1 /*
   2  * Copyright (c) 1998, 2016, 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/interfaceSupport.hpp"
  28 #include "runtime/mutex.hpp"
  29 #include "runtime/orderAccess.inline.hpp"
  30 #include "runtime/osThread.hpp"
  31 #include "runtime/thread.inline.hpp"
  32 #include "utilities/events.hpp"
  33 #ifdef TARGET_OS_FAMILY_linux
  34 # include "mutex_linux.inline.hpp"
  35 #endif
  36 #ifdef TARGET_OS_FAMILY_solaris
  37 # include "mutex_solaris.inline.hpp"
  38 #endif
  39 #ifdef TARGET_OS_FAMILY_windows
  40 # include "mutex_windows.inline.hpp"
  41 #endif
  42 #ifdef TARGET_OS_FAMILY_bsd
  43 # include "mutex_bsd.inline.hpp"
  44 #endif
  45 
  46 // 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
  47 //
  48 // Native Monitor-Mutex locking - theory of operations
  49 //
  50 // * Native Monitors are completely unrelated to Java-level monitors,
  51 //   although the "back-end" slow-path implementations share a common lineage.
  52 //   See objectMonitor:: in synchronizer.cpp.
  53 //   Native Monitors do *not* support nesting or recursion but otherwise
  54 //   they're basically Hoare-flavor monitors.
  55 //
  56 // * A thread acquires ownership of a Monitor/Mutex by CASing the LockByte
  57 //   in the _LockWord from zero to non-zero.  Note that the _Owner field
  58 //   is advisory and is used only to verify that the thread calling unlock()
  59 //   is indeed the last thread to have acquired the lock.
  60 //
  61 // * Contending threads "push" themselves onto the front of the contention
  62 //   queue -- called the cxq -- with CAS and then spin/park.
  63 //   The _LockWord contains the LockByte as well as the pointer to the head
  64 //   of the cxq.  Colocating the LockByte with the cxq precludes certain races.
  65 //
  66 // * Using a separately addressable LockByte allows for CAS:MEMBAR or CAS:0
  67 //   idioms.  We currently use MEMBAR in the uncontended unlock() path, as
  68 //   MEMBAR often has less latency than CAS.  If warranted, we could switch to
  69 //   a CAS:0 mode, using timers to close the resultant race, as is done
  70 //   with Java Monitors in synchronizer.cpp.
  71 //
  72 //   See the following for a discussion of the relative cost of atomics (CAS)
  73 //   MEMBAR, and ways to eliminate such instructions from the common-case paths:
  74 //   -- http://blogs.sun.com/dave/entry/biased_locking_in_hotspot
  75 //   -- http://blogs.sun.com/dave/resource/MustangSync.pdf
  76 //   -- http://blogs.sun.com/dave/resource/synchronization-public2.pdf
  77 //   -- synchronizer.cpp
  78 //
  79 // * Overall goals - desiderata
  80 //   1. Minimize context switching
  81 //   2. Minimize lock migration
  82 //   3. Minimize CPI -- affinity and locality
  83 //   4. Minimize the execution of high-latency instructions such as CAS or MEMBAR
  84 //   5. Minimize outer lock hold times
  85 //   6. Behave gracefully on a loaded system
  86 //
  87 // * Thread flow and list residency:
  88 //
  89 //   Contention queue --> EntryList --> OnDeck --> Owner --> !Owner
  90 //   [..resident on monitor list..]
  91 //   [...........contending..................]
  92 //
  93 //   -- The contention queue (cxq) contains recently-arrived threads (RATs).
  94 //      Threads on the cxq eventually drain into the EntryList.
  95 //   -- Invariant: a thread appears on at most one list -- cxq, EntryList
  96 //      or WaitSet -- at any one time.
  97 //   -- For a given monitor there can be at most one "OnDeck" thread at any
  98 //      given time but if needbe this particular invariant could be relaxed.
  99 //
 100 // * The WaitSet and EntryList linked lists are composed of ParkEvents.
 101 //   I use ParkEvent instead of threads as ParkEvents are immortal and
 102 //   type-stable, meaning we can safely unpark() a possibly stale
 103 //   list element in the unlock()-path.  (That's benign).
 104 //
 105 // * Succession policy - providing for progress:
 106 //
 107 //   As necessary, the unlock()ing thread identifies, unlinks, and unparks
 108 //   an "heir presumptive" tentative successor thread from the EntryList.
 109 //   This becomes the so-called "OnDeck" thread, of which there can be only
 110 //   one at any given time for a given monitor.  The wakee will recontend
 111 //   for ownership of monitor.
 112 //
 113 //   Succession is provided for by a policy of competitive handoff.
 114 //   The exiting thread does _not_ grant or pass ownership to the
 115 //   successor thread.  (This is also referred to as "handoff" succession").
 116 //   Instead the exiting thread releases ownership and possibly wakes
 117 //   a successor, so the successor can (re)compete for ownership of the lock.
 118 //
 119 //   Competitive handoff provides excellent overall throughput at the expense
 120 //   of short-term fairness.  If fairness is a concern then one remedy might
 121 //   be to add an AcquireCounter field to the monitor.  After a thread acquires
 122 //   the lock it will decrement the AcquireCounter field.  When the count
 123 //   reaches 0 the thread would reset the AcquireCounter variable, abdicate
 124 //   the lock directly to some thread on the EntryList, and then move itself to the
 125 //   tail of the EntryList.
 126 //
 127 //   But in practice most threads engage or otherwise participate in resource
 128 //   bounded producer-consumer relationships, so lock domination is not usually
 129 //   a practical concern.  Recall too, that in general it's easier to construct
 130 //   a fair lock from a fast lock, but not vice-versa.
 131 //
 132 // * The cxq can have multiple concurrent "pushers" but only one concurrent
 133 //   detaching thread.  This mechanism is immune from the ABA corruption.
 134 //   More precisely, the CAS-based "push" onto cxq is ABA-oblivious.
 135 //   We use OnDeck as a pseudo-lock to enforce the at-most-one detaching
 136 //   thread constraint.
 137 //
 138 // * Taken together, the cxq and the EntryList constitute or form a
 139 //   single logical queue of threads stalled trying to acquire the lock.
 140 //   We use two distinct lists to reduce heat on the list ends.
 141 //   Threads in lock() enqueue onto cxq while threads in unlock() will
 142 //   dequeue from the EntryList.  (c.f. Michael Scott's "2Q" algorithm).
 143 //   A key desideratum is to minimize queue & monitor metadata manipulation
 144 //   that occurs while holding the "outer" monitor lock -- that is, we want to
 145 //   minimize monitor lock holds times.
 146 //
 147 //   The EntryList is ordered by the prevailing queue discipline and
 148 //   can be organized in any convenient fashion, such as a doubly-linked list or
 149 //   a circular doubly-linked list.  If we need a priority queue then something akin
 150 //   to Solaris' sleepq would work nicely.  Viz.,
 151 //   -- http://agg.eng/ws/on10_nightly/source/usr/src/uts/common/os/sleepq.c.
 152 //   -- http://cvs.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/uts/common/os/sleepq.c
 153 //   Queue discipline is enforced at ::unlock() time, when the unlocking thread
 154 //   drains the cxq into the EntryList, and orders or reorders the threads on the
 155 //   EntryList accordingly.
 156 //
 157 //   Barring "lock barging", this mechanism provides fair cyclic ordering,
 158 //   somewhat similar to an elevator-scan.
 159 //
 160 // * OnDeck
 161 //   --  For a given monitor there can be at most one OnDeck thread at any given
 162 //       instant.  The OnDeck thread is contending for the lock, but has been
 163 //       unlinked from the EntryList and cxq by some previous unlock() operations.
 164 //       Once a thread has been designated the OnDeck thread it will remain so
 165 //       until it manages to acquire the lock -- being OnDeck is a stable property.
 166 //   --  Threads on the EntryList or cxq are _not allowed to attempt lock acquisition.
 167 //   --  OnDeck also serves as an "inner lock" as follows.  Threads in unlock() will, after
 168 //       having cleared the LockByte and dropped the outer lock,  attempt to "trylock"
 169 //       OnDeck by CASing the field from null to non-null.  If successful, that thread
 170 //       is then responsible for progress and succession and can use CAS to detach and
 171 //       drain the cxq into the EntryList.  By convention, only this thread, the holder of
 172 //       the OnDeck inner lock, can manipulate the EntryList or detach and drain the
 173 //       RATs on the cxq into the EntryList.  This avoids ABA corruption on the cxq as
 174 //       we allow multiple concurrent "push" operations but restrict detach concurrency
 175 //       to at most one thread.  Having selected and detached a successor, the thread then
 176 //       changes the OnDeck to refer to that successor, and then unparks the successor.
