1 /*
   2  * Copyright (c) 1998, 2014, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "runtime/atomic.inline.hpp"
  27 #include "runtime/mutex.hpp"
  28 #include "runtime/orderAccess.inline.hpp"
  29 #include "runtime/osThread.hpp"
  30 #include "runtime/thread.inline.hpp"
  31 #include "utilities/events.hpp"
  32 #ifdef TARGET_OS_FAMILY_linux
  33 # include "mutex_linux.inline.hpp"
  34 #endif
  35 #ifdef TARGET_OS_FAMILY_solaris
  36 # include "mutex_solaris.inline.hpp"
  37 #endif
  38 #ifdef TARGET_OS_FAMILY_windows
  39 # include "mutex_windows.inline.hpp"
  40 #endif
  41 #ifdef TARGET_OS_FAMILY_bsd
  42 # include "mutex_bsd.inline.hpp"
  43 #endif
  44 
  45 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
  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)  \
 271   intptr_t(Atomic::cmpxchg_ptr((void *)(s), (void *)(a), (void *)(c)))
 272 #define UNS(x) (uintptr_t(x))
 273 #define TRACE(m)                   \
 274   {                                \
 275     static volatile int ctr = 0;   \
 276     int x = ++ctr;                 \
 277     if ((x & (x - 1)) == 0) {      \
 278       ::printf("%d:%s\n", x, #m);  \
 279       ::fflush(stdout);            \
 280     }                              \
 281   }
 282 
 283 // Simplistic low-quality Marsaglia SHIFT-XOR RNG.
 284 // Bijective except for the trailing mask operation.
 285 // Useful for spin loops as the compiler can't optimize it away.
 286 
 287 static inline jint MarsagliaXORV(jint x) {
 288   if (x == 0) x = 1|os::random();
 289   x ^= x << 6;
 290   x ^= ((unsigned)x) >> 21;
 291   x ^= x << 7;
 292   return x & 0x7FFFFFFF;
 293 }
 294 
 295 static int Stall(int its) {
 296   static volatile jint rv = 1;
 297   volatile int OnFrame = 0;
 298   jint v = rv ^ UNS(OnFrame);
 299   while (--its >= 0) {
 300     v = MarsagliaXORV(v);
 301   }
 302   // Make this impossible for the compiler to optimize away,
 303   // but (mostly) avoid W coherency sharing on MP systems.
 304   if (v == 0x12345) rv = v;
 305   return v;
 306 }
 307 
 308 int Monitor::TryLock() {
 309   intptr_t v = _LockWord.FullWord;
 310   for (;;) {
 311     if ((v & _LBIT) != 0) return 0;
 312     const intptr_t u = CASPTR(&_LockWord, v, v|_LBIT);
 313     if (v == u) return 1;
 314     v = u;
 315   }
 316 }
 317 
 318 int Monitor::TryFast() {
 319   // Optimistic fast-path form ...
 320   // Fast-path attempt for the common uncontended case.
 321   // Avoid RTS->RTO $ coherence upgrade on typical SMP systems.
 322   intptr_t v = CASPTR(&_LockWord, 0, _LBIT);  // agro ...
 323   if (v == 0) return 1;
 324 
 325   for (;;) {
 326     if ((v & _LBIT) != 0) return 0;
 327     const intptr_t u = CASPTR(&_LockWord, v, v|_LBIT);
 328     if (v == u) return 1;
 329     v = u;
 330   }
 331 }
 332 
 333 int Monitor::ILocked() {
 334   const intptr_t w = _LockWord.FullWord & 0xFF;
 335   assert(w == 0 || w == _LBIT, "invariant");
 336   return w == _LBIT;
 337 }
 338 
 339 // Polite TATAS spinlock with exponential backoff - bounded spin.
 340 // Ideally we'd use processor cycles, time or vtime to control
 341 // the loop, but we currently use iterations.
 342 // All the constants within were derived empirically but work over
 343 // over the spectrum of J2SE reference platforms.
 344 // On Niagara-class systems the back-off is unnecessary but
 345 // is relatively harmless.  (At worst it'll slightly retard
 346 // acquisition times).  The back-off is critical for older SMP systems
 347 // where constant fetching of the LockWord would otherwise impair
 348 // scalability.
 349 //
 350 // Clamp spinning at approximately 1/2 of a context-switch round-trip.
 351 // See synchronizer.cpp for details and rationale.
 352 
 353 int Monitor::TrySpin(Thread * const Self) {
 354   if (TryLock())    return 1;
 355   if (!os::is_MP()) return 0;
 356 
 357   int Probes  = 0;
 358   int Delay   = 0;
 359   int Steps   = 0;
 360   int SpinMax = NativeMonitorSpinLimit;
 361   int flgs    = NativeMonitorFlags;
 362   for (;;) {
 363     intptr_t v = _LockWord.FullWord;
 364     if ((v & _LBIT) == 0) {
 365       if (CASPTR (&_LockWord, v, v|_LBIT) == v) {
 366         return 1;
 367       }
 368       continue;
 369     }
 370 
 371     if ((flgs & 8) == 0) {
 372       SpinPause();
 373     }
 374 
 375     // Periodically increase Delay -- variable Delay form
 376     // conceptually: delay *= 1 + 1/Exponent
 377     ++Probes;
 378     if (Probes > SpinMax) return 0;
 379 
 380     if ((Probes & 0x7) == 0) {
 381       Delay = ((Delay << 1)|1) & 0x7FF;
 382       // CONSIDER: Delay += 1 + (Delay/4); Delay &= 0x7FF ;
 383     }
 384 
 385     if (flgs & 2) continue;
 386 
 387     // Consider checking _owner's schedctl state, if OFFPROC abort spin.
 388     // If the owner is OFFPROC then it's unlike that the lock will be dropped
 389     // in a timely fashion, which suggests that spinning would not be fruitful
 390     // or profitable.
 391 
 392     // Stall for "Delay" time units - iterations in the current implementation.
 393     // Avoid generating coherency traffic while stalled.
 394     // Possible ways to delay:
 395     //   PAUSE, SLEEP, MEMBAR #sync, MEMBAR #halt,
 396     //   wr %g0,%asi, gethrtime, rdstick, rdtick, rdtsc, etc. ...
 397     // Note that on Niagara-class systems we want to minimize STs in the
 398     // spin loop.  N1 and brethren write-around the L1$ over the xbar into the L2$.
 399     // Furthermore, they don't have a W$ like traditional SPARC processors.
 400     // We currently use a Marsaglia Shift-Xor RNG loop.
