ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/jsr166/jsr166/src/jsr166y/ForkJoinPool.java
(Generate patch)

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.53 by dl, Mon Apr 5 15:52:26 2010 UTC vs.
Revision 1.85 by dl, Sun Nov 21 13:55:04 2010 UTC

# Line 6 | Line 6
6  
7   package jsr166y;
8  
9 import java.util.concurrent.*;
10
9   import java.util.ArrayList;
10   import java.util.Arrays;
11   import java.util.Collection;
12   import java.util.Collections;
13   import java.util.List;
14 + import java.util.concurrent.AbstractExecutorService;
15 + import java.util.concurrent.Callable;
16 + import java.util.concurrent.ExecutorService;
17 + import java.util.concurrent.Future;
18 + import java.util.concurrent.RejectedExecutionException;
19 + import java.util.concurrent.RunnableFuture;
20 + import java.util.concurrent.TimeUnit;
21 + import java.util.concurrent.TimeoutException;
22 + import java.util.concurrent.atomic.AtomicInteger;
23   import java.util.concurrent.locks.LockSupport;
24   import java.util.concurrent.locks.ReentrantLock;
18 import java.util.concurrent.atomic.AtomicInteger;
19 import java.util.concurrent.CountDownLatch;
25  
26   /**
27   * An {@link ExecutorService} for running {@link ForkJoinTask}s.
28   * A {@code ForkJoinPool} provides the entry point for submissions
29 < * from non-{@code ForkJoinTask}s, as well as management and
29 > * from non-{@code ForkJoinTask} clients, as well as management and
30   * monitoring operations.
31   *
32   * <p>A {@code ForkJoinPool} differs from other kinds of {@link
# Line 30 | Line 35 | import java.util.concurrent.CountDownLat
35   * execute subtasks created by other active tasks (eventually blocking
36   * waiting for work if none exist). This enables efficient processing
37   * when most tasks spawn other subtasks (as do most {@code
38 < * ForkJoinTask}s). A {@code ForkJoinPool} may also be used for mixed
39 < * execution of some plain {@code Runnable}- or {@code Callable}-
40 < * based activities along with {@code ForkJoinTask}s. When setting
36 < * {@linkplain #setAsyncMode async mode}, a {@code ForkJoinPool} may
37 < * also be appropriate for use with fine-grained tasks of any form
38 < * that are never joined. Otherwise, other {@code ExecutorService}
39 < * implementations are typically more appropriate choices.
38 > * ForkJoinTask}s). When setting <em>asyncMode</em> to true in
39 > * constructors, {@code ForkJoinPool}s may also be appropriate for use
40 > * with event-style tasks that are never joined.
41   *
42   * <p>A {@code ForkJoinPool} is constructed with a given target
43   * parallelism level; by default, equal to the number of available
44 < * processors. Unless configured otherwise via {@link
45 < * #setMaintainsParallelism}, the pool attempts to maintain this
46 < * number of active (or available) threads by dynamically adding,
47 < * suspending, or resuming internal worker threads, even if some tasks
48 < * are stalled waiting to join others. However, no such adjustments
49 < * are performed in the face of blocked IO or other unmanaged
50 < * synchronization. The nested {@link ManagedBlocker} interface
50 < * enables extension of the kinds of synchronization accommodated.
51 < * The target parallelism level may also be changed dynamically
52 < * ({@link #setParallelism}). The total number of threads may be
53 < * limited using method {@link #setMaximumPoolSize}, in which case it
54 < * may become possible for the activities of a pool to stall due to
55 < * the lack of available threads to process new tasks. When the pool
56 < * is executing tasks, these and other configuration setting methods
57 < * may only gradually affect actual pool sizes. It is normally best
58 < * practice to invoke these methods only when the pool is known to be
59 < * quiescent.
44 > * processors. The pool attempts to maintain enough active (or
45 > * available) threads by dynamically adding, suspending, or resuming
46 > * internal worker threads, even if some tasks are stalled waiting to
47 > * join others. However, no such adjustments are guaranteed in the
48 > * face of blocked IO or other unmanaged synchronization. The nested
49 > * {@link ManagedBlocker} interface enables extension of the kinds of
50 > * synchronization accommodated.
51   *
52   * <p>In addition to execution and lifecycle control methods, this
53   * class provides status check methods (for example
# Line 65 | Line 56 | import java.util.concurrent.CountDownLat
56   * {@link #toString} returns indications of pool state in a
57   * convenient form for informal monitoring.
58   *
59 + * <p> As is the case with other ExecutorServices, there are three
60 + * main task execution methods summarized in the following
61 + * table. These are designed to be used by clients not already engaged
62 + * in fork/join computations in the current pool.  The main forms of
63 + * these methods accept instances of {@code ForkJoinTask}, but
64 + * overloaded forms also allow mixed execution of plain {@code
65 + * Runnable}- or {@code Callable}- based activities as well.  However,
66 + * tasks that are already executing in a pool should normally
67 + * <em>NOT</em> use these pool execution methods, but instead use the
68 + * within-computation forms listed in the table.
69 + *
70 + * <table BORDER CELLPADDING=3 CELLSPACING=1>
71 + *  <tr>
72 + *    <td></td>
73 + *    <td ALIGN=CENTER> <b>Call from non-fork/join clients</b></td>
74 + *    <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td>
75 + *  </tr>
76 + *  <tr>
77 + *    <td> <b>Arrange async execution</td>
78 + *    <td> {@link #execute(ForkJoinTask)}</td>
79 + *    <td> {@link ForkJoinTask#fork}</td>
80 + *  </tr>
81 + *  <tr>
82 + *    <td> <b>Await and obtain result</td>
83 + *    <td> {@link #invoke(ForkJoinTask)}</td>
84 + *    <td> {@link ForkJoinTask#invoke}</td>
85 + *  </tr>
86 + *  <tr>
87 + *    <td> <b>Arrange exec and obtain Future</td>
88 + *    <td> {@link #submit(ForkJoinTask)}</td>
89 + *    <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
90 + *  </tr>
91 + * </table>
92 + *
93   * <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is
94   * used for all parallel task execution in a program or subsystem.
95   * Otherwise, use would not usually outweigh the construction and
# Line 89 | Line 114 | import java.util.concurrent.CountDownLat
114   * {@code IllegalArgumentException}.
115   *
116   * <p>This implementation rejects submitted tasks (that is, by throwing
117 < * {@link RejectedExecutionException}) only when the pool is shut down.
117 > * {@link RejectedExecutionException}) only when the pool is shut down
118 > * or internal resources have been exhausted.
119   *
120   * @since 1.7
121   * @author Doug Lea
# Line 116 | Line 142 | public class ForkJoinPool extends Abstra
142       * of tasks profit from cache affinities, but others are harmed by
143       * cache pollution effects.)
144       *
145 +     * Beyond work-stealing support and essential bookkeeping, the
146 +     * main responsibility of this framework is to take actions when
147 +     * one worker is waiting to join a task stolen (or always held by)
148 +     * another.  Because we are multiplexing many tasks on to a pool
149 +     * of workers, we can't just let them block (as in Thread.join).
150 +     * We also cannot just reassign the joiner's run-time stack with
151 +     * another and replace it later, which would be a form of
152 +     * "continuation", that even if possible is not necessarily a good
153 +     * idea. Given that the creation costs of most threads on most
154 +     * systems mainly surrounds setting up runtime stacks, thread
155 +     * creation and switching is usually not much more expensive than
156 +     * stack creation and switching, and is more flexible). Instead we
157 +     * combine two tactics:
158 +     *
159 +     *   Helping: Arranging for the joiner to execute some task that it
160 +     *      would be running if the steal had not occurred.  Method
161 +     *      ForkJoinWorkerThread.helpJoinTask tracks joining->stealing
162 +     *      links to try to find such a task.
163 +     *
164 +     *   Compensating: Unless there are already enough live threads,
165 +     *      method helpMaintainParallelism() may create or
166 +     *      re-activate a spare thread to compensate for blocked
167 +     *      joiners until they unblock.
168 +     *
169 +     * It is impossible to keep exactly the target (parallelism)
170 +     * number of threads running at any given time.  Determining
171 +     * existence of conservatively safe helping targets, the
172 +     * availability of already-created spares, and the apparent need
173 +     * to create new spares are all racy and require heuristic
174 +     * guidance, so we rely on multiple retries of each.  Compensation
175 +     * occurs in slow-motion. It is triggered only upon timeouts of
176 +     * Object.wait used for joins. This reduces poor decisions that
177 +     * would otherwise be made when threads are waiting for others
178 +     * that are stalled because of unrelated activities such as
179 +     * garbage collection.
180 +     *
181 +     * The ManagedBlocker extension API can't use helping so relies
182 +     * only on compensation in method awaitBlocker.
183 +     *
184       * The main throughput advantages of work-stealing stem from
185       * decentralized control -- workers mostly steal tasks from each
186       * other. We do not want to negate this by creating bottlenecks
187 <     * implementing the management responsibilities of this class. So
188 <     * we use a collection of techniques that avoid, reduce, or cope
189 <     * well with contention. These entail several instances of
190 <     * bit-packing into CASable fields to maintain only the minimally
191 <     * required atomicity. To enable such packing, we restrict maximum
192 <     * parallelism to (1<<15)-1 (enabling twice this to fit into a 16
193 <     * bit field), which is far in excess of normal operating range.
194 <     * Even though updates to some of these bookkeeping fields do
195 <     * sometimes contend with each other, they don't normally
196 <     * cache-contend with updates to others enough to warrant memory
197 <     * padding or isolation. So they are all held as fields of
198 <     * ForkJoinPool objects.  The main capabilities are as follows:
187 >     * implementing other management responsibilities. So we use a
188 >     * collection of techniques that avoid, reduce, or cope well with
189 >     * contention. These entail several instances of bit-packing into
190 >     * CASable fields to maintain only the minimally required
191 >     * atomicity. To enable such packing, we restrict maximum
192 >     * parallelism to (1<<15)-1 (enabling twice this (to accommodate
193 >     * unbalanced increments and decrements) to fit into a 16 bit
194 >     * field, which is far in excess of normal operating range.  Even
195 >     * though updates to some of these bookkeeping fields do sometimes
196 >     * contend with each other, they don't normally cache-contend with
197 >     * updates to others enough to warrant memory padding or
198 >     * isolation. So they are all held as fields of ForkJoinPool
199 >     * objects.  The main capabilities are as follows:
200       *
201       * 1. Creating and removing workers. Workers are recorded in the
202       * "workers" array. This is an array as opposed to some other data
# Line 140 | Line 206 | public class ForkJoinPool extends Abstra
206       * (workerLock) but the array is otherwise concurrently readable,
207       * and accessed directly by workers. To simplify index-based
208       * operations, the array size is always a power of two, and all
209 <     * readers must tolerate null slots. Currently, all but the first
210 <     * worker thread creation is on-demand, triggered by task
211 <     * submissions, replacement of terminated workers, and/or
212 <     * compensation for blocked workers. However, all other support
213 <     * code is set up to work with other policies.
209 >     * readers must tolerate null slots. Currently, all worker thread
210 >     * creation is on-demand, triggered by task submissions,
211 >     * replacement of terminated workers, and/or compensation for
212 >     * blocked workers. However, all other support code is set up to
213 >     * work with other policies.
214 >     *
215 >     * To ensure that we do not hold on to worker references that
216 >     * would prevent GC, ALL accesses to workers are via indices into
217 >     * the workers array (which is one source of some of the unusual
218 >     * code constructions here). In essence, the workers array serves
219 >     * as a WeakReference mechanism. Thus for example the event queue
220 >     * stores worker indices, not worker references. Access to the
221 >     * workers in associated methods (for example releaseEventWaiters)
222 >     * must both index-check and null-check the IDs. All such accesses
223 >     * ignore bad IDs by returning out early from what they are doing,
224 >     * since this can only be associated with shutdown, in which case
225 >     * it is OK to give up. On termination, we just clobber these
226 >     * data structures without trying to use them.
227       *
228       * 2. Bookkeeping for dynamically adding and removing workers. We
229 <     * maintain a given level of parallelism (or, if
230 <     * maintainsParallelism is false, at least avoid starvation). When
152 <     * some workers are known to be blocked (on joins or via
229 >     * aim to approximately maintain the given level of parallelism.