 177 //       That successor will eventually acquire the lock and clear OnDeck.  Beware
 178 //       that the OnDeck usage as a lock is asymmetric.  A thread in unlock() transiently
 179 //       "acquires" OnDeck, performs queue manipulations, passes OnDeck to some successor,
 180 //       and then the successor eventually "drops" OnDeck.  Note that there's never
 181 //       any sense of contention on the inner lock, however.  Threads never contend
 182 //       or wait for the inner lock.
 183 //   --  OnDeck provides for futile wakeup throttling a described in section 3.3 of
 184 //       See http://www.usenix.org/events/jvm01/full_papers/dice/dice.pdf
 185 //       In a sense, OnDeck subsumes the ObjectMonitor _Succ and ObjectWaiter
 186 //       TState fields found in Java-level objectMonitors.  (See synchronizer.cpp).
 187 //
 188 // * Waiting threads reside on the WaitSet list -- wait() puts
 189 //   the caller onto the WaitSet.  Notify() or notifyAll() simply
 190 //   transfers threads from the WaitSet to either the EntryList or cxq.
 191 //   Subsequent unlock() operations will eventually unpark the notifyee.
 192 //   Unparking a notifee in notify() proper is inefficient - if we were to do so
 193 //   it's likely the notifyee would simply impale itself on the lock held
 194 //   by the notifier.
 195 //
 196 // * The mechanism is obstruction-free in that if the holder of the transient
 197 //   OnDeck lock in unlock() is preempted or otherwise stalls, other threads
 198 //   can still acquire and release the outer lock and continue to make progress.
 199 //   At worst, waking of already blocked contending threads may be delayed,
 200 //   but nothing worse.  (We only use "trylock" operations on the inner OnDeck
 201 //   lock).
 202 //
 203 // * Note that thread-local storage must be initialized before a thread
 204 //   uses Native monitors or mutexes.  The native monitor-mutex subsystem
 205 //   depends on Thread::current().
 206 //
 207 // * The monitor synchronization subsystem avoids the use of native
 208 //   synchronization primitives except for the narrow platform-specific
 209 //   park-unpark abstraction.  See the comments in os_solaris.cpp regarding
 210 //   the semantics of park-unpark.  Put another way, this monitor implementation
 211 //   depends only on atomic operations and park-unpark.  The monitor subsystem
 212 //   manages all RUNNING->BLOCKED and BLOCKED->READY transitions while the
 213 //   underlying OS manages the READY<->RUN transitions.
 214 //
 215 // * The memory consistency model provide by lock()-unlock() is at least as
 216 //   strong or stronger than the Java Memory model defined by JSR-133.
 217 //   That is, we guarantee at least entry consistency, if not stronger.
 218 //   See http://g.oswego.edu/dl/jmm/cookbook.html.
 219 //
 220 // * Thread:: currently contains a set of purpose-specific ParkEvents:
 221 //   _MutexEvent, _ParkEvent, etc.  A better approach might be to do away with
 222 //   the purpose-specific ParkEvents and instead implement a general per-thread
 223 //   stack of available ParkEvents which we could provision on-demand.  The
 224 //   stack acts as a local cache to avoid excessive calls to ParkEvent::Allocate()
 225 //   and ::Release().  A thread would simply pop an element from the local stack before it
 226 //   enqueued or park()ed.  When the contention was over the thread would
 227 //   push the no-longer-needed ParkEvent back onto its stack.
 228 //
 229 // * A slightly reduced form of ILock() and IUnlock() have been partially
 230 //   model-checked (Murphi) for safety and progress at T=1,2,3 and 4.
 231 //   It'd be interesting to see if TLA/TLC could be useful as well.
 232 //
 233 // * Mutex-Monitor is a low-level "leaf" subsystem.  That is, the monitor
 234 //   code should never call other code in the JVM that might itself need to
 235 //   acquire monitors or mutexes.  That's true *except* in the case of the
 236 //   ThreadBlockInVM state transition wrappers.  The ThreadBlockInVM DTOR handles
 237 //   mutator reentry (ingress) by checking for a pending safepoint in which case it will
 238 //   call SafepointSynchronize::block(), which in turn may call Safepoint_lock->lock(), etc.
 239 //   In that particular case a call to lock() for a given Monitor can end up recursively
 240 //   calling lock() on another monitor.   While distasteful, this is largely benign
 241 //   as the calls come from jacket that wraps lock(), and not from deep within lock() itself.
 242 //
 243 //   It's unfortunate that native mutexes and thread state transitions were convolved.
 244 //   They're really separate concerns and should have remained that way.  Melding
 245 //   them together was facile -- a bit too facile.   The current implementation badly
 246 //   conflates the two concerns.
 247 //
 248 // * TODO-FIXME:
 249 //
 250 //   -- Add DTRACE probes for contended acquire, contended acquired, contended unlock
 251 //      We should also add DTRACE probes in the ParkEvent subsystem for
 252 //      Park-entry, Park-exit, and Unpark.
 253 //
 254 //   -- We have an excess of mutex-like constructs in the JVM, namely:
 255 //      1. objectMonitors for Java-level synchronization (synchronizer.cpp)
 256 //      2. low-level muxAcquire and muxRelease
 257 //      3. low-level spinAcquire and spinRelease
 258 //      4. native Mutex:: and Monitor::
 259 //      5. jvm_raw_lock() and _unlock()
 260 //      6. JVMTI raw monitors -- distinct from (5) despite having a confusingly
 261 //         similar name.
 262 //
 263 // 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
 264 
 265 
 266 // CASPTR() uses the canonical argument order that dominates in the literature.
 267 // Our internal cmpxchg_ptr() uses a bastardized ordering to accommodate Sun .il templates.
 268 
 269 #define CASPTR(a, c, s)  \
 270   intptr_t(Atomic::cmpxchg_ptr((void *)(s), (void *)(a), (void *)(c)))
 271 #define UNS(x) (uintptr_t(x))
 272 #define TRACE(m)                   \
 273   {                                \
 274     static volatile int ctr = 0;   \
 275     int x = ++ctr;                 \
 276     if ((x & (x - 1)) == 0) {      \
 277       ::printf("%d:%s\n", x, #m);  \
 278       ::fflush(stdout);            \
 279     }                              \
 280   }
 281 
 282 // Simplistic low-quality Marsaglia SHIFT-XOR RNG.
 283 // Bijective except for the trailing mask operation.
 284 // Useful for spin loops as the compiler can't optimize it away.
 285 
 286 static inline jint MarsagliaXORV(jint x) {
 287   if (x == 0) x = 1|os::random();
 288   x ^= x << 6;
 289   x ^= ((unsigned)x) >> 21;
 290   x ^= x << 7;
 291   return x & 0x7FFFFFFF;
 292 }
 293 
 294 static int Stall(int its) {
 295   static volatile jint rv = 1;
 296   volatile int OnFrame = 0;
 297   jint v = rv ^ UNS(OnFrame);
 298   while (--its >= 0) {
 299     v = MarsagliaXORV(v);
 300   }
 301   // Make this impossible for the compiler to optimize away,
 302   // but (mostly) avoid W coherency sharing on MP systems.
 303   if (v == 0x12345) rv = v;
 304   return v;
 305 }
 306 
 307 int Monitor::TryLock() {
 308   intptr_t v = _LockWord.FullWord;
 309   for (;;) {
 310     if ((v & _LBIT) != 0) return 0;
 311     const intptr_t u = CASPTR(&_LockWord, v, v|_LBIT);
 312     if (v == u) return 1;
 313     v = u;
 314   }
 315 }
 316 
 317 int Monitor::TryFast() {
 318   // Optimistic fast-path form ...
 319   // Fast-path attempt for the common uncontended case.
 320   // Avoid RTS->RTO $ coherence upgrade on typical SMP systems.
 321   intptr_t v = CASPTR(&_LockWord, 0, _LBIT);  // agro ...
