1 /*
2 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
3 *
4 * This code is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License version 2 only, as
6 * published by the Free Software Foundation. Oracle designates this
7 * particular file as subject to the "Classpath" exception as provided
8 * by Oracle in the LICENSE file that accompanied this code.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 */
24
25 /*
26 * This file is available under and governed by the GNU General Public
27 * License version 2 only, as published by the Free Software Foundation.
28 * However, the following notice accompanied the original version of this
29 * file:
30 *
31 * Written by Doug Lea with assistance from members of JCP JSR-166
32 * Expert Group and released to the public domain, as explained at
33 * http://creativecommons.org/publicdomain/zero/1.0/
34 */
35
36 package java.util.concurrent;
37
38 import java.lang.Thread.UncaughtExceptionHandler;
39 import java.lang.invoke.MethodHandles;
40 import java.lang.invoke.VarHandle;
41 import java.security.AccessController;
42 import java.security.AccessControlContext;
43 import java.security.Permission;
44 import java.security.Permissions;
45 import java.security.PrivilegedAction;
46 import java.security.ProtectionDomain;
47 import java.util.ArrayList;
48 import java.util.Collection;
49 import java.util.Collections;
50 import java.util.List;
51 import java.util.function.Predicate;
52 import java.util.concurrent.locks.LockSupport;
53
54 /**
55 * An {@link ExecutorService} for running {@link ForkJoinTask}s.
56 * A {@code ForkJoinPool} provides the entry point for submissions
57 * from non-{@code ForkJoinTask} clients, as well as management and
58 * monitoring operations.
59 *
60 * <p>A {@code ForkJoinPool} differs from other kinds of {@link
61 * ExecutorService} mainly by virtue of employing
62 * <em>work-stealing</em>: all threads in the pool attempt to find and
63 * execute tasks submitted to the pool and/or created by other active
64 * tasks (eventually blocking waiting for work if none exist). This
65 * enables efficient processing when most tasks spawn other subtasks
66 * (as do most {@code ForkJoinTask}s), as well as when many small
67 * tasks are submitted to the pool from external clients. Especially
68 * when setting <em>asyncMode</em> to true in constructors, {@code
69 * ForkJoinPool}s may also be appropriate for use with event-style
70 * tasks that are never joined. All worker threads are initialized
71 * with {@link Thread#isDaemon} set {@code true}.
72 *
73 * <p>A static {@link #commonPool()} is available and appropriate for
74 * most applications. The common pool is used by any ForkJoinTask that
75 * is not explicitly submitted to a specified pool. Using the common
76 * pool normally reduces resource usage (its threads are slowly
77 * reclaimed during periods of non-use, and reinstated upon subsequent
78 * use).
79 *
80 * <p>For applications that require separate or custom pools, a {@code
81 * ForkJoinPool} may be constructed with a given target parallelism
82 * level; by default, equal to the number of available processors.
83 * The pool attempts to maintain enough active (or available) threads
84 * by dynamically adding, suspending, or resuming internal worker
85 * threads, even if some tasks are stalled waiting to join others.
86 * However, no such adjustments are guaranteed in the face of blocked
87 * I/O or other unmanaged synchronization. The nested {@link
88 * ManagedBlocker} interface enables extension of the kinds of
89 * synchronization accommodated. The default policies may be
90 * overridden using a constructor with parameters corresponding to
91 * those documented in class {@link ThreadPoolExecutor}.
92 *
93 * <p>In addition to execution and lifecycle control methods, this
94 * class provides status check methods (for example
95 * {@link #getStealCount}) that are intended to aid in developing,
96 * tuning, and monitoring fork/join applications. Also, method
97 * {@link #toString} returns indications of pool state in a
98 * convenient form for informal monitoring.
99 *
100 * <p>As is the case with other ExecutorServices, there are three
101 * main task execution methods summarized in the following table.
102 * These are designed to be used primarily by clients not already
103 * engaged in fork/join computations in the current pool. The main
104 * forms of these methods accept instances of {@code ForkJoinTask},
105 * but overloaded forms also allow mixed execution of plain {@code
106 * Runnable}- or {@code Callable}- based activities as well. However,
107 * tasks that are already executing in a pool should normally instead
108 * use the within-computation forms listed in the table unless using
109 * async event-style tasks that are not usually joined, in which case
110 * there is little difference among choice of methods.
111 *
112 * <table class="plain">
113 * <caption>Summary of task execution methods</caption>
114 * <tr>
115 * <td></td>
116 * <th scope="col"> Call from non-fork/join clients</th>
117 * <th scope="col"> Call from within fork/join computations</th>
118 * </tr>
119 * <tr>
120 * <th scope="row" style="text-align:left"> Arrange async execution</th>
121 * <td> {@link #execute(ForkJoinTask)}</td>
122 * <td> {@link ForkJoinTask#fork}</td>
123 * </tr>
124 * <tr>
125 * <th scope="row" style="text-align:left"> Await and obtain result</th>
126 * <td> {@link #invoke(ForkJoinTask)}</td>
127 * <td> {@link ForkJoinTask#invoke}</td>
128 * </tr>
129 * <tr>
130 * <th scope="row" style="text-align:left"> Arrange exec and obtain Future</th>
131 * <td> {@link #submit(ForkJoinTask)}</td>
132 * <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
133 * </tr>
134 * </table>
135 *
136 * <p>The parameters used to construct the common pool may be controlled by
137 * setting the following {@linkplain System#getProperty system properties}:
138 * <ul>
139 * <li>{@systemProperty java.util.concurrent.ForkJoinPool.common.parallelism}
140 * - the parallelism level, a non-negative integer
141 * <li>{@systemProperty java.util.concurrent.ForkJoinPool.common.threadFactory}
142 * - the class name of a {@link ForkJoinWorkerThreadFactory}.
143 * The {@linkplain ClassLoader#getSystemClassLoader() system class loader}
144 * is used to load this class.
145 * <li>{@systemProperty java.util.concurrent.ForkJoinPool.common.exceptionHandler}
146 * - the class name of a {@link UncaughtExceptionHandler}.
147 * The {@linkplain ClassLoader#getSystemClassLoader() system class loader}
148 * is used to load this class.
149 * <li>{@systemProperty java.util.concurrent.ForkJoinPool.common.maximumSpares}
150 * - the maximum number of allowed extra threads to maintain target
151 * parallelism (default 256).
152 * </ul>
153 * If no thread factory is supplied via a system property, then the
154 * common pool uses a factory that uses the system class loader as the
155 * {@linkplain Thread#getContextClassLoader() thread context class loader}.
156 * In addition, if a {@link SecurityManager} is present, then
157 * the common pool uses a factory supplying threads that have no
158 * {@link Permissions} enabled.
159 *
160 * Upon any error in establishing these settings, default parameters
161 * are used. It is possible to disable or limit the use of threads in
162 * the common pool by setting the parallelism property to zero, and/or
163 * using a factory that may return {@code null}. However doing so may
164 * cause unjoined tasks to never be executed.
165 *
166 * <p><b>Implementation notes</b>: This implementation restricts the
167 * maximum number of running threads to 32767. Attempts to create
168 * pools with greater than the maximum number result in
169 * {@code IllegalArgumentException}.
170 *
171 * <p>This implementation rejects submitted tasks (that is, by throwing
172 * {@link RejectedExecutionException}) only when the pool is shut down
173 * or internal resources have been exhausted.
174 *
175 * @since 1.7
176 * @author Doug Lea
177 */
178 public class ForkJoinPool extends AbstractExecutorService {
179
180 /*
181 * Implementation Overview
182 *
183 * This class and its nested classes provide the main
184 * functionality and control for a set of worker threads:
185 * Submissions from non-FJ threads enter into submission queues.
186 * Workers take these tasks and typically split them into subtasks
187 * that may be stolen by other workers. Work-stealing based on
188 * randomized scans generally leads to better throughput than
189 * "work dealing" in which producers assign tasks to idle threads,
190 * in part because threads that have finished other tasks before
191 * the signalled thread wakes up (which can be a long time) can
192 * take the task instead. Preference rules give first priority to
193 * processing tasks from their own queues (LIFO or FIFO, depending
194 * on mode), then to randomized FIFO steals of tasks in other
195 * queues. This framework began as vehicle for supporting
196 * tree-structured parallelism using work-stealing. Over time,
197 * its scalability advantages led to extensions and changes to
198 * better support more diverse usage contexts. Because most
199 * internal methods and nested classes are interrelated, their
200 * main rationale and descriptions are presented here; individual
201 * methods and nested classes contain only brief comments about
202 * details.
203 *
204 * WorkQueues
205 * ==========
206 *
207 * Most operations occur within work-stealing queues (in nested
208 * class WorkQueue). These are special forms of Deques that
209 * support only three of the four possible end-operations -- push,
210 * pop, and poll (aka steal), under the further constraints that
211 * push and pop are called only from the owning thread (or, as
212 * extended here, under a lock), while poll may be called from
213 * other threads. (If you are unfamiliar with them, you probably
214 * want to read Herlihy and Shavit's book "The Art of
215 * Multiprocessor programming", chapter 16 describing these in
216 * more detail before proceeding.) The main work-stealing queue
217 * design is roughly similar to those in the papers "Dynamic
218 * Circular Work-Stealing Deque" by Chase and Lev, SPAA 2005
219 * (http://research.sun.com/scalable/pubs/index.html) and
220 * "Idempotent work stealing" by Michael, Saraswat, and Vechev,
221 * PPoPP 2009 (http://portal.acm.org/citation.cfm?id=1504186).
222 * The main differences ultimately stem from GC requirements that
223 * we null out taken slots as soon as we can, to maintain as small
224 * a footprint as possible even in programs generating huge
225 * numbers of tasks. To accomplish this, we shift the CAS
226 * arbitrating pop vs poll (steal) from being on the indices
227 * ("base" and "top") to the slots themselves.
228 *
229 * Adding tasks then takes the form of a classic array push(task)
230 * in a circular buffer:
231 * q.array[q.top++ % length] = task;
232 *
233 * (The actual code needs to null-check and size-check the array,
234 * uses masking, not mod, for indexing a power-of-two-sized array,
235 * adds a release fence for publication, and possibly signals
236 * waiting workers to start scanning -- see below.) Both a
237 * successful pop and poll mainly entail a CAS of a slot from
238 * non-null to null.
239 *
240 * The pop operation (always performed by owner) is:
241 * if ((the task at top slot is not null) and
242 * (CAS slot to null))
243 * decrement top and return task;
244 *
245 * And the poll operation (usually by a stealer) is
246 * if ((the task at base slot is not null) and
247 * (CAS slot to null))
248 * increment base and return task;
249 *
250 * There are several variants of each of these. Most uses occur
251 * within operations that also interleave contention or emptiness
252 * tracking or inspection of elements before extracting them, so
253 * must interleave these with the above code. When performed by
254 * owner, getAndSet is used instead of CAS (see for example method
255 * nextLocalTask) which is usually more efficient, and possible
256 * because the top index cannot independently change during the
257 * operation.
258 *
259 * Memory ordering. See "Correct and Efficient Work-Stealing for
260 * Weak Memory Models" by Le, Pop, Cohen, and Nardelli, PPoPP 2013
261 * (http://www.di.ens.fr/~zappa/readings/ppopp13.pdf) for an
262 * analysis of memory ordering requirements in work-stealing
263 * algorithms similar to (but different than) the one used here.
264 * Extracting tasks in array slots via (fully fenced) CAS provides
265 * primary synchronization. The base and top indices imprecisely
266 * guide where to extract from. We do not usually require strict
267 * orderings of array and index updates. Many index accesses use
268 * plain mode, with ordering constrained by surrounding context
269 * (usually with respect to element CASes or the two WorkQueue
270 * volatile fields source and phase). When not otherwise already
271 * constrained, reads of "base" by queue owners use acquire-mode,
272 * and some externally callable methods preface accesses with
273 * acquire fences. Additionally, to ensure that index update
274 * writes are not coalesced or postponed in loops etc, "opaque"
275 * mode is used in a few cases where timely writes are not
276 * otherwise ensured. The "locked" versions of push- and pop-
277 * based methods for shared queues differ from owned versions
278 * because locking already forces some of the ordering.
279 *
280 * Because indices and slot contents cannot always be consistent,
281 * a check that base == top indicates (momentary) emptiness, but
282 * otherwise may err on the side of possibly making the queue
283 * appear nonempty when a push, pop, or poll have not fully
284 * committed, or making it appear empty when an update of top has
285 * not yet been visibly written. (Method isEmpty() checks the
286 * case of a partially completed removal of the last element.)
287 * Because of this, the poll operation, considered individually,
288 * is not wait-free. One thief cannot successfully continue until
289 * another in-progress one (or, if previously empty, a push)
290 * visibly completes. This can stall threads when required to
291 * consume from a given queue (see method poll()). However, in
292 * the aggregate, we ensure at least probabilistic
293 * non-blockingness. If an attempted steal fails, a scanning
294 * thief chooses a different random victim target to try next. So,
295 * in order for one thief to progress, it suffices for any
296 * in-progress poll or new push on any empty queue to complete.
297 *
298 * This approach also enables support of a user mode in which
299 * local task processing is in FIFO, not LIFO order, simply by
300 * using poll rather than pop. This can be useful in
301 * message-passing frameworks in which tasks are never joined.
302 *
303 * WorkQueues are also used in a similar way for tasks submitted
304 * to the pool. We cannot mix these tasks in the same queues used
305 * by workers. Instead, we randomly associate submission queues
306 * with submitting threads, using a form of hashing. The
307 * ThreadLocalRandom probe value serves as a hash code for
308 * choosing existing queues, and may be randomly repositioned upon
309 * contention with other submitters. In essence, submitters act
310 * like workers except that they are restricted to executing local
311 * tasks that they submitted. Insertion of tasks in shared mode
312 * requires a lock but we use only a simple spinlock (using field
313 * phase), because submitters encountering a busy queue move to a
314 * different position to use or create other queues -- they block
315 * only when creating and registering new queues. Because it is
316 * used only as a spinlock, unlocking requires only a "releasing"
317 * store (using setRelease) unless otherwise signalling.
318 *
319 * Management
320 * ==========
321 *
322 * The main throughput advantages of work-stealing stem from
323 * decentralized control -- workers mostly take tasks from
324 * themselves or each other, at rates that can exceed a billion
325 * per second. The pool itself creates, activates (enables
326 * scanning for and running tasks), deactivates, blocks, and
327 * terminates threads, all with minimal central information.
328 * There are only a few properties that we can globally track or
329 * maintain, so we pack them into a small number of variables,
330 * often maintaining atomicity without blocking or locking.
331 * Nearly all essentially atomic control state is held in a few
332 * volatile variables that are by far most often read (not
333 * written) as status and consistency checks. We pack as much
334 * information into them as we can.
335 *
336 * Field "ctl" contains 64 bits holding information needed to
337 * atomically decide to add, enqueue (on an event queue), and
338 * dequeue and release workers. To enable this packing, we
339 * restrict maximum parallelism to (1<<15)-1 (which is far in
340 * excess of normal operating range) to allow ids, counts, and
341 * their negations (used for thresholding) to fit into 16bit
342 * subfields.
343 *
344 * Field "mode" holds configuration parameters as well as lifetime
345 * status, atomically and monotonically setting SHUTDOWN, STOP,
346 * and finally TERMINATED bits.
347 *
348 * Field "workQueues" holds references to WorkQueues. It is
349 * updated (only during worker creation and termination) under
350 * lock (using field workerNamePrefix as lock), but is otherwise
351 * concurrently readable, and accessed directly. We also ensure
352 * that uses of the array reference itself never become too stale
353 * in case of resizing, by arranging that (re-)reads are separated
354 * by at least one acquiring read access. To simplify index-based
355 * operations, the array size is always a power of two, and all
356 * readers must tolerate null slots. Worker queues are at odd
357 * indices. Shared (submission) queues are at even indices, up to
358 * a maximum of 64 slots, to limit growth even if the array needs
359 * to expand to add more workers. Grouping them together in this
360 * way simplifies and speeds up task scanning.