230 >     * When some workers are known to be blocked (on joins or via
231       * ManagedBlocker), we may create or resume others to take their
232       * place until they unblock (see below). Implementing this
233       * requires counts of the number of "running" threads (i.e., those
234 <     * that are neither blocked nor artifically suspended) as well as
234 >     * that are neither blocked nor artificially suspended) as well as
235       * the total number.  These two values are packed into one field,
236       * "workerCounts" because we need accurate snapshots when deciding
237 <     * to create, resume or suspend.  To support these decisions,
238 <     * updates must be prospective (not retrospective).  For example,
239 <     * the running count is decremented before blocking by a thread
240 <     * about to block, but incremented by the thread about to unblock
163 <     * it. (In a few cases, these prospective updates may need to be
164 <     * rolled back, for example when deciding to create a new worker
165 <     * but the thread factory fails or returns null. In these cases,
166 <     * we are no worse off wrt other decisions than we would be
167 <     * otherwise.)  Updates to the workerCounts field sometimes
168 <     * transiently encounter a fair amount of contention when join
169 <     * dependencies are such that many threads block or unblock at
170 <     * about the same time. We alleviate this by sometimes bundling
171 <     * updates (for example blocking one thread on join and resuming a
172 <     * spare cancel each other out), and in most other cases
173 <     * performing an alternative action (like releasing waiters and
174 <     * finding spares; see below) as a more productive form of
175 <     * backoff.
237 >     * to create, resume or suspend.  Note however that the
238 >     * correspondence of these counts to reality is not guaranteed. In
239 >     * particular updates for unblocked threads may lag until they
240 >     * actually wake up.
241       *
242       * 3. Maintaining global run state. The run state of the pool
243       * consists of a runLevel (SHUTDOWN, TERMINATING, etc) similar to
# Line 201 | Line 266 | public class ForkJoinPool extends Abstra
266       * workers that previously could not find a task to now find one:
267       * Submission of a new task to the pool, or another worker pushing
268       * a task onto a previously empty queue.  (We also use this
269 <     * mechanism for termination and reconfiguration actions that
269 >     * mechanism for configuration and termination actions that
270       * require wakeups of idle workers).  Each worker maintains its
271       * last known event count, and blocks when a scan for work did not
272       * find a task AND its lastEventCount matches the current
# Line 212 | Line 277 | public class ForkJoinPool extends Abstra
277       * a record (field nextEventWaiter) for the next waiting worker.
278       * In addition to allowing simpler decisions about need for
279       * wakeup, the event count bits in eventWaiters serve the role of
280 <     * tags to avoid ABA errors in Treiber stacks.  To reduce delays
281 <     * in task diffusion, workers not otherwise occupied may invoke
282 <     * method releaseWaiters, that removes and signals (unparks)
283 <     * workers not waiting on current count. To minimize task
284 <     * production stalls associate with signalling, any worker pushing
285 <     * a task on an empty queue invokes the weaker method signalWork,
221 <     * that only releases idle workers until it detects interference
222 <     * by other threads trying to release, and lets them take
223 <     * over. The net effect is a tree-like diffusion of signals, where
224 <     * released threads and possibly others) help with unparks.  To
225 <     * further reduce contention effects a bit, failed CASes to
226 <     * increment field eventCount are tolerated without retries.
280 >     * tags to avoid ABA errors in Treiber stacks. Upon any wakeup,
281 >     * released threads also try to release at most two others.  The
282 >     * net effect is a tree-like diffusion of signals, where released
283 >     * threads (and possibly others) help with unparks.  To further
284 >     * reduce contention effects a bit, failed CASes to increment
285 >     * field eventCount are tolerated without retries in signalWork.
286       * Conceptually they are merged into the same event, which is OK
287       * when their only purpose is to enable workers to scan for work.
288       *
289 <     * 5. Managing suspension of extra workers. When a worker is about
290 <     * to block waiting for a join (or via ManagedBlockers), we may
291 <     * create a new thread to maintain parallelism level, or at least
292 <     * avoid starvation (see below). Usually, extra threads are needed
293 <     * for only very short periods, yet join dependencies are such
294 <     * that we sometimes need them in bursts. Rather than create new
295 <     * threads each time this happens, we suspend no-longer-needed
296 <     * extra ones as "spares". For most purposes, we don't distinguish
297 <     * "extra" spare threads from normal "core" threads: On each call
298 <     * to preStep (the only point at which we can do this) a worker
299 <     * checks to see if there are now too many running workers, and if
300 <     * so, suspends itself.  Methods preJoin and doBlock look for
301 <     * suspended threads to resume before considering creating a new
302 <     * replacement. We don't need a special data structure to maintain
303 <     * spares; simply scanning the workers array looking for
304 <     * worker.isSuspended() is fine because the calling thread is
305 <     * otherwise not doing anything useful anyway; we are at least as
306 <     * happy if after locating a spare, the caller doesn't actually
307 <     * block because the join is ready before we try to adjust and
308 <     * compensate.  Note that this is intrinsically racy.  One thread
309 <     * may become a spare at about the same time as another is
310 <     * needlessly being created. We counteract this and related slop
311 <     * in part by requiring resumed spares to immediately recheck (in
312 <     * preStep) to see whether they they should re-suspend. The only
313 <     * effective difference between "extra" and "core" threads is that
314 <     * we allow the "extra" ones to time out and die if they are not
315 <     * resumed within a keep-alive interval of a few seconds. This is
316 <     * implemented mainly within ForkJoinWorkerThread, but requires
317 <     * some coordination (isTrimmed() -- meaning killed while
318 <     * suspended) to correctly maintain pool counts.
319 <     *
320 <     * 6. Deciding when to create new workers. The main dynamic
321 <     * control in this class is deciding when to create extra threads,
322 <     * in methods preJoin and doBlock. We always need to create one
323 <     * when the number of running threads becomes zero. But because
324 <     * blocked joins are typically dependent, we don't necessarily
325 <     * need or want one-to-one replacement. Using a one-to-one
326 <     * compensation rule often leads to enough useless overhead
327 <     * creating, suspending, resuming, and/or killing threads to
328 <     * signficantly degrade throughput.  We use a rule reflecting the
329 <     * idea that, the more spare threads you already have, the more
330 <     * evidence you need to create another one; where "evidence" is
331 <     * expressed as the current deficit -- target minus running
332 <     * threads. To reduce flickering and drift around target values,
333 <     * the relation is quadratic: adding a spare if (dc*dc)>=(sc*pc)
334 <     * (where dc is deficit, sc is number of spare threads and pc is
335 <     * target parallelism.)  This effectively reduces churn at the
336 <     * price of systematically undershooting target parallelism when
337 <     * many threads are blocked.  However, biasing toward undeshooting
338 <     * partially compensates for the above mechanics to suspend extra
339 <     * threads, that normally lead to overshoot because we can only
281 <     * suspend workers in-between top-level actions. It also better
282 <     * copes with the fact that some of the methods in this class tend
283 <     * to never become compiled (but are interpreted), so some
284 <     * components of the entire set of controls might execute many
285 <     * times faster than others. And similarly for cases where the
286 <     * apparent lack of work is just due to GC stalls and other
287 <     * transient system activity.
288 <     *
289 <     * 7. Maintaining other configuration parameters and monitoring
290 <     * statistics. Updates to fields controlling parallelism level,
291 <     * max size, etc can only meaningfully take effect for individual
292 <     * threads upon their next top-level actions; i.e., between
293 <     * stealing/running tasks/submission, which are separated by calls
294 <     * to preStep.  Memory ordering for these (assumed infrequent)
295 <     * reconfiguration calls is ensured by using reads and writes to
296 <     * volatile field workerCounts (that must be read in preStep anyway)
297 <     * as "fences" -- user-level reads are preceded by reads of
298 <     * workCounts, and writes are followed by no-op CAS to
299 <     * workerCounts. The values reported by other management and
300 <     * monitoring methods are either computed on demand, or are kept
301 <     * in fields that are only updated when threads are otherwise
302 <     * idle.
289 >     * 5. Managing suspension of extra workers. When a worker notices
290 >     * (usually upon timeout of a wait()) that there are too few
291 >     * running threads, we may create a new thread to maintain
292 >     * parallelism level, or at least avoid starvation. Usually, extra
293 >     * threads are needed for only very short periods, yet join
294 >     * dependencies are such that we sometimes need them in
295 >     * bursts. Rather than create new threads each time this happens,
296 >     * we suspend no-longer-needed extra ones as "spares". For most
297 >     * purposes, we don't distinguish "extra" spare threads from
298 >     * normal "core" threads: On each call to preStep (the only point
299 >     * at which we can do this) a worker checks to see if there are
300 >     * now too many running workers, and if so, suspends itself.
301 >     * Method helpMaintainParallelism looks for suspended threads to
302 >     * resume before considering creating a new replacement. The
303 >     * spares themselves are encoded on another variant of a Treiber
304 >     * Stack, headed at field "spareWaiters".  Note that the use of
305 >     * spares is intrinsically racy.  One thread may become a spare at
306 >     * about the same time as another is needlessly being created. We
307 >     * counteract this and related slop in part by requiring resumed
308 >     * spares to immediately recheck (in preStep) to see whether they
309 >     * should re-suspend.
310 >     *
311 >     * 6. Killing off unneeded workers. A timeout mechanism is used to
312 >     * shed unused workers: The oldest (first) event queue waiter uses
313 >     * a timed rather than hard wait. When this wait times out without
314 >     * a normal wakeup, it tries to shutdown any one (for convenience
315 >     * the newest) other spare or event waiter via
316 >     * tryShutdownUnusedWorker. This eventually reduces the number of
317 >     * worker threads to a minimum of one after a long enough period
318 >     * without use.
319 >     *
320 >     * 7. Deciding when to create new workers. The main dynamic
321 >     * control in this class is deciding when to create extra threads
322 >     * in method helpMaintainParallelism. We would like to keep
323 >     * exactly #parallelism threads running, which is an impossible
324 >     * task. We always need to create one when the number of running
325 >     * threads would become zero and all workers are busy. Beyond
326 >     * this, we must rely on heuristics that work well in the
327 >     * presence of transient phenomena such as GC stalls, dynamic
328 >     * compilation, and wake-up lags. These transients are extremely
329 >     * common -- we are normally trying to fully saturate the CPUs on
330 >     * a machine, so almost any activity other than running tasks
331 >     * impedes accuracy. Our main defense is to allow parallelism to
332 >     * lapse for a while during joins, and use a timeout to see if,
333 >     * after the resulting settling, there is still a need for
334 >     * additional workers.  This also better copes with the fact that
335 >     * some of the methods in this class tend to never become compiled
336 >     * (but are interpreted), so some components of the entire set of
337 >     * controls might execute 100 times faster than others. And
338 >     * similarly for cases where the apparent lack of work is just due
339 >     * to GC stalls and other transient system activity.
340       *
341       * Beware that there is a lot of representation-level coupling
342       * among classes ForkJoinPool, ForkJoinWorkerThread, and
# Line 312 | Line 349 | public class ForkJoinPool extends Abstra
349       *
350       * Style notes: There are lots of inline assignments (of form
351       * "while ((local = field) != 0)") which are usually the simplest
352 <     * way to ensure read orderings. Also several occurrences of the
353 <     * unusual "do {} while(!cas...)" which is the simplest way to
354 <     * force an update of a CAS'ed variable. There are also a few
355 <     * other coding oddities that help some methods perform reasonably
356 <     * even when interpreted (not compiled).
352 >     * way to ensure the required read orderings (which are sometimes
353 >     * critical). Also several occurrences of the unusual "do {}
354 >     * while (!cas...)" which is the simplest way to force an update of
355 >     * a CAS'ed variable. There are also other coding oddities that
356 >     * help some methods perform reasonably even when interpreted (not
357 >     * compiled), at the expense of some messy constructions that
358 >     * reduce byte code counts.
359       *
360       * The order of declarations in this file is: (1) statics (2)
361       * fields (along with constants used when unpacking some of them)
# Line 345 | Line 384 | public class ForkJoinPool extends Abstra
384       * Default ForkJoinWorkerThreadFactory implementation; creates a
385       * new ForkJoinWorkerThread.