 401     Steps += Delay;
 402     if (Self != NULL) {
 403       jint rv = Self->rng[0];
 404       for (int k = Delay; --k >= 0;) {
 405         rv = MarsagliaXORV(rv);
 406         if ((flgs & 4) == 0 && SafepointSynchronize::do_call_back()) return 0;
 407       }
 408       Self->rng[0] = rv;
 409     } else {
 410       Stall(Delay);
 411     }
 412   }
 413 }
 414 
 415 static int ParkCommon(ParkEvent * ev, jlong timo) {
 416   // Diagnostic support - periodically unwedge blocked threads
 417   intx nmt = NativeMonitorTimeout;
 418   if (nmt > 0 && (nmt < timo || timo <= 0)) {
 419     timo = nmt;
 420   }
 421   int err = OS_OK;
 422   if (0 == timo) {
 423     ev->park();
 424   } else {
 425     err = ev->park(timo);
 426   }
 427   return err;
 428 }
 429 
 430 inline int Monitor::AcquireOrPush(ParkEvent * ESelf) {
 431   intptr_t v = _LockWord.FullWord;
 432   for (;;) {
 433     if ((v & _LBIT) == 0) {
 434       const intptr_t u = CASPTR(&_LockWord, v, v|_LBIT);
 435       if (u == v) return 1;        // indicate acquired
 436       v = u;
 437     } else {
 438       // Anticipate success ...
 439       ESelf->ListNext = (ParkEvent *)(v & ~_LBIT);
 440       const intptr_t u = CASPTR(&_LockWord, v, intptr_t(ESelf)|_LBIT);
 441       if (u == v) return 0;        // indicate pushed onto cxq
 442       v = u;
 443     }
 444     // Interference - LockWord change - just retry
 445   }
 446 }
 447 
 448 // ILock and IWait are the lowest level primitive internal blocking
 449 // synchronization functions.  The callers of IWait and ILock must have
 450 // performed any needed state transitions beforehand.
 451 // IWait and ILock may directly call park() without any concern for thread state.
 452 // Note that ILock and IWait do *not* access _owner.
 453 // _owner is a higher-level logical concept.
 454 
 455 void Monitor::ILock(Thread * Self) {
 456   assert(_OnDeck != Self->_MutexEvent, "invariant");
 457 
 458   if (TryFast()) {
 459  Exeunt:
 460     assert(ILocked(), "invariant");
 461     return;
 462   }
 463 
 464   ParkEvent * const ESelf = Self->_MutexEvent;
 465   assert(_OnDeck != ESelf, "invariant");
 466 
 467   // As an optimization, spinners could conditionally try to set ONDECK to _LBIT
 468   // Synchronizer.cpp uses a similar optimization.
 469   if (TrySpin(Self)) goto Exeunt;
 470 
 471   // Slow-path - the lock is contended.
 472   // Either Enqueue Self on cxq or acquire the outer lock.
 473   // LockWord encoding = (cxq,LOCKBYTE)
 474   ESelf->reset();
 475   OrderAccess::fence();
 476 
 477   // Optional optimization ... try barging on the inner lock
 478   if ((NativeMonitorFlags & 32) && CASPTR (&_OnDeck, NULL, UNS(Self)) == 0) {
 479     goto OnDeck_LOOP;
 480   }
 481 
 482   if (AcquireOrPush(ESelf)) goto Exeunt;
 483 
 484   // At any given time there is at most one ondeck thread.
 485   // ondeck implies not resident on cxq and not resident on EntryList
 486   // Only the OnDeck thread can try to acquire -- contended for -- the lock.
 487   // CONSIDER: use Self->OnDeck instead of m->OnDeck.
 488   // Deschedule Self so that others may run.
 489   while (_OnDeck != ESelf) {
 490     ParkCommon(ESelf, 0);
 491   }
 492 
 493   // Self is now in the ONDECK position and will remain so until it
 494   // manages to acquire the lock.
 495  OnDeck_LOOP:
 496   for (;;) {
 497     assert(_OnDeck == ESelf, "invariant");
 498     if (TrySpin(Self)) break;
 499     // It's probably wise to spin only if we *actually* blocked
 500     // CONSIDER: check the lockbyte, if it remains set then
 501     // preemptively drain the cxq into the EntryList.
 502     // The best place and time to perform queue operations -- lock metadata --
 503     // is _before having acquired the outer lock, while waiting for the lock to drop.
 504     ParkCommon(ESelf, 0);
 505   }
 506 
 507   assert(_OnDeck == ESelf, "invariant");
 508   _OnDeck = NULL;
 509 
 510   // Note that we current drop the inner lock (clear OnDeck) in the slow-path
 511   // epilogue immediately after having acquired the outer lock.
 512   // But instead we could consider the following optimizations:
 513   // A. Shift or defer dropping the inner lock until the subsequent IUnlock() operation.
 514   //    This might avoid potential reacquisition of the inner lock in IUlock().
 515   // B. While still holding the inner lock, attempt to opportunistically select
 516   //    and unlink the next ONDECK thread from the EntryList.
 517   //    If successful, set ONDECK to refer to that thread, otherwise clear ONDECK.
 518   //    It's critical that the select-and-unlink operation run in constant-time as
 519   //    it executes when holding the outer lock and may artificially increase the
 520   //    effective length of the critical section.
 521   // Note that (A) and (B) are tantamount to succession by direct handoff for
 522   // the inner lock.
 523   goto Exeunt;
 524 }
 525 
 526 void Monitor::IUnlock(bool RelaxAssert) {
 527   assert(ILocked(), "invariant");
 528   // Conceptually we need a MEMBAR #storestore|#loadstore barrier or fence immediately
 529   // before the store that releases the lock.  Crucially, all the stores and loads in the
 530   // critical section must be globally visible before the store of 0 into the lock-word
 531   // that releases the lock becomes globally visible.  That is, memory accesses in the
 532   // critical section should not be allowed to bypass or overtake the following ST that
 533   // releases the lock.  As such, to prevent accesses within the critical section
 534   // from "leaking" out, we need a release fence between the critical section and the
 535   // store that releases the lock.  In practice that release barrier is elided on
 536   // platforms with strong memory models such as TSO.
 537   //
 538   // Note that the OrderAccess::storeload() fence that appears after unlock store
 539   // provides for progress conditions and succession and is _not related to exclusion
 540   // safety or lock release consistency.
 541   OrderAccess::release_store(&_LockWord.Bytes[_LSBINDEX], 0); // drop outer lock
 542 
 543   OrderAccess::storeload();
 544   ParkEvent * const w = _OnDeck;
 545   assert(RelaxAssert || w != Thread::current()->_MutexEvent, "invariant");
 546   if (w != NULL) {
 547     // Either we have a valid ondeck thread or ondeck is transiently "locked"
 548     // by some exiting thread as it arranges for succession.  The LSBit of
 549     // OnDeck allows us to discriminate two cases.  If the latter, the
 550     // responsibility for progress and succession lies with that other thread.