361 *
362 * All worker thread creation is on-demand, triggered by task
363 * submissions, replacement of terminated workers, and/or
364 * compensation for blocked workers. However, all other support
365 * code is set up to work with other policies. To ensure that we
366 * do not hold on to worker references that would prevent GC, all
367 * accesses to workQueues are via indices into the workQueues
368 * array (which is one source of some of the messy code
369 * constructions here). In essence, the workQueues array serves as
370 * a weak reference mechanism. Thus for example the stack top
371 * subfield of ctl stores indices, not references.
372 *
373 * Queuing Idle Workers. Unlike HPC work-stealing frameworks, we
374 * cannot let workers spin indefinitely scanning for tasks when
375 * none can be found immediately, and we cannot start/resume
376 * workers unless there appear to be tasks available. On the
377 * other hand, we must quickly prod them into action when new
378 * tasks are submitted or generated. In many usages, ramp-up time
379 * is the main limiting factor in overall performance, which is
380 * compounded at program start-up by JIT compilation and
381 * allocation. So we streamline this as much as possible.
382 *
383 * The "ctl" field atomically maintains total worker and
384 * "released" worker counts, plus the head of the available worker
385 * queue (actually stack, represented by the lower 32bit subfield
386 * of ctl). Released workers are those known to be scanning for
387 * and/or running tasks. Unreleased ("available") workers are
388 * recorded in the ctl stack. These workers are made available for
389 * signalling by enqueuing in ctl (see method runWorker). The
390 * "queue" is a form of Treiber stack. This is ideal for
391 * activating threads in most-recently used order, and improves
392 * performance and locality, outweighing the disadvantages of
393 * being prone to contention and inability to release a worker
394 * unless it is topmost on stack. To avoid missed signal problems
395 * inherent in any wait/signal design, available workers rescan
396 * for (and if found run) tasks after enqueuing. Normally their
397 * release status will be updated while doing so, but the released
398 * worker ctl count may underestimate the number of active
399 * threads. (However, it is still possible to determine quiescence
400 * via a validation traversal -- see isQuiescent). After an
401 * unsuccessful rescan, available workers are blocked until
402 * signalled (see signalWork). The top stack state holds the
403 * value of the "phase" field of the worker: its index and status,
404 * plus a version counter that, in addition to the count subfields
405 * (also serving as version stamps) provide protection against
406 * Treiber stack ABA effects.
407 *
408 * Creating workers. To create a worker, we pre-increment counts
409 * (serving as a reservation), and attempt to construct a
410 * ForkJoinWorkerThread via its factory. Upon construction, the
411 * new thread invokes registerWorker, where it constructs a
412 * WorkQueue and is assigned an index in the workQueues array
413 * (expanding the array if necessary). The thread is then started.
414 * Upon any exception across these steps, or null return from
415 * factory, deregisterWorker adjusts counts and records
416 * accordingly. If a null return, the pool continues running with
417 * fewer than the target number workers. If exceptional, the
418 * exception is propagated, generally to some external caller.
419 * Worker index assignment avoids the bias in scanning that would
420 * occur if entries were sequentially packed starting at the front
421 * of the workQueues array. We treat the array as a simple
422 * power-of-two hash table, expanding as needed. The seedIndex
423 * increment ensures no collisions until a resize is needed or a
424 * worker is deregistered and replaced, and thereafter keeps
425 * probability of collision low. We cannot use
426 * ThreadLocalRandom.getProbe() for similar purposes here because
427 * the thread has not started yet, but do so for creating
428 * submission queues for existing external threads (see
429 * externalPush).
430 *
431 * WorkQueue field "phase" is used by both workers and the pool to
432 * manage and track whether a worker is UNSIGNALLED (possibly
433 * blocked waiting for a signal). When a worker is enqueued its
434 * phase field is set. Note that phase field updates lag queue CAS
435 * releases so usage requires care -- seeing a negative phase does
436 * not guarantee that the worker is available. When queued, the
437 * lower 16 bits of scanState must hold its pool index. So we
438 * place the index there upon initialization and otherwise keep it
439 * there or restore it when necessary.
440 *
441 * The ctl field also serves as the basis for memory
442 * synchronization surrounding activation. This uses a more
443 * efficient version of a Dekker-like rule that task producers and
444 * consumers sync with each other by both writing/CASing ctl (even
445 * if to its current value). This would be extremely costly. So
446 * we relax it in several ways: (1) Producers only signal when
447 * their queue is possibly empty at some point during a push
448 * operation. (2) Other workers propagate this signal
449 * when they find tasks in a queue with size greater than one. (3)
450 * Workers only enqueue after scanning (see below) and not finding
451 * any tasks. (4) Rather than CASing ctl to its current value in
452 * the common case where no action is required, we reduce write
453 * contention by equivalently prefacing signalWork when called by
454 * an external task producer using a memory access with
455 * full-volatile semantics or a "fullFence".
456 *
457 * Almost always, too many signals are issued, in part because a
458 * task producer cannot tell if some existing worker is in the
459 * midst of finishing one task (or already scanning) and ready to
460 * take another without being signalled. So the producer might
461 * instead activate a different worker that does not find any
462 * work, and then inactivates. This scarcely matters in
463 * steady-state computations involving all workers, but can create
464 * contention and bookkeeping bottlenecks during ramp-up,
465 * ramp-down, and small computations involving only a few workers.
466 *
467 * Scanning. Method scan (from runWorker) performs top-level
468 * scanning for tasks. (Similar scans appear in helpQuiesce and
469 * pollScan.) Each scan traverses and tries to poll from each
470 * queue starting at a random index. Scans are not performed in
471 * ideal random permutation order, to reduce cacheline
472 * contention. The pseudorandom generator need not have
473 * high-quality statistical properties in the long term, but just
474 * within computations; We use Marsaglia XorShifts (often via
475 * ThreadLocalRandom.nextSecondarySeed), which are cheap and
476 * suffice. Scanning also includes contention reduction: When
477 * scanning workers fail to extract an apparently existing task,
478 * they soon restart at a different pseudorandom index. This form
479 * of backoff improves throughput when many threads are trying to
480 * take tasks from few queues, which can be common in some usages.
481 * Scans do not otherwise explicitly take into account core
482 * affinities, loads, cache localities, etc, However, they do
483 * exploit temporal locality (which usually approximates these) by
484 * preferring to re-poll from the same queue after a successful
485 * poll before trying others (see method topLevelExec). However
486 * this preference is bounded (see TOP_BOUND_SHIFT) as a safeguard
487 * against infinitely unfair looping under unbounded user task
488 * recursion, and also to reduce long-term contention when many
489 * threads poll few queues holding many small tasks. The bound is
490 * high enough to avoid much impact on locality and scheduling
491 * overhead.
492 *
493 * Trimming workers. To release resources after periods of lack of
494 * use, a worker starting to wait when the pool is quiescent will
495 * time out and terminate (see method runWorker) if the pool has
496 * remained quiescent for period given by field keepAlive.
497 *
498 * Shutdown and Termination. A call to shutdownNow invokes
499 * tryTerminate to atomically set a runState bit. The calling
500 * thread, as well as every other worker thereafter terminating,
501 * helps terminate others by cancelling their unprocessed tasks,
502 * and waking them up, doing so repeatedly until stable. Calls to
503 * non-abrupt shutdown() preface this by checking whether
504 * termination should commence by sweeping through queues (until
505 * stable) to ensure lack of in-flight submissions and workers
506 * about to process them before triggering the "STOP" phase of
507 * termination.
508 *
509 * Joining Tasks
510 * =============
511 *
512 * Any of several actions may be taken when one worker is waiting
513 * to join a task stolen (or always held) by another. Because we
514 * are multiplexing many tasks on to a pool of workers, we can't
515 * always just let them block (as in Thread.join). We also cannot
516 * just reassign the joiner's run-time stack with another and
517 * replace it later, which would be a form of "continuation", that
518 * even if possible is not necessarily a good idea since we may
519 * need both an unblocked task and its continuation to progress.
520 * Instead we combine two tactics:
521 *
522 * Helping: Arranging for the joiner to execute some task that it
523 * would be running if the steal had not occurred.
524 *
525 * Compensating: Unless there are already enough live threads,
526 * method tryCompensate() may create or re-activate a spare
527 * thread to compensate for blocked joiners until they unblock.
528 *
529 * A third form (implemented in tryRemoveAndExec) amounts to
530 * helping a hypothetical compensator: If we can readily tell that
531 * a possible action of a compensator is to steal and execute the
532 * task being joined, the joining thread can do so directly,
533 * without the need for a compensation thread.
534 *
535 * The ManagedBlocker extension API can't use helping so relies
536 * only on compensation in method awaitBlocker.
537 *
538 * The algorithm in awaitJoin entails a form of "linear helping".
539 * Each worker records (in field source) the id of the queue from
540 * which it last stole a task. The scan in method awaitJoin uses
541 * these markers to try to find a worker to help (i.e., steal back
542 * a task from and execute it) that could hasten completion of the
543 * actively joined task. Thus, the joiner executes a task that
544 * would be on its own local deque if the to-be-joined task had
545 * not been stolen. This is a conservative variant of the approach
546 * described in Wagner & Calder "Leapfrogging: a portable
547 * technique for implementing efficient futures" SIGPLAN Notices,
548 * 1993 (http://portal.acm.org/citation.cfm?id=155354). It differs
549 * mainly in that we only record queue ids, not full dependency
550 * links. This requires a linear scan of the workQueues array to
551 * locate stealers, but isolates cost to when it is needed, rather
552 * than adding to per-task overhead. Searches can fail to locate
553 * stealers GC stalls and the like delay recording sources.
554 * Further, even when accurately identified, stealers might not
555 * ever produce a task that the joiner can in turn help with. So,
556 * compensation is tried upon failure to find tasks to run.
557 *
558 * Compensation does not by default aim to keep exactly the target
559 * parallelism number of unblocked threads running at any given
560 * time. Some previous versions of this class employed immediate
561 * compensations for any blocked join. However, in practice, the
562 * vast majority of blockages are transient byproducts of GC and
563 * other JVM or OS activities that are made worse by replacement
564 * when they cause longer-term oversubscription. Rather than
565 * impose arbitrary policies, we allow users to override the
566 * default of only adding threads upon apparent starvation. The
567 * compensation mechanism may also be bounded. Bounds for the
568 * commonPool (see COMMON_MAX_SPARES) better enable JVMs to cope
569 * with programming errors and abuse before running out of
570 * resources to do so.
571 *
572 * Common Pool
573 * ===========
574 *
575 * The static common pool always exists after static
576 * initialization. Since it (or any other created pool) need
577 * never be used, we minimize initial construction overhead and
578 * footprint to the setup of about a dozen fields.
579 *
580 * When external threads submit to the common pool, they can
581 * perform subtask processing (see externalHelpComplete and
582 * related methods) upon joins. This caller-helps policy makes it
583 * sensible to set common pool parallelism level to one (or more)
584 * less than the total number of available cores, or even zero for
585 * pure caller-runs. We do not need to record whether external
586 * submissions are to the common pool -- if not, external help
587 * methods return quickly. These submitters would otherwise be
588 * blocked waiting for completion, so the extra effort (with
589 * liberally sprinkled task status checks) in inapplicable cases
590 * amounts to an odd form of limited spin-wait before blocking in
591 * ForkJoinTask.join.
592 *
593 * As a more appropriate default in managed environments, unless
594 * overridden by system properties, we use workers of subclass
595 * InnocuousForkJoinWorkerThread when there is a SecurityManager
596 * present. These workers have no permissions set, do not belong
597 * to any user-defined ThreadGroup, and erase all ThreadLocals
598 * after executing any top-level task (see
599 * WorkQueue.afterTopLevelExec). The associated mechanics (mainly
600 * in ForkJoinWorkerThread) may be JVM-dependent and must access
601 * particular Thread class fields to achieve this effect.
602 *
603 * Memory placement
604 * ================
605 *
606 * Performance can be very sensitive to placement of instances of
607 * ForkJoinPool and WorkQueues and their queue arrays. To reduce
608 * false-sharing impact, the @Contended annotation isolates
609 * adjacent WorkQueue instances, as well as the ForkJoinPool.ctl
610 * field. WorkQueue arrays are allocated (by their threads) with
611 * larger initial sizes than most ever need, mostly to reduce
612 * false sharing with current garbage collectors that use cardmark
613 * tables.
614 *
615 * Style notes
616 * ===========
617 *
618 * Memory ordering relies mainly on VarHandles. This can be
619 * awkward and ugly, but also reflects the need to control
620 * outcomes across the unusual cases that arise in very racy code
621 * with very few invariants. All fields are read into locals
622 * before use, and null-checked if they are references. Array
623 * accesses using masked indices include checks (that are always
624 * true) that the array length is non-zero to avoid compilers
625 * inserting more expensive traps. This is usually done in a
626 * "C"-like style of listing declarations at the heads of methods
627 * or blocks, and using inline assignments on first encounter.
628 * Nearly all explicit checks lead to bypass/return, not exception
629 * throws, because they may legitimately arise due to
630 * cancellation/revocation during shutdown.
631 *
632 * There is a lot of representation-level coupling among classes
633 * ForkJoinPool, ForkJoinWorkerThread, and ForkJoinTask. The
634 * fields of WorkQueue maintain data structures managed by
635 * ForkJoinPool, so are directly accessed. There is little point
636 * trying to reduce this, since any associated future changes in
637 * representations will need to be accompanied by algorithmic
638 * changes anyway. Several methods intrinsically sprawl because
639 * they must accumulate sets of consistent reads of fields held in
640 * local variables. Some others are artificially broken up to
641 * reduce producer/consumer imbalances due to dynamic compilation.
642 * There are also other coding oddities (including several
643 * unnecessary-looking hoisted null checks) that help some methods
644 * perform reasonably even when interpreted (not compiled).
645 *
646 * The order of declarations in this file is (with a few exceptions):
647 * (1) Static utility functions
648 * (2) Nested (static) classes
649 * (3) Static fields
650 * (4) Fields, along with constants used when unpacking some of them
651 * (5) Internal control methods
652 * (6) Callbacks and other support for ForkJoinTask methods
653 * (7) Exported methods
654 * (8) Static block initializing statics in minimally dependent order
655 */
656
657 // Static utilities
658
659 /**
660 * If there is a security manager, makes sure caller has
661 * permission to modify threads.
662 */
663 private static void checkPermission() {
664 SecurityManager security = System.getSecurityManager();
665 if (security != null)
666 security.checkPermission(modifyThreadPermission);
667 }
668
669 // Nested classes
670
671 /**
672 * Factory for creating new {@link ForkJoinWorkerThread}s.
673 * A {@code ForkJoinWorkerThreadFactory} must be defined and used
674 * for {@code ForkJoinWorkerThread} subclasses that extend base
675 * functionality or initialize threads with different contexts.
676 */
677 public static interface ForkJoinWorkerThreadFactory {
678 /**
679 * Returns a new worker thread operating in the given pool.
680 * Returning null or throwing an exception may result in tasks
681 * never being executed. If this method throws an exception,
682 * it is relayed to the caller of the method (for example
683 * {@code execute}) causing attempted thread creation. If this
684 * method returns null or throws an exception, it is not
685 * retried until the next attempted creation (for example
686 * another call to {@code execute}).
687 *
688 * @param pool the pool this thread works in
689 * @return the new worker thread, or {@code null} if the request
690 * to create a thread is rejected
691 * @throws NullPointerException if the pool is null
692 */
693 public ForkJoinWorkerThread newThread(ForkJoinPool pool);
694 }
695
696 static AccessControlContext contextWithPermissions(Permission ... perms) {
697 Permissions permissions = new Permissions();
698 for (Permission perm : perms)
699 permissions.add(perm);
700 return new AccessControlContext(
701 new ProtectionDomain[] { new ProtectionDomain(null, permissions) });
702 }
703
704 /**
705 * Default ForkJoinWorkerThreadFactory implementation; creates a
706 * new ForkJoinWorkerThread using the system class loader as the
707 * thread context class loader.