386       */
387 <    static class  DefaultForkJoinWorkerThreadFactory
387 >    static class DefaultForkJoinWorkerThreadFactory
388          implements ForkJoinWorkerThreadFactory {
389          public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
390              return new ForkJoinWorkerThread(pool);
# Line 384 | Line 423 | public class ForkJoinPool extends Abstra
423          new AtomicInteger();
424  
425      /**
426 <     * Absolute bound for parallelism level. Twice this number must
427 <     * fit into a 16bit field to enable word-packing for some counts.
426 >     * The time to block in a join (see awaitJoin) before checking if
427 >     * a new worker should be (re)started to maintain parallelism
428 >     * level. The value should be short enough to maintain global
429 >     * responsiveness and progress but long enough to avoid
430 >     * counterproductive firings during GC stalls or unrelated system
431 >     * activity, and to not bog down systems with continual re-firings
432 >     * on GCs or legitimately long waits.
433       */
434 <    private static final int MAX_THREADS = 0x7fff;
434 >    private static final long JOIN_TIMEOUT_MILLIS = 250L; // 4 per second
435 >
436 >    /**
437 >     * The wakeup interval (in nanoseconds) for the oldest worker
438 >     * waiting for an event to invoke tryShutdownUnusedWorker to
439 >     * shrink the number of workers.  The exact value does not matter
440 >     * too much. It must be short enough to release resources during
441 >     * sustained periods of idleness, but not so short that threads
442 >     * are continually re-created.
443 >     */
444 >    private static final long SHRINK_RATE_NANOS =
445 >        30L * 1000L * 1000L * 1000L; // 2 per minute
446 >
447 >    /**
448 >     * Absolute bound for parallelism level. Twice this number plus
449 >     * one (i.e., 0xfff) must fit into a 16bit field to enable
450 >     * word-packing for some counts and indices.
451 >     */
452 >    private static final int MAX_WORKERS   = 0x7fff;
453  
454      /**
455       * Array holding all worker threads in the pool.  Array size must
# Line 413 | Line 475 | public class ForkJoinPool extends Abstra
475      /**
476       * Latch released upon termination.
477       */
478 <    private final CountDownLatch terminationLatch;
478 >    private final Phaser termination;
479  
480      /**
481       * Creation factory for worker threads.
# Line 427 | Line 489 | public class ForkJoinPool extends Abstra
489      private volatile long stealCount;
490  
491      /**
492 <     * Encoded record of top of treiber stack of threads waiting for
492 >     * Encoded record of top of Treiber stack of threads waiting for
493       * events. The top 32 bits contain the count being waited for. The
494 <     * bottom word contains one plus the pool index of waiting worker
495 <     * thread.
494 >     * bottom 16 bits contains one plus the pool index of waiting
495 >     * worker thread. (Bits 16-31 are unused.)
496       */
497      private volatile long eventWaiters;
498  
499      private static final int  EVENT_COUNT_SHIFT = 32;
500 <    private static final long WAITER_INDEX_MASK = (1L << EVENT_COUNT_SHIFT)-1L;
500 >    private static final long WAITER_ID_MASK    = (1L << 16) - 1L;
501  
502      /**
503       * A counter for events that may wake up worker threads:
504       *   - Submission of a new task to the pool
505       *   - A worker pushing a task on an empty queue
506 <     *   - termination and reconfiguration
506 >     *   - termination
507       */
508      private volatile int eventCount;
509  
510      /**
511 +     * Encoded record of top of Treiber stack of spare threads waiting
512 +     * for resumption. The top 16 bits contain an arbitrary count to
513 +     * avoid ABA effects. The bottom 16bits contains one plus the pool
514 +     * index of waiting worker thread.
515 +     */
516 +    private volatile int spareWaiters;
517 +
518 +    private static final int SPARE_COUNT_SHIFT = 16;
519 +    private static final int SPARE_ID_MASK     = (1 << 16) - 1;
520 +
521 +    /**
522       * Lifecycle control. The low word contains the number of workers
523       * that are (probably) executing tasks. This value is atomically
524       * incremented before a worker gets a task to run, and decremented
525 <     * when worker has no tasks and cannot find any.  Bits 16-18
525 >     * when a worker has no tasks and cannot find any.  Bits 16-18
526       * contain runLevel value. When all are zero, the pool is
527       * running. Level transitions are monotonic (running -> shutdown
528       * -> terminating -> terminated) so each transition adds a bit.
529       * These are bundled together to ensure consistent read for
530       * termination checks (i.e., that runLevel is at least SHUTDOWN
531       * and active threads is zero).
532 +     *
533 +     * Notes: Most direct CASes are dependent on these bitfield
534 +     * positions.  Also, this field is non-private to enable direct
535 +     * performance-sensitive CASes in ForkJoinWorkerThread.
536       */
537 <    private volatile int runState;
537 >    volatile int runState;
538  
539      // Note: The order among run level values matters.
540      private static final int RUNLEVEL_SHIFT     = 16;
# Line 465 | Line 542 | public class ForkJoinPool extends Abstra
542      private static final int TERMINATING        = 1 << (RUNLEVEL_SHIFT + 1);
543      private static final int TERMINATED         = 1 << (RUNLEVEL_SHIFT + 2);
544      private static final int ACTIVE_COUNT_MASK  = (1 << RUNLEVEL_SHIFT) - 1;
468    private static final int ONE_ACTIVE         = 1; // active update delta
545  
546      /**
547       * Holds number of total (i.e., created and not yet terminated)
# Line 474 | Line 550 | public class ForkJoinPool extends Abstra
550       * making decisions about creating and suspending spare
551       * threads. Updated only by CAS. Note that adding a new worker
552       * requires incrementing both counts, since workers start off in
553 <     * running state.  This field is also used for memory-fencing
478 <     * configuration parameters.
553 >     * running state.
554       */
555      private volatile int workerCounts;
556  
# Line 484 | Line 559 | public class ForkJoinPool extends Abstra
559      private static final int ONE_RUNNING        = 1;
560      private static final int ONE_TOTAL          = 1 << TOTAL_COUNT_SHIFT;
561  
487    /*
488     * Fields parallelism. maxPoolSize, locallyFifo,
489     * maintainsParallelism, and ueh are non-volatile, but external
490     * reads/writes use workerCount fences to ensure visability.
491     */
492
562      /**
563       * The target parallelism level.
564 +     * Accessed directly by ForkJoinWorkerThreads.
565       */
566 <    private int parallelism;
497 <
498 <    /**
499 <     * The maximum allowed pool size.
500 <     */
501 <    private int maxPoolSize;
566 >    final int parallelism;
567  
568      /**
569       * True if use local fifo, not default lifo, for local polling
570 <     * Replicated by ForkJoinWorkerThreads
506 <     */
507 <    private boolean locallyFifo;
508 <
509 <    /**
510 <     * Controls whether to add spares to maintain parallelism
570 >     * Read by, and replicated by ForkJoinWorkerThreads
571       */
572 <    private boolean maintainsParallelism;
572 >    final boolean locallyFifo;
573  
574      /**
575 <     * The uncaught exception handler used when any worker
576 <     * abruptly terminates
575 >     * The uncaught exception handler used when any worker abruptly
576 >     * terminates.
577       */
578 <    private Thread.UncaughtExceptionHandler ueh;
578 >    private final Thread.UncaughtExceptionHandler ueh;
579  
580      /**
581       * Pool number, just for assigning useful names to worker threads
582       */
583      private final int poolNumber;
584  
585 <    // utilities for updating fields
585 >    // Utilities for CASing fields. Note that most of these
586 >    // are usually manually inlined by callers
587  
588      /**
589 <     * Adds delta to running count.  Used mainly by ForkJoinTask.
529 <     *
530 <     * @param delta the number to add
589 >     * Increments running count part of workerCounts
590       */
591 <    final void updateRunningCount(int delta) {
592 <        int wc;
591 >    final void incrementRunningCount() {
592 >        int c;
593          do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
594 <                                               wc = workerCounts,
595 <                                               wc + delta));
594 >                                               c = workerCounts,
595 >                                               c + ONE_RUNNING));
596      }
597  
598      /**
599 <     * Write fence for user modifications of pool parameters
541 <     * (parallelism. etc).  Note that it doesn't matter if CAS fails.
599 >     * Tries to increment running count part of workerCounts
600       */
601 <    private void workerCountWriteFence() {
602 <        int wc;
603 <        UNSAFE.compareAndSwapInt(this, workerCountsOffset,
604 <                                 wc = workerCounts, wc);
601 >    final boolean tryIncrementRunningCount() {
602 >        int c;
603 >        return UNSAFE.compareAndSwapInt(this, workerCountsOffset,
604 >                                        c = workerCounts,
605 >                                        c + ONE_RUNNING);
606      }
607  
608      /**
609 <     * Read fence for external reads of pool parameters
551 <     * (parallelism. maxPoolSize, etc).
609 >     * Tries to decrement running count unless already zero
610       */
611 <    private void workerCountReadFence() {
612 <        int ignore = workerCounts;
611 >    final boolean tryDecrementRunningCount() {
612 >        int wc = workerCounts;
613 >        if ((wc & RUNNING_COUNT_MASK) == 0)
614 >            return false;
615 >        return UNSAFE.compareAndSwapInt(this, workerCountsOffset,
616 >                                        wc, wc - ONE_RUNNING);
617      }
618  
619      /**
620 <     * Tries incrementing active count; fails on contention.
621 <     * Called by workers before executing tasks.
620 >     * Forces decrement of encoded workerCounts, awaiting nonzero if
621 >     * (rarely) necessary when other count updates lag.
622       *
623 <     * @return true on success
623 >     * @param dr -- either zero or ONE_RUNNING
624 >     * @param dt -- either zero or ONE_TOTAL
625       */
626 <    final boolean tryIncrementActiveCount() {
627 <        int c;
628 <        return UNSAFE.compareAndSwapInt(this, runStateOffset,
629 <                                        c = runState, c + ONE_ACTIVE);
626 >    private void decrementWorkerCounts(int dr, int dt) {
627 >        for (;;) {
628 >            int wc = workerCounts;
629 >            if ((wc & RUNNING_COUNT_MASK)  - dr < 0 ||
630 >                (wc >>> TOTAL_COUNT_SHIFT) - dt < 0) {
631 >                if ((runState & TERMINATED) != 0)
632 >                    return; // lagging termination on a backout
633 >                Thread.yield();
634 >            }
635 >            if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
636 >                                         wc, wc - (dr + dt)))
637 >                return;
638 >        }
639      }
640  
641      /**
# Line 573 | Line 645 | public class ForkJoinPool extends Abstra
645      final boolean tryDecrementActiveCount() {
646          int c;
647          return UNSAFE.compareAndSwapInt(this, runStateOffset,
648 <                                        c = runState, c - ONE_ACTIVE);
648 >                                        c = runState, c - 1);
649      }
650  
651      /**
# Line 602 | Line 674 | public class ForkJoinPool extends Abstra
674          lock.lock();
675          try {
676              ForkJoinWorkerThread[] ws = workers;
677 <            int len = ws.length;
678 <            if (k < 0 || k >= len || ws[k] != null) {
679 <                for (k = 0; k < len && ws[k] != null; ++k)
677 >            int n = ws.length;
678 >            if (k < 0 || k >= n || ws[k] != null) {
679 >                for (k = 0; k < n && ws[k] != null; ++k)
680                      ;
681 <                if (k == len)
682 <                    ws = Arrays.copyOf(ws, len << 1);
681 >                if (k == n)
682 >                    ws = workers = Arrays.copyOf(ws, n << 1);
683              }
684              ws[k] = w;
685 <            workers = ws; // volatile array write ensures slot visibility
685 >            int c = eventCount; // advance event count to ensure visibility
686 >            UNSAFE.compareAndSwapInt(this, eventCountOffset, c, c+1);
687          } finally {
688              lock.unlock();
689          }
# Line 618 | Line 691 | public class ForkJoinPool extends Abstra
691      }
692  
693      /**
694 <     * Nulls out record of worker in workers array
694 >     * Nulls out record of worker in workers array.
695       */
696      private void forgetWorker(ForkJoinWorkerThread w) {
697          int idx = w.poolIndex;
698 <        // Locking helps method recordWorker avoid unecessary expansion
698 >        // Locking helps method recordWorker avoid unnecessary expansion
699          final ReentrantLock lock = this.workerLock;
700          lock.lock();
701          try {
# Line 634 | Line 707 | public class ForkJoinPool extends Abstra
707          }
708      }
709  
637    // adding and removing workers
638
710      /**
711 <     * Tries to create and add new worker. Assumes that worker counts
712 <     * are already updated to accommodate the worker, so adjusts on
713 <     * failure.
