepoint_check logically implies java_thread
1093   guarantee (no_safepoint_check || Self->is_Java_thread(), "invariant") ;
1094 
1095   #ifdef ASSERT
1096     Monitor * least = get_least_ranked_lock_besides_this(Self->owned_locks());
1097     assert(least != this, "Specification of get_least_... call above");
1098     if (least != NULL && least->rank() <= special) {
1099       tty->print("Attempting to wait on monitor %s/%d while holding"
1100                  " lock %s/%d -- possible deadlock",
1101                  name(), rank(), least->name(), least->rank());
1102       assert(false, "Shouldn't block(wait) while holding a lock of rank special");
1103     }
1104   #endif // ASSERT
1105 
1106   int wait_status ;
1107   // conceptually set the owner to NULL in anticipation of
1108   // abdicating the lock in wait
1109   set_owner(NULL);
1110   if (no_safepoint_check) {
1111     wait_status = IWait (Self, timeout) ;
1112   } else {
1113     assert (Self->is_Java_thread(), "invariant") ;
1114     JavaThread *jt = (JavaThread *)Self;
1115 
1116     // Enter safepoint region - ornate and Rococo ...
1117     ThreadBlockInVM tbivm(jt);
1118     OSThreadWaitState osts(Self->osthread(), false /* not Object.wait() */);
1119 
1120     if (as_suspend_equivalent) {
1121       jt->set_suspend_equivalent();
1122       // cleared by handle_special_suspend_equivalent_condition() or
1123       // java_suspend_self()
1124     }
1125 
1126     wait_status = IWait (Self, timeout) ;
1127 
1128     // were we externally suspended while we were waiting?
1129     if (as_suspend_equivalent && jt->handle_special_suspend_equivalent_condition()) {
1130       // Our event wait has finished and we own the lock, but
1131       // while we were waiting another thread suspended us. We don't
1132       // want to hold the lock while suspended because that
1133       // would surprise the thread that suspended us.
1134       assert (ILocked(), "invariant") ;
1135       IUnlock (true) ;
1136       jt->java_suspend_self();
1137       ILock (Self) ;
1138       assert (ILocked(), "invariant") ;
1139     }
1140   }
1141 
1142   // Conceptually reestablish ownership of the lock.
1143   // The "real" lock -- the LockByte -- was reacquired by IWait().
1144   assert (ILocked(), "invariant") ;
1145   assert (_owner == NULL, "invariant") ;
1146   set_owner (Self) ;
1147   return wait_status != 0 ;          // return true IFF timeout
1148 }
1149 
1150 Monitor::~Monitor() {
1151   assert ((UNS(_owner)|UNS(_LockWord.FullWord)|UNS(_EntryList)|UNS(_WaitSet)|UNS(_OnDeck)) == 0, "") ;
1152 }
1153 
1154 void Monitor::ClearMonitor (Monitor * m, const char *name) {
1155   m->_owner             = NULL ;
1156   m->_snuck             = false ;
1157   if (name == NULL) {
1158     strcpy(m->_name, "UNKNOWN") ;
1159   } else {
1160     strncpy(m->_name, name, MONITOR_NAME_LEN - 1);
1161     m->_name[MONITOR_NAME_LEN - 1] = '\0';
1162   }
1163   m->_LockWord.FullWord = 0 ;
1164   m->_EntryList         = NULL ;
1165   m->_OnDeck            = NULL ;
1166   m->_WaitSet           = NULL ;
1167   m->_WaitLock[0]       = 0 ;
1168 }
1169 
1170 Monitor::Monitor() { ClearMonitor(this); }
1171 
1172 Monitor::Monitor (int Rank, const char * name, bool allow_vm_block) {
1173   ClearMonitor (this, name) ;
1174 #ifdef ASSERT
1175   _allow_vm_block  = allow_vm_block;
1176   _rank            = Rank ;
1177 #endif
1178 }
1179 
1180 Mutex::~Mutex() {
1181   assert ((UNS(_owner)|UNS(_LockWord.FullWord)|UNS(_EntryList)|UNS(_WaitSet)|UNS(_OnDeck)) == 0, "") ;