 551     // For good performance, we also depend on the fact that redundant unpark()
 552     // operations are cheap.  That is, repeated Unpark()ing of the ONDECK thread
 553     // is inexpensive.  This approach provides implicit futile wakeup throttling.
 554     // Note that the referent "w" might be stale with respect to the lock.
 555     // In that case the following unpark() is harmless and the worst that'll happen
 556     // is a spurious return from a park() operation.  Critically, if "w" _is stale,
 557     // then progress is known to have occurred as that means the thread associated
 558     // with "w" acquired the lock.  In that case this thread need take no further
 559     // action to guarantee progress.
 560     if ((UNS(w) & _LBIT) == 0) w->unpark();
 561     return;
 562   }
 563 
 564   intptr_t cxq = _LockWord.FullWord;
 565   if (((cxq & ~_LBIT)|UNS(_EntryList)) == 0) {
 566     return;      // normal fast-path exit - cxq and EntryList both empty
 567   }
 568   if (cxq & _LBIT) {
 569     // Optional optimization ...
 570     // Some other thread acquired the lock in the window since this
 571     // thread released it.  Succession is now that thread's responsibility.
 572     return;
 573   }
 574 
 575  Succession:
 576   // Slow-path exit - this thread must ensure succession and progress.
 577   // OnDeck serves as lock to protect cxq and EntryList.
 578   // Only the holder of OnDeck can manipulate EntryList or detach the RATs from cxq.
 579   // Avoid ABA - allow multiple concurrent producers (enqueue via push-CAS)
 580   // but only one concurrent consumer (detacher of RATs).
 581   // Consider protecting this critical section with schedctl on Solaris.
 582   // Unlike a normal lock, however, the exiting thread "locks" OnDeck,
 583   // picks a successor and marks that thread as OnDeck.  That successor
 584   // thread will then clear OnDeck once it eventually acquires the outer lock.
 585   if (CASPTR (&_OnDeck, NULL, _LBIT) != UNS(NULL)) {
 586     return;
 587   }
 588 
 589   ParkEvent * List = _EntryList;
 590   if (List != NULL) {
 591     // Transfer the head of the EntryList to the OnDeck position.
 592     // Once OnDeck, a thread stays OnDeck until it acquires the lock.
 593     // For a given lock there is at most OnDeck thread at any one instant.
 594    WakeOne:
 595     assert(List == _EntryList, "invariant");
 596     ParkEvent * const w = List;
 597     assert(RelaxAssert || w != Thread::current()->_MutexEvent, "invariant");
 598     _EntryList = w->ListNext;
 599     // as a diagnostic measure consider setting w->_ListNext = BAD
 600     assert(UNS(_OnDeck) == _LBIT, "invariant");
 601     _OnDeck = w;  // pass OnDeck to w.
 602                   // w will clear OnDeck once it acquires the outer lock
 603 
 604     // Another optional optimization ...
 605     // For heavily contended locks it's not uncommon that some other
 606     // thread acquired the lock while this thread was arranging succession.
 607     // Try to defer the unpark() operation - Delegate the responsibility
 608     // for unpark()ing the OnDeck thread to the current or subsequent owners
 609     // That is, the new owner is responsible for unparking the OnDeck thread.
 610     OrderAccess::storeload();
 611     cxq = _LockWord.FullWord;
 612     if (cxq & _LBIT) return;
 613 
 614     w->unpark();
 615     return;
 616   }
 617 
 618   cxq = _LockWord.FullWord;
 619   if ((cxq & ~_LBIT) != 0) {
 620     // The EntryList is empty but the cxq is populated.
 621     // drain RATs from cxq into EntryList
 622     // Detach RATs segment with CAS and then merge into EntryList
 623     for (;;) {
 624       // optional optimization - if locked, the owner is responsible for succession
 625       if (cxq & _LBIT) goto Punt;
 626       const intptr_t vfy = CASPTR(&_LockWord, cxq, cxq & _LBIT);
 627       if (vfy == cxq) break;
 628       cxq = vfy;
 629       // Interference - LockWord changed - Just retry
 630       // We can see concurrent interference from contending threads
 631       // pushing themselves onto the cxq or from lock-unlock operations.
 632       // From the perspective of this thread, EntryList is stable and
 633       // the cxq is prepend-only -- the head is volatile but the interior
 634       // of the cxq is stable.  In theory if we encounter interference from threads
 635       // pushing onto cxq we could simply break off the original cxq suffix and
 636       // move that segment to the EntryList, avoiding a 2nd or multiple CAS attempts
 637       // on the high-traffic LockWord variable.   For instance lets say the cxq is "ABCD"
 638       // when we first fetch cxq above.  Between the fetch -- where we observed "A"
 639       // -- and CAS -- where we attempt to CAS null over A -- "PQR" arrive,
 640       // yielding cxq = "PQRABCD".  In this case we could simply set A.ListNext
 641       // null, leaving cxq = "PQRA" and transfer the "BCD" segment to the EntryList.
 642       // Note too, that it's safe for this thread to traverse the cxq
 643       // without taking any special concurrency precautions.
 644     }
 645 
 646     // We don't currently reorder the cxq segment as we move it onto
 647     // the EntryList, but it might make sense to reverse the order
 648     // or perhaps sort by thread priority.  See the comments in
 649     // synchronizer.cpp objectMonitor::exit().
 650     assert(_EntryList == NULL, "invariant");
 651     _EntryList = List = (ParkEvent *)(cxq & ~_LBIT);
 652     assert(List != NULL, "invariant");
 653     goto WakeOne;
 654   }
 655 
 656   // cxq|EntryList is empty.
 657   // w == NULL implies that cxq|EntryList == NULL in the past.
 658   // Possible race - rare inopportune interleaving.
 659   // A thread could have added itself to cxq since this thread previously checked.
 660   // Detect and recover by refetching cxq.
 661  Punt:
 662   assert(UNS(_OnDeck) == _LBIT, "invariant");
 663   _OnDeck = NULL;            // Release inner lock.
 664   OrderAccess::storeload();   // Dekker duality - pivot point
 665 
 666   // Resample LockWord/cxq to recover from possible race.
 667   // For instance, while this thread T1 held OnDeck, some other thread T2 might
 668   // acquire the outer lock.  Another thread T3 might try to acquire the outer
 669   // lock, but encounter contention and enqueue itself on cxq.  T2 then drops the
 670   // outer lock, but skips succession as this thread T1 still holds OnDeck.
 671   // T1 is and remains responsible for ensuring succession of T3.
 672   //
 673   // Note that we don't need to recheck EntryList, just cxq.
 674   // If threads moved onto EntryList since we dropped OnDeck
 675   // that implies some other thread forced succession.