711 >     * Final callback from terminating worker.  Removes record of
712 >     * worker from array, and adjusts counts. If pool is shutting
713 >     * down, tries to complete termination.
714       *
715 <     * @return new worker or null if creation failed
715 >     * @param w the worker
716       */
717 <    private ForkJoinWorkerThread addWorker() {
718 <        ForkJoinWorkerThread w = null;
719 <        try {
720 <            w = factory.newThread(this);
721 <        } finally { // Adjust on either null or exceptional factory return
722 <            if (w == null) {
723 <                onWorkerCreationFailure();
724 <                return null;
717 >    final void workerTerminated(ForkJoinWorkerThread w) {
718 >        forgetWorker(w);
719 >        decrementWorkerCounts(w.isTrimmed() ? 0 : ONE_RUNNING, ONE_TOTAL);
720 >        while (w.stealCount != 0) // collect final count
721 >            tryAccumulateStealCount(w);
722 >        tryTerminate(false);
723 >    }
724 >
725 >    // Waiting for and signalling events
726 >
727 >    /**
728 >     * Releases workers blocked on a count not equal to current count.
729 >     * Normally called after precheck that eventWaiters isn't zero to
730 >     * avoid wasted array checks. Gives up upon a change in count or
731 >     * upon releasing two workers, letting others take over.
732 >     */
733 >    private void releaseEventWaiters() {
734 >        ForkJoinWorkerThread[] ws = workers;
735 >        int n = ws.length;
736 >        long h = eventWaiters;
737 >        int ec = eventCount;
738 >        boolean releasedOne = false;
739 >        ForkJoinWorkerThread w; int id;
740 >        while ((id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 &&
741 >               (int)(h >>> EVENT_COUNT_SHIFT) != ec &&
742 >               id < n && (w = ws[id]) != null) {
743 >            if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
744 >                                          h,  w.nextWaiter)) {
745 >                LockSupport.unpark(w);
746 >                if (releasedOne) // exit on second release
747 >                    break;
748 >                releasedOne = true;
749              }
750 +            if (eventCount != ec)
751 +                break;
752 +            h = eventWaiters;
753          }
656        w.start(recordWorker(w), locallyFifo, ueh);
657        return w;
754      }
755  
756      /**
757 <     * Adjusts counts upon failure to create worker
757 >     * Tries to advance eventCount and releases waiters. Called only
758 >     * from workers.
759       */
760 <    private void onWorkerCreationFailure() {
761 <        int c;
762 <        do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
763 <                                               c = workerCounts,
764 <                                               c - (ONE_RUNNING|ONE_TOTAL)));
668 <        tryTerminate(false); // in case of failure during shutdown
760 >    final void signalWork() {
761 >        int c; // try to increment event count -- CAS failure OK
762 >        UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
763 >        if (eventWaiters != 0L)
764 >            releaseEventWaiters();
765      }
766  
767      /**
768 <     * Create enough total workers to establish target parallelism,
769 <     * giving up if terminating or addWorker fails
768 >     * Adds the given worker to event queue and blocks until
769 >     * terminating or event count advances from the given value
770 >     *
771 >     * @param w the calling worker thread
772 >     * @param ec the count
773       */
774 <    private void ensureEnoughTotalWorkers() {
775 <        int wc;
776 <        while (runState < TERMINATING &&
777 <               ((wc = workerCounts) >>> TOTAL_COUNT_SHIFT) < parallelism) {
778 <            if ((UNSAFE.compareAndSwapInt(this, workerCountsOffset,
779 <                                          wc, wc + (ONE_RUNNING|ONE_TOTAL)) &&
780 <                 addWorker() == null))
774 >    private void eventSync(ForkJoinWorkerThread w, int ec) {
775 >        long nh = (((long)ec) << EVENT_COUNT_SHIFT) | ((long)(w.poolIndex+1));
776 >        long h;
777 >        while ((runState < SHUTDOWN || !tryTerminate(false)) &&
778 >               (((int)((h = eventWaiters) & WAITER_ID_MASK)) == 0 ||
779 >                (int)(h >>> EVENT_COUNT_SHIFT) == ec) &&
780 >               eventCount == ec) {
781 >            if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
782 >                                          w.nextWaiter = h, nh)) {
783 >                awaitEvent(w, ec);
784                  break;
785 +            }
786          }
787      }
788  
789      /**
790 <     * Final callback from terminating worker.  Removes record of
791 <     * worker from array, and adjusts counts. If pool is shutting
792 <     * down, tries to complete terminatation, else possibly replaces
793 <     * the worker.
790 >     * Blocks the given worker (that has already been entered as an
791 >     * event waiter) until terminating or event count advances from
792 >     * the given value. The oldest (first) waiter uses a timed wait to
793 >     * occasionally one-by-one shrink the number of workers (to a
794 >     * minimum of one) if the pool has not been used for extended
795 >     * periods.
796       *
797 <     * @param w the worker
797 >     * @param w the calling worker thread
798 >     * @param ec the count
799       */
800 <    final void workerTerminated(ForkJoinWorkerThread w) {
801 <        if (w.active) { // force inactive
802 <            w.active = false;
803 <            do {} while (!tryDecrementActiveCount());
800 >    private void awaitEvent(ForkJoinWorkerThread w, int ec) {
801 >        while (eventCount == ec) {
802 >            if (tryAccumulateStealCount(w)) { // transfer while idle
803 >                boolean untimed = (w.nextWaiter != 0L ||
804 >                                   (workerCounts & RUNNING_COUNT_MASK) <= 1);
805 >                long startTime = untimed ? 0 : System.nanoTime();
806 >                Thread.interrupted();         // clear/ignore interrupt
807 >                if (w.isTerminating() || eventCount != ec)
808 >                    break;                    // recheck after clear
809 >                if (untimed)
810 >                    LockSupport.park(w);
811 >                else {
812 >                    LockSupport.parkNanos(w, SHRINK_RATE_NANOS);
813 >                    if (eventCount != ec || w.isTerminating())
814 >                        break;
815 >                    if (System.nanoTime() - startTime >= SHRINK_RATE_NANOS)
816 >                        tryShutdownUnusedWorker(ec);
817 >                }
818 >            }
819          }
699        forgetWorker(w);
700
701        // decrement total count, and if was running, running count
702        int unit = w.isTrimmed()? ONE_TOTAL : (ONE_RUNNING|ONE_TOTAL);
703        int wc;
704        do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
705                                               wc = workerCounts, wc - unit));
706
707        accumulateStealCount(w); // collect final count
708        if (!tryTerminate(false))
709            ensureEnoughTotalWorkers();
820      }
821  
822 <    // Waiting for and signalling events
822 >    // Maintaining parallelism
823  
824      /**
825 <     * Ensures eventCount on exit is different (mod 2^32) than on
716 <     * entry.  CAS failures are OK -- any change in count suffices.
825 >     * Pushes worker onto the spare stack.
826       */
827 <    private void advanceEventCount() {
828 <        int c;
829 <        UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
827 >    final void pushSpare(ForkJoinWorkerThread w) {
828 >        int ns = (++w.spareCount << SPARE_COUNT_SHIFT) | (w.poolIndex + 1);
829 >        do {} while (!UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
830 >                                               w.nextSpare = spareWaiters,ns));
831      }
832  
833      /**
834 <     * Releases workers blocked on a count not equal to current count.
835 <     */
836 <    final void releaseWaiters() {
837 <        long top;
838 <        int id;
839 <        while ((id = (int)((top = eventWaiters) & WAITER_INDEX_MASK)) > 0 &&
840 <               (int)(top >>> EVENT_COUNT_SHIFT) != eventCount) {
841 <            ForkJoinWorkerThread[] ws = workers;
842 <            ForkJoinWorkerThread w;
843 <            if (ws.length >= id && (w = ws[id - 1]) != null &&
844 <                UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
845 <                                          top, w.nextWaiter))
834 >     * Tries (once) to resume a spare if the number of running
835 >     * threads is less than target.
836 >     */
837 >    private void tryResumeSpare() {
838 >        int sw, id;
839 >        ForkJoinWorkerThread[] ws = workers;
840 >        int n = ws.length;
841 >        ForkJoinWorkerThread w;
842 >        if ((sw = spareWaiters) != 0 &&
843 >            (id = (sw & SPARE_ID_MASK) - 1) >= 0 &&
844 >            id < n && (w = ws[id]) != null &&
845 >            (runState >= TERMINATING ||
846 >             (workerCounts & RUNNING_COUNT_MASK) < parallelism) &&
847 >            spareWaiters == sw &&
848 >            UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
849 >                                     sw, w.nextSpare)) {
850 >            int c; // increment running count before resume
851 >            do {} while (!UNSAFE.compareAndSwapInt
852 >                         (this, workerCountsOffset,
853 >                          c = workerCounts, c + ONE_RUNNING));
854 >            if (w.tryUnsuspend())
855                  LockSupport.unpark(w);
856 +            else   // back out if w was shutdown
857 +                decrementWorkerCounts(ONE_RUNNING, 0);
858          }
859      }
860  
861      /**
862 <     * Advances eventCount and releases waiters until interference by
863 <     * other releasing threads is detected.
862 >     * Tries to increase the number of running workers if below target
863 >     * parallelism: If a spare exists tries to resume it via
864 >     * tryResumeSpare.  Otherwise, if not enough total workers or all
865 >     * existing workers are busy, adds a new worker. In all cases also
866 >     * helps wake up releasable workers waiting for work.
867       */
868 <    final void signalWork() {
869 <        int ec;
870 <        UNSAFE.compareAndSwapInt(this, eventCountOffset, ec=eventCount, ec+1);
871 <        outer:for (;;) {
872 <            long top = eventWaiters;
873 <            ec = eventCount;
874 <            for (;;) {
875 <                ForkJoinWorkerThread[] ws; ForkJoinWorkerThread w;
876 <                int id = (int)(top & WAITER_INDEX_MASK);
877 <                if (id <= 0 || (int)(top >>> EVENT_COUNT_SHIFT) == ec)
878 <                    return;
879 <                if ((ws = workers).length < id || (w = ws[id - 1]) == null ||
880 <                    !UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
881 <                                               top, top = w.nextWaiter))
882 <                    continue outer;      // possibly stale; reread
883 <                LockSupport.unpark(w);
884 <                if (top != eventWaiters) // let someone else take over
885 <                    return;
868 >    private void helpMaintainParallelism() {
869 >        int pc = parallelism;
870 >        int wc, rs, tc;
871 >        while (((wc = workerCounts) & RUNNING_COUNT_MASK) < pc &&
872 >               (rs = runState) < TERMINATING) {
873 >            if (spareWaiters != 0)
874 >                tryResumeSpare();
875 >            else if ((tc = wc >>> TOTAL_COUNT_SHIFT) >= MAX_WORKERS ||
876 >                     (tc >= pc && (rs & ACTIVE_COUNT_MASK) != tc))
877 >                break;   // enough total
878 >            else if (runState == rs && workerCounts == wc &&
879 >                     UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
880 >                                              wc + (ONE_RUNNING|ONE_TOTAL))) {
881 >                ForkJoinWorkerThread w = null;
882 >                Throwable fail = null;
883 >                try {
884 >                    w = factory.newThread(this);
885 >                } catch (Throwable ex) {
886 >                    fail = ex;
887 >                }
888 >                if (w == null) { // null or exceptional factory return
889 >                    decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL);
890 >                    tryTerminate(false); // handle failure during shutdown
891 >                    // If originating from an external caller,
892 >                    // propagate exception, else ignore
893 >                    if (fail != null && runState < TERMINATING &&
894 >                        !(Thread.currentThread() instanceof
895 >                          ForkJoinWorkerThread))
896 >                        UNSAFE.throwException(fail);
897 >                    break;
898 >                }
899 >                w.start(recordWorker(w), ueh);
900 >                if ((workerCounts >>> TOTAL_COUNT_SHIFT) >= pc)
901 >                    break; // add at most one unless total below target
902              }
903          }
904 +        if (eventWaiters != 0L)
905 +            releaseEventWaiters();
906      }
907  
908      /**
909 <     * If worker is inactive, blocks until terminating or event count
910 <     * advances from last value held by worker; in any case helps
911 <     * release others.