708 */
709 private static final class DefaultForkJoinWorkerThreadFactory
710 implements ForkJoinWorkerThreadFactory {
711 private static final AccessControlContext ACC = contextWithPermissions(
712 new RuntimePermission("getClassLoader"),
713 new RuntimePermission("setContextClassLoader"));
714
715 public final ForkJoinWorkerThread newThread(ForkJoinPool pool) {
716 return AccessController.doPrivileged(
717 new PrivilegedAction<>() {
718 public ForkJoinWorkerThread run() {
719 return new ForkJoinWorkerThread(
720 pool, ClassLoader.getSystemClassLoader()); }},
721 ACC);
722 }
723 }
724
725 // Constants shared across ForkJoinPool and WorkQueue
726
727 // Bounds
728 static final int SWIDTH = 16; // width of short
729 static final int SMASK = 0xffff; // short bits == max index
730 static final int MAX_CAP = 0x7fff; // max #workers - 1
731 static final int SQMASK = 0x007e; // max 64 (even) slots
732
733 // Masks and units for WorkQueue.phase and ctl sp subfield
734 static final int UNSIGNALLED = 1 << 31; // must be negative
735 static final int SS_SEQ = 1 << 16; // version count
736 static final int QLOCK = 1; // must be 1
737
738 // Mode bits and sentinels, some also used in WorkQueue id and.source fields
739 static final int OWNED = 1; // queue has owner thread
740 static final int FIFO = 1 << 16; // fifo queue or access mode
741 static final int SHUTDOWN = 1 << 18;
742 static final int TERMINATED = 1 << 19;
743 static final int STOP = 1 << 31; // must be negative
744 static final int QUIET = 1 << 30; // not scanning or working
745 static final int DORMANT = QUIET | UNSIGNALLED;
746
747 /**
748 * Initial capacity of work-stealing queue array.
749 * Must be a power of two, at least 2.
750 */
751 static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
752
753 /**
754 * Maximum capacity for queue arrays. Must be a power of two less
755 * than or equal to 1 << (31 - width of array entry) to ensure
756 * lack of wraparound of index calculations, but defined to a
757 * value a bit less than this to help users trap runaway programs
758 * before saturating systems.
759 */
760 static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M
761
762 /**
763 * The maximum number of top-level polls per worker before
764 * checking other queues, expressed as a bit shift. See above for
765 * rationale.
766 */
767 static final int TOP_BOUND_SHIFT = 10;
768
769 /**
770 * Queues supporting work-stealing as well as external task
771 * submission. See above for descriptions and algorithms.
772 */
773 @jdk.internal.vm.annotation.Contended
774 static final class WorkQueue {
775 volatile int source; // source queue id, or sentinel
776 int id; // pool index, mode, tag
777 int base; // index of next slot for poll
778 int top; // index of next slot for push
779 volatile int phase; // versioned, negative: queued, 1: locked
780 int stackPred; // pool stack (ctl) predecessor link
781 int nsteals; // number of steals
782 ForkJoinTask<?>[] array; // the queued tasks; power of 2 size
783 final ForkJoinPool pool; // the containing pool (may be null)
784 final ForkJoinWorkerThread owner; // owning thread or null if shared
785
786 WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner) {
787 this.pool = pool;
788 this.owner = owner;
789 // Place indices in the center of array (that is not yet allocated)
790 base = top = INITIAL_QUEUE_CAPACITY >>> 1;
791 }
792
793 /**
794 * Tries to lock shared queue by CASing phase field.
795 */
796 final boolean tryLockPhase() {
797 return PHASE.compareAndSet(this, 0, 1);
798 }
799
800 final void releasePhaseLock() {
801 PHASE.setRelease(this, 0);
802 }
803
804 /**
805 * Returns an exportable index (used by ForkJoinWorkerThread).
806 */
807 final int getPoolIndex() {
808 return (id & 0xffff) >>> 1; // ignore odd/even tag bit
809 }
810
811 /**
812 * Returns the approximate number of tasks in the queue.
813 */
814 final int queueSize() {
815 int n = (int)BASE.getAcquire(this) - top;
816 return (n >= 0) ? 0 : -n; // ignore transient negative
817 }
818
819 /**
820 * Provides a more accurate estimate of whether this queue has
821 * any tasks than does queueSize, by checking whether a
822 * near-empty queue has at least one unclaimed task.
823 */
824 final boolean isEmpty() {
825 ForkJoinTask<?>[] a; int n, cap, b;
826 VarHandle.acquireFence(); // needed by external callers
827 return ((n = (b = base) - top) >= 0 || // possibly one task
828 (n == -1 && ((a = array) == null ||
829 (cap = a.length) == 0 ||
830 a[(cap - 1) & b] == null)));
831 }
832
833 /**
834 * Pushes a task. Call only by owner in unshared queues.
835 *
836 * @param task the task. Caller must ensure non-null.
837 * @throws RejectedExecutionException if array cannot be resized
838 */
839 final void push(ForkJoinTask<?> task) {
840 ForkJoinTask<?>[] a;
841 int s = top, d = s - base, cap, m;
842 ForkJoinPool p = pool;
843 if ((a = array) != null && (cap = a.length) > 0) {
844 QA.setRelease(a, (m = cap - 1) & s, task);
845 top = s + 1;
846 if (d == m)
847 growArray(false);
848 else if (QA.getAcquire(a, m & (s - 1)) == null && p != null) {
849 VarHandle.fullFence(); // was empty
850 p.signalWork(null);
851 }
852 }
853 }
854
855 /**
856 * Version of push for shared queues. Call only with phase lock held.
857 * @return true if should signal work
858 */
859 final boolean lockedPush(ForkJoinTask<?> task) {
860 ForkJoinTask<?>[] a;
861 boolean signal = false;
862 int s = top, d = s - base, cap, m;
863 if ((a = array) != null && (cap = a.length) > 0) {
864 a[(m = (cap - 1)) & s] = task;
865 top = s + 1;
866 if (d == m)
867 growArray(true);
868 else {
869 phase = 0; // full volatile unlock
870 if (((s - base) & ~1) == 0) // size 0 or 1
871 signal = true;
872 }
873 }
874 return signal;
875 }
876
877 /**
878 * Doubles the capacity of array. Call either by owner or with
879 * lock held -- it is OK for base, but not top, to move while
880 * resizings are in progress.
881 */
882 final void growArray(boolean locked) {
883 ForkJoinTask<?>[] newA = null;
884 try {
885 ForkJoinTask<?>[] oldA; int oldSize, newSize;
886 if ((oldA = array) != null && (oldSize = oldA.length) > 0 &&
887 (newSize = oldSize << 1) <= MAXIMUM_QUEUE_CAPACITY &&
888 newSize > 0) {
889 try {
890 newA = new ForkJoinTask<?>[newSize];
891 } catch (OutOfMemoryError ex) {
892 }
893 if (newA != null) { // poll from old array, push to new
894 int oldMask = oldSize - 1, newMask = newSize - 1;
895 for (int s = top - 1, k = oldMask; k >= 0; --k) {
896 ForkJoinTask<?> x = (ForkJoinTask<?>)
897 QA.getAndSet(oldA, s & oldMask, null);
898 if (x != null)
899 newA[s-- & newMask] = x;
900 else
901 break;
902 }
903 array = newA;
904 VarHandle.releaseFence();
905 }
906 }
907 } finally {
908 if (locked)
909 phase = 0;
910 }
911 if (newA == null)
912 throw new RejectedExecutionException("Queue capacity exceeded");
913 }
914
915 /**
916 * Takes next task, if one exists, in FIFO order.
917 */
918 final ForkJoinTask<?> poll() {
919 int b, k, cap; ForkJoinTask<?>[] a;
920 while ((a = array) != null && (cap = a.length) > 0 &&
921 top - (b = base) > 0) {
922 ForkJoinTask<?> t = (ForkJoinTask<?>)
923 QA.getAcquire(a, k = (cap - 1) & b);
924 if (base == b++) {
925 if (t == null)
926 Thread.yield(); // await index advance
927 else if (QA.compareAndSet(a, k, t, null)) {
928 BASE.setOpaque(this, b);
929 return t;
930 }
931 }
932 }
933 return null;
934 }
935
936 /**
937 * Takes next task, if one exists, in order specified by mode.
938 */
939 final ForkJoinTask<?> nextLocalTask() {
940 ForkJoinTask<?> t = null;
941 int md = id, b, s, d, cap; ForkJoinTask<?>[] a;
942 if ((a = array) != null && (cap = a.length) > 0 &&
943 (d = (s = top) - (b = base)) > 0) {
944 if ((md & FIFO) == 0 || d == 1) {
945 if ((t = (ForkJoinTask<?>)
946 QA.getAndSet(a, (cap - 1) & --s, null)) != null)
947 TOP.setOpaque(this, s);
948 }
949 else if ((t = (ForkJoinTask<?>)
950 QA.getAndSet(a, (cap - 1) & b++, null)) != null) {
951 BASE.setOpaque(this, b);
952 }
953 else // on contention in FIFO mode, use regular poll
954 t = poll();
955 }
956 return t;
957 }
958
959 /**
960 * Returns next task, if one exists, in order specified by mode.
961 */
962 final ForkJoinTask<?> peek() {
963 int cap; ForkJoinTask<?>[] a;
964 return ((a = array) != null && (cap = a.length) > 0) ?
965 a[(cap - 1) & ((id & FIFO) != 0 ? base : top - 1)] : null;
966 }
967
968 /**
969 * Pops the given task only if it is at the current top.
970 */
971 final boolean tryUnpush(ForkJoinTask<?> task) {
972 boolean popped = false;
973 int s, cap; ForkJoinTask<?>[] a;
974 if ((a = array) != null && (cap = a.length) > 0 &&
975 (s = top) != base &&
976 (popped = QA.compareAndSet(a, (cap - 1) & --s, task, null)))
977 TOP.setOpaque(this, s);
978 return popped;
979 }
980
981 /**
982 * Shared version of tryUnpush.
983 */
984 final boolean tryLockedUnpush(ForkJoinTask<?> task) {
985 boolean popped = false;
986 int s = top - 1, k, cap; ForkJoinTask<?>[] a;
987 if ((a = array) != null && (cap = a.length) > 0 &&
988 a[k = (cap - 1) & s] == task && tryLockPhase()) {
989 if (top == s + 1 && array == a &&
990 (popped = QA.compareAndSet(a, k, task, null)))
991 top = s;
992 releasePhaseLock();
993 }
994 return popped;
995 }
996
997 /**
998 * Removes and cancels all known tasks, ignoring any exceptions.
999 */
1000 final void cancelAll() {
1001 for (ForkJoinTask<?> t; (t = poll()) != null; )
1002 ForkJoinTask.cancelIgnoringExceptions(t);
1003 }
1004
1005 // Specialized execution methods
1006
1007 /**
1008 * Runs the given (stolen) task if nonnull, as well as
1009 * remaining local tasks and others available from the given
1010 * queue, up to bound n (to avoid infinite unfairness).
1011 */
1012 final void topLevelExec(ForkJoinTask<?> t, WorkQueue q, int n) {
1013 int nstolen = 1;
1014 for (int j = 0;;) {
1015 if (t != null)
1016 t.doExec();
1017 if (j++ <= n)
1018 t = nextLocalTask();
1019 else {
1020 j = 0;
1021 t = null;
1022 }
1023 if (t == null) {
1024 if (q != null && (t = q.poll()) != null) {
1025 ++nstolen;
1026 j = 0;
1027 }
1028 else if (j != 0)
1029 break;
1030 }
1031 }
1032 ForkJoinWorkerThread thread = owner;
1033 nsteals += nstolen;
1034 source = 0;
1035 if (thread != null)
1036 thread.afterTopLevelExec();
1037 }
1038
1039 /**
1040 * If present, removes task from queue and executes it.
1041 */
1042 final void tryRemoveAndExec(ForkJoinTask<?> task) {
1043 ForkJoinTask<?>[] a; int s, cap;
1044 if ((a = array) != null && (cap = a.length) > 0 &&
1045 (s = top) - base > 0) { // traverse from top
1046 for (int m = cap - 1, ns = s - 1, i = ns; ; --i) {
1047 int index = i & m;
1048 ForkJoinTask<?> t = (ForkJoinTask<?>)QA.get(a, index);
1049 if (t == null)
1050 break;
1051 else if (t == task) {
1052 if (QA.compareAndSet(a, index, t, null)) {
1053 top = ns; // safely shift down
1054 for (int j = i; j != ns; ++j) {
1055 ForkJoinTask<?> f;
1056 int pindex = (j + 1) & m;
1057 f = (ForkJoinTask<?>)QA.get(a, pindex);
1058 QA.setVolatile(a, pindex, null);
1059 int jindex = j & m;
1060 QA.setRelease(a, jindex, f);
1061 }
1062 VarHandle.releaseFence();
1063 t.doExec();
1064 }
1065 break;
1066 }
1067 }
1068 }
1069 }
1070
1071 /**
1072 * Tries to pop and run tasks within the target's computation
1073 * until done, not found, or limit exceeded.
1074 *
1075 * @param task root of CountedCompleter computation
1076 * @param limit max runs, or zero for no limit
1077 * @param shared true if must lock to extract task
1078 * @return task status on exit
1079 */
1080 final int helpCC(CountedCompleter<?> task, int limit, boolean shared) {
1081 int status = 0;
1082 if (task != null && (status = task.status) >= 0) {
1083 int s, k, cap; ForkJoinTask<?>[] a;
1084 while ((a = array) != null && (cap = a.length) > 0 &&
1085 (s = top) - base > 0) {
1086 CountedCompleter<?> v = null;
1087 ForkJoinTask<?> o = a[k = (cap - 1) & (s - 1)];
1088 if (o instanceof CountedCompleter) {
1089 CountedCompleter<?> t = (CountedCompleter<?>)o;
1090 for (CountedCompleter<?> f = t;;) {
1091 if (f != task) {
1092 if ((f = f.completer) == null)
1093 break;
1094 }
1095 else if (shared) {
1096 if (tryLockPhase()) {
1097 if (top == s && array == a &&
1098 QA.compareAndSet(a, k, t, null)) {
1099 top = s - 1;
1100 v = t;
1101 }
1102 releasePhaseLock();
1103 }
1104 break;
1105 }
1106 else {
1107 if (QA.compareAndSet(a, k, t, null)) {
1108 top = s - 1;
1109 v = t;
1110 }
1111 break;
1112 }
1113 }
1114 }
1115 if (v != null)
1116 v.doExec();
1117 if ((status = task.status) < 0 || v == null ||
1118 (limit != 0 && --limit == 0))
1119 break;
1120 }
1121 }
1122 return status;
1123 }
1124
1125 /**
1126 * Tries to poll and run AsynchronousCompletionTasks until
1127 * none found or blocker is released
1128 *
1129 * @param blocker the blocker
1130 */
1131 final void helpAsyncBlocker(ManagedBlocker blocker) {
1132 if (blocker != null) {
1133 int b, k, cap; ForkJoinTask<?>[] a; ForkJoinTask<?> t;
1134 while ((a = array) != null && (cap = a.length) > 0 &&
1135 top - (b = base) > 0) {
1136 t = (ForkJoinTask<?>)QA.getAcquire(a, k = (cap - 1) & b);
1137 if (blocker.isReleasable())
1138 break;
1139 else if (base == b++ && t != null) {
1140 if (!(t instanceof CompletableFuture.
1141 AsynchronousCompletionTask))
1142 break;
1143 else if (QA.compareAndSet(a, k, t, null)) {
1144 BASE.setOpaque(this, b);
1145 t.doExec();
1146 }
1147 }
1148 }
1149 }
1150 }
1151
1152 /**
1153 * Returns true if owned and not known to be blocked.
1154 */
1155 final boolean isApparentlyUnblocked() {
1156 Thread wt; Thread.State s;
1157 return ((wt = owner) != null &&
1158 (s = wt.getState()) != Thread.State.BLOCKED &&
1159 s != Thread.State.WAITING &&
1160 s != Thread.State.TIMED_WAITING);
1161 }
1162
1163 // VarHandle mechanics.
1164 static final VarHandle PHASE;
1165 static final VarHandle BASE;
1166 static final VarHandle TOP;
1167 static {
1168 try {
1169 MethodHandles.Lookup l = MethodHandles.lookup();
1170 PHASE = l.findVarHandle(WorkQueue.class, "phase", int.class);
1171 BASE = l.findVarHandle(WorkQueue.class, "base", int.class);
1172 TOP = l.findVarHandle(WorkQueue.class, "top", int.class);
1173 } catch (ReflectiveOperationException e) {
1174 throw new ExceptionInInitializerError(e);
1175 }
1176 }
1177 }
1178
1179 // static fields (initialized in static initializer below)
1180
1181 /**
1182 * Creates a new ForkJoinWorkerThread. This factory is used unless
1183 * overridden in ForkJoinPool constructors.