 676   cxq = _LockWord.FullWord;
 677   if ((cxq & ~_LBIT) != 0 && (cxq & _LBIT) == 0) {
 678     goto Succession;         // potential race -- re-run succession
 679   }
 680   return;
 681 }
 682 
 683 bool Monitor::notify() {
 684   assert(_owner == Thread::current(), "invariant");
 685   assert(ILocked(), "invariant");
 686   if (_WaitSet == NULL) return true;
 687   NotifyCount++;
 688 
 689   // Transfer one thread from the WaitSet to the EntryList or cxq.
 690   // Currently we just unlink the head of the WaitSet and prepend to the cxq.
 691   // And of course we could just unlink it and unpark it, too, but
 692   // in that case it'd likely impale itself on the reentry.
 693   Thread::muxAcquire(_WaitLock, "notify:WaitLock");
 694   ParkEvent * nfy = _WaitSet;
 695   if (nfy != NULL) {                  // DCL idiom
 696     _WaitSet = nfy->ListNext;
 697     assert(nfy->Notified == 0, "invariant");
 698     // push nfy onto the cxq
 699     for (;;) {
 700       const intptr_t v = _LockWord.FullWord;
 701       assert((v & 0xFF) == _LBIT, "invariant");
 702       nfy->ListNext = (ParkEvent *)(v & ~_LBIT);
 703       if (CASPTR (&_LockWord, v, UNS(nfy)|_LBIT) == v) break;
 704       // interference - _LockWord changed -- just retry
 705     }
 706     // Note that setting Notified before pushing nfy onto the cxq is
 707     // also legal and safe, but the safety properties are much more
 708     // subtle, so for the sake of code stewardship ...
 709     OrderAccess::fence();
 710     nfy->Notified = 1;
 711   }
 712   Thread::muxRelease(_WaitLock);
 713   if (nfy != NULL && (NativeMonitorFlags & 16)) {
 714     // Experimental code ... light up the wakee in the hope that this thread (the owner)
 715     // will drop the lock just about the time the wakee comes ONPROC.
 716     nfy->unpark();
 717   }
 718   assert(ILocked(), "invariant");
 719   return true;
 720 }
 721 
 722 // Currently notifyAll() transfers the waiters one-at-a-time from the waitset
 723 // to the cxq.  This could be done more efficiently with a single bulk en-mass transfer,
 724 // but in practice notifyAll() for large #s of threads is rare and not time-critical.
 725 // Beware too, that we invert the order of the waiters.  Lets say that the
 726 // waitset is "ABCD" and the cxq is "XYZ".  After a notifyAll() the waitset
 727 // will be empty and the cxq will be "DCBAXYZ".  This is benign, of course.
 728 
 729 bool Monitor::notify_all() {
 730   assert(_owner == Thread::current(), "invariant");
 731   assert(ILocked(), "invariant");
 732   while (_WaitSet != NULL) notify();
 733   return true;
 734 }
 735 
 736 int Monitor::IWait(Thread * Self, jlong timo) {
 737   assert(ILocked(), "invariant");
 738 
 739   // Phases:
 740   // 1. Enqueue Self on WaitSet - currently prepend
 741   // 2. unlock - drop the outer lock
 742   // 3. wait for either notification or timeout
 743   // 4. lock - reentry - reacquire the outer lock
 744 
 745   ParkEvent * const ESelf = Self->_MutexEvent;
 746   ESelf->Notified = 0;
 747   ESelf->reset();
 748   OrderAccess::fence();
 749 
 750   // Add Self to WaitSet
 751   // Ideally only the holder of the outer lock would manipulate the WaitSet -
 752   // That is, the outer lock would implicitly protect the WaitSet.
 753   // But if a thread in wait() encounters a timeout it will need to dequeue itself
 754   // from the WaitSet _before it becomes the owner of the lock.  We need to dequeue
 755   // as the ParkEvent -- which serves as a proxy for the thread -- can't reside
 756   // on both the WaitSet and the EntryList|cxq at the same time..  That is, a thread
 757   // on the WaitSet can't be allowed to compete for the lock until it has managed to
 758   // unlink its ParkEvent from WaitSet.  Thus the need for WaitLock.
 759   // Contention on the WaitLock is minimal.
 760   //
 761   // Another viable approach would be add another ParkEvent, "WaitEvent" to the
 762   // thread class.  The WaitSet would be composed of WaitEvents.  Only the
 763   // owner of the outer lock would manipulate the WaitSet.  A thread in wait()
 764   // could then compete for the outer lock, and then, if necessary, unlink itself
 765   // from the WaitSet only after having acquired the outer lock.  More precisely,
 766   // there would be no WaitLock.  A thread in in wait() would enqueue its WaitEvent
 767   // on the WaitSet; release the outer lock; wait for either notification or timeout;
 768   // reacquire the inner lock; and then, if needed, unlink itself from the WaitSet.
 769   //
 770   // Alternatively, a 2nd set of list link fields in the ParkEvent might suffice.
 771   // One set would be for the WaitSet and one for the EntryList.
 772   // We could also deconstruct the ParkEvent into a "pure" event and add a
 773   // new immortal/TSM "ListElement" class that referred to ParkEvents.
 774   // In that case we could have one ListElement on the WaitSet and another
 775   // on the EntryList, with both referring to the same pure Event.
 776 
 777   Thread::muxAcquire(_WaitLock, "wait:WaitLock:Add");
 778   ESelf->ListNext = _WaitSet;
 779   _WaitSet = ESelf;
 780   Thread::muxRelease(_WaitLock);
 781 
 782   // Release the outer lock
 783   // We call IUnlock (RelaxAssert=true) as a thread T1 might
 784   // enqueue itself on the WaitSet, call IUnlock(), drop the lock,
 785   // and then stall before it can attempt to wake a successor.
 786   // Some other thread T2 acquires the lock, and calls notify(), moving
 787   // T1 from the WaitSet to the cxq.  T2 then drops the lock.  T1 resumes,
 788   // and then finds *itself* on the cxq.  During the course of a normal
 789   // IUnlock() call a thread should _never find itself on the EntryList
 790   // or cxq, but in the case of wait() it's possible.
 791   // See synchronizer.cpp objectMonitor::wait().
 792   IUnlock(true);
 793 
 794   // Wait for either notification or timeout
 795   // Beware that in some circumstances we might propagate
 796   // spurious wakeups back to the caller.
 797 
 798   for (;;) {
 799     if (ESelf->Notified) break;
 800     int err = ParkCommon(ESelf, timo);
 801     if (err == OS_TIMEOUT || (NativeMonitorFlags & 1)) break;
 802   }
 803 
 804   // Prepare for reentry - if necessary, remove ESelf from WaitSet
 805   // ESelf can be:
 806   // 1. Still on the WaitSet.  This can happen if we exited the loop by timeout.
 807   // 2. On the cxq or EntryList
 808   // 3. Not resident on cxq, EntryList or WaitSet, but in the OnDeck position.