909 >     * Callback from the oldest waiter in awaitEvent waking up after a
910 >     * period of non-use. If all workers are idle, tries (once) to
911 >     * shutdown an event waiter or a spare, if one exists. Note that
912 >     * we don't need CAS or locks here because the method is called
913 >     * only from one thread occasionally waking (and even misfires are
914 >     * OK). Note that until the shutdown worker fully terminates,
915 >     * workerCounts will overestimate total count, which is tolerable.
916       *
917 <     * @param w the calling worker thread
917 >     * @param ec the event count waited on by caller (to abort
918 >     * attempt if count has since changed).
919       */
920 <    private void eventSync(ForkJoinWorkerThread w) {
921 <        if (!w.active) {
922 <            int prev = w.lastEventCount;
923 <            long nextTop = (((long)prev << EVENT_COUNT_SHIFT) |
924 <                            ((long)(w.poolIndex + 1)));
925 <            long top;
926 <            while ((runState < SHUTDOWN || !tryTerminate(false)) &&
927 <                   (((int)(top = eventWaiters) & WAITER_INDEX_MASK) == 0 ||
928 <                    (int)(top >>> EVENT_COUNT_SHIFT) == prev) &&
929 <                   eventCount == prev) {
930 <                if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset,
931 <                                              w.nextWaiter = top, nextTop)) {
932 <                    accumulateStealCount(w); // transfer steals while idle
933 <                    Thread.interrupted();    // clear/ignore interrupt
934 <                    while (eventCount == prev)
935 <                        w.doPark();
936 <                    break;
937 <                }
920 >    private void tryShutdownUnusedWorker(int ec) {
921 >        if (runState == 0 && eventCount == ec) { // only trigger if all idle
922 >            ForkJoinWorkerThread[] ws = workers;
923 >            int n = ws.length;
924 >            ForkJoinWorkerThread w = null;
925 >            boolean shutdown = false;
926 >            int sw;
927 >            long h;
928 >            if ((sw = spareWaiters) != 0) { // prefer killing spares
929 >                int id = (sw & SPARE_ID_MASK) - 1;
930 >                if (id >= 0 && id < n && (w = ws[id]) != null &&
931 >                    UNSAFE.compareAndSwapInt(this, spareWaitersOffset,
932 >                                             sw, w.nextSpare))
933 >                    shutdown = true;
934 >            }
935 >            else if ((h = eventWaiters) != 0L) {
936 >                long nh;
937 >                int id = ((int)(h & WAITER_ID_MASK)) - 1;
938 >                if (id >= 0 && id < n && (w = ws[id]) != null &&
939 >                    (nh = w.nextWaiter) != 0L && // keep at least one worker
940 >                    UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh))
941 >                    shutdown = true;
942 >            }
943 >            if (w != null && shutdown) {
944 >                w.shutdown();
945 >                LockSupport.unpark(w);
946              }
792            w.lastEventCount = eventCount;
947          }
948 <        releaseWaiters();
948 >        releaseEventWaiters(); // in case of interference
949      }
950  
951      /**
952       * Callback from workers invoked upon each top-level action (i.e.,
953 <     * stealing a task or taking a submission and running
954 <     * it). Performs one or both of the following:
953 >     * stealing a task or taking a submission and running it).
954 >     * Performs one or more of the following:
955       *
956 <     * * If the worker cannot find work, updates its active status to
957 <     * inactive and updates activeCount unless there is contention, in
958 <     * which case it may try again (either in this or a subsequent
959 <     * call).  Additionally, awaits the next task event and/or helps
960 <     * wake up other releasable waiters.
961 <     *
962 <     * * If there are too many running threads, suspends this worker
963 <     * (first forcing inactivation if necessary).  If it is not
964 <     * resumed before a keepAlive elapses, the worker may be "trimmed"
965 <     * -- killed while suspended within suspendAsSpare. Otherwise,
966 <     * upon resume it rechecks to make sure that it is still needed.
956 >     * 1. If the worker is active and either did not run a task
957 >     *    or there are too many workers, try to set its active status
958 >     *    to inactive and update activeCount. On contention, we may
959 >     *    try again in this or a subsequent call.
960 >     *
961 >     * 2. If not enough total workers, help create some.
962 >     *
963 >     * 3. If there are too many running workers, suspend this worker
964 >     *    (first forcing inactive if necessary).  If it is not needed,
965 >     *    it may be shutdown while suspended (via
966 >     *    tryShutdownUnusedWorker).  Otherwise, upon resume it
967 >     *    rechecks running thread count and need for event sync.
968 >     *
969 >     * 4. If worker did not run a task, await the next task event via
970 >     *    eventSync if necessary (first forcing inactivation), upon
971 >     *    which the worker may be shutdown via
972 >     *    tryShutdownUnusedWorker.  Otherwise, help release any
973 >     *    existing event waiters that are now releasable,
974       *
975       * @param w the worker
976 <     * @param worked false if the worker scanned for work but didn't
816 <     * find any (in which case it may block waiting for work).
976 >     * @param ran true if worker ran a task since last call to this method
977       */
978 <    final void preStep(ForkJoinWorkerThread w, boolean worked) {
978 >    final void preStep(ForkJoinWorkerThread w, boolean ran) {
979 >        int wec = w.lastEventCount;
980          boolean active = w.active;
981 <        boolean inactivate = !worked & active;
982 <        for (;;) {
983 <            if (inactivate) {
984 <                int c = runState;
985 <                if (UNSAFE.compareAndSwapInt(this, runStateOffset,
986 <                                             c, c - ONE_ACTIVE))
987 <                    inactivate = active = w.active = false;
827 <            }
828 <            int wc = workerCounts;
829 <            if ((wc & RUNNING_COUNT_MASK) <= parallelism) {
830 <                if (!worked)
831 <                    eventSync(w);
832 <                return;
981 >        boolean inactivate = false;
982 >        int pc = parallelism;
983 >        while (w.runState == 0) {
984 >            int rs = runState;
985 >            if (rs >= TERMINATING) {           // propagate shutdown
986 >                w.shutdown();
987 >                break;
988              }
989 <            if (!(inactivate |= active) &&  // must inactivate to suspend
990 <                UNSAFE.compareAndSwapInt(this, workerCountsOffset,
991 <                                         wc, wc - ONE_RUNNING) &&
992 <                !w.suspendAsSpare())        // false if trimmed
993 <                return;
994 <        }
840 <    }
841 <
842 <    /**
843 <     * Adjusts counts and creates or resumes compensating threads for
844 <     * a worker about to block on task joinMe, returning early if
845 <     * joinMe becomes ready. First tries resuming an existing spare
846 <     * (which usually also avoids any count adjustment), but must then
847 <     * decrement running count to determine whether a new thread is
848 <     * needed. See above for fuller explanation.
849 <     */
850 <    final void preJoin(ForkJoinTask<?> joinMe) {
851 <        boolean dec = false;       // true when running count decremented
852 <        for (;;) {
853 <            releaseWaiters();      // help other threads progress
854 <
855 <            if (joinMe.status < 0) // surround spare search with done checks
856 <                return;
857 <            ForkJoinWorkerThread spare = null;
858 <            for (ForkJoinWorkerThread w : workers) {
859 <                if (w != null && w.isSuspended()) {
860 <                    spare = w;
861 <                    break;
989 >            if ((inactivate || (active && (rs & ACTIVE_COUNT_MASK) >= pc)) &&
990 >                UNSAFE.compareAndSwapInt(this, runStateOffset, rs, --rs)) {
991 >                inactivate = active = w.active = false;
992 >                if (rs == SHUTDOWN) {          // all inactive and shut down
993 >                    tryTerminate(false);
994 >                    continue;
995                  }
996              }
997 <            if (joinMe.status < 0)
998 <                return;
999 <
1000 <            if (spare != null && spare.tryUnsuspend()) {
1001 <                if (dec || joinMe.requestSignal() < 0) {
1002 <                    int c;
1003 <                    do {} while (!UNSAFE.compareAndSwapInt(this,
871 <                                                           workerCountsOffset,
872 <                                                           c = workerCounts,
873 <                                                           c + ONE_RUNNING));
874 <                } // else no net count change
875 <                LockSupport.unpark(spare);
876 <                return;
877 <            }
878 <
879 <            int wc = workerCounts; // decrement running count
880 <            if (!dec && (wc & RUNNING_COUNT_MASK) != 0 &&
881 <                (dec = UNSAFE.compareAndSwapInt(this, workerCountsOffset,
882 <                                                wc, wc -= ONE_RUNNING)) &&
883 <                joinMe.requestSignal() < 0) { // cannot block
884 <                int c;                        // back out
885 <                do {} while (!UNSAFE.compareAndSwapInt(this,
886 <                                                       workerCountsOffset,
887 <                                                       c = workerCounts,
888 <                                                       c + ONE_RUNNING));
889 <                return;
997 >            int wc = workerCounts;             // try to suspend as spare
998 >            if ((wc & RUNNING_COUNT_MASK) > pc) {
999 >                if (!(inactivate |= active) && // must inactivate to suspend
1000 >                    workerCounts == wc &&
1001 >                    UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1002 >                                             wc, wc - ONE_RUNNING))
1003 >                    w.suspendAsSpare();
1004              }
1005 <
1006 <            if (dec) {
1007 <                int tc = wc >>> TOTAL_COUNT_SHIFT;
1008 <                int pc = parallelism;
1009 <                int dc = pc - (wc & RUNNING_COUNT_MASK); // deficit count
1010 <                if ((dc < pc && (dc <= 0 || (dc * dc < (tc - pc) * pc) ||
1011 <                                 !maintainsParallelism)) ||
1012 <                    tc >= maxPoolSize) // cannot add
1013 <                    return;
1014 <                if (spare == null &&
1015 <                    UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
1016 <                                             wc + (ONE_RUNNING|ONE_TOTAL))) {
903 <                    addWorker();
904 <                    return;
1005 >            else if ((wc >>> TOTAL_COUNT_SHIFT) < pc)
1006 >                helpMaintainParallelism();     // not enough workers
1007 >            else if (ran)
1008 >                break;
1009 >            else {
1010 >                long h = eventWaiters;
1011 >                int ec = eventCount;
1012 >                if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != ec)
1013 >                    releaseEventWaiters();     // release others before waiting
1014 >                else if (ec != wec) {
1015 >                    w.lastEventCount = ec;     // no need to wait
1016 >                    break;
1017                  }
1018 +                else if (!(inactivate |= active))
1019 +                    eventSync(w, wec);         // must inactivate before sync
1020              }
1021          }
1022      }
1023  
1024      /**
1025 <     * Same idea as preJoin but with too many differing details to
1026 <     * integrate: There are no task-based signal counts, and only one
1027 <     * way to do the actual blocking. So for simplicity it is directly
1028 <     * incorporated into this method.
1029 <     */
1030 <    final void doBlock(ManagedBlocker blocker, boolean maintainPar)
1031 <        throws InterruptedException {
1032 <        maintainPar &= maintainsParallelism; // override
1033 <        boolean dec = false;
1034 <        boolean done = false;
1035 <        for (;;) {
1036 <            releaseWaiters();
1037 <            if (done = blocker.isReleasable())
1025 >     * Helps and/or blocks awaiting join of the given task.
1026 >     * See above for explanation.