1184 */
1185 public static final ForkJoinWorkerThreadFactory
1186 defaultForkJoinWorkerThreadFactory;
1187
1188 /**
1189 * Permission required for callers of methods that may start or
1190 * kill threads.
1191 */
1192 static final RuntimePermission modifyThreadPermission;
1193
1194 /**
1195 * Common (static) pool. Non-null for public use unless a static
1196 * construction exception, but internal usages null-check on use
1197 * to paranoically avoid potential initialization circularities
1198 * as well as to simplify generated code.
1199 */
1200 static final ForkJoinPool common;
1201
1202 /**
1203 * Common pool parallelism. To allow simpler use and management
1204 * when common pool threads are disabled, we allow the underlying
1205 * common.parallelism field to be zero, but in that case still report
1206 * parallelism as 1 to reflect resulting caller-runs mechanics.
1207 */
1208 static final int COMMON_PARALLELISM;
1209
1210 /**
1211 * Limit on spare thread construction in tryCompensate.
1212 */
1213 private static final int COMMON_MAX_SPARES;
1214
1215 /**
1216 * Sequence number for creating workerNamePrefix.
1217 */
1218 private static int poolNumberSequence;
1219
1220 /**
1221 * Returns the next sequence number. We don't expect this to
1222 * ever contend, so use simple builtin sync.
1223 */
1224 private static final synchronized int nextPoolId() {
1225 return ++poolNumberSequence;
1226 }
1227
1228 // static configuration constants
1229
1230 /**
1231 * Default idle timeout value (in milliseconds) for the thread
1232 * triggering quiescence to park waiting for new work
1233 */
1234 private static final long DEFAULT_KEEPALIVE = 60_000L;
1235
1236 /**
1237 * Undershoot tolerance for idle timeouts
1238 */
1239 private static final long TIMEOUT_SLOP = 20L;
1240
1241 /**
1242 * The default value for COMMON_MAX_SPARES. Overridable using the
1243 * "java.util.concurrent.ForkJoinPool.common.maximumSpares" system
1244 * property. The default value is far in excess of normal
1245 * requirements, but also far short of MAX_CAP and typical OS
1246 * thread limits, so allows JVMs to catch misuse/abuse before
1247 * running out of resources needed to do so.
1248 */
1249 private static final int DEFAULT_COMMON_MAX_SPARES = 256;
1250
1251 /**
1252 * Increment for seed generators. See class ThreadLocal for
1253 * explanation.
1254 */
1255 private static final int SEED_INCREMENT = 0x9e3779b9;
1256
1257 /*
1258 * Bits and masks for field ctl, packed with 4 16 bit subfields:
1259 * RC: Number of released (unqueued) workers minus target parallelism
1260 * TC: Number of total workers minus target parallelism
1261 * SS: version count and status of top waiting thread
1262 * ID: poolIndex of top of Treiber stack of waiters
1263 *
1264 * When convenient, we can extract the lower 32 stack top bits
1265 * (including version bits) as sp=(int)ctl. The offsets of counts
1266 * by the target parallelism and the positionings of fields makes
1267 * it possible to perform the most common checks via sign tests of
1268 * fields: When ac is negative, there are not enough unqueued
1269 * workers, when tc is negative, there are not enough total
1270 * workers. When sp is non-zero, there are waiting workers. To
1271 * deal with possibly negative fields, we use casts in and out of
1272 * "short" and/or signed shifts to maintain signedness.
1273 *
1274 * Because it occupies uppermost bits, we can add one release count
1275 * using getAndAddLong of RC_UNIT, rather than CAS, when returning
1276 * from a blocked join. Other updates entail multiple subfields
1277 * and masking, requiring CAS.
1278 *
1279 * The limits packed in field "bounds" are also offset by the
1280 * parallelism level to make them comparable to the ctl rc and tc
1281 * fields.
1282 */
1283
1284 // Lower and upper word masks
1285 private static final long SP_MASK = 0xffffffffL;
1286 private static final long UC_MASK = ~SP_MASK;
1287
1288 // Release counts
1289 private static final int RC_SHIFT = 48;
1290 private static final long RC_UNIT = 0x0001L << RC_SHIFT;
1291 private static final long RC_MASK = 0xffffL << RC_SHIFT;
1292
1293 // Total counts
1294 private static final int TC_SHIFT = 32;
1295 private static final long TC_UNIT = 0x0001L << TC_SHIFT;
1296 private static final long TC_MASK = 0xffffL << TC_SHIFT;
1297 private static final long ADD_WORKER = 0x0001L << (TC_SHIFT + 15); // sign
1298
1299 // Instance fields
1300
1301 volatile long stealCount; // collects worker nsteals
1302 final long keepAlive; // milliseconds before dropping if idle
1303 int indexSeed; // next worker index
1304 final int bounds; // min, max threads packed as shorts
1305 volatile int mode; // parallelism, runstate, queue mode
1306 WorkQueue[] workQueues; // main registry
1307 final String workerNamePrefix; // for worker thread string; sync lock
1308 final ForkJoinWorkerThreadFactory factory;
1309 final UncaughtExceptionHandler ueh; // per-worker UEH
1310 final Predicate<? super ForkJoinPool> saturate;
1311
1312 @jdk.internal.vm.annotation.Contended("fjpctl") // segregate
1313 volatile long ctl; // main pool control
1314
1315 // Creating, registering and deregistering workers
1316
1317 /**
1318 * Tries to construct and start one worker. Assumes that total
1319 * count has already been incremented as a reservation. Invokes
1320 * deregisterWorker on any failure.
1321 *
1322 * @return true if successful
1323 */
1324 private boolean createWorker() {
1325 ForkJoinWorkerThreadFactory fac = factory;
1326 Throwable ex = null;
1327 ForkJoinWorkerThread wt = null;
1328 try {
1329 if (fac != null && (wt = fac.newThread(this)) != null) {
1330 wt.start();
1331 return true;
1332 }
1333 } catch (Throwable rex) {
1334 ex = rex;
1335 }
1336 deregisterWorker(wt, ex);
1337 return false;
1338 }
1339
1340 /**
1341 * Tries to add one worker, incrementing ctl counts before doing
1342 * so, relying on createWorker to back out on failure.
1343 *
1344 * @param c incoming ctl value, with total count negative and no
1345 * idle workers. On CAS failure, c is refreshed and retried if
1346 * this holds (otherwise, a new worker is not needed).
1347 */
1348 private void tryAddWorker(long c) {
1349 do {
1350 long nc = ((RC_MASK & (c + RC_UNIT)) |
1351 (TC_MASK & (c + TC_UNIT)));
1352 if (ctl == c && CTL.compareAndSet(this, c, nc)) {
1353 createWorker();
1354 break;
1355 }
1356 } while (((c = ctl) & ADD_WORKER) != 0L && (int)c == 0);
1357 }
1358
1359 /**
1360 * Callback from ForkJoinWorkerThread constructor to establish and
1361 * record its WorkQueue.
1362 *
1363 * @param wt the worker thread
1364 * @return the worker's queue
1365 */
1366 final WorkQueue registerWorker(ForkJoinWorkerThread wt) {
1367 UncaughtExceptionHandler handler;
1368 wt.setDaemon(true); // configure thread
1369 if ((handler = ueh) != null)
1370 wt.setUncaughtExceptionHandler(handler);
1371 int tid = 0; // for thread name
1372 int idbits = mode & FIFO;
1373 String prefix = workerNamePrefix;
1374 WorkQueue w = new WorkQueue(this, wt);
1375 if (prefix != null) {
1376 synchronized (prefix) {
1377 WorkQueue[] ws = workQueues; int n;
1378 int s = indexSeed += SEED_INCREMENT;
1379 idbits |= (s & ~(SMASK | FIFO | DORMANT));
1380 if (ws != null && (n = ws.length) > 1) {
1381 int m = n - 1;
1382 tid = m & ((s << 1) | 1); // odd-numbered indices
1383 for (int probes = n >>> 1;;) { // find empty slot
1384 WorkQueue q;
1385 if ((q = ws[tid]) == null || q.phase == QUIET)
1386 break;
1387 else if (--probes == 0) {
1388 tid = n | 1; // resize below
1389 break;
1390 }
1391 else
1392 tid = (tid + 2) & m;
1393 }
1394 w.phase = w.id = tid | idbits; // now publishable
1395
1396 if (tid < n)
1397 ws[tid] = w;
1398 else { // expand array
1399 int an = n << 1;
1400 WorkQueue[] as = new WorkQueue[an];
1401 as[tid] = w;
1402 int am = an - 1;
1403 for (int j = 0; j < n; ++j) {
1404 WorkQueue v; // copy external queue
1405 if ((v = ws[j]) != null) // position may change
1406 as[v.id & am & SQMASK] = v;
1407 if (++j >= n)
1408 break;
1409 as[j] = ws[j]; // copy worker
1410 }
1411 workQueues = as;
1412 }
1413 }
1414 }
1415 wt.setName(prefix.concat(Integer.toString(tid)));
1416 }
1417 return w;
1418 }
1419
1420 /**
1421 * Final callback from terminating worker, as well as upon failure
1422 * to construct or start a worker. Removes record of worker from
1423 * array, and adjusts counts. If pool is shutting down, tries to
1424 * complete termination.
1425 *
1426 * @param wt the worker thread, or null if construction failed
1427 * @param ex the exception causing failure, or null if none
1428 */
1429 final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
1430 WorkQueue w = null;
1431 int phase = 0;
1432 if (wt != null && (w = wt.workQueue) != null) {
1433 Object lock = workerNamePrefix;
1434 int wid = w.id;
1435 long ns = (long)w.nsteals & 0xffffffffL;
1436 if (lock != null) {
1437 synchronized (lock) {
1438 WorkQueue[] ws; int n, i; // remove index from array
1439 if ((ws = workQueues) != null && (n = ws.length) > 0 &&
1440 ws[i = wid & (n - 1)] == w)
1441 ws[i] = null;
1442 stealCount += ns;
1443 }
1444 }
1445 phase = w.phase;
1446 }
1447 if (phase != QUIET) { // else pre-adjusted
1448 long c; // decrement counts
1449 do {} while (!CTL.weakCompareAndSet
1450 (this, c = ctl, ((RC_MASK & (c - RC_UNIT)) |
1451 (TC_MASK & (c - TC_UNIT)) |
1452 (SP_MASK & c))));
1453 }
1454 if (w != null)
1455 w.cancelAll(); // cancel remaining tasks
1456
1457 if (!tryTerminate(false, false) && // possibly replace worker
1458 w != null && w.array != null) // avoid repeated failures
1459 signalWork(null);
1460
1461 if (ex == null) // help clean on way out
1462 ForkJoinTask.helpExpungeStaleExceptions();
1463 else // rethrow
1464 ForkJoinTask.rethrow(ex);
1465 }
1466
1467 /**
1468 * Tries to create or release a worker if too few are running.
1469 * @param q if non-null recheck if empty on CAS failure
1470 */
1471 final void signalWork(WorkQueue q) {
1472 for (;;) {
1473 long c; int sp; WorkQueue[] ws; int i; WorkQueue v;
1474 if ((c = ctl) >= 0L) // enough workers
1475 break;
1476 else if ((sp = (int)c) == 0) { // no idle workers
1477 if ((c & ADD_WORKER) != 0L) // too few workers
1478 tryAddWorker(c);
1479 break;
1480 }
1481 else if ((ws = workQueues) == null)
1482 break; // unstarted/terminated
1483 else if (ws.length <= (i = sp & SMASK))
1484 break; // terminated
1485 else if ((v = ws[i]) == null)
1486 break; // terminating
1487 else {
1488 int np = sp & ~UNSIGNALLED;
1489 int vp = v.phase;
1490 long nc = (v.stackPred & SP_MASK) | (UC_MASK & (c + RC_UNIT));
1491 Thread vt = v.owner;
1492 if (sp == vp && CTL.compareAndSet(this, c, nc)) {
1493 v.phase = np;
1494 if (vt != null && v.source < 0)
1495 LockSupport.unpark(vt);
1496 break;
1497 }
1498 else if (q != null && q.isEmpty()) // no need to retry
1499 break;
1500 }
1501 }
1502 }
1503
1504 /**
1505 * Tries to decrement counts (sometimes implicitly) and possibly
1506 * arrange for a compensating worker in preparation for blocking:
1507 * If not all core workers yet exist, creates one, else if any are
1508 * unreleased (possibly including caller) releases one, else if
1509 * fewer than the minimum allowed number of workers running,
1510 * checks to see that they are all active, and if so creates an
1511 * extra worker unless over maximum limit and policy is to
1512 * saturate. Most of these steps can fail due to interference, in
1513 * which case 0 is returned so caller will retry. A negative
1514 * return value indicates that the caller doesn't need to
1515 * re-adjust counts when later unblocked.
1516 *
1517 * @return 1: block then adjust, -1: block without adjust, 0 : retry
1518 */
1519 private int tryCompensate(WorkQueue w) {
1520 int t, n, sp;
1521 long c = ctl;
1522 WorkQueue[] ws = workQueues;
1523 if ((t = (short)(c >>> TC_SHIFT)) >= 0) {
1524 if (ws == null || (n = ws.length) <= 0 || w == null)
1525 return 0; // disabled
1526 else if ((sp = (int)c) != 0) { // replace or release
1527 WorkQueue v = ws[sp & (n - 1)];
1528 int wp = w.phase;
1529 long uc = UC_MASK & ((wp < 0) ? c + RC_UNIT : c);
1530 int np = sp & ~UNSIGNALLED;
1531 if (v != null) {
1532 int vp = v.phase;
1533 Thread vt = v.owner;
1534 long nc = ((long)v.stackPred & SP_MASK) | uc;
1535 if (vp == sp && CTL.compareAndSet(this, c, nc)) {
1536 v.phase = np;
1537 if (vt != null && v.source < 0)
1538 LockSupport.unpark(vt);
1539 return (wp < 0) ? -1 : 1;
1540 }
1541 }
1542 return 0;
1543 }
1544 else if ((int)(c >> RC_SHIFT) - // reduce parallelism
1545 (short)(bounds & SMASK) > 0) {
1546 long nc = ((RC_MASK & (c - RC_UNIT)) | (~RC_MASK & c));
1547 return CTL.compareAndSet(this, c, nc) ? 1 : 0;
1548 }
1549 else { // validate
1550 int md = mode, pc = md & SMASK, tc = pc + t, bc = 0;
1551 boolean unstable = false;
1552 for (int i = 1; i < n; i += 2) {
1553 WorkQueue q; Thread wt; Thread.State ts;
1554 if ((q = ws[i]) != null) {
1555 if (q.source == 0) {
1556 unstable = true;
1557 break;
1558 }
1559 else {
1560 --tc;
1561 if ((wt = q.owner) != null &&
1562 ((ts = wt.getState()) == Thread.State.BLOCKED ||
1563 ts == Thread.State.WAITING))
1564 ++bc; // worker is blocking
1565 }
1566 }
1567 }
1568 if (unstable || tc != 0 || ctl != c)
1569 return 0; // inconsistent
1570 else if (t + pc >= MAX_CAP || t >= (bounds >>> SWIDTH)) {
1571 Predicate<? super ForkJoinPool> sat;
1572 if ((sat = saturate) != null && sat.test(this))
1573 return -1;
1574 else if (bc < pc) { // lagging
1575 Thread.yield(); // for retry spins
1576 return 0;
1577 }
1578 else
1579 throw new RejectedExecutionException(
1580 "Thread limit exceeded replacing blocked worker");
1581 }
1582 }
1583 }
1584
1585 long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK); // expand pool
1586 return CTL.compareAndSet(this, c, nc) && createWorker() ? 1 : 0;
1587 }
1588
1589 /**
1590 * Top-level runloop for workers, called by ForkJoinWorkerThread.run.
1591 * See above for explanation.