 809 
 810   OrderAccess::fence();
 811   int WasOnWaitSet = 0;
 812   if (ESelf->Notified == 0) {
 813     Thread::muxAcquire(_WaitLock, "wait:WaitLock:remove");
 814     if (ESelf->Notified == 0) {     // DCL idiom
 815       assert(_OnDeck != ESelf, "invariant");   // can't be both OnDeck and on WaitSet
 816       // ESelf is resident on the WaitSet -- unlink it.
 817       // A doubly-linked list would be better here so we can unlink in constant-time.
 818       // We have to unlink before we potentially recontend as ESelf might otherwise
 819       // end up on the cxq|EntryList -- it can't be on two lists at once.
 820       ParkEvent * p = _WaitSet;
 821       ParkEvent * q = NULL;            // classic q chases p
 822       while (p != NULL && p != ESelf) {
 823         q = p;
 824         p = p->ListNext;
 825       }
 826       assert(p == ESelf, "invariant");
 827       if (p == _WaitSet) {      // found at head
 828         assert(q == NULL, "invariant");
 829         _WaitSet = p->ListNext;
 830       } else {                  // found in interior
 831         assert(q->ListNext == p, "invariant");
 832         q->ListNext = p->ListNext;
 833       }
 834       WasOnWaitSet = 1;        // We were *not* notified but instead encountered timeout
 835     }
 836     Thread::muxRelease(_WaitLock);
 837   }
 838 
 839   // Reentry phase - reacquire the lock
 840   if (WasOnWaitSet) {
 841     // ESelf was previously on the WaitSet but we just unlinked it above
 842     // because of a timeout.  ESelf is not resident on any list and is not OnDeck
 843     assert(_OnDeck != ESelf, "invariant");
 844     ILock(Self);
 845   } else {
 846     // A prior notify() operation moved ESelf from the WaitSet to the cxq.
 847     // ESelf is now on the cxq, EntryList or at the OnDeck position.
 848     // The following fragment is extracted from Monitor::ILock()
 849     for (;;) {
 850       if (_OnDeck == ESelf && TrySpin(Self)) break;
 851       ParkCommon(ESelf, 0);
 852     }
 853     assert(_OnDeck == ESelf, "invariant");
 854     _OnDeck = NULL;
 855   }
 856 
 857   assert(ILocked(), "invariant");
 858   return WasOnWaitSet != 0;        // return true IFF timeout
 859 }
 860 
 861 
 862 // ON THE VMTHREAD SNEAKING PAST HELD LOCKS:
 863 // In particular, there are certain types of global lock that may be held
 864 // by a Java thread while it is blocked at a safepoint but before it has
 865 // written the _owner field. These locks may be sneakily acquired by the
 866 // VM thread during a safepoint to avoid deadlocks. Alternatively, one should
 867 // identify all such locks, and ensure that Java threads never block at
 868 // safepoints while holding them (_no_safepoint_check_flag). While it
 869 // seems as though this could increase the time to reach a safepoint
 870 // (or at least increase the mean, if not the variance), the latter
 871 // approach might make for a cleaner, more maintainable JVM design.
 872 //
 873 // Sneaking is vile and reprehensible and should be excised at the 1st
 874 // opportunity.  It's possible that the need for sneaking could be obviated
 875 // as follows.  Currently, a thread might (a) while TBIVM, call pthread_mutex_lock
 876 // or ILock() thus acquiring the "physical" lock underlying Monitor/Mutex.
 877 // (b) stall at the TBIVM exit point as a safepoint is in effect.  Critically,
 878 // it'll stall at the TBIVM reentry state transition after having acquired the
 879 // underlying lock, but before having set _owner and having entered the actual
 880 // critical section.  The lock-sneaking facility leverages that fact and allowed the
 881 // VM thread to logically acquire locks that had already be physically locked by mutators
 882 // but where mutators were known blocked by the reentry thread state transition.
 883 //
 884 // If we were to modify the Monitor-Mutex so that TBIVM state transitions tightly
 885 // wrapped calls to park(), then we could likely do away with sneaking.  We'd
 886 // decouple lock acquisition and parking.  The critical invariant  to eliminating
 887 // sneaking is to ensure that we never "physically" acquire the lock while TBIVM.
 888 // An easy way to accomplish this is to wrap the park calls in a narrow TBIVM jacket.
 889 // One difficulty with this approach is that the TBIVM wrapper could recurse and
 890 // call lock() deep from within a lock() call, while the MutexEvent was already enqueued.
 891 // Using a stack (N=2 at minimum) of ParkEvents would take care of that problem.
 892 //
 893 // But of course the proper ultimate approach is to avoid schemes that require explicit
 894 // sneaking or dependence on any any clever invariants or subtle implementation properties
 895 // of Mutex-Monitor and instead directly address the underlying design flaw.
 896 
 897 void Monitor::lock(Thread * Self) {
 898   // Ensure that the Monitor requires/allows safepoint checks.
 899   assert(_safepoint_check_required != Monitor::_safepoint_check_never,
 900          err_msg("This lock should never have a safepoint check: %s",
 901                  name()));
 902 
 903 #ifdef CHECK_UNHANDLED_OOPS
 904   // Clear unhandled oops so we get a crash right away.  Only clear for non-vm
 905   // or GC threads.
 906   if (Self->is_Java_thread()) {
 907     Self->clear_unhandled_oops();
 908   }
 909 #endif // CHECK_UNHANDLED_OOPS
 910 
 911   debug_only(check_prelock_state(Self));
 912   assert(_owner != Self, "invariant");
 913   assert(_OnDeck != Self->_MutexEvent, "invariant");
 914 
 915   if (TryFast()) {
 916  Exeunt:
 917     assert(ILocked(), "invariant");
 918     assert(owner() == NULL, "invariant");
 919     set_owner(Self);
 920     return;
 921   }
 922 
 923   // The lock is contended ...
 924 
 925   bool can_sneak = Self->is_VM_thread() && SafepointSynchronize::is_at_safepoint();
 926   if (can_sneak && _owner == NULL) {
 927     // a java thread has locked the lock but has not entered the
 928     // critical region -- let's just pretend we've locked the lock
 929     // and go on.  we note this with _snuck so we can also
 930     // pretend to unlock when the time comes.
 931     _snuck = true;
 932     goto Exeunt;
 933   }
 934 
 935   // Try a brief spin to avoid passing thru thread state transition ...