 322   if (v == 0) return 1;
 323 
 324   for (;;) {
 325     if ((v & _LBIT) != 0) return 0;
 326     const intptr_t u = CASPTR(&_LockWord, v, v|_LBIT);
 327     if (v == u) return 1;
 328     v = u;
 329   }
 330 }
 331 
 332 int Monitor::ILocked() {
 333   const intptr_t w = _LockWord.FullWord & 0xFF;
 334   assert(w == 0 || w == _LBIT, "invariant");
 335   return w == _LBIT;
 336 }
 337 
 338 // Polite TATAS spinlock with exponential backoff - bounded spin.
 339 // Ideally we'd use processor cycles, time or vtime to control
 340 // the loop, but we currently use iterations.
 341 // All the constants within were derived empirically but work over
 342 // over the spectrum of J2SE reference platforms.
 343 // On Niagara-class systems the back-off is unnecessary but
 344 // is relatively harmless.  (At worst it'll slightly retard
 345 // acquisition times).  The back-off is critical for older SMP systems
 346 // where constant fetching of the LockWord would otherwise impair
 347 // scalability.
 348 //
 349 // Clamp spinning at approximately 1/2 of a context-switch round-trip.
 350 // See synchronizer.cpp for details and rationale.
 351 
 352 int Monitor::TrySpin(Thread * const Self) {
 353   if (TryLock())    return 1;
 354   if (!os::is_MP()) return 0;
 355 
 356   int Probes  = 0;
 357   int Delay   = 0;
 358   int Steps   = 0;
 359   int SpinMax = NativeMonitorSpinLimit;
 360   int flgs    = NativeMonitorFlags;
 361   for (;;) {
 362     intptr_t v = _LockWord.FullWord;
 363     if ((v & _LBIT) == 0) {
 364       if (CASPTR (&_LockWord, v, v|_LBIT) == v) {
 365         return 1;
 366       }
 367       continue;
 368     }
 369 
 370     if ((flgs & 8) == 0) {
 371       SpinPause();
 372     }
 373 
 374     // Periodically increase Delay -- variable Delay form
 375     // conceptually: delay *= 1 + 1/Exponent
 376     ++Probes;
 377     if (Probes > SpinMax) return 0;
 378 
 379     if ((Probes & 0x7) == 0) {
 380       Delay = ((Delay << 1)|1) & 0x7FF;
 381       // CONSIDER: Delay += 1 + (Delay/4); Delay &= 0x7FF ;
 382     }
 383 
 384     if (flgs & 2) continue;
 385 
 386     // Consider checking _owner's schedctl state, if OFFPROC abort spin.
 387     // If the owner is OFFPROC then it's unlike that the lock will be dropped
 388     // in a timely fashion, which suggests that spinning would not be fruitful
 389     // or profitable.
 390 
 391     // Stall for "Delay" time units - iterations in the current implementation.
 392     // Avoid generating coherency traffic while stalled.
 393     // Possible ways to delay:
 394     //   PAUSE, SLEEP, MEMBAR #sync, MEMBAR #halt,
 395     //   wr %g0,%asi, gethrtime, rdstick, rdtick, rdtsc, etc. ...
 396     // Note that on Niagara-class systems we want to minimize STs in the
 397     // spin loop.  N1 and brethren write-around the L1$ over the xbar into the L2$.
 398     // Furthermore, they don't have a W$ like traditional SPARC processors.
 399     // We currently use a Marsaglia Shift-Xor RNG loop.
 400     Steps += Delay;
 401     if (Self != NULL) {
 402       jint rv = Self->rng[0];
 403       for (int k = Delay; --k >= 0;) {
 404         rv = MarsagliaXORV(rv);
 405         if ((flgs & 4) == 0 && SafepointSynchronize::do_call_back()) return 0;
 406       }
 407       Self->rng[0] = rv;
 408     } else {
 409       Stall(Delay);
 410     }
 411   }
 412 }
 413 
 414 static int ParkCommon(ParkEvent * ev, jlong timo) {
 415   // Diagnostic support - periodically unwedge blocked threads
 416   intx nmt = NativeMonitorTimeout;
 417   if (nmt > 0 && (nmt < timo || timo <= 0)) {
 418     timo = nmt;
 419   }
 420   int err = OS_OK;
 421   if (0 == timo) {
 422     ev->park();
 423   } else {
 424     err = ev->park(timo);
 425   }
 426   return err;
 427 }
 428 
 429 inline int Monitor::AcquireOrPush(ParkEvent * ESelf) {
 430   intptr_t v = _LockWord.FullWord;
 431   for (;;) {
 432     if ((v & _LBIT) == 0) {
 433       const intptr_t u = CASPTR(&_LockWord, v, v|_LBIT);
 434       if (u == v) return 1;        // indicate acquired
 435       v = u;
 436     } else {
 437       // Anticipate success ...
 438       ESelf->ListNext = (ParkEvent *)(v & ~_LBIT);
 439       const intptr_t u = CASPTR(&_LockWord, v, intptr_t(ESelf)|_LBIT);
 440       if (u == v) return 0;        // indicate pushed onto cxq
 441       v = u;
 442     }
 443     // Interference - LockWord change - just retry
 444   }
 445 }
 446 
 447 // ILock and IWait are the lowest level primitive internal blocking
 448 // synchronization functions.  The callers of IWait and ILock must have
 449 // performed any needed state transitions beforehand.
 450 // IWait and ILock may directly call park() without any concern for thread state.
 451 // Note that ILock and IWait do *not* access _owner.
 452 // _owner is a higher-level logical concept.
 453 
 454 void Monitor::ILock(Thread * Self) {
 455   assert(_OnDeck != Self->_MutexEvent, "invariant");
 456 
 457   if (TryFast()) {
 458  Exeunt:
 459     assert(ILocked(), "invariant");
 460     return;
 461   }
 462 
 463   ParkEvent * const ESelf = Self->_MutexEvent;
 464   assert(_OnDeck != ESelf, "invariant");
 465 
 466   // As an optimization, spinners could conditionally try to set ONDECK to _LBIT
 467   // Synchronizer.cpp uses a similar optimization.
 468   if (TrySpin(Self)) goto Exeunt;
 469 
 470   // Slow-path - the lock is contended.
 471   // Either Enqueue Self on cxq or acquire the outer lock.
 472   // LockWord encoding = (cxq,LOCKBYTE)
 473   ESelf->reset();
 474   OrderAccess::fence();
 475 
 476   // Optional optimization ... try barging on the inner lock
 477   if ((NativeMonitorFlags & 32) && CASPTR (&_OnDeck, NULL, UNS(Self)) == 0) {
 478     goto OnDeck_LOOP;
 479   }
 480 
 481   if (AcquireOrPush(ESelf)) goto Exeunt;
 482 
 483   // At any given time there is at most one ondeck thread.
 484   // ondeck implies not resident on cxq and not resident on EntryList
 485   // Only the OnDeck thread can try to acquire -- contended for -- the lock.
 486   // CONSIDER: use Self->OnDeck instead of m->OnDeck.
 487   // Deschedule Self so that others may run.
 488   while (_OnDeck != ESelf) {
 489     ParkCommon(ESelf, 0);
 490   }
 491 
 492   // Self is now in the ONDECK position and will remain so until it
 493   // manages to acquire the lock.
 494  OnDeck_LOOP:
 495   for (;;) {
 496     assert(_OnDeck == ESelf, "invariant");
 497     if (TrySpin(Self)) break;
 498     // It's probably wise to spin only if we *actually* blocked
 499     // CONSIDER: check the lockbyte, if it remains set then
 500     // preemptively drain the cxq into the EntryList.
 501     // The best place and time to perform queue operations -- lock metadata --
 502     // is _before having acquired the outer lock, while waiting for the lock to drop.
 503     ParkCommon(ESelf, 0);
 504   }
 505 
 506   assert(_OnDeck == ESelf, "invariant");
 507   _OnDeck = NULL;
 508 
 509   // Note that we current drop the inner lock (clear OnDeck) in the slow-path
 510   // epilogue immediately after having acquired the outer lock.
 511   // But instead we could consider the following optimizations:
 512   // A. Shift or defer dropping the inner lock until the subsequent IUnlock() operation.
 513   //    This might avoid potential reacquisition of the inner lock in IUlock().
 514   // B. While still holding the inner lock, attempt to opportunistically select
 515   //    and unlink the next ONDECK thread from the EntryList.
 516   //    If successful, set ONDECK to refer to that thread, otherwise clear ONDECK.