1027 >     *
1028 >     * @param joinMe the task to join
1029 >     * @param worker the current worker thread
1030 >     * @param timed true if wait should time out
1031 >     * @param nanos timeout value if timed
1032 >     */
1033 >    final void awaitJoin(ForkJoinTask<?> joinMe, ForkJoinWorkerThread worker,
1034 >                         boolean timed, long nanos) {
1035 >        long startTime = timed? System.nanoTime() : 0L;
1036 >        int retries = 2 + (parallelism >> 2); // #helpJoins before blocking
1037 >        boolean running = true;               // false when count decremented
1038 >        while (joinMe.status >= 0) {
1039 >            if (runState >= TERMINATING) {
1040 >                joinMe.cancelIgnoringExceptions();
1041                  break;
925            ForkJoinWorkerThread spare = null;
926            for (ForkJoinWorkerThread w : workers) {
927                if (w != null && w.isSuspended()) {
928                    spare = w;
929                    break;
930                }
1042              }
1043 <            if (done = blocker.isReleasable())
1044 <                break;
934 <            if (spare != null && spare.tryUnsuspend()) {
935 <                if (dec) {
936 <                    int c;
937 <                    do {} while (!UNSAFE.compareAndSwapInt(this,
938 <                                                           workerCountsOffset,
939 <                                                           c = workerCounts,
940 <                                                           c + ONE_RUNNING));
941 <                }
942 <                LockSupport.unpark(spare);
1043 >            running = worker.helpJoinTask(joinMe, running);
1044 >            if (joinMe.status < 0)
1045                  break;
1046 +            if (retries > 0) {
1047 +                --retries;
1048 +                continue;
1049              }
1050              int wc = workerCounts;
1051 <            if (!dec && (wc & RUNNING_COUNT_MASK) != 0)
1052 <                dec = UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1053 <                                               wc, wc -= ONE_RUNNING);
1054 <            if (dec) {
1055 <                int tc = wc >>> TOTAL_COUNT_SHIFT;
1056 <                int pc = parallelism;
1057 <                int dc = pc - (wc & RUNNING_COUNT_MASK);
1058 <                if ((dc < pc && (dc <= 0 || (dc * dc < (tc - pc) * pc) ||
1059 <                                 !maintainPar)) ||
1060 <                    tc >= maxPoolSize)
1061 <                    break;
957 <                if (spare == null &&
958 <                    UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc,
959 <                                             wc + (ONE_RUNNING|ONE_TOTAL))){
960 <                    addWorker();
1051 >            if ((wc & RUNNING_COUNT_MASK) != 0) {
1052 >                if (running) {
1053 >                    if (!UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1054 >                                                  wc, wc - ONE_RUNNING))
1055 >                        continue;
1056 >                    running = false;
1057 >                }
1058 >                long h = eventWaiters;
1059 >                if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1060 >                    releaseEventWaiters();
1061 >                if (joinMe.status < 0)
1062                      break;
1063 +                if ((workerCounts & RUNNING_COUNT_MASK) != 0) {
1064 +                    long ms; int ns;
1065 +                    if (!timed) {
1066 +                        ms = JOIN_TIMEOUT_MILLIS;
1067 +                        ns = 0;
1068 +                    }
1069 +                    else { // at most JOIN_TIMEOUT_MILLIS per wait
1070 +                        long nt = nanos - (System.nanoTime() - startTime);
1071 +                        if (nt <= 0L)
1072 +                            break;
1073 +                        ms = nt / 1000000;
1074 +                        if (ms > JOIN_TIMEOUT_MILLIS) {
1075 +                            ms = JOIN_TIMEOUT_MILLIS;
1076 +                            ns = 0;
1077 +                        }
1078 +                        else
1079 +                            ns = (int) (nt % 1000000);
1080 +                    }
1081 +                    if (joinMe.internalAwaitDone(ms, ns) < 0)
1082 +                        break;
1083                  }
1084              }
1085 +            helpMaintainParallelism();
1086          }
1087 +        if (!running) {
1088 +            int c;
1089 +            do {} while (!UNSAFE.compareAndSwapInt
1090 +                         (this, workerCountsOffset,
1091 +                          c = workerCounts, c + ONE_RUNNING));
1092 +        }
1093 +    }
1094  
1095 <        try {
1096 <            if (!done)
1097 <                do {} while (!blocker.isReleasable() && !blocker.block());
1098 <        } finally {
1099 <            if (dec) {
1100 <                int c;
1101 <                do {} while (!UNSAFE.compareAndSwapInt(this,
1102 <                                                       workerCountsOffset,
1103 <                                                       c = workerCounts,
1104 <                                                       c + ONE_RUNNING));
1095 >    /**
1096 >     * Same idea as awaitJoin, but no helping, retries, or timeouts.
1097 >     */
1098 >    final void awaitBlocker(ManagedBlocker blocker)
1099 >        throws InterruptedException {
1100 >        while (!blocker.isReleasable()) {
1101 >            int wc = workerCounts;
1102 >            if ((wc & RUNNING_COUNT_MASK) == 0)
1103 >                helpMaintainParallelism();
1104 >            else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset,
1105 >                                              wc, wc - ONE_RUNNING)) {
1106 >                try {
1107 >                    while (!blocker.isReleasable()) {
1108 >                        long h = eventWaiters;
1109 >                        if (h != 0L &&
1110 >                            (int)(h >>> EVENT_COUNT_SHIFT) != eventCount)
1111 >                            releaseEventWaiters();
1112 >                        else if ((workerCounts & RUNNING_COUNT_MASK) == 0 &&
1113 >                                 runState < TERMINATING)
1114 >                            helpMaintainParallelism();
1115 >                        else if (blocker.block())
1116 >                            break;
1117 >                    }
1118 >                } finally {
1119 >                    int c;
1120 >                    do {} while (!UNSAFE.compareAndSwapInt
1121 >                                 (this, workerCountsOffset,
1122 >                                  c = workerCounts, c + ONE_RUNNING));
1123 >                }
1124 >                break;
1125              }
1126          }
1127      }
# Line 998 | Line 1147 | public class ForkJoinPool extends Abstra
1147          // Finish now if all threads terminated; else in some subsequent call
1148          if ((workerCounts >>> TOTAL_COUNT_SHIFT) == 0) {
1149              advanceRunLevel(TERMINATED);
1150 <            terminationLatch.countDown();
1150 >            termination.forceTermination();
1151          }
1152          return true;
1153      }
1154  
1155      /**
1156       * Actions on transition to TERMINATING
1157 +     *
1158 +     * Runs up to four passes through workers: (0) shutting down each
1159 +     * (without waking up if parked) to quickly spread notifications
1160 +     * without unnecessary bouncing around event queues etc (1) wake
1161 +     * up and help cancel tasks (2) interrupt (3) mop up races with
1162 +     * interrupted workers
1163       */
1164      private void startTerminating() {
1165 <        // Clear out and cancel submissions, ignoring exceptions
1165 >        cancelSubmissions();
1166 >        for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) {
1167 >            int c; // advance event count
1168 >            UNSAFE.compareAndSwapInt(this, eventCountOffset,
1169 >                                     c = eventCount, c+1);
1170 >            eventWaiters = 0L; // clobber lists
1171 >            spareWaiters = 0;
1172 >            for (ForkJoinWorkerThread w : workers) {
1173 >                if (w != null) {
1174 >                    w.shutdown();
1175 >                    if (passes > 0 && !w.isTerminated()) {
1176 >                        w.cancelTasks();
1177 >                        LockSupport.unpark(w);
1178 >                        if (passes > 1 && !w.isInterrupted()) {
1179 >                            try {
1180 >                                w.interrupt();
1181 >                            } catch (SecurityException ignore) {
1182 >                            }
1183 >                        }
1184 >                    }
1185 >                }
1186 >            }
1187 >        }
1188 >    }
1189 >
1190 >    /**
1191 >     * Clears out and cancels submissions, ignoring exceptions.
1192 >     */
1193 >    private void cancelSubmissions() {
1194          ForkJoinTask<?> task;
1195          while ((task = submissionQueue.poll()) != null) {
1196              try {
# Line 1015 | Line 1198 | public class ForkJoinPool extends Abstra
1198              } catch (Throwable ignore) {
1199              }
1200          }
1018        // Propagate run level
1019        for (ForkJoinWorkerThread w : workers) {
1020            if (w != null)
1021                w.shutdown();    // also resumes suspended workers
1022        }
1023        // Ensure no straggling local tasks
1024        for (ForkJoinWorkerThread w : workers) {
1025            if (w != null)
1026                w.cancelTasks();
1027        }
1028        // Wake up idle workers
1029        advanceEventCount();
1030        releaseWaiters();
1031        // Unstick pending joins
1032        for (ForkJoinWorkerThread w : workers) {
1033            if (w != null && !w.isTerminated()) {
1034                try {
1035                    w.interrupt();
1036                } catch (SecurityException ignore) {
1037                }
1038            }
1039        }
1201      }
1202  
1203      // misc support for ForkJoinWorkerThread
1204  
1205      /**
1206 <     * Returns pool number
1206 >     * Returns pool number.
1207       */
1208      final int getPoolNumber() {
1209          return poolNumber;
1210      }
1211  
1212      /**
1213 <     * Accumulates steal count from a worker, clearing
1214 <     * the worker's value
1213 >     * Tries to accumulate steal count from a worker, clearing
1214 >     * the worker's value if successful.
1215 >     *
1216 >     * @return true if worker steal count now zero
1217       */
1218 <    final void accumulateStealCount(ForkJoinWorkerThread w) {
1218 >    final boolean tryAccumulateStealCount(ForkJoinWorkerThread w) {
1219          int sc = w.stealCount;
1220 <        if (sc != 0) {
1221 <            long c;
1222 <            w.stealCount = 0;
1223 <            do {} while (!UNSAFE.compareAndSwapLong(this, stealCountOffset,
1224 <                                                    c = stealCount, c + sc));
1220 >        long c = stealCount;
1221 >        // CAS even if zero, for fence effects
1222 >        if (UNSAFE.compareAndSwapLong(this, stealCountOffset, c, c + sc)) {
1223 >            if (sc != 0)
1224 >                w.stealCount = 0;
1225 >            return true;
1226          }
1227 +        return sc == 0;
1228      }
1229  
1230      /**
# Line 1067 | Line 1232 | public class ForkJoinPool extends Abstra
1232       * active thread.
1233       */
1234      final int idlePerActive() {
1235 <        int ac = runState;    // no mask -- artifically boosts during shutdown
1236 <        int pc = parallelism; // use targeted parallelism, not rc
1235 >        int pc = parallelism; // use parallelism, not rc
1236 >        int ac = runState;    // no mask -- artificially boosts during shutdown
1237          // Use exact results for small values, saturate past 4
1238 <        return pc <= ac? 0 : pc >>> 1 <= ac? 1 : pc >>> 2 <= ac? 3 : pc >>> 3;
1239 <    }
1240 <
1241 <    /**
1077 <     * Returns the approximate (non-atomic) difference between running
1078 <     * and active counts.
1079 <     */
1080 <    final int inactiveCount() {
1081 <        return (workerCounts & RUNNING_COUNT_MASK) -
1082 <            (runState & ACTIVE_COUNT_MASK);
1238 >        return ((pc <= ac) ? 0 :
1239 >                (pc >>> 1 <= ac) ? 1 :
1240 >                (pc >>> 2 <= ac) ? 3 :
1241 >                pc >>> 3);
1242      }
1243  
1244      // Public and protected methods
# Line 1088 | Line 1247 | public class ForkJoinPool extends Abstra
1247  
1248      /**
1249       * Creates a {@code ForkJoinPool} with parallelism equal to {@link
1250 <     * java.lang.Runtime#availableProcessors}, and using the {@linkplain
1251 <     * #defaultForkJoinWorkerThreadFactory default thread factory}.
1250 >     * java.lang.Runtime#availableProcessors}, using the {@linkplain
1251 >     * #defaultForkJoinWorkerThreadFactory default thread factory},
1252 >     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
1253       *
1254       * @throws SecurityException if a security manager exists and
1255       *         the caller is not permitted to modify threads
# Line 1098 | Line 1258 | public class ForkJoinPool extends Abstra
1258       */
1259      public ForkJoinPool() {
1260          this(Runtime.getRuntime().availableProcessors(),
1261 <             defaultForkJoinWorkerThreadFactory);
1261 >             defaultForkJoinWorkerThreadFactory, null, false);
1262      }
1263  
1264      /**
1265       * Creates a {@code ForkJoinPool} with the indicated parallelism
1266 <     * level and using the {@linkplain
1267 <     * #defaultForkJoinWorkerThreadFactory default thread factory}.
1266 >     * level, the {@linkplain
1267 >     * #defaultForkJoinWorkerThreadFactory default thread factory},
1268 >     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
1269       *
1270       * @param parallelism the parallelism level
1271       * @throws IllegalArgumentException if parallelism less than or
# Line 1115 | Line 1276 | public class ForkJoinPool extends Abstra
1276       *         java.lang.RuntimePermission}{@code ("modifyThread")}
1277       */
1278      public ForkJoinPool(int parallelism) {
1279 <        this(parallelism, defaultForkJoinWorkerThreadFactory);
1279 >        this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
1280      }
1281  
1282      /**
1283 <     * Creates a {@code ForkJoinPool} with parallelism equal to {@link
1123 <     * java.lang.Runtime#availableProcessors}, and using the given
1124 <     * thread factory.