 936   if (TrySpin(Self)) goto Exeunt;
 937 
 938   check_block_state(Self);
 939   if (Self->is_Java_thread()) {
 940     // Horrible dictu - we suffer through a state transition
 941     assert(rank() > Mutex::special, "Potential deadlock with special or lesser rank mutex");
 942     ThreadBlockInVM tbivm((JavaThread *) Self);
 943     ILock(Self);
 944   } else {
 945     // Mirabile dictu
 946     ILock(Self);
 947   }
 948   goto Exeunt;
 949 }
 950 
 951 void Monitor::lock() {
 952   this->lock(Thread::current());
 953 }
 954 
 955 // Lock without safepoint check - a degenerate variant of lock().
 956 // Should ONLY be used by safepoint code and other code
 957 // that is guaranteed not to block while running inside the VM. If this is called with
 958 // thread state set to be in VM, the safepoint synchronization code will deadlock!
 959 
 960 void Monitor::lock_without_safepoint_check(Thread * Self) {
 961   // Ensure that the Monitor does not require or allow safepoint checks.
 962   assert(_safepoint_check_required != Monitor::_safepoint_check_always,
 963          err_msg("This lock should always have a safepoint check: %s",
 964                  name()));
 965   assert(_owner != Self, "invariant");
 966   ILock(Self);
 967   assert(_owner == NULL, "invariant");
 968   set_owner(Self);
 969 }
 970 
 971 void Monitor::lock_without_safepoint_check() {
 972   lock_without_safepoint_check(Thread::current());
 973 }
 974 
 975 
 976 // Returns true if thread succeeds in grabbing the lock, otherwise false.
 977 
 978 bool Monitor::try_lock() {
 979   Thread * const Self = Thread::current();
 980   debug_only(check_prelock_state(Self));
 981   // assert(!thread->is_inside_signal_handler(), "don't lock inside signal handler");
 982 
 983   // Special case, where all Java threads are stopped.
 984   // The lock may have been acquired but _owner is not yet set.
 985   // In that case the VM thread can safely grab the lock.
 986   // It strikes me this should appear _after the TryLock() fails, below.
 987   bool can_sneak = Self->is_VM_thread() && SafepointSynchronize::is_at_safepoint();
 988   if (can_sneak && _owner == NULL) {
 989     set_owner(Self); // Do not need to be atomic, since we are at a safepoint
 990     _snuck = true;
 991     return true;
 992   }
 993 
 994   if (TryLock()) {
 995     // We got the lock
 996     assert(_owner == NULL, "invariant");
 997     set_owner(Self);
 998     return true;
 999   }
1000   return false;
1001 }
1002 
1003 void Monitor::unlock() {
1004   assert(_owner == Thread::current(), "invariant");
1005   assert(_OnDeck != Thread::current()->_MutexEvent, "invariant");
1006   set_owner(NULL);
1007   if (_snuck) {
1008     assert(SafepointSynchronize::is_at_safepoint() && Thread::current()->is_VM_thread(), "sneak");
1009     _snuck = false;
1010     return;
1011   }
1012   IUnlock(false);
1013 }
1014 
1015 // Yet another degenerate version of Monitor::lock() or lock_without_safepoint_check()
1016 // jvm_raw_lock() and _unlock() can be called by non-Java threads via JVM_RawMonitorEnter.
1017 //
1018 // There's no expectation that JVM_RawMonitors will interoperate properly with the native
1019 // Mutex-Monitor constructs.  We happen to implement JVM_RawMonitors in terms of
1020 // native Mutex-Monitors simply as a matter of convenience.  A simple abstraction layer
1021 // over a pthread_mutex_t would work equally as well, but require more platform-specific
1022 // code -- a "PlatformMutex".  Alternatively, a simply layer over muxAcquire-muxRelease
1023 // would work too.
1024 //
1025 // Since the caller might be a foreign thread, we don't necessarily have a Thread.MutexEvent
1026 // instance available.  Instead, we transiently allocate a ParkEvent on-demand if
1027 // we encounter contention.  That ParkEvent remains associated with the thread
1028 // until it manages to acquire the lock, at which time we return the ParkEvent
1029 // to the global ParkEvent free list.  This is correct and suffices for our purposes.
1030 //
1031 // Beware that the original jvm_raw_unlock() had a "_snuck" test but that
1032 // jvm_raw_lock() didn't have the corresponding test.  I suspect that's an
1033 // oversight, but I've replicated the original suspect logic in the new code ...
1034 
1035 void Monitor::jvm_raw_lock() {
1036   assert(rank() == native, "invariant");
1037 
1038   if (TryLock()) {
1039  Exeunt:
1040     assert(ILocked(), "invariant");
1041     assert(_owner == NULL, "invariant");
1042     // This can potentially be called by non-java Threads. Thus, the ThreadLocalStorage
1043     // might return NULL. Don't call set_owner since it will break on an NULL owner
1044     // Consider installing a non-null "ANON" distinguished value instead of just NULL.
1045     _owner = ThreadLocalStorage::thread();
1046     return;
1047   }
1048 
1049   if (TrySpin(NULL)) goto Exeunt;
1050 
1051   // slow-path - apparent contention
1052   // Allocate a ParkEvent for transient use.
1053   // The ParkEvent remains associated with this thread until
1054   // the time the thread manages to acquire the lock.
1055   ParkEvent * const ESelf = ParkEvent::Allocate(NULL);
1056   ESelf->reset();
1057   OrderAccess::storeload();
1058 
1059   // Either Enqueue Self on cxq or acquire the outer lock.
1060   if (AcquireOrPush (ESelf)) {
1061     ParkEvent::Release(ESelf);      // surrender the ParkEvent
1062     goto Exeunt;
1063   }
1064 
1065   // At any given time there is at most one ondeck thread.
1066   // ondeck implies not resident on cxq and not resident on EntryList
1067   // Only the OnDeck thread can try to acquire -- contended for -- the lock.
1068   // CONSIDER: use Self->OnDeck instead of m->OnDeck.
1069   for (;;) {
1070     if (_OnDeck == ESelf && TrySpin(NULL)) break;
1071     ParkCommon(ESelf, 0);
1072   }
1073 
1074   assert(_OnDeck == ESelf, "invariant");
1075   _OnDeck = NULL;
1076   ParkEvent::Release(ESelf);      // surrender the ParkEvent
1077   goto Exeunt;
1078 }
1079 
1080 void Monitor::jvm_raw_unlock() {
1081   // Nearly the same as Monitor::unlock() ...
1082   // directly set _owner instead of using set_owner(null)
1083   _owner = NULL;
1084   if (_snuck) {         // ???
1085     assert(SafepointSynchronize::is_at_safepoint() && Thread::current()->is_VM_thread(), "sneak");
1086     _snuck = false;
1087     return;
1088   }
1089   IUnlock(false);
1090 }
1091 
1092 bool Monitor::wait(bool no_safepoint_check, long timeout,
1093                    bool as_suspend_equivalent) {
1094   // Make sure safepoint checking is used properly.