1592 */
1593 final void runWorker(WorkQueue w) {
1594 int r = (w.id ^ ThreadLocalRandom.nextSecondarySeed()) | FIFO; // rng
1595 w.array = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY]; // initialize
1596 for (;;) {
1597 int phase;
1598 if (scan(w, r)) { // scan until apparently empty
1599 r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // move (xorshift)
1600 }
1601 else if ((phase = w.phase) >= 0) { // enqueue, then rescan
1602 long np = (w.phase = (phase + SS_SEQ) | UNSIGNALLED) & SP_MASK;
1603 long c, nc;
1604 do {
1605 w.stackPred = (int)(c = ctl);
1606 nc = ((c - RC_UNIT) & UC_MASK) | np;
1607 } while (!CTL.weakCompareAndSet(this, c, nc));
1608 }
1609 else { // already queued
1610 int pred = w.stackPred;
1611 Thread.interrupted(); // clear before park
1612 w.source = DORMANT; // enable signal
1613 long c = ctl;
1614 int md = mode, rc = (md & SMASK) + (int)(c >> RC_SHIFT);
1615 if (md < 0) // terminating
1616 break;
1617 else if (rc <= 0 && (md & SHUTDOWN) != 0 &&
1618 tryTerminate(false, false))
1619 break; // quiescent shutdown
1620 else if (w.phase < 0) {
1621 if (rc <= 0 && pred != 0 && phase == (int)c) {
1622 long nc = (UC_MASK & (c - TC_UNIT)) | (SP_MASK & pred);
1623 long d = keepAlive + System.currentTimeMillis();
1624 LockSupport.parkUntil(this, d);
1625 if (ctl == c && // drop on timeout if all idle
1626 d - System.currentTimeMillis() <= TIMEOUT_SLOP &&
1627 CTL.compareAndSet(this, c, nc)) {
1628 w.phase = QUIET;
1629 break;
1630 }
1631 }
1632 else {
1633 LockSupport.park(this);
1634 if (w.phase < 0) // one spurious wakeup check
1635 LockSupport.park(this);
1636 }
1637 }
1638 w.source = 0; // disable signal
1639 }
1640 }
1641 }
1642
1643 /**
1644 * Scans for and if found executes one or more top-level tasks from a queue.
1645 *
1646 * @return true if found an apparently non-empty queue, and
1647 * possibly ran task(s).
1648 */
1649 private boolean scan(WorkQueue w, int r) {
1650 WorkQueue[] ws; int n;
1651 if ((ws = workQueues) != null && (n = ws.length) > 0 && w != null) {
1652 for (int m = n - 1, j = r & m;;) {
1653 WorkQueue q; int b;
1654 if ((q = ws[j]) != null && q.top != (b = q.base)) {
1655 int qid = q.id;
1656 ForkJoinTask<?>[] a; int cap, k; ForkJoinTask<?> t;
1657 if ((a = q.array) != null && (cap = a.length) > 0) {
1658 t = (ForkJoinTask<?>)QA.getAcquire(a, k = (cap - 1) & b);
1659 if (q.base == b++ && t != null &&
1660 QA.compareAndSet(a, k, t, null)) {
1661 q.base = b;
1662 w.source = qid;
1663 if (a[(cap - 1) & b] != null)
1664 signalWork(q); // help signal if more tasks
1665 w.topLevelExec(t, q, // random fairness bound
1666 (r | (1 << TOP_BOUND_SHIFT)) & SMASK);
1667 }
1668 }
1669 return true;
1670 }
1671 else if (--n > 0)
1672 j = (j + 1) & m;
1673 else
1674 break;
1675 }
1676 }
1677 return false;
1678 }
1679
1680 /**
1681 * Helps and/or blocks until the given task is done or timeout.
1682 * First tries locally helping, then scans other queues for a task
1683 * produced by one of w's stealers; compensating and blocking if
1684 * none are found (rescanning if tryCompensate fails).
1685 *
1686 * @param w caller
1687 * @param task the task
1688 * @param deadline for timed waits, if nonzero
1689 * @return task status on exit
1690 */
1691 final int awaitJoin(WorkQueue w, ForkJoinTask<?> task, long deadline) {
1692 int s = 0;
1693 int seed = ThreadLocalRandom.nextSecondarySeed();
1694 if (w != null && task != null &&
1695 (!(task instanceof CountedCompleter) ||
1696 (s = w.helpCC((CountedCompleter<?>)task, 0, false)) >= 0)) {
1697 w.tryRemoveAndExec(task);
1698 int src = w.source, id = w.id;
1699 int r = (seed >>> 16) | 1, step = (seed & ~1) | 2;
1700 s = task.status;
1701 while (s >= 0) {
1702 WorkQueue[] ws;
1703 int n = (ws = workQueues) == null ? 0 : ws.length, m = n - 1;
1704 while (n > 0) {
1705 WorkQueue q; int b;
1706 if ((q = ws[r & m]) != null && q.source == id &&
1707 q.top != (b = q.base)) {
1708 ForkJoinTask<?>[] a; int cap, k;
1709 int qid = q.id;
1710 if ((a = q.array) != null && (cap = a.length) > 0) {
1711 ForkJoinTask<?> t = (ForkJoinTask<?>)
1712 QA.getAcquire(a, k = (cap - 1) & b);
1713 if (q.source == id && q.base == b++ &&
1714 t != null && QA.compareAndSet(a, k, t, null)) {
1715 q.base = b;
1716 w.source = qid;
1717 t.doExec();
1718 w.source = src;
1719 }
1720 }
1721 break;
1722 }
1723 else {
1724 r += step;
1725 --n;
1726 }
1727 }
1728 if ((s = task.status) < 0)
1729 break;
1730 else if (n == 0) { // empty scan
1731 long ms, ns; int block;
1732 if (deadline == 0L)
1733 ms = 0L; // untimed
1734 else if ((ns = deadline - System.nanoTime()) <= 0L)
1735 break; // timeout
1736 else if ((ms = TimeUnit.NANOSECONDS.toMillis(ns)) <= 0L)
1737 ms = 1L; // avoid 0 for timed wait
1738 if ((block = tryCompensate(w)) != 0) {
1739 task.internalWait(ms);
1740 CTL.getAndAdd(this, (block > 0) ? RC_UNIT : 0L);
1741 }
1742 s = task.status;
1743 }
1744 }
1745 }
1746 return s;
1747 }
1748
1749 /**
1750 * Runs tasks until {@code isQuiescent()}. Rather than blocking
1751 * when tasks cannot be found, rescans until all others cannot
1752 * find tasks either.
1753 */
1754 final void helpQuiescePool(WorkQueue w) {
1755 int prevSrc = w.source;
1756 int seed = ThreadLocalRandom.nextSecondarySeed();
1757 int r = seed >>> 16, step = r | 1;
1758 for (int source = prevSrc, released = -1;;) { // -1 until known
1759 ForkJoinTask<?> localTask; WorkQueue[] ws;
1760 while ((localTask = w.nextLocalTask()) != null)
1761 localTask.doExec();
1762 if (w.phase >= 0 && released == -1)
1763 released = 1;
1764 boolean quiet = true, empty = true;
1765 int n = (ws = workQueues) == null ? 0 : ws.length;
1766 for (int m = n - 1; n > 0; r += step, --n) {
1767 WorkQueue q; int b;
1768 if ((q = ws[r & m]) != null) {
1769 int qs = q.source;
1770 if (q.top != (b = q.base)) {
1771 quiet = empty = false;
1772 ForkJoinTask<?>[] a; int cap, k;
1773 int qid = q.id;
1774 if ((a = q.array) != null && (cap = a.length) > 0) {
1775 if (released == 0) { // increment
1776 released = 1;
1777 CTL.getAndAdd(this, RC_UNIT);
1778 }
1779 ForkJoinTask<?> t = (ForkJoinTask<?>)
1780 QA.getAcquire(a, k = (cap - 1) & b);
1781 if (q.base == b++ && t != null &&
1782 QA.compareAndSet(a, k, t, null)) {
1783 q.base = b;
1784 w.source = qid;
1785 t.doExec();
1786 w.source = source = prevSrc;
1787 }
1788 }
1789 break;
1790 }
1791 else if ((qs & QUIET) == 0)
1792 quiet = false;
1793 }
1794 }
1795 if (quiet) {
1796 if (released == 0)
1797 CTL.getAndAdd(this, RC_UNIT);
1798 w.source = prevSrc;
1799 break;
1800 }
1801 else if (empty) {
1802 if (source != QUIET)
1803 w.source = source = QUIET;
1804 if (released == 1) { // decrement
1805 released = 0;
1806 CTL.getAndAdd(this, RC_MASK & -RC_UNIT);
1807 }
1808 }
1809 }
1810 }
1811
1812 /**
1813 * Scans for and returns a polled task, if available.
1814 * Used only for untracked polls.
1815 *
1816 * @param submissionsOnly if true, only scan submission queues
1817 */
1818 private ForkJoinTask<?> pollScan(boolean submissionsOnly) {
1819 WorkQueue[] ws; int n;
1820 rescan: while ((mode & STOP) == 0 && (ws = workQueues) != null &&
1821 (n = ws.length) > 0) {
1822 int m = n - 1;
1823 int r = ThreadLocalRandom.nextSecondarySeed();
1824 int h = r >>> 16;
1825 int origin, step;
1826 if (submissionsOnly) {
1827 origin = (r & ~1) & m; // even indices and steps
1828 step = (h & ~1) | 2;
1829 }
1830 else {
1831 origin = r & m;
1832 step = h | 1;
1833 }
1834 boolean nonempty = false;
1835 for (int i = origin, oldSum = 0, checkSum = 0;;) {
1836 WorkQueue q;
1837 if ((q = ws[i]) != null) {
1838 int b; ForkJoinTask<?> t;
1839 if (q.top - (b = q.base) > 0) {
1840 nonempty = true;
1841 if ((t = q.poll()) != null)
1842 return t;
1843 }
1844 else
1845 checkSum += b + q.id;
1846 }
1847 if ((i = (i + step) & m) == origin) {
1848 if (!nonempty && oldSum == (oldSum = checkSum))
1849 break rescan;
1850 checkSum = 0;
1851 nonempty = false;
1852 }
1853 }
1854 }
1855 return null;
1856 }
1857
1858 /**
1859 * Gets and removes a local or stolen task for the given worker.
1860 *
1861 * @return a task, if available
1862 */
1863 final ForkJoinTask<?> nextTaskFor(WorkQueue w) {
1864 ForkJoinTask<?> t;
1865 if (w == null || (t = w.nextLocalTask()) == null)
1866 t = pollScan(false);
1867 return t;
1868 }
1869
1870 // External operations
1871
1872 /**
1873 * Adds the given task to a submission queue at submitter's
1874 * current queue, creating one if null or contended.
1875 *
1876 * @param task the task. Caller must ensure non-null.
1877 */
1878 final void externalPush(ForkJoinTask<?> task) {
1879 int r; // initialize caller's probe
1880 if ((r = ThreadLocalRandom.getProbe()) == 0) {
1881 ThreadLocalRandom.localInit();
1882 r = ThreadLocalRandom.getProbe();
1883 }
1884 for (;;) {
1885 WorkQueue q;
1886 int md = mode, n;
1887 WorkQueue[] ws = workQueues;
1888 if ((md & SHUTDOWN) != 0 || ws == null || (n = ws.length) <= 0)
1889 throw new RejectedExecutionException();
1890 else if ((q = ws[(n - 1) & r & SQMASK]) == null) { // add queue
1891 int qid = (r | QUIET) & ~(FIFO | OWNED);
1892 Object lock = workerNamePrefix;
1893 ForkJoinTask<?>[] qa =
1894 new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY];
1895 q = new WorkQueue(this, null);
1896 q.array = qa;
1897 q.id = qid;
1898 q.source = QUIET;
1899 if (lock != null) { // unless disabled, lock pool to install
1900 synchronized (lock) {
1901 WorkQueue[] vs; int i, vn;
1902 if ((vs = workQueues) != null && (vn = vs.length) > 0 &&
1903 vs[i = qid & (vn - 1) & SQMASK] == null)
1904 vs[i] = q; // else another thread already installed
1905 }
1906 }
1907 }
1908 else if (!q.tryLockPhase()) // move if busy
1909 r = ThreadLocalRandom.advanceProbe(r);
1910 else {
1911 if (q.lockedPush(task))
1912 signalWork(null);
1913 return;
1914 }
1915 }
1916 }
1917
1918 /**
1919 * Pushes a possibly-external submission.
1920 */
1921 private <T> ForkJoinTask<T> externalSubmit(ForkJoinTask<T> task) {
1922 Thread t; ForkJoinWorkerThread w; WorkQueue q;
1923 if (task == null)
1924 throw new NullPointerException();
1925 if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread) &&
1926 (w = (ForkJoinWorkerThread)t).pool == this &&
1927 (q = w.workQueue) != null)
1928 q.push(task);
1929 else
1930 externalPush(task);
1931 return task;
1932 }
1933
1934 /**
1935 * Returns common pool queue for an external thread.
1936 */
1937 static WorkQueue commonSubmitterQueue() {
1938 ForkJoinPool p = common;
1939 int r = ThreadLocalRandom.getProbe();
1940 WorkQueue[] ws; int n;
1941 return (p != null && (ws = p.workQueues) != null &&
1942 (n = ws.length) > 0) ?
1943 ws[(n - 1) & r & SQMASK] : null;
1944 }
1945
1946 /**
1947 * Performs tryUnpush for an external submitter.
1948 */
1949 final boolean tryExternalUnpush(ForkJoinTask<?> task) {
1950 int r = ThreadLocalRandom.getProbe();
1951 WorkQueue[] ws; WorkQueue w; int n;
1952 return ((ws = workQueues) != null &&
1953 (n = ws.length) > 0 &&
1954 (w = ws[(n - 1) & r & SQMASK]) != null &&
1955 w.tryLockedUnpush(task));
1956 }
1957
1958 /**
1959 * Performs helpComplete for an external submitter.
1960 */
1961 final int externalHelpComplete(CountedCompleter<?> task, int maxTasks) {
1962 int r = ThreadLocalRandom.getProbe();
1963 WorkQueue[] ws; WorkQueue w; int n;
1964 return ((ws = workQueues) != null && (n = ws.length) > 0 &&
1965 (w = ws[(n - 1) & r & SQMASK]) != null) ?
1966 w.helpCC(task, maxTasks, true) : 0;
1967 }
1968
1969 /**
1970 * Tries to steal and run tasks within the target's computation.
1971 * The maxTasks argument supports external usages; internal calls
1972 * use zero, allowing unbounded steps (external calls trap
1973 * non-positive values).
1974 *
1975 * @param w caller
1976 * @param maxTasks if non-zero, the maximum number of other tasks to run
1977 * @return task status on exit
1978 */
1979 final int helpComplete(WorkQueue w, CountedCompleter<?> task,
1980 int maxTasks) {
1981 return (w == null) ? 0 : w.helpCC(task, maxTasks, false);
1982 }
1983
1984 /**
1985 * Returns a cheap heuristic guide for task partitioning when
1986 * programmers, frameworks, tools, or languages have little or no
1987 * idea about task granularity. In essence, by offering this
1988 * method, we ask users only about tradeoffs in overhead vs
1989 * expected throughput and its variance, rather than how finely to
1990 * partition tasks.
1991 *
1992 * In a steady state strict (tree-structured) computation, each
1993 * thread makes available for stealing enough tasks for other
1994 * threads to remain active. Inductively, if all threads play by
1995 * the same rules, each thread should make available only a
1996 * constant number of tasks.
1997 *
1998 * The minimum useful constant is just 1. But using a value of 1
1999 * would require immediate replenishment upon each steal to
2000 * maintain enough tasks, which is infeasible. Further,
2001 * partitionings/granularities of offered tasks should minimize
2002 * steal rates, which in general means that threads nearer the top
2003 * of computation tree should generate more than those nearer the
2004 * bottom. In perfect steady state, each thread is at
2005 * approximately the same level of computation tree. However,
2006 * producing extra tasks amortizes the uncertainty of progress and
2007 * diffusion assumptions.