 517   //    It's critical that the select-and-unlink operation run in constant-time as
 518   //    it executes when holding the outer lock and may artificially increase the
 519   //    effective length of the critical section.
 520   // Note that (A) and (B) are tantamount to succession by direct handoff for
 521   // the inner lock.
 522   goto Exeunt;
 523 }
 524 
 525 void Monitor::IUnlock(bool RelaxAssert) {
 526   assert(ILocked(), "invariant");
 527   // Conceptually we need a MEMBAR #storestore|#loadstore barrier or fence immediately
 528   // before the store that releases the lock.  Crucially, all the stores and loads in the
 529   // critical section must be globally visible before the store of 0 into the lock-word
 530   // that releases the lock becomes globally visible.  That is, memory accesses in the
 531   // critical section should not be allowed to bypass or overtake the following ST that
 532   // releases the lock.  As such, to prevent accesses within the critical section
 533   // from "leaking" out, we need a release fence between the critical section and the
 534   // store that releases the lock.  In practice that release barrier is elided on
 535   // platforms with strong memory models such as TSO.
 536   //
 537   // Note that the OrderAccess::storeload() fence that appears after unlock store
 538   // provides for progress conditions and succession and is _not related to exclusion
 539   // safety or lock release consistency.
 540   OrderAccess::release_store(&_LockWord.Bytes[_LSBINDEX], 0); // drop outer lock
 541 
 542   OrderAccess::storeload();
 543   ParkEvent * const w = _OnDeck;
 544   assert(RelaxAssert || w != Thread::current()->_MutexEvent, "invariant");
 545   if (w != NULL) {
 546     // Either we have a valid ondeck thread or ondeck is transiently "locked"
 547     // by some exiting thread as it arranges for succession.  The LSBit of
 548     // OnDeck allows us to discriminate two cases.  If the latter, the
 549     // responsibility for progress and succession lies with that other thread.
 550     // For good performance, we also depend on the fact that redundant unpark()
 551     // operations are cheap.  That is, repeated Unpark()ing of the ONDECK thread
 552     // is inexpensive.  This approach provides implicit futile wakeup throttling.
 553     // Note that the referent "w" might be stale with respect to the lock.
 554     // In that case the following unpark() is harmless and the worst that'll happen
 555     // is a spurious return from a park() operation.  Critically, if "w" _is stale,
 556     // then progress is known to have occurred as that means the thread associated
 557     // with "w" acquired the lock.  In that case this thread need take no further
 558     // action to guarantee progress.
 559     if ((UNS(w) & _LBIT) == 0) w->unpark();
 560     return;
 561   }
 562 
 563   intptr_t cxq = _LockWord.FullWord;
 564   if (((cxq & ~_LBIT)|UNS(_EntryList)) == 0) {
 565     return;      // normal fast-path exit - cxq and EntryList both empty
 566   }
 567   if (cxq & _LBIT) {
 568     // Optional optimization ...
 569     // Some other thread acquired the lock in the window since this
 570     // thread released it.  Succession is now that thread's responsibility.
 571     return;
 572   }
 573 
 574  Succession:
 575   // Slow-path exit - this thread must ensure succession and progress.
 576   // OnDeck serves as lock to protect cxq and EntryList.
 577   // Only the holder of OnDeck can manipulate EntryList or detach the RATs from cxq.
 578   // Avoid ABA - allow multiple concurrent producers (enqueue via push-CAS)
 579   // but only one concurrent consumer (detacher of RATs).
 580   // Consider protecting this critical section with schedctl on Solaris.
 581   // Unlike a normal lock, however, the exiting thread "locks" OnDeck,
 582   // picks a successor and marks that thread as OnDeck.  That successor
 583   // thread will then clear OnDeck once it eventually acquires the outer lock.
 584   if (CASPTR (&_OnDeck, NULL, _LBIT) != UNS(NULL)) {
 585     return;
 586   }
 587 
 588   ParkEvent * List = _EntryList;
 589   if (List != NULL) {
 590     // Transfer the head of the EntryList to the OnDeck position.
 591     // Once OnDeck, a thread stays OnDeck until it acquires the lock.
 592     // For a given lock there is at most OnDeck thread at any one instant.
 593    WakeOne:
 594     assert(List == _EntryList, "invariant");
 595     ParkEvent * const w = List;
 596     assert(RelaxAssert || w != Thread::current()->_MutexEvent, "invariant");
 597     _EntryList = w->ListNext;
 598     // as a diagnostic measure consider setting w->_ListNext = BAD
 599     assert(UNS(_OnDeck) == _LBIT, "invariant");
 600     _OnDeck = w;  // pass OnDeck to w.
 601                   // w will clear OnDeck once it acquires the outer lock
 602 
 603     // Another optional optimization ...
 604     // For heavily contended locks it's not uncommon that some other
 605     // thread acquired the lock while this thread was arranging succession.
 606     // Try to defer the unpark() operation - Delegate the responsibility
 607     // for unpark()ing the OnDeck thread to the current or subsequent owners
 608     // That is, the new owner is responsible for unparking the OnDeck thread.
 609     OrderAccess::storeload();
 610     cxq = _LockWord.FullWord;
 611     if (cxq & _LBIT) return;
 612 
 613     w->unpark();
 614     return;
 615   }
 616 
 617   cxq = _LockWord.FullWord;
 618   if ((cxq & ~_LBIT) != 0) {
 619     // The EntryList is empty but the cxq is populated.
 620     // drain RATs from cxq into EntryList
 621     // Detach RATs segment with CAS and then merge into EntryList
 622     for (;;) {
 623       // optional optimization - if locked, the owner is responsible for succession
 624       if (cxq & _LBIT) goto Punt;
 625       const intptr_t vfy = CASPTR(&_LockWord, cxq, cxq & _LBIT);
 626       if (vfy == cxq) break;
 627       cxq = vfy;
 628       // Interference - LockWord changed - Just retry
 629       // We can see concurrent interference from contending threads
 630       // pushing themselves onto the cxq or from lock-unlock operations.
 631       // From the perspective of this thread, EntryList is stable and
 632       // the cxq is prepend-only -- the head is volatile but the interior
 633       // of the cxq is stable.  In theory if we encounter interference from threads
 634       // pushing onto cxq we could simply break off the original cxq suffix and
 635       // move that segment to the EntryList, avoiding a 2nd or multiple CAS attempts
 636       // on the high-traffic LockWord variable.   For instance lets say the cxq is "ABCD"
 637       // when we first fetch cxq above.  Between the fetch -- where we observed "A"
 638       // -- and CAS -- where we attempt to CAS null over A -- "PQR" arrive,
 639       // yielding cxq = "PQRABCD".  In this case we could simply set A.ListNext
 640       // null, leaving cxq = "PQRA" and transfer the "BCD" segment to the EntryList.
 641       // Note too, that it's safe for this thread to traverse the cxq
 642       // without taking any special concurrency precautions.
 643     }
 644 
 645     // We don't currently reorder the cxq segment as we move it onto
 646     // the EntryList, but it might make sense to reverse the order
 647     // or perhaps sort by thread priority.  See the comments in
 648     // synchronizer.cpp objectMonitor::exit().
 649     assert(_EntryList == NULL, "invariant");
 650     _EntryList = List = (ParkEvent *)(cxq & ~_LBIT);
 651     assert(List != NULL, "invariant");
 652     goto WakeOne;
 653   }
 654 
 655   // cxq|EntryList is empty.
 656   // w == NULL implies that cxq|EntryList == NULL in the past.
 657   // Possible race - rare inopportune interleaving.
 658   // A thread could have added itself to cxq since this thread previously checked.
 659   // Detect and recover by refetching cxq.
 660  Punt:
 661   assert(UNS(_OnDeck) == _LBIT, "invariant");
 662   _OnDeck = NULL;            // Release inner lock.
 663   OrderAccess::storeload();   // Dekker duality - pivot point
 664 
 665   // Resample LockWord/cxq to recover from possible race.
 666   // For instance, while this thread T1 held OnDeck, some other thread T2 might
 667   // acquire the outer lock.  Another thread T3 might try to acquire the outer
 668   // lock, but encounter contention and enqueue itself on cxq.  T2 then drops the
 669   // outer lock, but skips succession as this thread T1 still holds OnDeck.
 670   // T1 is and remains responsible for ensuring succession of T3.
 671   //
 672   // Note that we don't need to recheck EntryList, just cxq.
 673   // If threads moved onto EntryList since we dropped OnDeck
 674   // that implies some other thread forced succession.