1182 }
1183 
1184 Mutex::Mutex (int Rank, const char * name, bool allow_vm_block) {
1185   ClearMonitor ((Monitor *) this, name) ;
1186 #ifdef ASSERT
1187  _allow_vm_block   = allow_vm_block;
1188  _rank             = Rank ;
1189 #endif
1190 }
1191 
1192 bool Monitor::owned_by_self() const {
1193   bool ret = _owner == Thread::current();
1194   assert (!ret || _LockWord.Bytes[_LSBINDEX] != 0, "invariant") ;
1195   return ret;
1196 }
1197 
1198 void Monitor::print_on_error(outputStream* st) const {
1199   st->print("[" PTR_FORMAT, this);
1200   st->print("] %s", _name);
1201   st->print(" - owner thread: " PTR_FORMAT, _owner);
1202 }
1203 
1204 
1205 
1206 
1207 // ----------------------------------------------------------------------------------
1208 // Non-product code
1209 
1210 #ifndef PRODUCT
1211 void Monitor::print_on(outputStream* st) const {
1212   st->print_cr("Mutex: [0x%lx/0x%lx] %s - owner: 0x%lx", this, _LockWord.FullWord, _name, _owner);
1213 }
1214 #endif
1215 
1216 #ifndef PRODUCT
1217 #ifdef ASSERT
1218 Monitor * Monitor::get_least_ranked_lock(Monitor * locks) {
1219   Monitor *res, *tmp;
1220   for (res = tmp = locks; tmp != NULL; tmp = tmp->next()) {
1221     if (tmp->rank() < res->rank()) {
1222       res = tmp;
1223     }
1224   }
1225   if (!SafepointSynchronize::is_at_safepoint()) {
1226     // In this case, we expect the held locks to be
1227     // in increasing rank order (modulo any native ranks)
1228     for (tmp = locks; tmp != NULL; tmp = tmp->next()) {
1229       if (tmp->next() != NULL) {
1230         assert(tmp->rank() == Mutex::native ||
1231                tmp->rank() <= tmp->next()->rank(), "mutex rank anomaly?");
1232       }
1233     }
1234   }
1235   return res;
1236 }
1237 
1238 Monitor* Monitor::get_least_ranked_lock_besides_this(Monitor* locks) {
1239   Monitor *res, *tmp;
1240   for (res = NULL, tmp = locks; tmp != NULL; tmp = tmp->next()) {
1241     if (tmp != this && (res == NULL || tmp->rank() < res->rank())) {
1242       res = tmp;
1243     }
1244   }
1245   if (!SafepointSynchronize::is_at_safepoint()) {
1246     // In this case, we expect the held locks to be
1247     // in increasing rank order (modulo any native ranks)
1248     for (tmp = locks; tmp != NULL; tmp = tmp->next()) {
1249       if (tmp->next() != NULL) {
1250         assert(tmp->rank() == Mutex::native ||
1251                tmp->rank() <= tmp->next()->rank(), "mutex rank anomaly?");
1252       }
1253     }
1254   }
1255   return res;
1256 }
1257 
1258 
1259 bool Monitor::contains(Monitor* locks, Monitor * lock) {
1260   for (; locks != NULL; locks = locks->next()) {
1261     if (locks == lock)
1262       return true;
1263   }
1264   return false;
1265 }
1266 #endif
1267 
1268 // Called immediately after lock acquisition or release as a diagnostic
1269 // to track the lock-set of the thread and test for rank violations that
1270 // might indicate exposure to deadlock.
1271 // Rather like an EventListener for _owner (:>).
1272 
1273 void Monitor::set_owner_implementation(Thread *new_owner) {
1274   // This function is solely responsible for maintaining
1275   // and checking the invariant that threads and locks
1276   // are in a 1/N relation, with some some locks unowned.
1277   // It uses the Mutex::_owner, Mutex::_next, and
1278   // Thread::_owned_locks fields, and no other function
1279   // changes those fields.
1280   // It is illegal to set the mutex from one non-NULL
1281   // owner to another--it must be owned by NULL as an
1282   // intermediate state.