1095   assert(!(_safepoint_check_required == Monitor::_safepoint_check_never && no_safepoint_check == false),
1096          err_msg("This lock should never have a safepoint check: %s", name()));
1097   assert(!(_safepoint_check_required == Monitor::_safepoint_check_always && no_safepoint_check == true),
1098          err_msg("This lock should always have a safepoint check: %s", name()));
1099 
1100   Thread * const Self = Thread::current();
1101   assert(_owner == Self, "invariant");
1102   assert(ILocked(), "invariant");
1103 
1104   // as_suspend_equivalent logically implies !no_safepoint_check
1105   guarantee(!as_suspend_equivalent || !no_safepoint_check, "invariant");
1106   // !no_safepoint_check logically implies java_thread
1107   guarantee(no_safepoint_check || Self->is_Java_thread(), "invariant");
1108 
1109   #ifdef ASSERT
1110   Monitor * least = get_least_ranked_lock_besides_this(Self->owned_locks());
1111   assert(least != this, "Specification of get_least_... call above");
1112   if (least != NULL && least->rank() <= special) {
1113     tty->print("Attempting to wait on monitor %s/%d while holding"
1114                " lock %s/%d -- possible deadlock",
1115                name(), rank(), least->name(), least->rank());
1116     assert(false, "Shouldn't block(wait) while holding a lock of rank special");
1117   }
1118   #endif // ASSERT
1119 
1120   int wait_status;
1121   // conceptually set the owner to NULL in anticipation of
1122   // abdicating the lock in wait
1123   set_owner(NULL);
1124   if (no_safepoint_check) {
1125     wait_status = IWait(Self, timeout);
1126   } else {
1127     assert(Self->is_Java_thread(), "invariant");
1128     JavaThread *jt = (JavaThread *)Self;
1129 
1130     // Enter safepoint region - ornate and Rococo ...
1131     ThreadBlockInVM tbivm(jt);
1132     OSThreadWaitState osts(Self->osthread(), false /* not Object.wait() */);
1133 
1134     if (as_suspend_equivalent) {
1135       jt->set_suspend_equivalent();
1136       // cleared by handle_special_suspend_equivalent_condition() or
1137       // java_suspend_self()
1138     }
1139 
1140     wait_status = IWait(Self, timeout);
1141 
1142     // were we externally suspended while we were waiting?
1143     if (as_suspend_equivalent && jt->handle_special_suspend_equivalent_condition()) {
1144       // Our event wait has finished and we own the lock, but
1145       // while we were waiting another thread suspended us. We don't
1146       // want to hold the lock while suspended because that
1147       // would surprise the thread that suspended us.
1148       assert(ILocked(), "invariant");
1149       IUnlock(true);
1150       jt->java_suspend_self();
1151       ILock(Self);
1152       assert(ILocked(), "invariant");
1153     }
1154   }
1155 
1156   // Conceptually reestablish ownership of the lock.
1157   // The "real" lock -- the LockByte -- was reacquired by IWait().
1158   assert(ILocked(), "invariant");
1159   assert(_owner == NULL, "invariant");
1160   set_owner(Self);
1161   return wait_status != 0;          // return true IFF timeout
1162 }
1163 
1164 Monitor::~Monitor() {
1165   assert((UNS(_owner)|UNS(_LockWord.FullWord)|UNS(_EntryList)|UNS(_WaitSet)|UNS(_OnDeck)) == 0, "");
1166 }
1167 
1168 void Monitor::ClearMonitor(Monitor * m, const char *name) {
1169   m->_owner             = NULL;
1170   m->_snuck             = false;
1171   if (name == NULL) {
1172     strcpy(m->_name, "UNKNOWN");
1173   } else {
1174     strncpy(m->_name, name, MONITOR_NAME_LEN - 1);
1175     m->_name[MONITOR_NAME_LEN - 1] = '\0';
1176   }
1177   m->_LockWord.FullWord = 0;
1178   m->_EntryList         = NULL;
1179   m->_OnDeck            = NULL;
1180   m->_WaitSet           = NULL;
1181   m->_WaitLock[0]       = 0;
1182 }
1183 
1184 Monitor::Monitor() { ClearMonitor(this); }
1185 
1186 Monitor::Monitor(int Rank, const char * name, bool allow_vm_block,
1187                  SafepointCheckRequired safepoint_check_required) {
1188   ClearMonitor(this, name);
1189 #ifdef ASSERT
1190   _allow_vm_block  = allow_vm_block;
1191   _rank            = Rank;
1192   NOT_PRODUCT(_safepoint_check_required = safepoint_check_required;)
1193 #endif
1194 }
1195 
1196 Mutex::~Mutex() {
1197   assert((UNS(_owner)|UNS(_LockWord.FullWord)|UNS(_EntryList)|UNS(_WaitSet)|UNS(_OnDeck)) == 0, "");
1198 }
1199 
1200 Mutex::Mutex(int Rank, const char * name, bool allow_vm_block,
1201              SafepointCheckRequired safepoint_check_required) {
1202   ClearMonitor((Monitor *) this, name);
1203 #ifdef ASSERT
1204   _allow_vm_block   = allow_vm_block;
1205   _rank             = Rank;
1206   NOT_PRODUCT(_safepoint_check_required = safepoint_check_required;)
1207 #endif
1208 }
1209 
1210 bool Monitor::owned_by_self() const {
1211   bool ret = _owner == Thread::current();
1212   assert(!ret || _LockWord.Bytes[_LSBINDEX] != 0, "invariant");
1213   return ret;
1214 }
1215 
1216 void Monitor::print_on_error(outputStream* st) const {
1217   st->print("[" PTR_FORMAT, this);
1218   st->print("] %s", _name);
1219   st->print(" - owner thread: " PTR_FORMAT, _owner);
1220 }
1221 
1222 
1223 
1224 
1225 // ----------------------------------------------------------------------------------
1226 // Non-product code
1227 
1228 #ifndef PRODUCT
1229 void Monitor::print_on(outputStream* st) const {
1230   st->print_cr("Mutex: [0x%lx/0x%lx] %s - owner: 0x%lx", this, _LockWord.FullWord, _name, _owner);
1231 }
1232 #endif
1233 
1234 #ifndef PRODUCT
1235 #ifdef ASSERT
1236 Monitor * Monitor::get_least_ranked_lock(Monitor * locks) {
1237   Monitor *res, *tmp;
1238   for (res = tmp = locks; tmp != NULL; tmp = tmp->next()) {
1239     if (tmp->rank() < res->rank()) {
1240       res = tmp;
1241     }
1242   }
1243   if (!SafepointSynchronize::is_at_safepoint()) {
1244     // In this case, we expect the held locks to be
1245     // in increasing rank order (modulo any native ranks)
1246     for (tmp = locks; tmp != NULL; tmp = tmp->next()) {
1247       if (tmp->next() != NULL) {
1248         assert(tmp->rank() == Mutex::native ||
1249                tmp->rank() <= tmp->next()->rank(), "mutex rank anomaly?");
1250       }
1251     }
1252   }
1253   return res;
1254 }
1255 
1256 Monitor* Monitor::get_least_ranked_lock_besides_this(Monitor* locks) {
1257   Monitor *res, *tmp;
1258   for (res = NULL, tmp = locks; tmp != NULL; tmp = tmp->next()) {
1259     if (tmp != this && (res == NULL || tmp->rank() < res->rank())) {
1260       res = tmp;
1261     }
1262   }
1263   if (!SafepointSynchronize::is_at_safepoint()) {
1264     // In this case, we expect the held locks to be
1265     // in increasing rank order (modulo any native ranks)
1266     for (tmp = locks; tmp != NULL; tmp = tmp->next()) {
1267       if (tmp->next() != NULL) {
1268         assert(tmp->rank() == Mutex::native ||
1269                tmp->rank() <= tmp->next()->rank(), "mutex rank anomaly?");
1270       }
1271     }
1272   }
1273   return res;
1274 }
1275 
1276 
1277 bool Monitor::contains(Monitor* locks, Monitor * lock) {
1278   for (; locks != NULL; locks = locks->next()) {
1279     if (locks == lock) {
1280       return true;
1281     }
1282   }
1283   return false;
1284 }
1285 #endif
1286 
1287 // Called immediately after lock acquisition or release as a diagnostic
1288 // to track the lock-set of the thread and test for rank violations that
1289 // might indicate exposure to deadlock.