 675   cxq = _LockWord.FullWord;
 676   if ((cxq & ~_LBIT) != 0 && (cxq & _LBIT) == 0) {
 677     goto Succession;         // potential race -- re-run succession
 678   }
 679   return;
 680 }
 681 
 682 bool Monitor::notify() {
 683   assert(_owner == Thread::current(), "invariant");
 684   assert(ILocked(), "invariant");
 685   if (_WaitSet == NULL) return true;
 686   NotifyCount++;
 687 
 688   // Transfer one thread from the WaitSet to the EntryList or cxq.
 689   // Currently we just unlink the head of the WaitSet and prepend to the cxq.
 690   // And of course we could just unlink it and unpark it, too, but
 691   // in that case it'd likely impale itself on the reentry.
 692   Thread::muxAcquire(_WaitLock, "notify:WaitLock");
 693   ParkEvent * nfy = _WaitSet;
 694   if (nfy != NULL) {                  // DCL idiom
 695     _WaitSet = nfy->ListNext;
 696     assert(nfy->Notified == 0, "invariant");
 697     // push nfy onto the cxq
 698     for (;;) {
 699       const intptr_t v = _LockWord.FullWord;
 700       assert((v & 0xFF) == _LBIT, "invariant");
 701       nfy->ListNext = (ParkEvent *)(v & ~_LBIT);
 702       if (CASPTR (&_LockWord, v, UNS(nfy)|_LBIT) == v) break;
 703       // interference - _LockWord changed -- just retry
 704     }
 705     // Note that setting Notified before pushing nfy onto the cxq is
 706     // also legal and safe, but the safety properties are much more
 707     // subtle, so for the sake of code stewardship ...
 708     OrderAccess::fence();
 709     nfy->Notified = 1;
 710   }
 711   Thread::muxRelease(_WaitLock);
 712   if (nfy != NULL && (NativeMonitorFlags & 16)) {
 713     // Experimental code ... light up the wakee in the hope that this thread (the owner)
 714     // will drop the lock just about the time the wakee comes ONPROC.
 715     nfy->unpark();
 716   }
 717   assert(ILocked(), "invariant");
 718   return true;
 719 }
 720 
 721 // Currently notifyAll() transfers the waiters one-at-a-time from the waitset
 722 // to the cxq.  This could be done more efficiently with a single bulk en-mass transfer,
 723 // but in practice notifyAll() for large #s of threads is rare and not time-critical.
 724 // Beware too, that we invert the order of the waiters.  Lets say that the
 725 // waitset is "ABCD" and the cxq is "XYZ".  After a notifyAll() the waitset
 726 // will be empty and the cxq will be "DCBAXYZ".  This is benign, of course.
 727 
 728 bool Monitor::notify_all() {
 729   assert(_owner == Thread::current(), "invariant");
 730   assert(ILocked(), "invariant");
 731   while (_WaitSet != NULL) notify();
 732   return true;
 733 }
 734 
 735 int Monitor::IWait(Thread * Self, jlong timo) {
 736   assert(ILocked(), "invariant");
 737 
 738   // Phases:
 739   // 1. Enqueue Self on WaitSet - currently prepend
 740   // 2. unlock - drop the outer lock
 741   // 3. wait for either notification or timeout
 742   // 4. lock - reentry - reacquire the outer lock
 743 
 744   ParkEvent * const ESelf = Self->_MutexEvent;
 745   ESelf->Notified = 0;
 746   ESelf->reset();
 747   OrderAccess::fence();
 748 
 749   // Add Self to WaitSet
 750   // Ideally only the holder of the outer lock would manipulate the WaitSet -
 751   // That is, the outer lock would implicitly protect the WaitSet.
 752   // But if a thread in wait() encounters a timeout it will need to dequeue itself
 753   // from the WaitSet _before it becomes the owner of the lock.  We need to dequeue
 754   // as the ParkEvent -- which serves as a proxy for the thread -- can't reside
 755   // on both the WaitSet and the EntryList|cxq at the same time..  That is, a thread
 756   // on the WaitSet can't be allowed to compete for the lock until it has managed to
 757   // unlink its ParkEvent from WaitSet.  Thus the need for WaitLock.
 758   // Contention on the WaitLock is minimal.
 759   //
 760   // Another viable approach would be add another ParkEvent, "WaitEvent" to the
 761   // thread class.  The WaitSet would be composed of WaitEvents.  Only the
 762   // owner of the outer lock would manipulate the WaitSet.  A thread in wait()
 763   // could then compete for the outer lock, and then, if necessary, unlink itself
 764   // from the WaitSet only after having acquired the outer lock.  More precisely,
 765   // there would be no WaitLock.  A thread in in wait() would enqueue its WaitEvent
 766   // on the WaitSet; release the outer lock; wait for either notification or timeout;
 767   // reacquire the inner lock; and then, if needed, unlink itself from the WaitSet.
 768   //
 769   // Alternatively, a 2nd set of list link fields in the ParkEvent might suffice.
 770   // One set would be for the WaitSet and one for the EntryList.
 771   // We could also deconstruct the ParkEvent into a "pure" event and add a
 772   // new immortal/TSM "ListElement" class that referred to ParkEvents.
 773   // In that case we could have one ListElement on the WaitSet and another
 774   // on the EntryList, with both referring to the same pure Event.
 775 
 776   Thread::muxAcquire(_WaitLock, "wait:WaitLock:Add");
 777   ESelf->ListNext = _WaitSet;
 778   _WaitSet = ESelf;
 779   Thread::muxRelease(_WaitLock);
 780 
 781   // Release the outer lock
 782   // We call IUnlock (RelaxAssert=true) as a thread T1 might
 783   // enqueue itself on the WaitSet, call IUnlock(), drop the lock,
 784   // and then stall before it can attempt to wake a successor.
 785   // Some other thread T2 acquires the lock, and calls notify(), moving
 786   // T1 from the WaitSet to the cxq.  T2 then drops the lock.  T1 resumes,
 787   // and then finds *itself* on the cxq.  During the course of a normal
 788   // IUnlock() call a thread should _never find itself on the EntryList
 789   // or cxq, but in the case of wait() it's possible.
 790   // See synchronizer.cpp objectMonitor::wait().
 791   IUnlock(true);
 792 
 793   // Wait for either notification or timeout
 794   // Beware that in some circumstances we might propagate
 795   // spurious wakeups back to the caller.
 796 
 797   for (;;) {
 798     if (ESelf->Notified) break;
 799     int err = ParkCommon(ESelf, timo);
 800     if (err == OS_TIMEOUT || (NativeMonitorFlags & 1)) break;
 801   }
 802 
 803   // Prepare for reentry - if necessary, remove ESelf from WaitSet
 804   // ESelf can be:
 805   // 1. Still on the WaitSet.  This can happen if we exited the loop by timeout.
 806   // 2. On the cxq or EntryList
 807   // 3. Not resident on cxq, EntryList or WaitSet, but in the OnDeck position.
 808 
 809   OrderAccess::fence();
 810   int WasOnWaitSet = 0;
 811   if (ESelf->Notified == 0) {
 812     Thread::muxAcquire(_WaitLock, "wait:WaitLock:remove");
 813     if (ESelf->Notified == 0) {     // DCL idiom
 814       assert(_OnDeck != ESelf, "invariant");   // can't be both OnDeck and on WaitSet
 815       // ESelf is resident on the WaitSet -- unlink it.
 816       // A doubly-linked list would be better here so we can unlink in constant-time.
 817       // We have to unlink before we potentially recontend as ESelf might otherwise
 818       // end up on the cxq|EntryList -- it can't be on two lists at once.
 819       ParkEvent * p = _WaitSet;
 820       ParkEvent * q = NULL;            // classic q chases p
 821       while (p != NULL && p != ESelf) {
 822         q = p;
 823         p = p->ListNext;
 824       }
 825       assert(p == ESelf, "invariant");
 826       if (p == _WaitSet) {      // found at head
 827         assert(q == NULL, "invariant");
 828         _WaitSet = p->ListNext;
 829       } else {                  // found in interior
 830         assert(q->ListNext == p, "invariant");
 831         q->ListNext = p->ListNext;
 832       }
 833       WasOnWaitSet = 1;        // We were *not* notified but instead encountered timeout
 834     }
 835     Thread::muxRelease(_WaitLock);
 836   }
 837 
 838   // Reentry phase - reacquire the lock
 839   if (WasOnWaitSet) {
 840     // ESelf was previously on the WaitSet but we just unlinked it above
 841     // because of a timeout.  ESelf is not resident on any list and is not OnDeck
 842     assert(_OnDeck != ESelf, "invariant");
 843     ILock(Self);
 844   } else {
 845     // A prior notify() operation moved ESelf from the WaitSet to the cxq.
 846     // ESelf is now on the cxq, EntryList or at the OnDeck position.
 847     // The following fragment is extracted from Monitor::ILock()
 848     for (;;) {
 849       if (_OnDeck == ESelf && TrySpin(Self)) break;
 850       ParkCommon(ESelf, 0);
 851     }
 852     assert(_OnDeck == ESelf, "invariant");
 853     _OnDeck = NULL;
 854   }
 855 
 856   assert(ILocked(), "invariant");
 857   return WasOnWaitSet != 0;        // return true IFF timeout
 858 }
 859 
 860 
 861 // ON THE VMTHREAD SNEAKING PAST HELD LOCKS:
 862 // In particular, there are certain types of global lock that may be held
 863 // by a Java thread while it is blocked at a safepoint but before it has
 864 // written the _owner field. These locks may be sneakily acquired by the
 865 // VM thread during a safepoint to avoid deadlocks. Alternatively, one should
 866 // identify all such locks, and ensure that Java threads never block at
 867 // safepoints while holding them (_no_safepoint_check_flag). While it
 868 // seems as though this could increase the time to reach a safepoint
 869 // (or at least increase the mean, if not the variance), the latter
 870 // approach might make for a cleaner, more maintainable JVM design.
 871 //
 872 // Sneaking is vile and reprehensible and should be excised at the 1st
 873 // opportunity.  It's possible that the need for sneaking could be obviated
 874 // as follows.  Currently, a thread might (a) while TBIVM, call pthread_mutex_lock
 875 // or ILock() thus acquiring the "physical" lock underlying Monitor/Mutex.
 876 // (b) stall at the TBIVM exit point as a safepoint is in effect.  Critically,
 877 // it'll stall at the TBIVM reentry state transition after having acquired the
 878 // underlying lock, but before having set _owner and having entered the actual
 879 // critical section.  The lock-sneaking facility leverages that fact and allowed the
 880 // VM thread to logically acquire locks that had already be physically locked by mutators
 881 // but where mutators were known blocked by the reentry thread state transition.
 882 //
 883 // If we were to modify the Monitor-Mutex so that TBIVM state transitions tightly
 884 // wrapped calls to park(), then we could likely do away with sneaking.  We'd
 885 // decouple lock acquisition and parking.  The critical invariant  to eliminating
 886 // sneaking is to ensure that we never "physically" acquire the lock while TBIVM.
 887 // An easy way to accomplish this is to wrap the park calls in a narrow TBIVM jacket.
 888 // One difficulty with this approach is that the TBIVM wrapper could recurse and
 889 // call lock() deep from within a lock() call, while the MutexEvent was already enqueued.
 890 // Using a stack (N=2 at minimum) of ParkEvents would take care of that problem.
 891 //
 892 // But of course the proper ultimate approach is to avoid schemes that require explicit
 893 // sneaking or dependence on any any clever invariants or subtle implementation properties
 894 // of Mutex-Monitor and instead directly address the underlying design flaw.
 895 
 896 void Monitor::lock(Thread * Self) {
 897   // Ensure that the Monitor requires/allows safepoint checks.
 898   assert(_safepoint_check_required != Monitor::_safepoint_check_never,
 899          "This lock should never have a safepoint check: %s", name());
 900 
 901 #ifdef CHECK_UNHANDLED_OOPS
 902   // Clear unhandled oops so we get a crash right away.  Only clear for non-vm
 903   // or GC threads.
 904   if (Self->is_Java_thread()) {
 905     Self->clear_unhandled_oops();
 906   }
 907 #endif // CHECK_UNHANDLED_OOPS
 908 
 909   debug_only(check_prelock_state(Self));
 910   assert(_owner != Self, "invariant");
 911   assert(_OnDeck != Self->_MutexEvent, "invariant");
 912 
 913   if (TryFast()) {
 914  Exeunt:
 915     assert(ILocked(), "invariant");
 916     assert(owner() == NULL, "invariant");
 917     set_owner(Self);
 918     return;
 919   }
 920 
 921   // The lock is contended ...
 922 
 923   bool can_sneak = Self->is_VM_thread() && SafepointSynchronize::is_at_safepoint();
 924   if (can_sneak && _owner == NULL) {
 925     // a java thread has locked the lock but has not entered the
 926     // critical region -- let's just pretend we've locked the lock
 927     // and go on.  we note this with _snuck so we can also
 928     // pretend to unlock when the time comes.
 929     _snuck = true;
 930     goto Exeunt;
 931   }
 932 
 933   // Try a brief spin to avoid passing thru thread state transition ...
 934   if (TrySpin(Self)) goto Exeunt;
 935 
 936   check_block_state(Self);
 937   if (Self->is_Java_thread()) {
 938     // Horrible dictu - we suffer through a state transition
 939     assert(rank() > Mutex::special, "Potential deadlock with special or lesser rank mutex");
 940     ThreadBlockInVM tbivm((JavaThread *) Self);
 941     ILock(Self);
 942   } else {
 943     // Mirabile dictu
 944     ILock(Self);
 945   }
 946   goto Exeunt;
 947 }
 948 
 949 void Monitor::lock() {
 950   this->lock(Thread::current());
 951 }
 952 
 953 // Lock without safepoint check - a degenerate variant of lock().
 954 // Should ONLY be used by safepoint code and other code
 955 // that is guaranteed not to block while running inside the VM. If this is called with
 956 // thread state set to be in VM, the safepoint synchronization code will deadlock!
 957 
 958 void Monitor::lock_without_safepoint_check(Thread * Self) {
 959   // Ensure that the Monitor does not require or allow safepoint checks.
 960   assert(_safepoint_check_required != Monitor::_safepoint_check_always,
 961          "This lock should always have a safepoint check: %s", name());
 962   assert(_owner != Self, "invariant");
 963   ILock(Self);
 964   assert(_owner == NULL, "invariant");
 965   set_owner(Self);
 966 }
 967 
 968 void Monitor::lock_without_safepoint_check() {
 969   lock_without_safepoint_check(Thread::current());
 970 }
 971 
 972 
 973 // Returns true if thread succeeds in grabbing the lock, otherwise false.
 974 
 975 bool Monitor::try_lock() {
 976   Thread * const Self = Thread::current();
 977   debug_only(check_prelock_state(Self));
 978   // assert(!thread->is_inside_signal_handler(), "don't lock inside signal handler");
 979 
 980   // Special case, where all Java threads are stopped.
 981   // The lock may have been acquired but _owner is not yet set.
 982   // In that case the VM thread can safely grab the lock.
 983   // It strikes me this should appear _after the TryLock() fails, below.
 984   bool can_sneak = Self->is_VM_thread() && SafepointSynchronize::is_at_safepoint();
 985   if (can_sneak && _owner == NULL) {
 986     set_owner(Self); // Do not need to be atomic, since we are at a safepoint
 987     _snuck = true;
 988     return true;
 989   }
 990 
 991   if (TryLock()) {
 992     // We got the lock
 993     assert(_owner == NULL, "invariant");
 994     set_owner(Self);
 995     return true;
 996   }
 997   return false;
 998 }
 999 
1000 void Monitor::unlock() {
1001   assert(_owner == Thread::current(), "invariant");
1002   assert(_OnDeck != Thread::current()->_MutexEvent, "invariant");
1003   set_owner(NULL);
1004   if (_snuck) {
1005     assert(SafepointSynchronize::is_at_safepoint() && Thread::current()->is_VM_thread(), "sneak");
1006     _snuck = false;
1007     return;
1008   }
1009   IUnlock(false);
1010 }
1011 
1012 // Yet another degenerate version of Monitor::lock() or lock_without_safepoint_check()
1013 // jvm_raw_lock() and _unlock() can be called by non-Java threads via JVM_RawMonitorEnter.
1014 //
1015 // There's no expectation that JVM_RawMonitors will interoperate properly with the native
1016 // Mutex-Monitor constructs.  We happen to implement JVM_RawMonitors in terms of
1017 // native Mutex-Monitors simply as a matter of convenience.  A simple abstraction layer
1018 // over a pthread_mutex_t would work equally as well, but require more platform-specific
1019 // code -- a "PlatformMutex".  Alternatively, a simply layer over muxAcquire-muxRelease
1020 // would work too.
1021 //
1022 // Since the caller might be a foreign thread, we don't necessarily have a Thread.MutexEvent
1023 // instance available.  Instead, we transiently allocate a ParkEvent on-demand if
1024 // we encounter contention.  That ParkEvent remains associated with the thread
1025 // until it manages to acquire the lock, at which time we return the ParkEvent
1026 // to the global ParkEvent free list.  This is correct and suffices for our purposes.
1027 //
1028 // Beware that the original jvm_raw_unlock() had a "_snuck" test but that
1029 // jvm_raw_lock() didn't have the corresponding test.  I suspect that's an
1030 // oversight, but I've replicated the original suspect logic in the new code ...
1031 
1032 void Monitor::jvm_raw_lock() {
1033   assert(rank() == native, "invariant");
1034 
1035   if (TryLock()) {
1036  Exeunt:
1037     assert(ILocked(), "invariant");
1038     assert(_owner == NULL, "invariant");
1039     // This can potentially be called by non-java Threads. Thus, the Thread::current_or_null()
1040     // might return NULL. Don't call set_owner since it will break on an NULL owner
1041     // Consider installing a non-null "ANON" distinguished value instead of just NULL.
1042     _owner = Thread::current_or_null();
1043     return;
1044   }
1045 
1046   if (TrySpin(NULL)) goto Exeunt;
1047 
1048   // slow-path - apparent contention
1049   // Allocate a ParkEvent for transient use.
1050   // The ParkEvent remains associated with this thread until
1051   // the time the thread manages to acquire the lock.
1052   ParkEvent * const ESelf = ParkEvent::Allocate(NULL);
1053   ESelf->reset();
1054   OrderAccess::storeload();
1055 
1056   // Either Enqueue Self on cxq or acquire the outer lock.
1057   if (AcquireOrPush (ESelf)) {
1058     ParkEvent::Release(ESelf);      // surrender the ParkEvent
1059     goto Exeunt;
1060   }
1061 
1062   // At any given time there is at most one ondeck thread.
1063   // ondeck implies not resident on cxq and not resident on EntryList
1064   // Only the OnDeck thread can try to acquire -- contended for -- the lock.
1065   // CONSIDER: use Self->OnDeck instead of m->OnDeck.
1066   for (;;) {
1067     if (_OnDeck == ESelf && TrySpin(NULL)) break;
1068     ParkCommon(ESelf, 0);
1069   }
1070 
1071   assert(_OnDeck == ESelf, "invariant");
1072   _OnDeck = NULL;
1073   ParkEvent::Release(ESelf);      // surrender the ParkEvent
1074   goto Exeunt;
1075 }
1076 
1077 void Monitor::jvm_raw_unlock() {
1078   // Nearly the same as Monitor::unlock() ...
1079   // directly set _owner instead of using set_owner(null)
1080   _owner = NULL;
1081   if (_snuck) {         // ???
1082     assert(SafepointSynchronize::is_at_safepoint() && Thread::current()->is_VM_thread(), "sneak");
1083     _snuck = false;
1084     return;
1085   }
1086   IUnlock(false);
1087 }
1088 
1089 bool Monitor::wait(bool no_safepoint_check, long timeout,
1090                    bool as_suspend_equivalent) {
1091   // Make sure safepoint checking is used properly.
1092   assert(!(_safepoint_check_required == Monitor::_safepoint_check_never && no_safepoint_check == false),
1093          "This lock should never have a safepoint check: %s", name());
1094   assert(!(_safepoint_check_required == Monitor::_safepoint_check_always && no_safepoint_check == true),
1095          "This lock should always have a safepoint check: %s", name());
1096 
1097   Thread * const Self = Thread::current();
1098   assert(_owner == Self, "invariant");
1099   assert(ILocked(), "invariant");
1100 
1101   // as_suspend_equivalent logically implies !no_safepoint_check
1102   guarantee(!as_suspend_equivalent || !no_safepoint_check, "invariant");
1103   // !no_safepoint_check logically implies java_thread
1104   guarantee(no_safepoint_check || Self->is_Java_thread(), "invariant");
1105 
1106   #ifdef ASSERT
1107   Monitor * least = get_least_ranked_lock_besides_this(Self->owned_locks());
1108   assert(least != this, "Specification of get_least_... call above");
1109   if (least != NULL && least->rank() <= special) {
1110     tty->print("Attempting to wait on monitor %s/%d while holding"
1111                " lock %s/%d -- possible deadlock",
1112                name(), rank(), least->name(), least->rank());
1113     assert(false, "Shouldn't block(wait) while holding a lock of rank special");
1114   }
1115   #endif // ASSERT
1116 
1117   int wait_status;
1118   // conceptually set the owner to NULL in anticipation of
1119   // abdicating the lock in wait
1120   set_owner(NULL);
1121   if (no_safepoint_check) {
1122     wait_status = IWait(Self, timeout);
1123   } else {
1124     assert(Self->is_Java_thread(), "invariant");
1125     JavaThread *jt = (JavaThread *)Self;
1126 
1127     // Enter safepoint region - ornate and Rococo ...
1128     ThreadBlockInVM tbivm(jt);
1129     OSThreadWaitState osts(Self->osthread(), false /* not Object.wait() */);
1130 
1131     if (as_suspend_equivalent) {
1132       jt->set_suspend_equivalent();
1133       // cleared by handle_special_suspend_equivalent_condition() or
1134       // java_suspend_self()
1135     }
1136 
1137     wait_status = IWait(Self, timeout);
1138 
1139     // were we externally suspended while we were waiting?
1140     if (as_suspend_equivalent && jt->handle_special_suspend_equivalent_condition()) {
1141       // Our event wait has finished and we own the lock, but
1142       // while we were waiting another thread suspended us. We don't
1143       // want to hold the lock while suspended because that
1144       // would surprise the thread that suspended us.
1145       assert(ILocked(), "invariant");
1146       IUnlock(true);
1147       jt->java_suspend_self();
1148       ILock(Self);
1149       assert(ILocked(), "invariant");
1150     }
1151   }
1152 
1153   // Conceptually reestablish ownership of the lock.
1154   // The "real" lock -- the LockByte -- was reacquired by IWait().