2008 *
2009 * So, users will want to use values larger (but not much larger)
2010 * than 1 to both smooth over transient shortages and hedge
2011 * against uneven progress; as traded off against the cost of
2012 * extra task overhead. We leave the user to pick a threshold
2013 * value to compare with the results of this call to guide
2014 * decisions, but recommend values such as 3.
2015 *
2016 * When all threads are active, it is on average OK to estimate
2017 * surplus strictly locally. In steady-state, if one thread is
2018 * maintaining say 2 surplus tasks, then so are others. So we can
2019 * just use estimated queue length. However, this strategy alone
2020 * leads to serious mis-estimates in some non-steady-state
2021 * conditions (ramp-up, ramp-down, other stalls). We can detect
2022 * many of these by further considering the number of "idle"
2023 * threads, that are known to have zero queued tasks, so
2024 * compensate by a factor of (#idle/#active) threads.
2025 */
2026 static int getSurplusQueuedTaskCount() {
2027 Thread t; ForkJoinWorkerThread wt; ForkJoinPool pool; WorkQueue q;
2028 if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread) &&
2029 (pool = (wt = (ForkJoinWorkerThread)t).pool) != null &&
2030 (q = wt.workQueue) != null) {
2031 int p = pool.mode & SMASK;
2032 int a = p + (int)(pool.ctl >> RC_SHIFT);
2033 int n = q.top - q.base;
2034 return n - (a > (p >>>= 1) ? 0 :
2035 a > (p >>>= 1) ? 1 :
2036 a > (p >>>= 1) ? 2 :
2037 a > (p >>>= 1) ? 4 :
2038 8);
2039 }
2040 return 0;
2041 }
2042
2043 // Termination
2044
2045 /**
2046 * Possibly initiates and/or completes termination.
2047 *
2048 * @param now if true, unconditionally terminate, else only
2049 * if no work and no active workers
2050 * @param enable if true, terminate when next possible
2051 * @return true if terminating or terminated
2052 */
2053 private boolean tryTerminate(boolean now, boolean enable) {
2054 int md; // 3 phases: try to set SHUTDOWN, then STOP, then TERMINATED
2055
2056 while (((md = mode) & SHUTDOWN) == 0) {
2057 if (!enable || this == common) // cannot shutdown
2058 return false;
2059 else
2060 MODE.compareAndSet(this, md, md | SHUTDOWN);
2061 }
2062
2063 while (((md = mode) & STOP) == 0) { // try to initiate termination
2064 if (!now) { // check if quiescent & empty
2065 for (long oldSum = 0L;;) { // repeat until stable
2066 boolean running = false;
2067 long checkSum = ctl;
2068 WorkQueue[] ws = workQueues;
2069 if ((md & SMASK) + (int)(checkSum >> RC_SHIFT) > 0)
2070 running = true;
2071 else if (ws != null) {
2072 WorkQueue w;
2073 for (int i = 0; i < ws.length; ++i) {
2074 if ((w = ws[i]) != null) {
2075 int s = w.source, p = w.phase;
2076 int d = w.id, b = w.base;
2077 if (b != w.top ||
2078 ((d & 1) == 1 && (s >= 0 || p >= 0))) {
2079 running = true;
2080 break; // working, scanning, or have work
2081 }
2082 checkSum += (((long)s << 48) + ((long)p << 32) +
2083 ((long)b << 16) + (long)d);
2084 }
2085 }
2086 }
2087 if (((md = mode) & STOP) != 0)
2088 break; // already triggered
2089 else if (running)
2090 return false;
2091 else if (workQueues == ws && oldSum == (oldSum = checkSum))
2092 break;
2093 }
2094 }
2095 if ((md & STOP) == 0)
2096 MODE.compareAndSet(this, md, md | STOP);
2097 }
2098
2099 while (((md = mode) & TERMINATED) == 0) { // help terminate others
2100 for (long oldSum = 0L;;) { // repeat until stable
2101 WorkQueue[] ws; WorkQueue w;
2102 long checkSum = ctl;
2103 if ((ws = workQueues) != null) {
2104 for (int i = 0; i < ws.length; ++i) {
2105 if ((w = ws[i]) != null) {
2106 ForkJoinWorkerThread wt = w.owner;
2107 w.cancelAll(); // clear queues
2108 if (wt != null) {
2109 try { // unblock join or park
2110 wt.interrupt();
2111 } catch (Throwable ignore) {
2112 }
2113 }
2114 checkSum += ((long)w.phase << 32) + w.base;
2115 }
2116 }
2117 }
2118 if (((md = mode) & TERMINATED) != 0 ||
2119 (workQueues == ws && oldSum == (oldSum = checkSum)))
2120 break;
2121 }
2122 if ((md & TERMINATED) != 0)
2123 break;
2124 else if ((md & SMASK) + (short)(ctl >>> TC_SHIFT) > 0)
2125 break;
2126 else if (MODE.compareAndSet(this, md, md | TERMINATED)) {
2127 synchronized (this) {
2128 notifyAll(); // for awaitTermination
2129 }
2130 break;
2131 }
2132 }
2133 return true;
2134 }
2135
2136 // Exported methods
2137
2138 // Constructors
2139
2140 /**
2141 * Creates a {@code ForkJoinPool} with parallelism equal to {@link
2142 * java.lang.Runtime#availableProcessors}, using defaults for all
2143 * other parameters (see {@link #ForkJoinPool(int,
2144 * ForkJoinWorkerThreadFactory, UncaughtExceptionHandler, boolean,
2145 * int, int, int, Predicate, long, TimeUnit)}).
2146 *
2147 * @throws SecurityException if a security manager exists and
2148 * the caller is not permitted to modify threads
2149 * because it does not hold {@link
2150 * java.lang.RuntimePermission}{@code ("modifyThread")}
2151 */
2152 public ForkJoinPool() {
2153 this(Math.min(MAX_CAP, Runtime.getRuntime().availableProcessors()),
2154 defaultForkJoinWorkerThreadFactory, null, false,
2155 0, MAX_CAP, 1, null, DEFAULT_KEEPALIVE, TimeUnit.MILLISECONDS);
2156 }
2157
2158 /**
2159 * Creates a {@code ForkJoinPool} with the indicated parallelism
2160 * level, using defaults for all other parameters (see {@link
2161 * #ForkJoinPool(int, ForkJoinWorkerThreadFactory,
2162 * UncaughtExceptionHandler, boolean, int, int, int, Predicate,
2163 * long, TimeUnit)}).
2164 *
2165 * @param parallelism the parallelism level
2166 * @throws IllegalArgumentException if parallelism less than or
2167 * equal to zero, or greater than implementation limit
2168 * @throws SecurityException if a security manager exists and
2169 * the caller is not permitted to modify threads
2170 * because it does not hold {@link
2171 * java.lang.RuntimePermission}{@code ("modifyThread")}
2172 */
2173 public ForkJoinPool(int parallelism) {
2174 this(parallelism, defaultForkJoinWorkerThreadFactory, null, false,
2175 0, MAX_CAP, 1, null, DEFAULT_KEEPALIVE, TimeUnit.MILLISECONDS);
2176 }
2177
2178 /**
2179 * Creates a {@code ForkJoinPool} with the given parameters (using
2180 * defaults for others -- see {@link #ForkJoinPool(int,
2181 * ForkJoinWorkerThreadFactory, UncaughtExceptionHandler, boolean,
2182 * int, int, int, Predicate, long, TimeUnit)}).
2183 *
2184 * @param parallelism the parallelism level. For default value,
2185 * use {@link java.lang.Runtime#availableProcessors}.
2186 * @param factory the factory for creating new threads. For default value,
2187 * use {@link #defaultForkJoinWorkerThreadFactory}.
2188 * @param handler the handler for internal worker threads that
2189 * terminate due to unrecoverable errors encountered while executing
2190 * tasks. For default value, use {@code null}.
2191 * @param asyncMode if true,
2192 * establishes local first-in-first-out scheduling mode for forked
2193 * tasks that are never joined. This mode may be more appropriate
2194 * than default locally stack-based mode in applications in which
2195 * worker threads only process event-style asynchronous tasks.
2196 * For default value, use {@code false}.
2197 * @throws IllegalArgumentException if parallelism less than or
2198 * equal to zero, or greater than implementation limit
2199 * @throws NullPointerException if the factory is null
2200 * @throws SecurityException if a security manager exists and
2201 * the caller is not permitted to modify threads
2202 * because it does not hold {@link
2203 * java.lang.RuntimePermission}{@code ("modifyThread")}
2204 */
2205 public ForkJoinPool(int parallelism,
2206 ForkJoinWorkerThreadFactory factory,
2207 UncaughtExceptionHandler handler,
2208 boolean asyncMode) {
2209 this(parallelism, factory, handler, asyncMode,
2210 0, MAX_CAP, 1, null, DEFAULT_KEEPALIVE, TimeUnit.MILLISECONDS);
2211 }
2212
2213 /**
2214 * Creates a {@code ForkJoinPool} with the given parameters.
2215 *
2216 * @param parallelism the parallelism level. For default value,
2217 * use {@link java.lang.Runtime#availableProcessors}.
2218 *
2219 * @param factory the factory for creating new threads. For
2220 * default value, use {@link #defaultForkJoinWorkerThreadFactory}.
2221 *
2222 * @param handler the handler for internal worker threads that
2223 * terminate due to unrecoverable errors encountered while
2224 * executing tasks. For default value, use {@code null}.
2225 *
2226 * @param asyncMode if true, establishes local first-in-first-out
2227 * scheduling mode for forked tasks that are never joined. This
2228 * mode may be more appropriate than default locally stack-based
2229 * mode in applications in which worker threads only process
2230 * event-style asynchronous tasks. For default value, use {@code
2231 * false}.
2232 *
2233 * @param corePoolSize the number of threads to keep in the pool
2234 * (unless timed out after an elapsed keep-alive). Normally (and
2235 * by default) this is the same value as the parallelism level,
2236 * but may be set to a larger value to reduce dynamic overhead if
2237 * tasks regularly block. Using a smaller value (for example
2238 * {@code 0}) has the same effect as the default.
2239 *
2240 * @param maximumPoolSize the maximum number of threads allowed.
2241 * When the maximum is reached, attempts to replace blocked
2242 * threads fail. (However, because creation and termination of
2243 * different threads may overlap, and may be managed by the given
2244 * thread factory, this value may be transiently exceeded.) To
2245 * arrange the same value as is used by default for the common
2246 * pool, use {@code 256} plus the {@code parallelism} level. (By
2247 * default, the common pool allows a maximum of 256 spare
2248 * threads.) Using a value (for example {@code
2249 * Integer.MAX_VALUE}) larger than the implementation's total
2250 * thread limit has the same effect as using this limit (which is
2251 * the default).
2252 *
2253 * @param minimumRunnable the minimum allowed number of core
2254 * threads not blocked by a join or {@link ManagedBlocker}. To
2255 * ensure progress, when too few unblocked threads exist and
2256 * unexecuted tasks may exist, new threads are constructed, up to
2257 * the given maximumPoolSize. For the default value, use {@code
2258 * 1}, that ensures liveness. A larger value might improve
2259 * throughput in the presence of blocked activities, but might
2260 * not, due to increased overhead. A value of zero may be
2261 * acceptable when submitted tasks cannot have dependencies
2262 * requiring additional threads.
2263 *
2264 * @param saturate if non-null, a predicate invoked upon attempts
2265 * to create more than the maximum total allowed threads. By
2266 * default, when a thread is about to block on a join or {@link
2267 * ManagedBlocker}, but cannot be replaced because the
2268 * maximumPoolSize would be exceeded, a {@link
2269 * RejectedExecutionException} is thrown. But if this predicate
2270 * returns {@code true}, then no exception is thrown, so the pool
2271 * continues to operate with fewer than the target number of
2272 * runnable threads, which might not ensure progress.
2273 *
2274 * @param keepAliveTime the elapsed time since last use before
2275 * a thread is terminated (and then later replaced if needed).
2276 * For the default value, use {@code 60, TimeUnit.SECONDS}.
2277 *
2278 * @param unit the time unit for the {@code keepAliveTime} argument
2279 *
2280 * @throws IllegalArgumentException if parallelism is less than or
2281 * equal to zero, or is greater than implementation limit,
2282 * or if maximumPoolSize is less than parallelism,
2283 * of if the keepAliveTime is less than or equal to zero.
2284 * @throws NullPointerException if the factory is null
2285 * @throws SecurityException if a security manager exists and
2286 * the caller is not permitted to modify threads
2287 * because it does not hold {@link
2288 * java.lang.RuntimePermission}{@code ("modifyThread")}
2289 * @since 9
2290 */
2291 public ForkJoinPool(int parallelism,
2292 ForkJoinWorkerThreadFactory factory,
2293 UncaughtExceptionHandler handler,
2294 boolean asyncMode,
2295 int corePoolSize,
2296 int maximumPoolSize,
2297 int minimumRunnable,
2298 Predicate<? super ForkJoinPool> saturate,
2299 long keepAliveTime,
2300 TimeUnit unit) {
2301 // check, encode, pack parameters
2302 if (parallelism <= 0 || parallelism > MAX_CAP ||
2303 maximumPoolSize < parallelism || keepAliveTime <= 0L)
2304 throw new IllegalArgumentException();
2305 if (factory == null)
2306 throw new NullPointerException();
2307 long ms = Math.max(unit.toMillis(keepAliveTime), TIMEOUT_SLOP);
2308
2309 int corep = Math.min(Math.max(corePoolSize, parallelism), MAX_CAP);
2310 long c = ((((long)(-corep) << TC_SHIFT) & TC_MASK) |
2311 (((long)(-parallelism) << RC_SHIFT) & RC_MASK));
2312 int m = parallelism | (asyncMode ? FIFO : 0);
2313 int maxSpares = Math.min(maximumPoolSize, MAX_CAP) - parallelism;
2314 int minAvail = Math.min(Math.max(minimumRunnable, 0), MAX_CAP);
2315 int b = ((minAvail - parallelism) & SMASK) | (maxSpares << SWIDTH);
2316 int n = (parallelism > 1) ? parallelism - 1 : 1; // at least 2 slots
2317 n |= n >>> 1; n |= n >>> 2; n |= n >>> 4; n |= n >>> 8; n |= n >>> 16;
2318 n = (n + 1) << 1; // power of two, including space for submission queues
2319
2320 this.workerNamePrefix = "ForkJoinPool-" + nextPoolId() + "-worker-";
2321 this.workQueues = new WorkQueue[n];
2322 this.factory = factory;
2323 this.ueh = handler;
2324 this.saturate = saturate;
2325 this.keepAlive = ms;
2326 this.bounds = b;
2327 this.mode = m;
2328 this.ctl = c;
2329 checkPermission();
2330 }
2331
2332 private static Object newInstanceFromSystemProperty(String property)
2333 throws ReflectiveOperationException {
2334 String className = System.getProperty(property);
2335 return (className == null)
2336 ? null
2337 : ClassLoader.getSystemClassLoader().loadClass(className)
2338 .getConstructor().newInstance();
2339 }
2340
2341 /**
2342 * Constructor for common pool using parameters possibly
2343 * overridden by system properties
2344 */
2345 private ForkJoinPool(byte forCommonPoolOnly) {
2346 int parallelism = -1;
2347 ForkJoinWorkerThreadFactory fac = null;
2348 UncaughtExceptionHandler handler = null;
2349 try { // ignore exceptions in accessing/parsing properties
2350 String pp = System.getProperty
2351 ("java.util.concurrent.ForkJoinPool.common.parallelism");
2352 if (pp != null)
2353 parallelism = Integer.parseInt(pp);
2354 fac = (ForkJoinWorkerThreadFactory) newInstanceFromSystemProperty(
2355 "java.util.concurrent.ForkJoinPool.common.threadFactory");
2356 handler = (UncaughtExceptionHandler) newInstanceFromSystemProperty(
2357 "java.util.concurrent.ForkJoinPool.common.exceptionHandler");
2358 } catch (Exception ignore) {
2359 }
2360
2361 if (fac == null) {
2362 if (System.getSecurityManager() == null)
2363 fac = defaultForkJoinWorkerThreadFactory;
2364 else // use security-managed default
2365 fac = new InnocuousForkJoinWorkerThreadFactory();
2366 }
2367 if (parallelism < 0 && // default 1 less than #cores
2368 (parallelism = Runtime.getRuntime().availableProcessors() - 1) <= 0)
2369 parallelism = 1;
2370 if (parallelism > MAX_CAP)
2371 parallelism = MAX_CAP;
2372
2373 long c = ((((long)(-parallelism) << TC_SHIFT) & TC_MASK) |
2374 (((long)(-parallelism) << RC_SHIFT) & RC_MASK));
2375 int b = ((1 - parallelism) & SMASK) | (COMMON_MAX_SPARES << SWIDTH);
2376 int n = (parallelism > 1) ? parallelism - 1 : 1;
2377 n |= n >>> 1; n |= n >>> 2; n |= n >>> 4; n |= n >>> 8; n |= n >>> 16;
2378 n = (n + 1) << 1;
2379
2380 this.workerNamePrefix = "ForkJoinPool.commonPool-worker-";
2381 this.workQueues = new WorkQueue[n];
2382 this.factory = fac;
2383 this.ueh = handler;
2384 this.saturate = null;
2385 this.keepAlive = DEFAULT_KEEPALIVE;
2386 this.bounds = b;
2387 this.mode = parallelism;
2388 this.ctl = c;
2389 }
2390
2391 /**
2392 * Returns the common pool instance. This pool is statically
2393 * constructed; its run state is unaffected by attempts to {@link
2394 * #shutdown} or {@link #shutdownNow}. However this pool and any
2395 * ongoing processing are automatically terminated upon program
2396 * {@link System#exit}. Any program that relies on asynchronous
2397 * task processing to complete before program termination should
2398 * invoke {@code commonPool().}{@link #awaitQuiescence awaitQuiescence},
2399 * before exit.