1290 // Rather like an EventListener for _owner (:>).
1291 
1292 void Monitor::set_owner_implementation(Thread *new_owner) {
1293   // This function is solely responsible for maintaining
1294   // and checking the invariant that threads and locks
1295   // are in a 1/N relation, with some some locks unowned.
1296   // It uses the Mutex::_owner, Mutex::_next, and
1297   // Thread::_owned_locks fields, and no other function
1298   // changes those fields.
1299   // It is illegal to set the mutex from one non-NULL
1300   // owner to another--it must be owned by NULL as an
1301   // intermediate state.
1302 
1303   if (new_owner != NULL) {
1304     // the thread is acquiring this lock
1305 
1306     assert(new_owner == Thread::current(), "Should I be doing this?");
1307     assert(_owner == NULL, "setting the owner thread of an already owned mutex");
1308     _owner = new_owner; // set the owner
1309 
1310     // link "this" into the owned locks list
1311 
1312 #ifdef ASSERT  // Thread::_owned_locks is under the same ifdef
1313     Monitor* locks = get_least_ranked_lock(new_owner->owned_locks());
1314     // Mutex::set_owner_implementation is a friend of Thread
1315 
1316     assert(this->rank() >= 0, "bad lock rank");
1317 
1318     // Deadlock avoidance rules require us to acquire Mutexes only in
1319     // a global total order. For example m1 is the lowest ranked mutex
1320     // that the thread holds and m2 is the mutex the thread is trying
1321     // to acquire, then  deadlock avoidance rules require that the rank
1322     // of m2 be less  than the rank of m1.
1323     // The rank Mutex::native  is an exception in that it is not subject
1324     // to the verification rules.
1325     // Here are some further notes relating to mutex acquisition anomalies:
1326     // . under Solaris, the interrupt lock gets acquired when doing
1327     //   profiling, so any lock could be held.
1328     // . it is also ok to acquire Safepoint_lock at the very end while we
1329     //   already hold Terminator_lock - may happen because of periodic safepoints
1330     if (this->rank() != Mutex::native &&
1331         this->rank() != Mutex::suspend_resume &&
1332         locks != NULL && locks->rank() <= this->rank() &&
1333         !SafepointSynchronize::is_at_safepoint() &&
1334         this != Interrupt_lock && this != ProfileVM_lock &&
1335         !(this == Safepoint_lock && contains(locks, Terminator_lock) &&
1336         SafepointSynchronize::is_synchronizing())) {
1337       new_owner->print_owned_locks();
1338       fatal(err_msg("acquiring lock %s/%d out of order with lock %s/%d -- "
1339                     "possible deadlock", this->name(), this->rank(),
1340                     locks->name(), locks->rank()));
1341     }
1342 
1343     this->_next = new_owner->_owned_locks;
1344     new_owner->_owned_locks = this;
1345 #endif
1346 
1347   } else {
1348     // the thread is releasing this lock
1349 
1350     Thread* old_owner = _owner;
1351     debug_only(_last_owner = old_owner);
1352 
1353     assert(old_owner != NULL, "removing the owner thread of an unowned mutex");
1354     assert(old_owner == Thread::current(), "removing the owner thread of an unowned mutex");
1355 
1356     _owner = NULL; // set the owner
1357 
1358 #ifdef ASSERT
1359     Monitor *locks = old_owner->owned_locks();
1360 
1361     // remove "this" from the owned locks list
1362 
1363     Monitor *prev = NULL;
1364     bool found = false;
1365     for (; locks != NULL; prev = locks, locks = locks->next()) {
1366       if (locks == this) {
1367         found = true;
1368         break;
1369       }
1370     }
1371     assert(found, "Removing a lock not owned");
1372     if (prev == NULL) {
1373       old_owner->_owned_locks = _next;
1374     } else {
1375       prev->_next = _next;
1376     }
1377     _next = NULL;
1378 #endif
1379   }
1380 }
1381 
1382 
1383 // Factored out common sanity checks for locking mutex'es. Used by lock() and try_lock()
1384 void Monitor::check_prelock_state(Thread *thread) {
1385   assert((!thread->is_Java_thread() || ((JavaThread *)thread)->thread_state() == _thread_in_vm)
1386          || rank() == Mutex::special, "wrong thread state for using locks");
1387   if (StrictSafepointChecks) {
1388     if (thread->is_VM_thread() && !allow_vm_block()) {
1389       fatal(err_msg("VM thread using lock %s (not allowed to block on)",
1390                     name()));
1391     }
1392     debug_only(if (rank() != Mutex::special) \
1393                thread->check_for_valid_safepoint_state(false);)
1394   }
1395   if (thread->is_Watcher_thread()) {
1396     assert(!WatcherThread::watcher_thread()->has_crash_protection(),
1397            "locking not allowed when crash protection is set");
1398   }
1399 }
1400 
1401 void Monitor::check_block_state(Thread *thread) {
1402   if (!_allow_vm_block && thread->is_VM_thread()) {
1403     warning("VM thread blocked on lock");
1404     print();
1405     BREAKPOINT;
1406   }
1407   assert(_owner != thread, "deadlock: blocking on monitor owned by current thread");
1408 }
1409 
1410 #endif // PRODUCT