1283 >     * Creates a {@code ForkJoinPool} with the given parameters.
1284       *
1285 <     * @param factory the factory for creating new threads
1286 <     * @throws NullPointerException if the factory is null
1287 <     * @throws SecurityException if a security manager exists and
1288 <     *         the caller is not permitted to modify threads
1289 <     *         because it does not hold {@link
1290 <     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1291 <     */
1292 <    public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
1293 <        this(Runtime.getRuntime().availableProcessors(), factory);
1294 <    }
1295 <
1296 <    /**
1297 <     * Creates a {@code ForkJoinPool} with the given parallelism and
1139 <     * thread factory.
1140 <     *
1141 <     * @param parallelism the parallelism level
1142 <     * @param factory the factory for creating new threads
1285 >     * @param parallelism the parallelism level. For default value,
1286 >     * use {@link java.lang.Runtime#availableProcessors}.
1287 >     * @param factory the factory for creating new threads. For default value,
1288 >     * use {@link #defaultForkJoinWorkerThreadFactory}.
1289 >     * @param handler the handler for internal worker threads that
1290 >     * terminate due to unrecoverable errors encountered while executing
1291 >     * tasks. For default value, use {@code null}.
1292 >     * @param asyncMode if true,
1293 >     * establishes local first-in-first-out scheduling mode for forked
1294 >     * tasks that are never joined. This mode may be more appropriate
1295 >     * than default locally stack-based mode in applications in which
1296 >     * worker threads only process event-style asynchronous tasks.
1297 >     * For default value, use {@code false}.
1298       * @throws IllegalArgumentException if parallelism less than or
1299       *         equal to zero, or greater than implementation limit
1300       * @throws NullPointerException if the factory is null
# Line 1148 | Line 1303 | public class ForkJoinPool extends Abstra
1303       *         because it does not hold {@link
1304       *         java.lang.RuntimePermission}{@code ("modifyThread")}
1305       */
1306 <    public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
1306 >    public ForkJoinPool(int parallelism,
1307 >                        ForkJoinWorkerThreadFactory factory,
1308 >                        Thread.UncaughtExceptionHandler handler,
1309 >                        boolean asyncMode) {
1310          checkPermission();
1311          if (factory == null)
1312              throw new NullPointerException();
1313 <        if (parallelism <= 0 || parallelism > MAX_THREADS)
1313 >        if (parallelism <= 0 || parallelism > MAX_WORKERS)
1314              throw new IllegalArgumentException();
1157        this.poolNumber = poolNumberGenerator.incrementAndGet();
1158        int arraySize = initialArraySizeFor(parallelism);
1315          this.parallelism = parallelism;
1316          this.factory = factory;
1317 <        this.maxPoolSize = MAX_THREADS;
1318 <        this.maintainsParallelism = true;
1317 >        this.ueh = handler;
1318 >        this.locallyFifo = asyncMode;
1319 >        int arraySize = initialArraySizeFor(parallelism);
1320          this.workers = new ForkJoinWorkerThread[arraySize];
1321          this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
1322          this.workerLock = new ReentrantLock();
1323 <        this.terminationLatch = new CountDownLatch(1);
1324 <        // Start first worker; remaining workers added upon first submission
1168 <        workerCounts = ONE_RUNNING | ONE_TOTAL;
1169 <        addWorker();
1323 >        this.termination = new Phaser(1);
1324 >        this.poolNumber = poolNumberGenerator.incrementAndGet();
1325      }
1326  
1327      /**
# Line 1174 | Line 1329 | public class ForkJoinPool extends Abstra
1329       * @param pc the initial parallelism level
1330       */
1331      private static int initialArraySizeFor(int pc) {
1332 <        // See Hackers Delight, sec 3.2. We know MAX_THREADS < (1 >>> 16)
1333 <        int size = pc < MAX_THREADS ? pc + 1 : MAX_THREADS;
1332 >        // If possible, initially allocate enough space for one spare
1333 >        int size = pc < MAX_WORKERS ? pc + 1 : MAX_WORKERS;
1334 >        // See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16)
1335          size |= size >>> 1;
1336          size |= size >>> 2;
1337          size |= size >>> 4;
# Line 1186 | Line 1342 | public class ForkJoinPool extends Abstra
1342      // Execution methods
1343  
1344      /**
1345 <     * Common code for execute, invoke and submit
1345 >     * Submits task and creates, starts, or resumes some workers if necessary
1346       */
1347      private <T> void doSubmit(ForkJoinTask<T> task) {
1192        if (task == null)
1193            throw new NullPointerException();
1194        if (runState >= SHUTDOWN)
1195            throw new RejectedExecutionException();
1348          submissionQueue.offer(task);
1349 <        advanceEventCount();
1350 <        releaseWaiters();
1351 <        if ((workerCounts >>> TOTAL_COUNT_SHIFT) < parallelism)
1200 <            ensureEnoughTotalWorkers();
1349 >        int c; // try to increment event count -- CAS failure OK
1350 >        UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1);
1351 >        helpMaintainParallelism();
1352      }
1353  
1354      /**
# Line 1210 | Line 1361 | public class ForkJoinPool extends Abstra
1361       *         scheduled for execution
1362       */
1363      public <T> T invoke(ForkJoinTask<T> task) {
1364 <        doSubmit(task);
1365 <        return task.join();
1364 >        if (task == null)
1365 >            throw new NullPointerException();
1366 >        if (runState >= SHUTDOWN)
1367 >            throw new RejectedExecutionException();
1368 >        Thread t = Thread.currentThread();
1369 >        if ((t instanceof ForkJoinWorkerThread) &&
1370 >            ((ForkJoinWorkerThread)t).pool == this)
1371 >            return task.invoke();  // bypass submit if in same pool
1372 >        else {
1373 >            doSubmit(task);
1374 >            return task.join();
1375 >        }
1376 >    }
1377 >
1378 >    /**
1379 >     * Unless terminating, forks task if within an ongoing FJ
1380 >     * computation in the current pool, else submits as external task.
1381 >     */
1382 >    private <T> void forkOrSubmit(ForkJoinTask<T> task) {
1383 >        if (runState >= SHUTDOWN)
1384 >            throw new RejectedExecutionException();
1385 >        Thread t = Thread.currentThread();
1386 >        if ((t instanceof ForkJoinWorkerThread) &&
1387 >            ((ForkJoinWorkerThread)t).pool == this)
1388 >            task.fork();
1389 >        else
1390 >            doSubmit(task);
1391      }
1392  
1393      /**
# Line 1223 | Line 1399 | public class ForkJoinPool extends Abstra
1399       *         scheduled for execution
1400       */
1401      public void execute(ForkJoinTask<?> task) {
1402 <        doSubmit(task);
1402 >        if (task == null)
1403 >            throw new NullPointerException();
1404 >        forkOrSubmit(task);
1405      }
1406  
1407      // AbstractExecutorService methods
# Line 1234 | Line 1412 | public class ForkJoinPool extends Abstra
1412       *         scheduled for execution
1413       */
1414      public void execute(Runnable task) {
1415 +        if (task == null)
1416 +            throw new NullPointerException();
1417          ForkJoinTask<?> job;
1418          if (task instanceof ForkJoinTask<?>) // avoid re-wrap
1419              job = (ForkJoinTask<?>) task;
1420          else
1421              job = ForkJoinTask.adapt(task, null);
1422 <        doSubmit(job);
1422 >        forkOrSubmit(job);
1423 >    }
1424 >
1425 >    /**
1426 >     * Submits a ForkJoinTask for execution.
1427 >     *
1428 >     * @param task the task to submit
1429 >     * @return the task
1430 >     * @throws NullPointerException if the task is null
1431 >     * @throws RejectedExecutionException if the task cannot be
1432 >     *         scheduled for execution
1433 >     */
1434 >    public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
1435 >        if (task == null)
1436 >            throw new NullPointerException();
1437 >        forkOrSubmit(task);
1438 >        return task;
1439      }
1440  
1441      /**
# Line 1248 | Line 1444 | public class ForkJoinPool extends Abstra
1444       *         scheduled for execution
1445       */
1446      public <T> ForkJoinTask<T> submit(Callable<T> task) {
1447 +        if (task == null)
1448 +            throw new NullPointerException();
1449          ForkJoinTask<T> job = ForkJoinTask.adapt(task);
1450 <        doSubmit(job);
1450 >        forkOrSubmit(job);
1451          return job;
1452      }
1453  
# Line 1259 | Line 1457 | public class ForkJoinPool extends Abstra
1457       *         scheduled for execution
1458       */
1459      public <T> ForkJoinTask<T> submit(Runnable task, T result) {
1460 +        if (task == null)
1461 +            throw new NullPointerException();
1462          ForkJoinTask<T> job = ForkJoinTask.adapt(task, result);
1463 <        doSubmit(job);
1463 >        forkOrSubmit(job);
1464          return job;
1465      }
1466  
# Line 1270 | Line 1470 | public class ForkJoinPool extends Abstra
1470       *         scheduled for execution
1471       */
1472      public ForkJoinTask<?> submit(Runnable task) {
1473 +        if (task == null)
1474 +            throw new NullPointerException();
1475          ForkJoinTask<?> job;
1476          if (task instanceof ForkJoinTask<?>) // avoid re-wrap
1477              job = (ForkJoinTask<?>) task;
1478          else
1479              job = ForkJoinTask.adapt(task, null);
1480 <        doSubmit(job);
1480 >        forkOrSubmit(job);
1481          return job;
1482      }
1483  
1484      /**
1283     * Submits a ForkJoinTask for execution.
1284     *
1285     * @param task the task to submit
1286     * @return the task
1287     * @throws NullPointerException if the task is null
1288     * @throws RejectedExecutionException if the task cannot be
1289     *         scheduled for execution
1290     */
1291    public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
1292        doSubmit(task);
1293        return task;
1294    }
1295
1296    /**
1485       * @throws NullPointerException       {@inheritDoc}
1486       * @throws RejectedExecutionException {@inheritDoc}
1487       */
# Line 1305 | Line 1493 | public class ForkJoinPool extends Abstra
1493          invoke(new InvokeAll<T>(forkJoinTasks));
1494  
1495          @SuppressWarnings({"unchecked", "rawtypes"})
1496 <        List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks;
1496 >            List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks;
1497          return futures;
1498      }
1499  
# Line 1335 | Line 1523 | public class ForkJoinPool extends Abstra
1523       * @return the handler, or {@code null} if none
1524       */
1525      public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
1338        workerCountReadFence();
1526          return ueh;
1527      }
1528  
1529      /**
1343     * Sets the handler for internal worker threads that terminate due
1344     * to unrecoverable errors encountered while executing tasks.
1345     * Unless set, the current default or ThreadGroup handler is used
1346     * as handler.
1347     *
1348     * @param h the new handler
1349     * @return the old handler, or {@code null} if none
1350     * @throws SecurityException if a security manager exists and
1351     *         the caller is not permitted to modify threads
1352     *         because it does not hold {@link
1353     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1354     */
1355    public Thread.UncaughtExceptionHandler
1356        setUncaughtExceptionHandler(Thread.UncaughtExceptionHandler h) {
1357        checkPermission();
1358        workerCountReadFence();
1359        Thread.UncaughtExceptionHandler old = ueh;
1360        if (h != old) {
1361            ueh = h;
1362            workerCountWriteFence();
1363            for (ForkJoinWorkerThread w : workers) {
1364                if (w != null)
1365                    w.setUncaughtExceptionHandler(h);
1366            }
1367        }
1368        return old;
1369    }
1370
1371    /**
1372     * Sets the target parallelism level of this pool.