2400 *
2401 * @return the common pool instance
2402 * @since 1.8
2403 */
2404 public static ForkJoinPool commonPool() {
2405 // assert common != null : "static init error";
2406 return common;
2407 }
2408
2409 // Execution methods
2410
2411 /**
2412 * Performs the given task, returning its result upon completion.
2413 * If the computation encounters an unchecked Exception or Error,
2414 * it is rethrown as the outcome of this invocation. Rethrown
2415 * exceptions behave in the same way as regular exceptions, but,
2416 * when possible, contain stack traces (as displayed for example
2417 * using {@code ex.printStackTrace()}) of both the current thread
2418 * as well as the thread actually encountering the exception;
2419 * minimally only the latter.
2420 *
2421 * @param task the task
2422 * @param <T> the type of the task's result
2423 * @return the task's result
2424 * @throws NullPointerException if the task is null
2425 * @throws RejectedExecutionException if the task cannot be
2426 * scheduled for execution
2427 */
2428 public <T> T invoke(ForkJoinTask<T> task) {
2429 if (task == null)
2430 throw new NullPointerException();
2431 externalSubmit(task);
2432 return task.join();
2433 }
2434
2435 /**
2436 * Arranges for (asynchronous) execution of the given task.
2437 *
2438 * @param task the task
2439 * @throws NullPointerException if the task is null
2440 * @throws RejectedExecutionException if the task cannot be
2441 * scheduled for execution
2442 */
2443 public void execute(ForkJoinTask<?> task) {
2444 externalSubmit(task);
2445 }
2446
2447 // AbstractExecutorService methods
2448
2449 /**
2450 * @throws NullPointerException if the task is null
2451 * @throws RejectedExecutionException if the task cannot be
2452 * scheduled for execution
2453 */
2454 public void execute(Runnable task) {
2455 if (task == null)
2456 throw new NullPointerException();
2457 ForkJoinTask<?> job;
2458 if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2459 job = (ForkJoinTask<?>) task;
2460 else
2461 job = new ForkJoinTask.RunnableExecuteAction(task);
2462 externalSubmit(job);
2463 }
2464
2465 /**
2466 * Submits a ForkJoinTask for execution.
2467 *
2468 * @param task the task to submit
2469 * @param <T> the type of the task's result
2470 * @return the task
2471 * @throws NullPointerException if the task is null
2472 * @throws RejectedExecutionException if the task cannot be
2473 * scheduled for execution
2474 */
2475 public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
2476 return externalSubmit(task);
2477 }
2478
2479 /**
2480 * @throws NullPointerException if the task is null
2481 * @throws RejectedExecutionException if the task cannot be
2482 * scheduled for execution
2483 */
2484 public <T> ForkJoinTask<T> submit(Callable<T> task) {
2485 return externalSubmit(new ForkJoinTask.AdaptedCallable<T>(task));
2486 }
2487
2488 /**
2489 * @throws NullPointerException if the task is null
2490 * @throws RejectedExecutionException if the task cannot be
2491 * scheduled for execution
2492 */
2493 public <T> ForkJoinTask<T> submit(Runnable task, T result) {
2494 return externalSubmit(new ForkJoinTask.AdaptedRunnable<T>(task, result));
2495 }
2496
2497 /**
2498 * @throws NullPointerException if the task is null
2499 * @throws RejectedExecutionException if the task cannot be
2500 * scheduled for execution
2501 */
2502 @SuppressWarnings("unchecked")
2503 public ForkJoinTask<?> submit(Runnable task) {
2504 if (task == null)
2505 throw new NullPointerException();
2506 return externalSubmit((task instanceof ForkJoinTask<?>)
2507 ? (ForkJoinTask<Void>) task // avoid re-wrap
2508 : new ForkJoinTask.AdaptedRunnableAction(task));
2509 }
2510
2511 /**
2512 * @throws NullPointerException {@inheritDoc}
2513 * @throws RejectedExecutionException {@inheritDoc}
2514 */
2515 public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
2516 // In previous versions of this class, this method constructed
2517 // a task to run ForkJoinTask.invokeAll, but now external
2518 // invocation of multiple tasks is at least as efficient.
2519 ArrayList<Future<T>> futures = new ArrayList<>(tasks.size());
2520
2521 try {
2522 for (Callable<T> t : tasks) {
2523 ForkJoinTask<T> f = new ForkJoinTask.AdaptedCallable<T>(t);
2524 futures.add(f);
2525 externalSubmit(f);
2526 }
2527 for (int i = 0, size = futures.size(); i < size; i++)
2528 ((ForkJoinTask<?>)futures.get(i)).quietlyJoin();
2529 return futures;
2530 } catch (Throwable t) {
2531 for (int i = 0, size = futures.size(); i < size; i++)
2532 futures.get(i).cancel(false);
2533 throw t;
2534 }
2535 }
2536
2537 /**
2538 * Returns the factory used for constructing new workers.
2539 *
2540 * @return the factory used for constructing new workers
2541 */
2542 public ForkJoinWorkerThreadFactory getFactory() {
2543 return factory;
2544 }
2545
2546 /**
2547 * Returns the handler for internal worker threads that terminate
2548 * due to unrecoverable errors encountered while executing tasks.
2549 *
2550 * @return the handler, or {@code null} if none
2551 */
2552 public UncaughtExceptionHandler getUncaughtExceptionHandler() {
2553 return ueh;
2554 }
2555
2556 /**
2557 * Returns the targeted parallelism level of this pool.
2558 *
2559 * @return the targeted parallelism level of this pool
2560 */
2561 public int getParallelism() {
2562 int par = mode & SMASK;
2563 return (par > 0) ? par : 1;
2564 }
2565
2566 /**
2567 * Returns the targeted parallelism level of the common pool.
2568 *
2569 * @return the targeted parallelism level of the common pool
2570 * @since 1.8
2571 */
2572 public static int getCommonPoolParallelism() {
2573 return COMMON_PARALLELISM;
2574 }
2575
2576 /**
2577 * Returns the number of worker threads that have started but not
2578 * yet terminated. The result returned by this method may differ
2579 * from {@link #getParallelism} when threads are created to
2580 * maintain parallelism when others are cooperatively blocked.
2581 *
2582 * @return the number of worker threads
2583 */
2584 public int getPoolSize() {
2585 return ((mode & SMASK) + (short)(ctl >>> TC_SHIFT));
2586 }
2587
2588 /**
2589 * Returns {@code true} if this pool uses local first-in-first-out
2590 * scheduling mode for forked tasks that are never joined.
2591 *
2592 * @return {@code true} if this pool uses async mode
2593 */
2594 public boolean getAsyncMode() {
2595 return (mode & FIFO) != 0;
2596 }
2597
2598 /**
2599 * Returns an estimate of the number of worker threads that are
2600 * not blocked waiting to join tasks or for other managed
2601 * synchronization. This method may overestimate the
2602 * number of running threads.
2603 *
2604 * @return the number of worker threads
2605 */
2606 public int getRunningThreadCount() {
2607 WorkQueue[] ws; WorkQueue w;
2608 VarHandle.acquireFence();
2609 int rc = 0;
2610 if ((ws = workQueues) != null) {
2611 for (int i = 1; i < ws.length; i += 2) {
2612 if ((w = ws[i]) != null && w.isApparentlyUnblocked())
2613 ++rc;
2614 }
2615 }
2616 return rc;
2617 }
2618
2619 /**
2620 * Returns an estimate of the number of threads that are currently
2621 * stealing or executing tasks. This method may overestimate the
2622 * number of active threads.
2623 *
2624 * @return the number of active threads
2625 */
2626 public int getActiveThreadCount() {
2627 int r = (mode & SMASK) + (int)(ctl >> RC_SHIFT);
2628 return (r <= 0) ? 0 : r; // suppress momentarily negative values
2629 }
2630
2631 /**
2632 * Returns {@code true} if all worker threads are currently idle.
2633 * An idle worker is one that cannot obtain a task to execute
2634 * because none are available to steal from other threads, and
2635 * there are no pending submissions to the pool. This method is
2636 * conservative; it might not return {@code true} immediately upon
2637 * idleness of all threads, but will eventually become true if
2638 * threads remain inactive.
2639 *
2640 * @return {@code true} if all threads are currently idle
2641 */
2642 public boolean isQuiescent() {
2643 for (;;) {
2644 long c = ctl;
2645 int md = mode, pc = md & SMASK;
2646 int tc = pc + (short)(c >>> TC_SHIFT);
2647 int rc = pc + (int)(c >> RC_SHIFT);
2648 if ((md & (STOP | TERMINATED)) != 0)
2649 return true;
2650 else if (rc > 0)
2651 return false;
2652 else {
2653 WorkQueue[] ws; WorkQueue v;
2654 if ((ws = workQueues) != null) {
2655 for (int i = 1; i < ws.length; i += 2) {
2656 if ((v = ws[i]) != null) {
2657 if (v.source > 0)
2658 return false;
2659 --tc;
2660 }
2661 }
2662 }
2663 if (tc == 0 && ctl == c)
2664 return true;
2665 }
2666 }
2667 }
2668
2669 /**
2670 * Returns an estimate of the total number of completed tasks that
2671 * were executed by a thread other than their submitter. The
2672 * reported value underestimates the actual total number of steals
2673 * when the pool is not quiescent. This value may be useful for
2674 * monitoring and tuning fork/join programs: in general, steal
2675 * counts should be high enough to keep threads busy, but low
2676 * enough to avoid overhead and contention across threads.
2677 *
2678 * @return the number of steals
2679 */
2680 public long getStealCount() {
2681 long count = stealCount;
2682 WorkQueue[] ws; WorkQueue w;
2683 if ((ws = workQueues) != null) {
2684 for (int i = 1; i < ws.length; i += 2) {
2685 if ((w = ws[i]) != null)
2686 count += (long)w.nsteals & 0xffffffffL;
2687 }
2688 }
2689 return count;
2690 }
2691
2692 /**
2693 * Returns an estimate of the total number of tasks currently held
2694 * in queues by worker threads (but not including tasks submitted
2695 * to the pool that have not begun executing). This value is only
2696 * an approximation, obtained by iterating across all threads in
2697 * the pool. This method may be useful for tuning task
2698 * granularities.
2699 *
2700 * @return the number of queued tasks
2701 */
2702 public long getQueuedTaskCount() {
2703 WorkQueue[] ws; WorkQueue w;
2704 VarHandle.acquireFence();
2705 int count = 0;
2706 if ((ws = workQueues) != null) {
2707 for (int i = 1; i < ws.length; i += 2) {
2708 if ((w = ws[i]) != null)
2709 count += w.queueSize();
2710 }
2711 }
2712 return count;
2713 }
2714
2715 /**
2716 * Returns an estimate of the number of tasks submitted to this
2717 * pool that have not yet begun executing. This method may take
2718 * time proportional to the number of submissions.
2719 *
2720 * @return the number of queued submissions
2721 */
2722 public int getQueuedSubmissionCount() {
2723 WorkQueue[] ws; WorkQueue w;
2724 VarHandle.acquireFence();
2725 int count = 0;
2726 if ((ws = workQueues) != null) {
2727 for (int i = 0; i < ws.length; i += 2) {
2728 if ((w = ws[i]) != null)
2729 count += w.queueSize();
2730 }
2731 }
2732 return count;
2733 }
2734
2735 /**
2736 * Returns {@code true} if there are any tasks submitted to this
2737 * pool that have not yet begun executing.
2738 *
2739 * @return {@code true} if there are any queued submissions
2740 */
2741 public boolean hasQueuedSubmissions() {
2742 WorkQueue[] ws; WorkQueue w;
2743 VarHandle.acquireFence();
2744 if ((ws = workQueues) != null) {
2745 for (int i = 0; i < ws.length; i += 2) {
2746 if ((w = ws[i]) != null && !w.isEmpty())
2747 return true;
2748 }
2749 }
2750 return false;
2751 }
2752
2753 /**
2754 * Removes and returns the next unexecuted submission if one is
2755 * available. This method may be useful in extensions to this
2756 * class that re-assign work in systems with multiple pools.
2757 *
2758 * @return the next submission, or {@code null} if none
2759 */
2760 protected ForkJoinTask<?> pollSubmission() {
2761 return pollScan(true);
2762 }
2763
2764 /**
2765 * Removes all available unexecuted submitted and forked tasks
2766 * from scheduling queues and adds them to the given collection,
2767 * without altering their execution status. These may include
2768 * artificially generated or wrapped tasks. This method is
2769 * designed to be invoked only when the pool is known to be
2770 * quiescent. Invocations at other times may not remove all
2771 * tasks. A failure encountered while attempting to add elements
2772 * to collection {@code c} may result in elements being in
2773 * neither, either or both collections when the associated
2774 * exception is thrown. The behavior of this operation is
2775 * undefined if the specified collection is modified while the
2776 * operation is in progress.
2777 *
2778 * @param c the collection to transfer elements into
2779 * @return the number of elements transferred
2780 */
2781 protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
2782 WorkQueue[] ws; WorkQueue w; ForkJoinTask<?> t;
2783 VarHandle.acquireFence();
2784 int count = 0;
2785 if ((ws = workQueues) != null) {
2786 for (int i = 0; i < ws.length; ++i) {
2787 if ((w = ws[i]) != null) {
2788 while ((t = w.poll()) != null) {
2789 c.add(t);
2790 ++count;
2791 }
2792 }
2793 }
2794 }
2795 return count;
2796 }
2797
2798 /**
2799 * Returns a string identifying this pool, as well as its state,
2800 * including indications of run state, parallelism level, and
2801 * worker and task counts.
2802 *
2803 * @return a string identifying this pool, as well as its state
2804 */
2805 public String toString() {
2806 // Use a single pass through workQueues to collect counts
2807 int md = mode; // read volatile fields first
2808 long c = ctl;
2809 long st = stealCount;
2810 long qt = 0L, qs = 0L; int rc = 0;
2811 WorkQueue[] ws; WorkQueue w;
2812 if ((ws = workQueues) != null) {
2813 for (int i = 0; i < ws.length; ++i) {
2814 if ((w = ws[i]) != null) {
2815 int size = w.queueSize();
2816 if ((i & 1) == 0)
2817 qs += size;
2818 else {
2819 qt += size;
2820 st += (long)w.nsteals & 0xffffffffL;
2821 if (w.isApparentlyUnblocked())
2822 ++rc;
2823 }
2824 }
2825 }
2826 }
2827
2828 int pc = (md & SMASK);
2829 int tc = pc + (short)(c >>> TC_SHIFT);
2830 int ac = pc + (int)(c >> RC_SHIFT);
2831 if (ac < 0) // ignore transient negative
2832 ac = 0;
2833 String level = ((md & TERMINATED) != 0 ? "Terminated" :
2834 (md & STOP) != 0 ? "Terminating" :
2835 (md & SHUTDOWN) != 0 ? "Shutting down" :
2836 "Running");
2837 return super.toString() +
2838 "[" + level +
2839 ", parallelism = " + pc +
2840 ", size = " + tc +
2841 ", active = " + ac +
2842 ", running = " + rc +
2843 ", steals = " + st +
2844 ", tasks = " + qt +
2845 ", submissions = " + qs +
2846 "]";
2847 }
2848
2849 /**
2850 * Possibly initiates an orderly shutdown in which previously
2851 * submitted tasks are executed, but no new tasks will be
2852 * accepted. Invocation has no effect on execution state if this
2853 * is the {@link #commonPool()}, and no additional effect if
2854 * already shut down. Tasks that are in the process of being
2855 * submitted concurrently during the course of this method may or
2856 * may not be rejected.
2857 *
2858 * @throws SecurityException if a security manager exists and
2859 * the caller is not permitted to modify threads
2860 * because it does not hold {@link
2861 * java.lang.RuntimePermission}{@code ("modifyThread")}
2862 */
2863 public void shutdown() {
2864 checkPermission();
2865 tryTerminate(false, true);
2866 }
2867
2868 /**
2869 * Possibly attempts to cancel and/or stop all tasks, and reject
2870 * all subsequently submitted tasks. Invocation has no effect on
2871 * execution state if this is the {@link #commonPool()}, and no
2872 * additional effect if already shut down. Otherwise, tasks that
2873 * are in the process of being submitted or executed concurrently
2874 * during the course of this method may or may not be
2875 * rejected. This method cancels both existing and unexecuted
2876 * tasks, in order to permit termination in the presence of task
2877 * dependencies. So the method always returns an empty list
2878 * (unlike the case for some other Executors).
2879 *
2880 * @return an empty list
2881 * @throws SecurityException if a security manager exists and
2882 * the caller is not permitted to modify threads
2883 * because it does not hold {@link
2884 * java.lang.RuntimePermission}{@code ("modifyThread")}
2885 */
2886 public List<Runnable> shutdownNow() {
2887 checkPermission();
2888 tryTerminate(true, true);
2889 return Collections.emptyList();
2890 }
2891
2892 /**
2893 * Returns {@code true} if all tasks have completed following shut down.
2894 *
2895 * @return {@code true} if all tasks have completed following shut down
2896 */
2897 public boolean isTerminated() {
2898 return (mode & TERMINATED) != 0;
2899 }
2900
2901 /**
2902 * Returns {@code true} if the process of termination has
2903 * commenced but not yet completed. This method may be useful for
2904 * debugging. A return of {@code true} reported a sufficient
2905 * period after shutdown may indicate that submitted tasks have
2906 * ignored or suppressed interruption, or are waiting for I/O,
2907 * causing this executor not to properly terminate. (See the
2908 * advisory notes for class {@link ForkJoinTask} stating that
2909 * tasks should not normally entail blocking operations. But if
2910 * they do, they must abort them on interrupt.)
2911 *
2912 * @return {@code true} if terminating but not yet terminated
2913 */
2914 public boolean isTerminating() {
2915 int md = mode;
2916 return (md & STOP) != 0 && (md & TERMINATED) == 0;
2917 }
2918
2919 /**
2920 * Returns {@code true} if this pool has been shut down.
2921 *
2922 * @return {@code true} if this pool has been shut down
2923 */
2924 public boolean isShutdown() {
2925 return (mode & SHUTDOWN) != 0;
2926 }
2927
2928 /**
2929 * Blocks until all tasks have completed execution after a
2930 * shutdown request, or the timeout occurs, or the current thread
2931 * is interrupted, whichever happens first. Because the {@link
2932 * #commonPool()} never terminates until program shutdown, when
2933 * applied to the common pool, this method is equivalent to {@link
2934 * #awaitQuiescence(long, TimeUnit)} but always returns {@code false}.
2935 *
2936 * @param timeout the maximum time to wait
2937 * @param unit the time unit of the timeout argument
2938 * @return {@code true} if this executor terminated and
2939 * {@code false} if the timeout elapsed before termination
2940 * @throws InterruptedException if interrupted while waiting
2941 */
2942 public boolean awaitTermination(long timeout, TimeUnit unit)
2943 throws InterruptedException {
2944 if (Thread.interrupted())
2945 throw new InterruptedException();
2946 if (this == common) {
2947 awaitQuiescence(timeout, unit);
2948 return false;
2949 }
2950 long nanos = unit.toNanos(timeout);
2951 if (isTerminated())
2952 return true;
2953 if (nanos <= 0L)
2954 return false;
2955 long deadline = System.nanoTime() + nanos;
2956 synchronized (this) {
2957 for (;;) {
2958 if (isTerminated())
2959 return true;
2960 if (nanos <= 0L)
2961 return false;
2962 long millis = TimeUnit.NANOSECONDS.toMillis(nanos);
2963 wait(millis > 0L ? millis : 1L);
2964 nanos = deadline - System.nanoTime();
2965 }
2966 }
2967 }
2968
2969 /**
2970 * If called by a ForkJoinTask operating in this pool, equivalent
2971 * in effect to {@link ForkJoinTask#helpQuiesce}. Otherwise,
2972 * waits and/or attempts to assist performing tasks until this
2973 * pool {@link #isQuiescent} or the indicated timeout elapses.
2974 *
2975 * @param timeout the maximum time to wait
2976 * @param unit the time unit of the timeout argument
2977 * @return {@code true} if quiescent; {@code false} if the
2978 * timeout elapsed.
2979 */
2980 public boolean awaitQuiescence(long timeout, TimeUnit unit) {
2981 long nanos = unit.toNanos(timeout);
2982 ForkJoinWorkerThread wt;
2983 Thread thread = Thread.currentThread();
2984 if ((thread instanceof ForkJoinWorkerThread) &&
2985 (wt = (ForkJoinWorkerThread)thread).pool == this) {
2986 helpQuiescePool(wt.workQueue);
2987 return true;
2988 }
2989 else {
2990 for (long startTime = System.nanoTime();;) {
2991 ForkJoinTask<?> t;
2992 if ((t = pollScan(false)) != null)
2993 t.doExec();
2994 else if (isQuiescent())
2995 return true;
2996 else if ((System.nanoTime() - startTime) > nanos)
2997 return false;
2998 else
2999 Thread.yield(); // cannot block
3000 }
3001 }
3002 }
3003
3004 /**
3005 * Waits and/or attempts to assist performing tasks indefinitely
3006 * until the {@link #commonPool()} {@link #isQuiescent}.
3007 */
3008 static void quiesceCommonPool() {
3009 common.awaitQuiescence(Long.MAX_VALUE, TimeUnit.NANOSECONDS);
3010 }
3011
3012 /**
3013 * Interface for extending managed parallelism for tasks running
3014 * in {@link ForkJoinPool}s.
3015 *
3016 * <p>A {@code ManagedBlocker} provides two methods. Method
3017 * {@link #isReleasable} must return {@code true} if blocking is
3018 * not necessary. Method {@link #block} blocks the current thread
3019 * if necessary (perhaps internally invoking {@code isReleasable}
3020 * before actually blocking). These actions are performed by any
3021 * thread invoking {@link ForkJoinPool#managedBlock(ManagedBlocker)}.
3022 * The unusual methods in this API accommodate synchronizers that
3023 * may, but don't usually, block for long periods. Similarly, they
3024 * allow more efficient internal handling of cases in which
3025 * additional workers may be, but usually are not, needed to
3026 * ensure sufficient parallelism. Toward this end,
3027 * implementations of method {@code isReleasable} must be amenable
3028 * to repeated invocation.
3029 *
3030 * <p>For example, here is a ManagedBlocker based on a
3031 * ReentrantLock:
3032 * <pre> {@code
3033 * class ManagedLocker implements ManagedBlocker {
3034 * final ReentrantLock lock;
3035 * boolean hasLock = false;
3036 * ManagedLocker(ReentrantLock lock) { this.lock = lock; }
3037 * public boolean block() {
3038 * if (!hasLock)
3039 * lock.lock();
3040 * return true;
3041 * }
3042 * public boolean isReleasable() {
3043 * return hasLock || (hasLock = lock.tryLock());
3044 * }
3045 * }}</pre>
3046 *
3047 * <p>Here is a class that possibly blocks waiting for an
3048 * item on a given queue:
3049 * <pre> {@code
3050 * class QueueTaker<E> implements ManagedBlocker {
3051 * final BlockingQueue<E> queue;
3052 * volatile E item = null;
3053 * QueueTaker(BlockingQueue<E> q) { this.queue = q; }
3054 * public boolean block() throws InterruptedException {
3055 * if (item == null)
3056 * item = queue.take();
3057 * return true;
3058 * }
3059 * public boolean isReleasable() {
3060 * return item != null || (item = queue.poll()) != null;
3061 * }
3062 * public E getItem() { // call after pool.managedBlock completes
3063 * return item;
3064 * }
3065 * }}</pre>
3066 */
3067 public static interface ManagedBlocker {
3068 /**
3069 * Possibly blocks the current thread, for example waiting for
3070 * a lock or condition.
3071 *
3072 * @return {@code true} if no additional blocking is necessary
3073 * (i.e., if isReleasable would return true)
3074 * @throws InterruptedException if interrupted while waiting
3075 * (the method is not required to do so, but is allowed to)
3076 */
3077 boolean block() throws InterruptedException;
3078
3079 /**
3080 * Returns {@code true} if blocking is unnecessary.
3081 * @return {@code true} if blocking is unnecessary
3082 */
3083 boolean isReleasable();
3084 }
3085
3086 /**
3087 * Runs the given possibly blocking task. When {@linkplain
3088 * ForkJoinTask#inForkJoinPool() running in a ForkJoinPool}, this
3089 * method possibly arranges for a spare thread to be activated if
3090 * necessary to ensure sufficient parallelism while the current
3091 * thread is blocked in {@link ManagedBlocker#block blocker.block()}.
3092 *
3093 * <p>This method repeatedly calls {@code blocker.isReleasable()} and
3094 * {@code blocker.block()} until either method returns {@code true}.
3095 * Every call to {@code blocker.block()} is preceded by a call to
3096 * {@code blocker.isReleasable()} that returned {@code false}.
3097 *
3098 * <p>If not running in a ForkJoinPool, this method is
3099 * behaviorally equivalent to
3100 * <pre> {@code
3101 * while (!blocker.isReleasable())
3102 * if (blocker.block())
3103 * break;}</pre>
3104 *
3105 * If running in a ForkJoinPool, the pool may first be expanded to
3106 * ensure sufficient parallelism available during the call to
3107 * {@code blocker.block()}.
3108 *
3109 * @param blocker the blocker task
3110 * @throws InterruptedException if {@code blocker.block()} did so
3111 */
3112 public static void managedBlock(ManagedBlocker blocker)
3113 throws InterruptedException {
3114 if (blocker == null) throw new NullPointerException();
3115 ForkJoinPool p;
3116 ForkJoinWorkerThread wt;
3117 WorkQueue w;
3118 Thread t = Thread.currentThread();
3119 if ((t instanceof ForkJoinWorkerThread) &&
3120 (p = (wt = (ForkJoinWorkerThread)t).pool) != null &&
3121 (w = wt.workQueue) != null) {
3122 int block;
3123 while (!blocker.isReleasable()) {
3124 if ((block = p.tryCompensate(w)) != 0) {
3125 try {
3126 do {} while (!blocker.isReleasable() &&
3127 !blocker.block());
3128 } finally {
3129 CTL.getAndAdd(p, (block > 0) ? RC_UNIT : 0L);
3130 }
3131 break;
3132 }
3133 }
3134 }
3135 else {
3136 do {} while (!blocker.isReleasable() &&
3137 !blocker.block());
3138 }
3139 }
3140
3141 /**
3142 * If the given executor is a ForkJoinPool, poll and execute
3143 * AsynchronousCompletionTasks from worker's queue until none are
3144 * available or blocker is released.
3145 */
3146 static void helpAsyncBlocker(Executor e, ManagedBlocker blocker) {
3147 if (e instanceof ForkJoinPool) {
3148 WorkQueue w; ForkJoinWorkerThread wt; WorkQueue[] ws; int r, n;
3149 ForkJoinPool p = (ForkJoinPool)e;
3150 Thread thread = Thread.currentThread();
3151 if (thread instanceof ForkJoinWorkerThread &&
3152 (wt = (ForkJoinWorkerThread)thread).pool == p)
3153 w = wt.workQueue;
3154 else if ((r = ThreadLocalRandom.getProbe()) != 0 &&
3155 (ws = p.workQueues) != null && (n = ws.length) > 0)
3156 w = ws[(n - 1) & r & SQMASK];
3157 else
3158 w = null;
3159 if (w != null)
3160 w.helpAsyncBlocker(blocker);
3161 }
3162 }
3163
3164 // AbstractExecutorService overrides. These rely on undocumented
3165 // fact that ForkJoinTask.adapt returns ForkJoinTasks that also
3166 // implement RunnableFuture.
3167
3168 protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
3169 return new ForkJoinTask.AdaptedRunnable<T>(runnable, value);
3170 }
3171
3172 protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
3173 return new ForkJoinTask.AdaptedCallable<T>(callable);
3174 }
3175
3176 // VarHandle mechanics
3177 private static final VarHandle CTL;
3178 private static final VarHandle MODE;
3179 static final VarHandle QA;
3180
3181 static {
3182 try {
3183 MethodHandles.Lookup l = MethodHandles.lookup();
3184 CTL = l.findVarHandle(ForkJoinPool.class, "ctl", long.class);
3185 MODE = l.findVarHandle(ForkJoinPool.class, "mode", int.class);
3186 QA = MethodHandles.arrayElementVarHandle(ForkJoinTask[].class);
3187 } catch (ReflectiveOperationException e) {
3188 throw new ExceptionInInitializerError(e);
3189 }
3190
3191 // Reduce the risk of rare disastrous classloading in first call to
3192 // LockSupport.park: https://bugs.openjdk.java.net/browse/JDK-8074773
3193 Class<?> ensureLoaded = LockSupport.class;
3194
3195 int commonMaxSpares = DEFAULT_COMMON_MAX_SPARES;
3196 try {
3197 String p = System.getProperty
3198 ("java.util.concurrent.ForkJoinPool.common.maximumSpares");
3199 if (p != null)
3200 commonMaxSpares = Integer.parseInt(p);
3201 } catch (Exception ignore) {}
3202 COMMON_MAX_SPARES = commonMaxSpares;
3203
3204 defaultForkJoinWorkerThreadFactory =
3205 new DefaultForkJoinWorkerThreadFactory();
3206 modifyThreadPermission = new RuntimePermission("modifyThread");
3207
3208 common = AccessController.doPrivileged(new PrivilegedAction<>() {
3209 public ForkJoinPool run() {
3210 return new ForkJoinPool((byte)0); }});
3211
3212 COMMON_PARALLELISM = Math.max(common.mode & SMASK, 1);
3213 }
3214
3215 /**
3216 * Factory for innocuous worker threads.
3217 */
3218 private static final class InnocuousForkJoinWorkerThreadFactory
3219 implements ForkJoinWorkerThreadFactory {
3220
3221 /**
3222 * An ACC to restrict permissions for the factory itself.
3223 * The constructed workers have no permissions set.
3224 */
3225 private static final AccessControlContext ACC = contextWithPermissions(
3226 modifyThreadPermission,
3227 new RuntimePermission("enableContextClassLoaderOverride"),
3228 new RuntimePermission("modifyThreadGroup"),
3229 new RuntimePermission("getClassLoader"),
3230 new RuntimePermission("setContextClassLoader"));
3231
3232 public final ForkJoinWorkerThread newThread(ForkJoinPool pool) {
3233 return AccessController.doPrivileged(
3234 new PrivilegedAction<>() {
3235 public ForkJoinWorkerThread run() {
3236 return new ForkJoinWorkerThread.
3237 InnocuousForkJoinWorkerThread(pool); }},
3238 ACC);
3239 }
3240 }
3241 }