1155   assert(ILocked(), "invariant");
1156   assert(_owner == NULL, "invariant");
1157   set_owner(Self);
1158   return wait_status != 0;          // return true IFF timeout
1159 }
1160 
1161 Monitor::~Monitor() {
1162   assert((UNS(_owner)|UNS(_LockWord.FullWord)|UNS(_EntryList)|UNS(_WaitSet)|UNS(_OnDeck)) == 0, "");
1163 }
1164 
1165 void Monitor::ClearMonitor(Monitor * m, const char *name) {
1166   m->_owner             = NULL;
1167   m->_snuck             = false;
1168   if (name == NULL) {
1169     strcpy(m->_name, "UNKNOWN");
1170   } else {
1171     strncpy(m->_name, name, MONITOR_NAME_LEN - 1);
1172     m->_name[MONITOR_NAME_LEN - 1] = '\0';
1173   }
1174   m->_LockWord.FullWord = 0;
1175   m->_EntryList         = NULL;
1176   m->_OnDeck            = NULL;
1177   m->_WaitSet           = NULL;
1178   m->_WaitLock[0]       = 0;
1179 }
1180 
1181 Monitor::Monitor() { ClearMonitor(this); }
1182 
1183 Monitor::Monitor(int Rank, const char * name, bool allow_vm_block,
1184                  SafepointCheckRequired safepoint_check_required) {
1185   ClearMonitor(this, name);
1186 #ifdef ASSERT
1187   _allow_vm_block  = allow_vm_block;
1188   _rank            = Rank;
1189   NOT_PRODUCT(_safepoint_check_required = safepoint_check_required;)
1190 #endif
1191 }
1192 
1193 Mutex::~Mutex() {
1194   assert((UNS(_owner)|UNS(_LockWord.FullWord)|UNS(_EntryList)|UNS(_WaitSet)|UNS(_OnDeck)) == 0, "");
1195 }
1196 
1197 Mutex::Mutex(int Rank, const char * name, bool allow_vm_block,
1198              SafepointCheckRequired safepoint_check_required) {
1199   ClearMonitor((Monitor *) this, name);
1200 #ifdef ASSERT
1201   _allow_vm_block   = allow_vm_block;
1202   _rank             = Rank;
1203   NOT_PRODUCT(_safepoint_check_required = safepoint_check_required;)
1204 #endif
1205 }
1206 
1207 bool Monitor::owned_by_self() const {
1208   bool ret = _owner == Thread::current();
1209   assert(!ret || _LockWord.Bytes[_LSBINDEX] != 0, "invariant");
1210   return ret;
1211 }
1212 
1213 void Monitor::print_on_error(outputStream* st) const {
1214   st->print("[" PTR_FORMAT, p2i(this));
1215   st->print("] %s", _name);
1216   st->print(" - owner thread: " PTR_FORMAT, p2i(_owner));
1217 }
1218 
1219 
1220 
1221 
1222 // ----------------------------------------------------------------------------------
1223 // Non-product code
1224 
1225 #ifndef PRODUCT
1226 void Monitor::print_on(outputStream* st) const {
1227   st->print_cr("Mutex: [" PTR_FORMAT "/" PTR_FORMAT "] %s - owner: " PTR_FORMAT,
1228                p2i(this), _LockWord.FullWord, _name, p2i(_owner));
1229 }
1230 #endif
1231 
1232 #ifndef PRODUCT
1233 #ifdef ASSERT
1234 Monitor * Monitor::get_least_ranked_lock(Monitor * locks) {
1235   Monitor *res, *tmp;
1236   for (res = tmp = locks; tmp != NULL; tmp = tmp->next()) {
1237     if (tmp->rank() < res->rank()) {
1238       res = tmp;
1239     }
1240   }
1241   if (!SafepointSynchronize::is_at_safepoint()) {
1242     // In this case, we expect the held locks to be
1243     // in increasing rank order (modulo any native ranks)
1244     for (tmp = locks; tmp != NULL; tmp = tmp->next()) {
1245       if (tmp->next() != NULL) {
1246         assert(tmp->rank() == Mutex::native ||
1247                tmp->rank() <= tmp->next()->rank(), "mutex rank anomaly?");
1248       }
1249     }
1250   }
1251   return res;
1252 }
1253 
1254 Monitor* Monitor::get_least_ranked_lock_besides_this(Monitor* locks) {
1255   Monitor *res, *tmp;
1256   for (res = NULL, tmp = locks; tmp != NULL; tmp = tmp->next()) {
1257     if (tmp != this && (res == NULL || tmp->rank() < res->rank())) {
1258       res = tmp;
1259     }
1260   }
1261   if (!SafepointSynchronize::is_at_safepoint()) {
1262     // In this case, we expect the held locks to be
1263     // in increasing rank order (modulo any native ranks)
1264     for (tmp = locks; tmp != NULL; tmp = tmp->next()) {
1265       if (tmp->next() != NULL) {
1266         assert(tmp->rank() == Mutex::native ||
1267                tmp->rank() <= tmp->next()->rank(), "mutex rank anomaly?");
1268       }
1269     }
1270   }
1271   return res;
1272 }
1273 
1274 
1275 bool Monitor::contains(Monitor* locks, Monitor * lock) {
1276   for (; locks != NULL; locks = locks->next()) {
1277     if (locks == lock) {
1278       return true;
1279     }
1280   }
1281   return false;
1282 }
1283 #endif
1284 
1285 // Called immediately after lock acquisition or release as a diagnostic
1286 // to track the lock-set of the thread and test for rank violations that
1287 // might indicate exposure to deadlock.
1288 // Rather like an EventListener for _owner (:>).
1289 
1290 void Monitor::set_owner_implementation(Thread *new_owner) {
1291   // This function is solely responsible for maintaining
1292   // and checking the invariant that threads and locks
1293   // are in a 1/N relation, with some some locks unowned.
1294   // It uses the Mutex::_owner, Mutex::_next, and
1295   // Thread::_owned_locks fields, and no other function
1296   // changes those fields.
1297   // It is illegal to set the mutex from one non-NULL
1298   // owner to another--it must be owned by NULL as an
1299   // intermediate state.
1300 
1301   if (new_owner != NULL) {
1302     // the thread is acquiring this lock
1303 
1304     assert(new_owner == Thread::current(), "Should I be doing this?");
1305     assert(_owner == NULL, "setting the owner thread of an already owned mutex");
1306     _owner = new_owner; // set the owner
1307 
1308     // link "this" into the owned locks list
1309 
1310 #ifdef ASSERT  // Thread::_owned_locks is under the same ifdef
1311     Monitor* locks = get_least_ranked_lock(new_owner->owned_locks());
1312     // Mutex::set_owner_implementation is a friend of Thread
1313 
1314     assert(this->rank() >= 0, "bad lock rank");
1315 
1316     // Deadlock avoidance rules require us to acquire Mutexes only in
1317     // a global total order. For example m1 is the lowest ranked mutex
1318     // that the thread holds and m2 is the mutex the thread is trying
1319     // to acquire, then  deadlock avoidance rules require that the rank
1320     // of m2 be less  than the rank of m1.
1321     // The rank Mutex::native  is an exception in that it is not subject
1322     // to the verification rules.
1323     // Here are some further notes relating to mutex acquisition anomalies:
1324     // . it is also ok to acquire Safepoint_lock at the very end while we
1325     //   already hold Terminator_lock - may happen because of periodic safepoints
1326     if (this->rank() != Mutex::native &&
1327         this->rank() != Mutex::suspend_resume &&
1328         locks != NULL && locks->rank() <= this->rank() &&
1329         !SafepointSynchronize::is_at_safepoint() &&
1330         !(this == Safepoint_lock && contains(locks, Terminator_lock) &&
1331         SafepointSynchronize::is_synchronizing())) {
1332       new_owner->print_owned_locks();
1333       fatal("acquiring lock %s/%d out of order with lock %s/%d -- "
1334             "possible deadlock", this->name(), this->rank(),
1335             locks->name(), locks->rank());
1336     }
1337 
1338     this->_next = new_owner->_owned_locks;
1339     new_owner->_owned_locks = this;
1340 #endif
1341 
1342   } else {
1343     // the thread is releasing this lock
1344 
1345     Thread* old_owner = _owner;
1346     debug_only(_last_owner = old_owner);
1347 
1348     assert(old_owner != NULL, "removing the owner thread of an unowned mutex");
1349     assert(old_owner == Thread::current(), "removing the owner thread of an unowned mutex");
1350 
1351     _owner = NULL; // set the owner
1352 
1353 #ifdef ASSERT
1354     Monitor *locks = old_owner->owned_locks();
1355 
1356     // remove "this" from the owned locks list
1357 
1358     Monitor *prev = NULL;
1359     bool found = false;
1360     for (; locks != NULL; prev = locks, locks = locks->next()) {
1361       if (locks == this) {
1362         found = true;
1363         break;
1364       }
1365     }
1366     assert(found, "Removing a lock not owned");
1367     if (prev == NULL) {
1368       old_owner->_owned_locks = _next;
1369     } else {
1370       prev->_next = _next;
1371     }
1372     _next = NULL;
1373 #endif
1374   }
1375 }
1376 
1377 
1378 // Factored out common sanity checks for locking mutex'es. Used by lock() and try_lock()
1379 void Monitor::check_prelock_state(Thread *thread) {
1380   assert((!thread->is_Java_thread() || ((JavaThread *)thread)->thread_state() == _thread_in_vm)
1381          || rank() == Mutex::special, "wrong thread state for using locks");
1382   if (StrictSafepointChecks) {
1383     if (thread->is_VM_thread() && !allow_vm_block()) {
1384       fatal("VM thread using lock %s (not allowed to block on)", name());
1385     }
1386     debug_only(if (rank() != Mutex::special) \
1387                thread->check_for_valid_safepoint_state(false);)
1388   }
1389   if (thread->is_Watcher_thread()) {
1390     assert(!WatcherThread::watcher_thread()->has_crash_protection(),
1391            "locking not allowed when crash protection is set");
1392   }
1393 }
1394 
1395 void Monitor::check_block_state(Thread *thread) {
1396   if (!_allow_vm_block && thread->is_VM_thread()) {
1397     warning("VM thread blocked on lock");
1398     print();
1399     BREAKPOINT;
1400   }
1401   assert(_owner != thread, "deadlock: blocking on monitor owned by current thread");
1402 }
1403 
1404 #endif // PRODUCT