1283 
1284   if (new_owner != NULL) {
1285     // the thread is acquiring this lock
1286 
1287     assert(new_owner == Thread::current(), "Should I be doing this?");
1288     assert(_owner == NULL, "setting the owner thread of an already owned mutex");
1289     _owner = new_owner; // set the owner
1290 
1291     // link "this" into the owned locks list
1292 
1293     #ifdef ASSERT  // Thread::_owned_locks is under the same ifdef
1294       Monitor* locks = get_least_ranked_lock(new_owner->owned_locks());
1295                     // Mutex::set_owner_implementation is a friend of Thread
1296 
1297       assert(this->rank() >= 0, "bad lock rank");
1298 
1299       // Deadlock avoidance rules require us to acquire Mutexes only in
1300       // a global total order. For example m1 is the lowest ranked mutex
1301       // that the thread holds and m2 is the mutex the thread is trying
1302       // to acquire, then  deadlock avoidance rules require that the rank
1303       // of m2 be less  than the rank of m1.
1304       // The rank Mutex::native  is an exception in that it is not subject
1305       // to the verification rules.
1306       // Here are some further notes relating to mutex acquisition anomalies:
1307       // . under Solaris, the interrupt lock gets acquired when doing
1308       //   profiling, so any lock could be held.
1309       // . it is also ok to acquire Safepoint_lock at the very end while we
1310       //   already hold Terminator_lock - may happen because of periodic safepoints
1311       if (this->rank() != Mutex::native &&
1312           this->rank() != Mutex::suspend_resume &&
1313           locks != NULL && locks->rank() <= this->rank() &&
1314           !SafepointSynchronize::is_at_safepoint() &&
1315           this != Interrupt_lock && 
1316           !(this == Safepoint_lock && contains(locks, Terminator_lock) && 
1317             SafepointSynchronize::is_synchronizing())) {
1318         new_owner->print_owned_locks();
1319         fatal(err_msg("acquiring lock %s/%d out of order with lock %s/%d -- "
1320                       "possible deadlock", this->name(), this->rank(),
1321                       locks->name(), locks->rank()));
1322       }
1323 
1324       this->_next = new_owner->_owned_locks;
1325       new_owner->_owned_locks = this;
1326     #endif
1327 
1328   } else {
1329     // the thread is releasing this lock
1330 
1331     Thread* old_owner = _owner;
1332     debug_only(_last_owner = old_owner);
1333 
1334     assert(old_owner != NULL, "removing the owner thread of an unowned mutex");
1335     assert(old_owner == Thread::current(), "removing the owner thread of an unowned mutex");
1336 
1337     _owner = NULL; // set the owner
1338 
1339     #ifdef ASSERT
1340       Monitor *locks = old_owner->owned_locks();
1341 
1342       // remove "this" from the owned locks list
1343 
1344       Monitor *prev = NULL;
1345       bool found = false;
1346       for (; locks != NULL; prev = locks, locks = locks->next()) {
1347         if (locks == this) {
1348           found = true;
1349           break;
1350         }
1351       }
1352       assert(found, "Removing a lock not owned");
1353       if (prev == NULL) {
1354         old_owner->_owned_locks = _next;
1355       } else {
1356         prev->_next = _next;
1357       }
1358       _next = NULL;
1359     #endif
1360   }
1361 }
1362 
1363 
1364 // Factored out common sanity checks for locking mutex'es. Used by lock() and try_lock()
1365 void Monitor::check_prelock_state(Thread *thread) {
1366   assert((!thread->is_Java_thread() || ((JavaThread *)thread)->thread_state() == _thread_in_vm)
1367          || rank() == Mutex::special, "wrong thread state for using locks");
1368   if (StrictSafepointChecks) {
1369     if (thread->is_VM_thread() && !allow_vm_block()) {
1370       fatal(err_msg("VM thread using lock %s (not allowed to block on)",
1371                     name()));
1372     }
1373     debug_only(if (rank() != Mutex::special) \
1374       thread->check_for_valid_safepoint_state(false);)
1375   }
1376 }
1377 
1378 void Monitor::check_block_state(Thread *thread) {
1379   if (!_allow_vm_block && thread->is_VM_thread()) {
1380     warning("VM thread blocked on lock");
1381     print();
1382     BREAKPOINT;
1383   }
1384   assert(_owner != thread, "deadlock: blocking on monitor owned by current thread");
1385 }
1386 
1387 #endif // PRODUCT