1373     *
1374     * @param parallelism the target parallelism
1375     * @throws IllegalArgumentException if parallelism less than or
1376     * equal to zero or greater than maximum size bounds
1377     * @throws SecurityException if a security manager exists and
1378     *         the caller is not permitted to modify threads
1379     *         because it does not hold {@link
1380     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1381     */
1382    public void setParallelism(int parallelism) {
1383        checkPermission();
1384        if (parallelism <= 0 || parallelism > maxPoolSize)
1385            throw new IllegalArgumentException();
1386        workerCountReadFence();
1387        int pc = this.parallelism;
1388        if (pc != parallelism) {
1389            this.parallelism = parallelism;
1390            workerCountWriteFence();
1391            // Release spares. If too many, some will die after re-suspend
1392            for (ForkJoinWorkerThread w : workers) {
1393                if (w != null && w.tryUnsuspend()) {
1394                    updateRunningCount(1);
1395                    LockSupport.unpark(w);
1396                }
1397            }
1398            ensureEnoughTotalWorkers();
1399            advanceEventCount();
1400            releaseWaiters(); // force config recheck by existing workers
1401        }
1402    }
1403
1404    /**
1530       * Returns the targeted parallelism level of this pool.
1531       *
1532       * @return the targeted parallelism level of this pool
1533       */
1534      public int getParallelism() {
1410        //        workerCountReadFence(); // inlined below
1411        int ignore = workerCounts;
1535          return parallelism;
1536      }
1537  
1538      /**
1539       * Returns the number of worker threads that have started but not
1540 <     * yet terminated.  This result returned by this method may differ
1540 >     * yet terminated.  The result returned by this method may differ
1541       * from {@link #getParallelism} when threads are created to
1542       * maintain parallelism when others are cooperatively blocked.
1543       *
# Line 1425 | Line 1548 | public class ForkJoinPool extends Abstra
1548      }
1549  
1550      /**
1428     * Returns the maximum number of threads allowed to exist in the
1429     * pool. Unless set using {@link #setMaximumPoolSize}, the
1430     * maximum is an implementation-defined value designed only to
1431     * prevent runaway growth.
1432     *
1433     * @return the maximum
1434     */
1435    public int getMaximumPoolSize() {
1436        workerCountReadFence();
1437        return maxPoolSize;
1438    }
1439
1440    /**
1441     * Sets the maximum number of threads allowed to exist in the
1442     * pool. The given value should normally be greater than or equal
1443     * to the {@link #getParallelism parallelism} level. Setting this
1444     * value has no effect on current pool size. It controls
1445     * construction of new threads.
1446     *
1447     * @throws IllegalArgumentException if negative or greater than
1448     * internal implementation limit
1449     */
1450    public void setMaximumPoolSize(int newMax) {
1451        if (newMax < 0 || newMax > MAX_THREADS)
1452            throw new IllegalArgumentException();
1453        maxPoolSize = newMax;
1454        workerCountWriteFence();
1455    }
1456
1457    /**
1458     * Returns {@code true} if this pool dynamically maintains its
1459     * target parallelism level. If false, new threads are added only
1460     * to avoid possible starvation.  This setting is by default true.
1461     *
1462     * @return {@code true} if maintains parallelism
1463     */
1464    public boolean getMaintainsParallelism() {
1465        workerCountReadFence();
1466        return maintainsParallelism;
1467    }
1468
1469    /**
1470     * Sets whether this pool dynamically maintains its target
1471     * parallelism level. If false, new threads are added only to
1472     * avoid possible starvation.
1473     *
1474     * @param enable {@code true} to maintain parallelism
1475     */
1476    public void setMaintainsParallelism(boolean enable) {
1477        maintainsParallelism = enable;
1478        workerCountWriteFence();
1479    }
1480
1481    /**
1482     * Establishes local first-in-first-out scheduling mode for forked
1483     * tasks that are never joined. This mode may be more appropriate
1484     * than default locally stack-based mode in applications in which
1485     * worker threads only process asynchronous tasks.  This method is
1486     * designed to be invoked only when the pool is quiescent, and
1487     * typically only before any tasks are submitted. The effects of
1488     * invocations at other times may be unpredictable.
1489     *
1490     * @param async if {@code true}, use locally FIFO scheduling
1491     * @return the previous mode
1492     * @see #getAsyncMode
1493     */
1494    public boolean setAsyncMode(boolean async) {
1495        workerCountReadFence();
1496        boolean oldMode = locallyFifo;
1497        if (oldMode != async) {
1498            locallyFifo = async;
1499            workerCountWriteFence();
1500            for (ForkJoinWorkerThread w : workers) {
1501                if (w != null)
1502                    w.setAsyncMode(async);
1503            }
1504        }
1505        return oldMode;
1506    }
1507
1508    /**
1551       * Returns {@code true} if this pool uses local first-in-first-out
1552       * scheduling mode for forked tasks that are never joined.
1553       *
1554       * @return {@code true} if this pool uses async mode
1513     * @see #setAsyncMode
1555       */
1556      public boolean getAsyncMode() {
1516        workerCountReadFence();
1557          return locallyFifo;
1558      }
1559  
# Line 1582 | Line 1622 | public class ForkJoinPool extends Abstra
1622       */
1623      public long getQueuedTaskCount() {
1624          long count = 0;
1625 <        for (ForkJoinWorkerThread w : workers) {
1625 >        for (ForkJoinWorkerThread w : workers)
1626              if (w != null)
1627                  count += w.getQueueSize();
1588        }
1628          return count;
1629      }
1630  
# Line 1639 | Line 1678 | public class ForkJoinPool extends Abstra
1678       * @return the number of elements transferred
1679       */
1680      protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
1681 <        int n = submissionQueue.drainTo(c);
1682 <        for (ForkJoinWorkerThread w : workers) {
1681 >        int count = submissionQueue.drainTo(c);
1682 >        for (ForkJoinWorkerThread w : workers)
1683              if (w != null)
1684 <                n += w.drainTasksTo(c);
1685 <        }
1647 <        return n;
1684 >                count += w.drainTasksTo(c);
1685 >        return count;
1686      }
1687  
1688      /**
# Line 1747 | Line 1785 | public class ForkJoinPool extends Abstra
1785      }
1786  
1787      /**
1788 +     * Returns true if terminating or terminated. Used by ForkJoinWorkerThread.
1789 +     */
1790 +    final boolean isAtLeastTerminating() {
1791 +        return runState >= TERMINATING;
1792 +    }
1793 +
1794 +    /**
1795       * Returns {@code true} if this pool has been shut down.
1796       *
1797       * @return {@code true} if this pool has been shut down
# Line 1768 | Line 1813 | public class ForkJoinPool extends Abstra
1813       */
1814      public boolean awaitTermination(long timeout, TimeUnit unit)
1815          throws InterruptedException {
1816 <        return terminationLatch.await(timeout, unit);
1816 >        try {
1817 >            termination.awaitAdvanceInterruptibly(0, timeout, unit);
1818 >        } catch (TimeoutException ex) {
1819 >            return false;
1820 >        }
1821 >        return true;
1822      }
1823  
1824      /**
1825       * Interface for extending managed parallelism for tasks running
1826       * in {@link ForkJoinPool}s.
1827       *
1828 <     * <p>A {@code ManagedBlocker} provides two methods.
1829 <     * Method {@code isReleasable} must return {@code true} if
1830 <     * blocking is not necessary. Method {@code block} blocks the
1831 <     * current thread if necessary (perhaps internally invoking
1832 <     * {@code isReleasable} before actually blocking).
1828 >     * <p>A {@code ManagedBlocker} provides two methods.  Method
1829 >     * {@code isReleasable} must return {@code true} if blocking is
1830 >     * not necessary. Method {@code block} blocks the current thread
1831 >     * if necessary (perhaps internally invoking {@code isReleasable}
1832 >     * before actually blocking). The unusual methods in this API
1833 >     * accommodate synchronizers that may, but don't usually, block
1834 >     * for long periods. Similarly, they allow more efficient internal
1835 >     * handling of cases in which additional workers may be, but
1836 >     * usually are not, needed to ensure sufficient parallelism.
1837 >     * Toward this end, implementations of method {@code isReleasable}
1838 >     * must be amenable to repeated invocation.
1839       *
1840       * <p>For example, here is a ManagedBlocker based on a
1841       * ReentrantLock:
# Line 1797 | Line 1853 | public class ForkJoinPool extends Abstra
1853       *     return hasLock || (hasLock = lock.tryLock());
1854       *   }
1855       * }}</pre>
1856 +     *
1857 +     * <p>Here is a class that possibly blocks waiting for an
1858 +     * item on a given queue:
1859 +     *  <pre> {@code
1860 +     * class QueueTaker<E> implements ManagedBlocker {
1861 +     *   final BlockingQueue<E> queue;
1862 +     *   volatile E item = null;
1863 +     *   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
1864 +     *   public boolean block() throws InterruptedException {
1865 +     *     if (item == null)
1866 +     *       item = queue.take();
1867 +     *     return true;
1868 +     *   }
1869 +     *   public boolean isReleasable() {
1870 +     *     return item != null || (item = queue.poll()) != null;
1871 +     *   }
1872 +     *   public E getItem() { // call after pool.managedBlock completes
1873 +     *     return item;
1874 +     *   }
1875 +     * }}</pre>
1876       */
1877      public static interface ManagedBlocker {
1878          /**
# Line 1820 | Line 1896 | public class ForkJoinPool extends Abstra
1896       * Blocks in accord with the given blocker.  If the current thread
1897       * is a {@link ForkJoinWorkerThread}, this method possibly
1898       * arranges for a spare thread to be activated if necessary to
1899 <     * ensure parallelism while the current thread is blocked.
1824 <     *
1825 <     * <p>If {@code maintainParallelism} is {@code true} and the pool
1826 <     * supports it ({@link #getMaintainsParallelism}), this method
1827 <     * attempts to maintain the pool's nominal parallelism. Otherwise
1828 <     * it activates a thread only if necessary to avoid complete
1829 <     * starvation. This option may be preferable when blockages use
1830 <     * timeouts, or are almost always brief.
1899 >     * ensure sufficient parallelism while the current thread is blocked.
1900       *
1901       * <p>If the caller is not a {@link ForkJoinTask}, this method is
1902       * behaviorally equivalent to
# Line 1841 | Line 1910 | public class ForkJoinPool extends Abstra
1910       * first be expanded to ensure parallelism, and later adjusted.
1911       *
1912       * @param blocker the blocker
1844     * @param maintainParallelism if {@code true} and supported by
1845     * this pool, attempt to maintain the pool's nominal parallelism;
1846     * otherwise activate a thread only if necessary to avoid
1847     * complete starvation.
1913       * @throws InterruptedException if blocker.block did so
1914       */
1915 <    public static void managedBlock(ManagedBlocker blocker,
1851 <                                    boolean maintainParallelism)
1915 >    public static void managedBlock(ManagedBlocker blocker)
1916          throws InterruptedException {
1917          Thread t = Thread.currentThread();
1918 <        if (t instanceof ForkJoinWorkerThread)
1919 <            ((ForkJoinWorkerThread) t).pool.
1920 <                doBlock(blocker, maintainParallelism);
1921 <        else
1922 <            awaitBlocker(blocker);
1923 <    }
1924 <
1861 <    /**
1862 <     * Performs Non-FJ blocking
1863 <     */
1864 <    private static void awaitBlocker(ManagedBlocker blocker)
1865 <        throws InterruptedException {
1866 <        do {} while (!blocker.isReleasable() && !blocker.block());
1918 >        if (t instanceof ForkJoinWorkerThread) {
1919 >            ForkJoinWorkerThread w = (ForkJoinWorkerThread) t;
1920 >            w.pool.awaitBlocker(blocker);
1921 >        }
1922 >        else {
1923 >            do {} while (!blocker.isReleasable() && !blocker.block());
1924 >        }
1925      }
1926  
1927      // AbstractExecutorService overrides.  These rely on undocumented
# Line 1888 | Line 1946 | public class ForkJoinPool extends Abstra
1946      private static final long eventCountOffset =
1947          objectFieldOffset("eventCount", ForkJoinPool.class);
1948      private static final long eventWaitersOffset =
1949 <        objectFieldOffset("eventWaiters",ForkJoinPool.class);
1949 >        objectFieldOffset("eventWaiters", ForkJoinPool.class);
1950      private static final long stealCountOffset =
1951 <        objectFieldOffset("stealCount",ForkJoinPool.class);
1952 <
1951 >        objectFieldOffset("stealCount", ForkJoinPool.class);
1952 >    private static final long spareWaitersOffset =
1953 >        objectFieldOffset("spareWaiters", ForkJoinPool.class);
1954  
1955      private static long objectFieldOffset(String field, Class<?> klazz) {
1956          try {

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines