ForkJoinPool.java 139.8 KB
Newer Older
D
dl 已提交
1 2 3 4 5
/*
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
6
 * published by the Free Software Foundation.  Oracle designates this
D
dl 已提交
7
 * particular file as subject to the "Classpath" exception as provided
8
 * by Oracle in the LICENSE file that accompanied this code.
D
dl 已提交
9 10 11 12 13 14 15 16 17 18 19
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
20 21 22
 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 * or visit www.oracle.com if you need additional information or have any
 * questions.
D
dl 已提交
23 24 25 26 27 28 29 30 31 32
 */

/*
 * This file is available under and governed by the GNU General Public
 * License version 2 only, as published by the Free Software Foundation.
 * However, the following notice accompanied the original version of this
 * file:
 *
 * Written by Doug Lea with assistance from members of JCP JSR-166
 * Expert Group and released to the public domain, as explained at
33
 * http://creativecommons.org/publicdomain/zero/1.0/
D
dl 已提交
34 35 36 37
 */

package java.util.concurrent;

38
import java.lang.Thread.UncaughtExceptionHandler;
D
dl 已提交
39 40 41 42 43
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collection;
import java.util.Collections;
import java.util.List;
D
dl 已提交
44 45 46 47 48 49
import java.util.concurrent.AbstractExecutorService;
import java.util.concurrent.Callable;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Future;
import java.util.concurrent.RejectedExecutionException;
import java.util.concurrent.RunnableFuture;
50
import java.util.concurrent.ThreadLocalRandom;
D
dl 已提交
51
import java.util.concurrent.TimeUnit;
D
dl 已提交
52 53 54 55

/**
 * An {@link ExecutorService} for running {@link ForkJoinTask}s.
 * A {@code ForkJoinPool} provides the entry point for submissions
D
dl 已提交
56
 * from non-{@code ForkJoinTask} clients, as well as management and
D
dl 已提交
57 58 59 60 61
 * monitoring operations.
 *
 * <p>A {@code ForkJoinPool} differs from other kinds of {@link
 * ExecutorService} mainly by virtue of employing
 * <em>work-stealing</em>: all threads in the pool attempt to find and
62 63 64 65 66 67 68 69
 * execute tasks submitted to the pool and/or created by other active
 * tasks (eventually blocking waiting for work if none exist). This
 * enables efficient processing when most tasks spawn other subtasks
 * (as do most {@code ForkJoinTask}s), as well as when many small
 * tasks are submitted to the pool from external clients.  Especially
 * when setting <em>asyncMode</em> to true in constructors, {@code
 * ForkJoinPool}s may also be appropriate for use with event-style
 * tasks that are never joined.
D
dl 已提交
70
 *
71 72 73 74 75 76 77 78 79 80 81 82
 * <p>A static {@link #commonPool()} is available and appropriate for
 * most applications. The common pool is used by any ForkJoinTask that
 * is not explicitly submitted to a specified pool. Using the common
 * pool normally reduces resource usage (its threads are slowly
 * reclaimed during periods of non-use, and reinstated upon subsequent
 * use).
 *
 * <p>For applications that require separate or custom pools, a {@code
 * ForkJoinPool} may be constructed with a given target parallelism
 * level; by default, equal to the number of available processors. The
 * pool attempts to maintain enough active (or available) threads by
 * dynamically adding, suspending, or resuming internal worker
83 84 85
 * threads, even if some tasks are stalled waiting to join others.
 * However, no such adjustments are guaranteed in the face of blocked
 * I/O or other unmanaged synchronization. The nested {@link
86
 * ManagedBlocker} interface enables extension of the kinds of
D
dl 已提交
87
 * synchronization accommodated.
D
dl 已提交
88 89 90 91 92 93 94 95
 *
 * <p>In addition to execution and lifecycle control methods, this
 * class provides status check methods (for example
 * {@link #getStealCount}) that are intended to aid in developing,
 * tuning, and monitoring fork/join applications. Also, method
 * {@link #toString} returns indications of pool state in a
 * convenient form for informal monitoring.
 *
96 97 98 99 100 101
 * <p>As is the case with other ExecutorServices, there are three
 * main task execution methods summarized in the following table.
 * These are designed to be used primarily by clients not already
 * engaged in fork/join computations in the current pool.  The main
 * forms of these methods accept instances of {@code ForkJoinTask},
 * but overloaded forms also allow mixed execution of plain {@code
D
dl 已提交
102
 * Runnable}- or {@code Callable}- based activities as well.  However,
103 104 105 106
 * tasks that are already executing in a pool should normally instead
 * use the within-computation forms listed in the table unless using
 * async event-style tasks that are not usually joined, in which case
 * there is little difference among choice of methods.
D
dl 已提交
107 108
 *
 * <table BORDER CELLPADDING=3 CELLSPACING=1>
109
 * <caption>Summary of task execution methods</caption>
D
dl 已提交
110 111 112 113 114 115
 *  <tr>
 *    <td></td>
 *    <td ALIGN=CENTER> <b>Call from non-fork/join clients</b></td>
 *    <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td>
 *  </tr>
 *  <tr>
116
 *    <td> <b>Arrange async execution</b></td>
D
dl 已提交
117 118 119 120
 *    <td> {@link #execute(ForkJoinTask)}</td>
 *    <td> {@link ForkJoinTask#fork}</td>
 *  </tr>
 *  <tr>
121
 *    <td> <b>Await and obtain result</b></td>
D
dl 已提交
122 123 124 125
 *    <td> {@link #invoke(ForkJoinTask)}</td>
 *    <td> {@link ForkJoinTask#invoke}</td>
 *  </tr>
 *  <tr>
126
 *    <td> <b>Arrange exec and obtain Future</b></td>
D
dl 已提交
127 128 129 130 131
 *    <td> {@link #submit(ForkJoinTask)}</td>
 *    <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
 *  </tr>
 * </table>
 *
132
 * <p>The common pool is by default constructed with default
133 134 135 136 137 138 139 140 141 142 143 144 145 146 147
 * parameters, but these may be controlled by setting three
 * {@linkplain System#getProperty system properties}:
 * <ul>
 * <li>{@code java.util.concurrent.ForkJoinPool.common.parallelism}
 * - the parallelism level, a non-negative integer
 * <li>{@code java.util.concurrent.ForkJoinPool.common.threadFactory}
 * - the class name of a {@link ForkJoinWorkerThreadFactory}
 * <li>{@code java.util.concurrent.ForkJoinPool.common.exceptionHandler}
 * - the class name of a {@link UncaughtExceptionHandler}
 * </ul>
 * The system class loader is used to load these classes.
 * Upon any error in establishing these settings, default parameters
 * are used. It is possible to disable or limit the use of threads in
 * the common pool by setting the parallelism property to zero, and/or
 * using a factory that may return {@code null}.
D
dl 已提交
148 149 150 151 152 153 154
 *
 * <p><b>Implementation notes</b>: This implementation restricts the
 * maximum number of running threads to 32767. Attempts to create
 * pools with greater than the maximum number result in
 * {@code IllegalArgumentException}.
 *
 * <p>This implementation rejects submitted tasks (that is, by throwing
D
dl 已提交
155 156
 * {@link RejectedExecutionException}) only when the pool is shut down
 * or internal resources have been exhausted.
D
dl 已提交
157 158 159 160
 *
 * @since 1.7
 * @author Doug Lea
 */
161
@sun.misc.Contended
D
dl 已提交
162 163 164
public class ForkJoinPool extends AbstractExecutorService {

    /*
D
dl 已提交
165 166
     * Implementation Overview
     *
167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193
     * This class and its nested classes provide the main
     * functionality and control for a set of worker threads:
     * Submissions from non-FJ threads enter into submission queues.
     * Workers take these tasks and typically split them into subtasks
     * that may be stolen by other workers.  Preference rules give
     * first priority to processing tasks from their own queues (LIFO
     * or FIFO, depending on mode), then to randomized FIFO steals of
     * tasks in other queues.
     *
     * WorkQueues
     * ==========
     *
     * Most operations occur within work-stealing queues (in nested
     * class WorkQueue).  These are special forms of Deques that
     * support only three of the four possible end-operations -- push,
     * pop, and poll (aka steal), under the further constraints that
     * push and pop are called only from the owning thread (or, as
     * extended here, under a lock), while poll may be called from
     * other threads.  (If you are unfamiliar with them, you probably
     * want to read Herlihy and Shavit's book "The Art of
     * Multiprocessor programming", chapter 16 describing these in
     * more detail before proceeding.)  The main work-stealing queue
     * design is roughly similar to those in the papers "Dynamic
     * Circular Work-Stealing Deque" by Chase and Lev, SPAA 2005
     * (http://research.sun.com/scalable/pubs/index.html) and
     * "Idempotent work stealing" by Michael, Saraswat, and Vechev,
     * PPoPP 2009 (http://portal.acm.org/citation.cfm?id=1504186).
194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222
     * See also "Correct and Efficient Work-Stealing for Weak Memory
     * Models" by Le, Pop, Cohen, and Nardelli, PPoPP 2013
     * (http://www.di.ens.fr/~zappa/readings/ppopp13.pdf) for an
     * analysis of memory ordering (atomic, volatile etc) issues.  The
     * main differences ultimately stem from GC requirements that we
     * null out taken slots as soon as we can, to maintain as small a
     * footprint as possible even in programs generating huge numbers
     * of tasks. To accomplish this, we shift the CAS arbitrating pop
     * vs poll (steal) from being on the indices ("base" and "top") to
     * the slots themselves.  So, both a successful pop and poll
     * mainly entail a CAS of a slot from non-null to null.  Because
     * we rely on CASes of references, we do not need tag bits on base
     * or top.  They are simple ints as used in any circular
     * array-based queue (see for example ArrayDeque).  Updates to the
     * indices must still be ordered in a way that guarantees that top
     * == base means the queue is empty, but otherwise may err on the
     * side of possibly making the queue appear nonempty when a push,
     * pop, or poll have not fully committed. Note that this means
     * that the poll operation, considered individually, is not
     * wait-free. One thief cannot successfully continue until another
     * in-progress one (or, if previously empty, a push) completes.
     * However, in the aggregate, we ensure at least probabilistic
     * non-blockingness.  If an attempted steal fails, a thief always
     * chooses a different random victim target to try next. So, in
     * order for one thief to progress, it suffices for any
     * in-progress poll or new push on any empty queue to
     * complete. (This is why we normally use method pollAt and its
     * variants that try once at the apparent base index, else
     * consider alternative actions, rather than method poll.)
223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240
     *
     * This approach also enables support of a user mode in which local
     * task processing is in FIFO, not LIFO order, simply by using
     * poll rather than pop.  This can be useful in message-passing
     * frameworks in which tasks are never joined.  However neither
     * mode considers affinities, loads, cache localities, etc, so
     * rarely provide the best possible performance on a given
     * machine, but portably provide good throughput by averaging over
     * these factors.  (Further, even if we did try to use such
     * information, we do not usually have a basis for exploiting it.
     * For example, some sets of tasks profit from cache affinities,
     * but others are harmed by cache pollution effects.)
     *
     * WorkQueues are also used in a similar way for tasks submitted
     * to the pool. We cannot mix these tasks in the same queues used
     * for work-stealing (this would contaminate lifo/fifo
     * processing). Instead, we randomly associate submission queues
     * with submitting threads, using a form of hashing.  The
241 242 243 244 245 246 247 248 249 250 251 252
     * ThreadLocalRandom probe value serves as a hash code for
     * choosing existing queues, and may be randomly repositioned upon
     * contention with other submitters.  In essence, submitters act
     * like workers except that they are restricted to executing local
     * tasks that they submitted (or in the case of CountedCompleters,
     * others with the same root task).  However, because most
     * shared/external queue operations are more expensive than
     * internal, and because, at steady state, external submitters
     * will compete for CPU with workers, ForkJoinTask.join and
     * related methods disable them from repeatedly helping to process
     * tasks if all workers are active.  Insertion of tasks in shared
     * mode requires a lock (mainly to protect in the case of
253 254 255 256 257 258 259
     * resizing) but we use only a simple spinlock (using bits in
     * field qlock), because submitters encountering a busy queue move
     * on to try or create other queues -- they block only when
     * creating and registering new queues.
     *
     * Management
     * ==========
260 261 262 263 264 265
     *
     * The main throughput advantages of work-stealing stem from
     * decentralized control -- workers mostly take tasks from
     * themselves or each other. We cannot negate this in the
     * implementation of other management responsibilities. The main
     * tactic for avoiding bottlenecks is packing nearly all
266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321
     * essentially atomic control state into two volatile variables
     * that are by far most often read (not written) as status and
     * consistency checks.
     *
     * Field "ctl" contains 64 bits holding all the information needed
     * to atomically decide to add, inactivate, enqueue (on an event
     * queue), dequeue, and/or re-activate workers.  To enable this
     * packing, we restrict maximum parallelism to (1<<15)-1 (which is
     * far in excess of normal operating range) to allow ids, counts,
     * and their negations (used for thresholding) to fit into 16bit
     * fields.
     *
     * Field "plock" is a form of sequence lock with a saturating
     * shutdown bit (similarly for per-queue "qlocks"), mainly
     * protecting updates to the workQueues array, as well as to
     * enable shutdown.  When used as a lock, it is normally only very
     * briefly held, so is nearly always available after at most a
     * brief spin, but we use a monitor-based backup strategy to
     * block when needed.
     *
     * Recording WorkQueues.  WorkQueues are recorded in the
     * "workQueues" array that is created upon first use and expanded
     * if necessary.  Updates to the array while recording new workers
     * and unrecording terminated ones are protected from each other
     * by a lock but the array is otherwise concurrently readable, and
     * accessed directly.  To simplify index-based operations, the
     * array size is always a power of two, and all readers must
     * tolerate null slots. Worker queues are at odd indices. Shared
     * (submission) queues are at even indices, up to a maximum of 64
     * slots, to limit growth even if array needs to expand to add
     * more workers. Grouping them together in this way simplifies and
     * speeds up task scanning.
     *
     * All worker thread creation is on-demand, triggered by task
     * submissions, replacement of terminated workers, and/or
     * compensation for blocked workers. However, all other support
     * code is set up to work with other policies.  To ensure that we
     * do not hold on to worker references that would prevent GC, ALL
     * accesses to workQueues are via indices into the workQueues
     * array (which is one source of some of the messy code
     * constructions here). In essence, the workQueues array serves as
     * a weak reference mechanism. Thus for example the wait queue
     * field of ctl stores indices, not references.  Access to the
     * workQueues in associated methods (for example signalWork) must
     * both index-check and null-check the IDs. All such accesses
     * ignore bad IDs by returning out early from what they are doing,
     * since this can only be associated with termination, in which
     * case it is OK to give up.  All uses of the workQueues array
     * also check that it is non-null (even if previously
     * non-null). This allows nulling during termination, which is
     * currently not necessary, but remains an option for
     * resource-revocation-based shutdown schemes. It also helps
     * reduce JIT issuance of uncommon-trap code, which tends to
     * unnecessarily complicate control flow in some methods.
     *
     * Event Queuing. Unlike HPC work-stealing frameworks, we cannot
322 323 324 325
     * let workers spin indefinitely scanning for tasks when none can
     * be found immediately, and we cannot start/resume workers unless
     * there appear to be tasks available.  On the other hand, we must
     * quickly prod them into action when new tasks are submitted or
326 327 328 329 330 331 332 333 334 335 336 337
     * generated. In many usages, ramp-up time to activate workers is
     * the main limiting factor in overall performance (this is
     * compounded at program start-up by JIT compilation and
     * allocation). So we try to streamline this as much as possible.
     * We park/unpark workers after placing in an event wait queue
     * when they cannot find work. This "queue" is actually a simple
     * Treiber stack, headed by the "id" field of ctl, plus a 15bit
     * counter value (that reflects the number of times a worker has
     * been inactivated) to avoid ABA effects (we need only as many
     * version numbers as worker threads). Successors are held in
     * field WorkQueue.nextWait.  Queuing deals with several intrinsic
     * races, mainly that a task-producing thread can miss seeing (and
338 339
     * signalling) another thread that gave up looking for work but
     * has not yet entered the wait queue. We solve this by requiring
340 341
     * a full sweep of all workers (via repeated calls to method
     * scan()) both before and after a newly waiting worker is added
342 343 344 345 346 347 348 349 350 351 352
     * to the wait queue.  Because enqueued workers may actually be
     * rescanning rather than waiting, we set and clear the "parker"
     * field of WorkQueues to reduce unnecessary calls to unpark.
     * (This requires a secondary recheck to avoid missed signals.)
     * Note the unusual conventions about Thread.interrupts
     * surrounding parking and other blocking: Because interrupts are
     * used solely to alert threads to check termination, which is
     * checked anyway upon blocking, we clear status (using
     * Thread.interrupted) before any call to park, so that park does
     * not immediately return due to status being set via some other
     * unrelated call to interrupt in user code.
353 354 355
     *
     * Signalling.  We create or wake up workers only when there
     * appears to be at least one task they might be able to find and
356 357 358 359 360 361 362 363
     * execute.  When a submission is added or another worker adds a
     * task to a queue that has fewer than two tasks, they signal
     * waiting workers (or trigger creation of new ones if fewer than
     * the given parallelism level -- signalWork).  These primary
     * signals are buttressed by others whenever other threads remove
     * a task from a queue and notice that there are other tasks there
     * as well.  So in general, pools will be over-signalled. On most
     * platforms, signalling (unpark) overhead time is noticeably
364 365 366
     * long, and the time between signalling a thread and it actually
     * making progress can be very noticeably long, so it is worth
     * offloading these delays from critical paths as much as
367 368 369 370
     * possible. Additionally, workers spin-down gradually, by staying
     * alive so long as they see the ctl state changing.  Similar
     * stability-sensing techniques are also used before blocking in
     * awaitJoin and helpComplete.
371 372 373
     *
     * Trimming workers. To release resources after periods of lack of
     * use, a worker starting to wait when the pool is quiescent will
374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401
     * time out and terminate if the pool has remained quiescent for a
     * given period -- a short period if there are more threads than
     * parallelism, longer as the number of threads decreases. This
     * will slowly propagate, eventually terminating all workers after
     * periods of non-use.
     *
     * Shutdown and Termination. A call to shutdownNow atomically sets
     * a plock bit and then (non-atomically) sets each worker's
     * qlock status, cancels all unprocessed tasks, and wakes up
     * all waiting workers.  Detecting whether termination should
     * commence after a non-abrupt shutdown() call requires more work
     * and bookkeeping. We need consensus about quiescence (i.e., that
     * there is no more work). The active count provides a primary
     * indication but non-abrupt shutdown still requires a rechecking
     * scan for any workers that are inactive but not queued.
     *
     * Joining Tasks
     * =============
     *
     * Any of several actions may be taken when one worker is waiting
     * to join a task stolen (or always held) by another.  Because we
     * are multiplexing many tasks on to a pool of workers, we can't
     * just let them block (as in Thread.join).  We also cannot just
     * reassign the joiner's run-time stack with another and replace
     * it later, which would be a form of "continuation", that even if
     * possible is not necessarily a good idea since we sometimes need
     * both an unblocked task and its continuation to progress.
     * Instead we combine two tactics:
D
dl 已提交
402 403
     *
     *   Helping: Arranging for the joiner to execute some task that it
404
     *      would be running if the steal had not occurred.
D
dl 已提交
405 406
     *
     *   Compensating: Unless there are already enough live threads,
407 408 409 410 411 412 413 414 415 416
     *      method tryCompensate() may create or re-activate a spare
     *      thread to compensate for blocked joiners until they unblock.
     *
     * A third form (implemented in tryRemoveAndExec) amounts to
     * helping a hypothetical compensator: If we can readily tell that
     * a possible action of a compensator is to steal and execute the
     * task being joined, the joining thread can do so directly,
     * without the need for a compensation thread (although at the
     * expense of larger run-time stacks, but the tradeoff is
     * typically worthwhile).
D
dl 已提交
417 418 419 420
     *
     * The ManagedBlocker extension API can't use helping so relies
     * only on compensation in method awaitBlocker.
     *
421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456
     * The algorithm in tryHelpStealer entails a form of "linear"
     * helping: Each worker records (in field currentSteal) the most
     * recent task it stole from some other worker. Plus, it records
     * (in field currentJoin) the task it is currently actively
     * joining. Method tryHelpStealer uses these markers to try to
     * find a worker to help (i.e., steal back a task from and execute
     * it) that could hasten completion of the actively joined task.
     * In essence, the joiner executes a task that would be on its own
     * local deque had the to-be-joined task not been stolen. This may
     * be seen as a conservative variant of the approach in Wagner &
     * Calder "Leapfrogging: a portable technique for implementing
     * efficient futures" SIGPLAN Notices, 1993
     * (http://portal.acm.org/citation.cfm?id=155354). It differs in
     * that: (1) We only maintain dependency links across workers upon
     * steals, rather than use per-task bookkeeping.  This sometimes
     * requires a linear scan of workQueues array to locate stealers,
     * but often doesn't because stealers leave hints (that may become
     * stale/wrong) of where to locate them.  It is only a hint
     * because a worker might have had multiple steals and the hint
     * records only one of them (usually the most current).  Hinting
     * isolates cost to when it is needed, rather than adding to
     * per-task overhead.  (2) It is "shallow", ignoring nesting and
     * potentially cyclic mutual steals.  (3) It is intentionally
     * racy: field currentJoin is updated only while actively joining,
     * which means that we miss links in the chain during long-lived
     * tasks, GC stalls etc (which is OK since blocking in such cases
     * is usually a good idea).  (4) We bound the number of attempts
     * to find work (see MAX_HELP) and fall back to suspending the
     * worker and if necessary replacing it with another.
     *
     * Helping actions for CountedCompleters are much simpler: Method
     * helpComplete can take and execute any task with the same root
     * as the task being waited on. However, this still entails some
     * traversal of completer chains, so is less efficient than using
     * CountedCompleters without explicit joins.
     *
457 458 459 460
     * It is impossible to keep exactly the target parallelism number
     * of threads running at any given time.  Determining the
     * existence of conservatively safe helping targets, the
     * availability of already-created spares, and the apparent need
461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482
     * to create new spares are all racy, so we rely on multiple
     * retries of each.  Compensation in the apparent absence of
     * helping opportunities is challenging to control on JVMs, where
     * GC and other activities can stall progress of tasks that in
     * turn stall out many other dependent tasks, without us being
     * able to determine whether they will ever require compensation.
     * Even though work-stealing otherwise encounters little
     * degradation in the presence of more threads than cores,
     * aggressively adding new threads in such cases entails risk of
     * unwanted positive feedback control loops in which more threads
     * cause more dependent stalls (as well as delayed progress of
     * unblocked threads to the point that we know they are available)
     * leading to more situations requiring more threads, and so
     * on. This aspect of control can be seen as an (analytically
     * intractable) game with an opponent that may choose the worst
     * (for us) active thread to stall at any time.  We take several
     * precautions to bound losses (and thus bound gains), mainly in
     * methods tryCompensate and awaitJoin.
     *
     * Common Pool
     * ===========
     *
483
     * The static common pool always exists after static
484 485 486 487 488 489 490
     * initialization.  Since it (or any other created pool) need
     * never be used, we minimize initial construction overhead and
     * footprint to the setup of about a dozen fields, with no nested
     * allocation. Most bootstrapping occurs within method
     * fullExternalPush during the first submission to the pool.
     *
     * When external threads submit to the common pool, they can
491 492 493 494 495
     * perform subtask processing (see externalHelpJoin and related
     * methods).  This caller-helps policy makes it sensible to set
     * common pool parallelism level to one (or more) less than the
     * total number of available cores, or even zero for pure
     * caller-runs.  We do not need to record whether external
496 497 498 499 500 501 502 503 504 505 506 507 508 509
     * submissions are to the common pool -- if not, externalHelpJoin
     * returns quickly (at the most helping to signal some common pool
     * workers). These submitters would otherwise be blocked waiting
     * for completion, so the extra effort (with liberally sprinkled
     * task status checks) in inapplicable cases amounts to an odd
     * form of limited spin-wait before blocking in ForkJoinTask.join.
     *
     * Style notes
     * ===========
     *
     * There is a lot of representation-level coupling among classes
     * ForkJoinPool, ForkJoinWorkerThread, and ForkJoinTask.  The
     * fields of WorkQueue maintain data structures managed by
     * ForkJoinPool, so are directly accessed.  There is little point
D
dl 已提交
510 511
     * trying to reduce this, since any associated future changes in
     * representations will need to be accompanied by algorithmic
512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543
     * changes anyway. Several methods intrinsically sprawl because
     * they must accumulate sets of consistent reads of volatiles held
     * in local variables.  Methods signalWork() and scan() are the
     * main bottlenecks, so are especially heavily
     * micro-optimized/mangled.  There are lots of inline assignments
     * (of form "while ((local = field) != 0)") which are usually the
     * simplest way to ensure the required read orderings (which are
     * sometimes critical). This leads to a "C"-like style of listing
     * declarations of these locals at the heads of methods or blocks.
     * There are several occurrences of the unusual "do {} while
     * (!cas...)"  which is the simplest way to force an update of a
     * CAS'ed variable. There are also other coding oddities (including
     * several unnecessary-looking hoisted null checks) that help
     * some methods perform reasonably even when interpreted (not
     * compiled).
     *
     * The order of declarations in this file is:
     * (1) Static utility functions
     * (2) Nested (static) classes
     * (3) Static fields
     * (4) Fields, along with constants used when unpacking some of them
     * (5) Internal control methods
     * (6) Callbacks and other support for ForkJoinTask methods
     * (7) Exported methods
     * (8) Static block initializing statics in minimally dependent order
     */

    // Static utilities

    /**
     * If there is a security manager, makes sure caller has
     * permission to modify threads.
D
dl 已提交
544
     */
545 546 547 548 549 550 551
    private static void checkPermission() {
        SecurityManager security = System.getSecurityManager();
        if (security != null)
            security.checkPermission(modifyThreadPermission);
    }

    // Nested classes
D
dl 已提交
552 553 554 555 556 557 558 559 560 561 562 563

    /**
     * Factory for creating new {@link ForkJoinWorkerThread}s.
     * A {@code ForkJoinWorkerThreadFactory} must be defined and used
     * for {@code ForkJoinWorkerThread} subclasses that extend base
     * functionality or initialize threads with different contexts.
     */
    public static interface ForkJoinWorkerThreadFactory {
        /**
         * Returns a new worker thread operating in the given pool.
         *
         * @param pool the pool this thread works in
564
         * @return the new worker thread
565
         * @throws NullPointerException if the pool is null
D
dl 已提交
566 567 568 569 570 571 572 573
         */
        public ForkJoinWorkerThread newThread(ForkJoinPool pool);
    }

    /**
     * Default ForkJoinWorkerThreadFactory implementation; creates a
     * new ForkJoinWorkerThread.
     */
574
    static final class DefaultForkJoinWorkerThreadFactory
D
dl 已提交
575
        implements ForkJoinWorkerThreadFactory {
576
        public final ForkJoinWorkerThread newThread(ForkJoinPool pool) {
D
dl 已提交
577
            return new ForkJoinWorkerThread(pool);
D
dl 已提交
578 579 580
        }
    }

581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639
    /**
     * Class for artificial tasks that are used to replace the target
     * of local joins if they are removed from an interior queue slot
     * in WorkQueue.tryRemoveAndExec. We don't need the proxy to
     * actually do anything beyond having a unique identity.
     */
    static final class EmptyTask extends ForkJoinTask<Void> {
        private static final long serialVersionUID = -7721805057305804111L;
        EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
        public final Void getRawResult() { return null; }
        public final void setRawResult(Void x) {}
        public final boolean exec() { return true; }
    }

    /**
     * Queues supporting work-stealing as well as external task
     * submission. See above for main rationale and algorithms.
     * Implementation relies heavily on "Unsafe" intrinsics
     * and selective use of "volatile":
     *
     * Field "base" is the index (mod array.length) of the least valid
     * queue slot, which is always the next position to steal (poll)
     * from if nonempty. Reads and writes require volatile orderings
     * but not CAS, because updates are only performed after slot
     * CASes.
     *
     * Field "top" is the index (mod array.length) of the next queue
     * slot to push to or pop from. It is written only by owner thread
     * for push, or under lock for external/shared push, and accessed
     * by other threads only after reading (volatile) base.  Both top
     * and base are allowed to wrap around on overflow, but (top -
     * base) (or more commonly -(base - top) to force volatile read of
     * base before top) still estimates size. The lock ("qlock") is
     * forced to -1 on termination, causing all further lock attempts
     * to fail. (Note: we don't need CAS for termination state because
     * upon pool shutdown, all shared-queues will stop being used
     * anyway.)  Nearly all lock bodies are set up so that exceptions
     * within lock bodies are "impossible" (modulo JVM errors that
     * would cause failure anyway.)
     *
     * The array slots are read and written using the emulation of
     * volatiles/atomics provided by Unsafe. Insertions must in
     * general use putOrderedObject as a form of releasing store to
     * ensure that all writes to the task object are ordered before
     * its publication in the queue.  All removals entail a CAS to
     * null.  The array is always a power of two. To ensure safety of
     * Unsafe array operations, all accesses perform explicit null
     * checks and implicit bounds checks via power-of-two masking.
     *
     * In addition to basic queuing support, this class contains
     * fields described elsewhere to control execution. It turns out
     * to work better memory-layout-wise to include them in this class
     * rather than a separate class.
     *
     * Performance on most platforms is very sensitive to placement of
     * instances of both WorkQueues and their arrays -- we absolutely
     * do not want multiple WorkQueue instances or multiple queue
     * arrays sharing cache lines. (It would be best for queue objects
     * and their arrays to share, but there is nothing available to
640 641
     * help arrange that). The @Contended annotation alerts JVMs to
     * try to keep instances apart.
642
     */
643
    @sun.misc.Contended
644 645 646 647 648 649 650 651 652 653 654
    static final class WorkQueue {
        /**
         * Capacity of work-stealing queue array upon initialization.
         * Must be a power of two; at least 4, but should be larger to
         * reduce or eliminate cacheline sharing among queues.
         * Currently, it is much larger, as a partial workaround for
         * the fact that JVMs often place arrays in locations that
         * share GC bookkeeping (especially cardmarks) such that
         * per-write accesses encounter serious memory contention.
         */
        static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
D
dl 已提交
655

656 657 658 659 660 661 662 663 664 665 666 667
        /**
         * Maximum size for queue arrays. Must be a power of two less
         * than or equal to 1 << (31 - width of array entry) to ensure
         * lack of wraparound of index calculations, but defined to a
         * value a bit less than this to help users trap runaway
         * programs before saturating systems.
         */
        static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M

        volatile int eventCount;   // encoded inactivation count; < 0 if inactive
        int nextWait;              // encoded record of next event waiter
        int nsteals;               // number of steals
668 669 670
        int hint;                  // steal index hint
        short poolIndex;           // index of this queue in pool
        final short mode;          // 0: lifo, > 0: fifo, < 0: shared
671 672 673 674 675 676 677 678 679 680 681 682 683 684
        volatile int qlock;        // 1: locked, -1: terminate; else 0
        volatile int base;         // index of next slot for poll
        int top;                   // index of next slot for push
        ForkJoinTask<?>[] array;   // the elements (initially unallocated)
        final ForkJoinPool pool;   // the containing pool (may be null)
        final ForkJoinWorkerThread owner; // owning thread or null if shared
        volatile Thread parker;    // == owner during call to park; else null
        volatile ForkJoinTask<?> currentJoin;  // task being joined in awaitJoin
        ForkJoinTask<?> currentSteal; // current non-local task being executed

        WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode,
                  int seed) {
            this.pool = pool;
            this.owner = owner;
685 686
            this.mode = (short)mode;
            this.hint = seed; // store initial seed for runWorker
687 688 689
            // Place indices in the center of array (that is not yet allocated)
            base = top = INITIAL_QUEUE_CAPACITY >>> 1;
        }
D
dl 已提交
690

691 692 693 694 695 696 697 698
        /**
         * Returns the approximate number of tasks in the queue.
         */
        final int queueSize() {
            int n = base - top;       // non-owner callers must read base first
            return (n >= 0) ? 0 : -n; // ignore transient negative
        }

699
        /**
700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719
         * Provides a more accurate estimate of whether this queue has
         * any tasks than does queueSize, by checking whether a
         * near-empty queue has at least one unclaimed task.
         */
        final boolean isEmpty() {
            ForkJoinTask<?>[] a; int m, s;
            int n = base - (s = top);
            return (n >= 0 ||
                    (n == -1 &&
                     ((a = array) == null ||
                      (m = a.length - 1) < 0 ||
                      U.getObject
                      (a, (long)((m & (s - 1)) << ASHIFT) + ABASE) == null)));
        }

        /**
         * Pushes a task. Call only by owner in unshared queues.  (The
         * shared-queue version is embedded in method externalPush.)
         *
         * @param task the task. Caller must ensure non-null.
720
         * @throws RejectedExecutionException if array cannot be resized
721 722 723
         */
        final void push(ForkJoinTask<?> task) {
            ForkJoinTask<?>[] a; ForkJoinPool p;
724
            int s = top, n;
725
            if ((a = array) != null) {    // ignore if queue removed
726 727 728 729
                int m = a.length - 1;
                U.putOrderedObject(a, ((m & s) << ASHIFT) + ABASE, task);
                if ((n = (top = s + 1) - base) <= 2)
                    (p = pool).signalWork(p.workQueues, this);
730 731 732 733 734
                else if (n >= m)
                    growArray();
            }
        }

735
        /**
736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792
         * Initializes or doubles the capacity of array. Call either
         * by owner or with lock held -- it is OK for base, but not
         * top, to move while resizings are in progress.
         */
        final ForkJoinTask<?>[] growArray() {
            ForkJoinTask<?>[] oldA = array;
            int size = oldA != null ? oldA.length << 1 : INITIAL_QUEUE_CAPACITY;
            if (size > MAXIMUM_QUEUE_CAPACITY)
                throw new RejectedExecutionException("Queue capacity exceeded");
            int oldMask, t, b;
            ForkJoinTask<?>[] a = array = new ForkJoinTask<?>[size];
            if (oldA != null && (oldMask = oldA.length - 1) >= 0 &&
                (t = top) - (b = base) > 0) {
                int mask = size - 1;
                do {
                    ForkJoinTask<?> x;
                    int oldj = ((b & oldMask) << ASHIFT) + ABASE;
                    int j    = ((b &    mask) << ASHIFT) + ABASE;
                    x = (ForkJoinTask<?>)U.getObjectVolatile(oldA, oldj);
                    if (x != null &&
                        U.compareAndSwapObject(oldA, oldj, x, null))
                        U.putObjectVolatile(a, j, x);
                } while (++b != t);
            }
            return a;
        }

        /**
         * Takes next task, if one exists, in LIFO order.  Call only
         * by owner in unshared queues.
         */
        final ForkJoinTask<?> pop() {
            ForkJoinTask<?>[] a; ForkJoinTask<?> t; int m;
            if ((a = array) != null && (m = a.length - 1) >= 0) {
                for (int s; (s = top - 1) - base >= 0;) {
                    long j = ((m & s) << ASHIFT) + ABASE;
                    if ((t = (ForkJoinTask<?>)U.getObject(a, j)) == null)
                        break;
                    if (U.compareAndSwapObject(a, j, t, null)) {
                        top = s;
                        return t;
                    }
                }
            }
            return null;
        }

        /**
         * Takes a task in FIFO order if b is base of queue and a task
         * can be claimed without contention. Specialized versions
         * appear in ForkJoinPool methods scan and tryHelpStealer.
         */
        final ForkJoinTask<?> pollAt(int b) {
            ForkJoinTask<?> t; ForkJoinTask<?>[] a;
            if ((a = array) != null) {
                int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
                if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) != null &&
793 794
                    base == b && U.compareAndSwapObject(a, j, t, null)) {
                    U.putOrderedInt(this, QBASE, b + 1);
795 796 797 798 799 800 801 802 803 804 805 806 807 808 809
                    return t;
                }
            }
            return null;
        }

        /**
         * Takes next task, if one exists, in FIFO order.
         */
        final ForkJoinTask<?> poll() {
            ForkJoinTask<?>[] a; int b; ForkJoinTask<?> t;
            while ((b = base) - top < 0 && (a = array) != null) {
                int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
                t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
                if (t != null) {
810 811
                    if (U.compareAndSwapObject(a, j, t, null)) {
                        U.putOrderedInt(this, QBASE, b + 1);
812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870
                        return t;
                    }
                }
                else if (base == b) {
                    if (b + 1 == top)
                        break;
                    Thread.yield(); // wait for lagging update (very rare)
                }
            }
            return null;
        }

        /**
         * Takes next task, if one exists, in order specified by mode.
         */
        final ForkJoinTask<?> nextLocalTask() {
            return mode == 0 ? pop() : poll();
        }

        /**
         * Returns next task, if one exists, in order specified by mode.
         */
        final ForkJoinTask<?> peek() {
            ForkJoinTask<?>[] a = array; int m;
            if (a == null || (m = a.length - 1) < 0)
                return null;
            int i = mode == 0 ? top - 1 : base;
            int j = ((i & m) << ASHIFT) + ABASE;
            return (ForkJoinTask<?>)U.getObjectVolatile(a, j);
        }

        /**
         * Pops the given task only if it is at the current top.
         * (A shared version is available only via FJP.tryExternalUnpush)
         */
        final boolean tryUnpush(ForkJoinTask<?> t) {
            ForkJoinTask<?>[] a; int s;
            if ((a = array) != null && (s = top) != base &&
                U.compareAndSwapObject
                (a, (((a.length - 1) & --s) << ASHIFT) + ABASE, t, null)) {
                top = s;
                return true;
            }
            return false;
        }

        /**
         * Removes and cancels all known tasks, ignoring any exceptions.
         */
        final void cancelAll() {
            ForkJoinTask.cancelIgnoringExceptions(currentJoin);
            ForkJoinTask.cancelIgnoringExceptions(currentSteal);
            for (ForkJoinTask<?> t; (t = poll()) != null; )
                ForkJoinTask.cancelIgnoringExceptions(t);
        }

        // Specialized execution methods

        /**
871
         * Polls and runs tasks until empty.
872
         */
873 874 875
        final void pollAndExecAll() {
            for (ForkJoinTask<?> t; (t = poll()) != null;)
                t.doExec();
876 877 878
        }

        /**
879 880
         * Executes a top-level task and any local tasks remaining
         * after execution.
881
         */
882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901
        final void runTask(ForkJoinTask<?> task) {
            if ((currentSteal = task) != null) {
                task.doExec();
                ForkJoinTask<?>[] a = array;
                int md = mode;
                ++nsteals;
                currentSteal = null;
                if (md != 0)
                    pollAndExecAll();
                else if (a != null) {
                    int s, m = a.length - 1;
                    ForkJoinTask<?> t;
                    while ((s = top - 1) - base >= 0 &&
                           (t = (ForkJoinTask<?>)U.getAndSetObject
                            (a, ((m & s) << ASHIFT) + ABASE, null)) != null) {
                        top = s;
                        t.doExec();
                    }
                }
            }
902 903 904 905 906 907 908
        }

        /**
         * If present, removes from queue and executes the given task,
         * or any other cancelled task. Returns (true) on any CAS
         * or consistency check failure so caller can retry.
         *
909
         * @return false if no progress can be made, else true
910 911
         */
        final boolean tryRemoveAndExec(ForkJoinTask<?> task) {
912
            boolean stat;
913
            ForkJoinTask<?>[] a; int m, s, b, n;
914
            if (task != null && (a = array) != null && (m = a.length - 1) >= 0 &&
915
                (n = (s = top) - (b = base)) > 0) {
916 917
                boolean removed = false, empty = true;
                stat = true;
918
                for (ForkJoinTask<?> t;;) {           // traverse from s to b
919 920
                    long j = ((--s & m) << ASHIFT) + ABASE;
                    t = (ForkJoinTask<?>)U.getObject(a, j);
921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947
                    if (t == null)                    // inconsistent length
                        break;
                    else if (t == task) {
                        if (s + 1 == top) {           // pop
                            if (!U.compareAndSwapObject(a, j, task, null))
                                break;
                            top = s;
                            removed = true;
                        }
                        else if (base == b)           // replace with proxy
                            removed = U.compareAndSwapObject(a, j, task,
                                                             new EmptyTask());
                        break;
                    }
                    else if (t.status >= 0)
                        empty = false;
                    else if (s + 1 == top) {          // pop and throw away
                        if (U.compareAndSwapObject(a, j, t, null))
                            top = s;
                        break;
                    }
                    if (--n == 0) {
                        if (!empty && base == b)
                            stat = false;
                        break;
                    }
                }
948 949
                if (removed)
                    task.doExec();
950
            }
951 952
            else
                stat = false;
953 954 955 956
            return stat;
        }

        /**
957 958
         * Tries to poll for and execute the given task or any other
         * task in its CountedCompleter computation.
959
         */
960 961 962
        final boolean pollAndExecCC(CountedCompleter<?> root) {
            ForkJoinTask<?>[] a; int b; Object o; CountedCompleter<?> t, r;
            if ((b = base) - top < 0 && (a = array) != null) {
963
                long j = (((a.length - 1) & b) << ASHIFT) + ABASE;
964 965 966 967 968 969 970 971 972 973
                if ((o = U.getObjectVolatile(a, j)) == null)
                    return true; // retry
                if (o instanceof CountedCompleter) {
                    for (t = (CountedCompleter<?>)o, r = t;;) {
                        if (r == root) {
                            if (base == b &&
                                U.compareAndSwapObject(a, j, t, null)) {
                                U.putOrderedInt(this, QBASE, b + 1);
                                t.doExec();
                            }
974 975
                            return true;
                        }
976 977
                        else if ((r = r.completer) == null)
                            break; // not part of root computation
978 979 980 981 982 983 984
                    }
                }
            }
            return false;
        }

        /**
985 986
         * Tries to pop and execute the given task or any other task
         * in its CountedCompleter computation.
987
         */
988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009
        final boolean externalPopAndExecCC(CountedCompleter<?> root) {
            ForkJoinTask<?>[] a; int s; Object o; CountedCompleter<?> t, r;
            if (base - (s = top) < 0 && (a = array) != null) {
                long j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE;
                if ((o = U.getObject(a, j)) instanceof CountedCompleter) {
                    for (t = (CountedCompleter<?>)o, r = t;;) {
                        if (r == root) {
                            if (U.compareAndSwapInt(this, QLOCK, 0, 1)) {
                                if (top == s && array == a &&
                                    U.compareAndSwapObject(a, j, t, null)) {
                                    top = s - 1;
                                    qlock = 0;
                                    t.doExec();
                                }
                                else
                                    qlock = 0;
                            }
                            return true;
                        }
                        else if ((r = r.completer) == null)
                            break;
                    }
1010 1011
                }
            }
1012
            return false;
1013 1014 1015
        }

        /**
1016
         * Internal version
1017
         */
1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034
        final boolean internalPopAndExecCC(CountedCompleter<?> root) {
            ForkJoinTask<?>[] a; int s; Object o; CountedCompleter<?> t, r;
            if (base - (s = top) < 0 && (a = array) != null) {
                long j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE;
                if ((o = U.getObject(a, j)) instanceof CountedCompleter) {
                    for (t = (CountedCompleter<?>)o, r = t;;) {
                        if (r == root) {
                            if (U.compareAndSwapObject(a, j, t, null)) {
                                top = s - 1;
                                t.doExec();
                            }
                            return true;
                        }
                        else if ((r = r.completer) == null)
                            break;
                    }
                }
1035
            }
1036
            return false;
1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052
        }

        /**
         * Returns true if owned and not known to be blocked.
         */
        final boolean isApparentlyUnblocked() {
            Thread wt; Thread.State s;
            return (eventCount >= 0 &&
                    (wt = owner) != null &&
                    (s = wt.getState()) != Thread.State.BLOCKED &&
                    s != Thread.State.WAITING &&
                    s != Thread.State.TIMED_WAITING);
        }

        // Unsafe mechanics
        private static final sun.misc.Unsafe U;
1053
        private static final long QBASE;
1054 1055 1056 1057 1058 1059 1060 1061
        private static final long QLOCK;
        private static final int ABASE;
        private static final int ASHIFT;
        static {
            try {
                U = sun.misc.Unsafe.getUnsafe();
                Class<?> k = WorkQueue.class;
                Class<?> ak = ForkJoinTask[].class;
1062 1063
                QBASE = U.objectFieldOffset
                    (k.getDeclaredField("base"));
1064 1065 1066
                QLOCK = U.objectFieldOffset
                    (k.getDeclaredField("qlock"));
                ABASE = U.arrayBaseOffset(ak);
1067 1068 1069 1070
                int scale = U.arrayIndexScale(ak);
                if ((scale & (scale - 1)) != 0)
                    throw new Error("data type scale not a power of two");
                ASHIFT = 31 - Integer.numberOfLeadingZeros(scale);
1071 1072 1073 1074
            } catch (Exception e) {
                throw new Error(e);
            }
        }
D
dl 已提交
1075 1076
    }

1077 1078
    // static fields (initialized in static initializer below)

D
dl 已提交
1079
    /**
1080 1081
     * Creates a new ForkJoinWorkerThread. This factory is used unless
     * overridden in ForkJoinPool constructors.
D
dl 已提交
1082
     */
1083 1084
    public static final ForkJoinWorkerThreadFactory
        defaultForkJoinWorkerThreadFactory;
D
dl 已提交
1085 1086

    /**
1087 1088
     * Permission required for callers of methods that may start or
     * kill threads.
D
dl 已提交
1089
     */
1090
    private static final RuntimePermission modifyThreadPermission;
D
dl 已提交
1091 1092

    /**
1093 1094 1095 1096
     * Common (static) pool. Non-null for public use unless a static
     * construction exception, but internal usages null-check on use
     * to paranoically avoid potential initialization circularities
     * as well as to simplify generated code.
D
dl 已提交
1097
     */
1098
    static final ForkJoinPool common;
D
dl 已提交
1099 1100

    /**
1101 1102
     * Common pool parallelism. To allow simpler use and management
     * when common pool threads are disabled, we allow the underlying
1103
     * common.parallelism field to be zero, but in that case still report
1104
     * parallelism as 1 to reflect resulting caller-runs mechanics.
D
dl 已提交
1105
     */
1106
    static final int commonParallelism;
D
dl 已提交
1107 1108

    /**
1109
     * Sequence number for creating workerNamePrefix.
D
dl 已提交
1110
     */
1111
    private static int poolNumberSequence;
D
dl 已提交
1112 1113

    /**
1114 1115
     * Returns the next sequence number. We don't expect this to
     * ever contend, so use simple builtin sync.
D
dl 已提交
1116
     */
1117 1118 1119 1120 1121
    private static final synchronized int nextPoolId() {
        return ++poolNumberSequence;
    }

    // static constants
D
dl 已提交
1122 1123

    /**
1124 1125 1126 1127 1128 1129
     * Initial timeout value (in nanoseconds) for the thread
     * triggering quiescence to park waiting for new work. On timeout,
     * the thread will instead try to shrink the number of
     * workers. The value should be large enough to avoid overly
     * aggressive shrinkage during most transient stalls (long GCs
     * etc).
D
dl 已提交
1130
     */
1131
    private static final long IDLE_TIMEOUT      = 2000L * 1000L * 1000L; // 2sec
D
dl 已提交
1132 1133

    /**
1134
     * Timeout value when there are more threads than parallelism level
D
dl 已提交
1135
     */
1136
    private static final long FAST_IDLE_TIMEOUT =  200L * 1000L * 1000L;
D
dl 已提交
1137 1138

    /**
1139
     * Tolerance for idle timeouts, to cope with timer undershoots
D
dl 已提交
1140
     */
1141
    private static final long TIMEOUT_SLOP = 2000000L;
D
dl 已提交
1142 1143

    /**
1144 1145 1146 1147 1148 1149
     * The maximum stolen->joining link depth allowed in method
     * tryHelpStealer.  Must be a power of two.  Depths for legitimate
     * chains are unbounded, but we use a fixed constant to avoid
     * (otherwise unchecked) cycles and to bound staleness of
     * traversal parameters at the expense of sometimes blocking when
     * we could be helping.
D
dl 已提交
1150
     */
1151
    private static final int MAX_HELP = 64;
D
dl 已提交
1152 1153

    /**
1154 1155
     * Increment for seed generators. See class ThreadLocal for
     * explanation.
D
dl 已提交
1156
     */
1157
    private static final int SEED_INCREMENT = 0x61c88647;
D
dl 已提交
1158

1159
    /*
1160 1161 1162
     * Bits and masks for control variables
     *
     * Field ctl is a long packed with:
1163
     * AC: Number of active running workers minus target parallelism (16 bits)
1164
     * TC: Number of total workers minus target parallelism (16 bits)
1165 1166
     * ST: true if pool is terminating (1 bit)
     * EC: the wait count of top waiting thread (15 bits)
1167
     * ID: poolIndex of top of Treiber stack of waiters (16 bits)
1168 1169 1170 1171 1172 1173 1174 1175
     *
     * When convenient, we can extract the upper 32 bits of counts and
     * the lower 32 bits of queue state, u = (int)(ctl >>> 32) and e =
     * (int)ctl.  The ec field is never accessed alone, but always
     * together with id and st. The offsets of counts by the target
     * parallelism and the positionings of fields makes it possible to
     * perform the most common checks via sign tests of fields: When
     * ac is negative, there are not enough active workers, when tc is
1176
     * negative, there are not enough total workers, and when e is
1177 1178 1179
     * negative, the pool is terminating.  To deal with these possibly
     * negative fields, we use casts in and out of "short" and/or
     * signed shifts to maintain signedness.
1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194
     *
     * When a thread is queued (inactivated), its eventCount field is
     * set negative, which is the only way to tell if a worker is
     * prevented from executing tasks, even though it must continue to
     * scan for them to avoid queuing races. Note however that
     * eventCount updates lag releases so usage requires care.
     *
     * Field plock is an int packed with:
     * SHUTDOWN: true if shutdown is enabled (1 bit)
     * SEQ:  a sequence lock, with PL_LOCK bit set if locked (30 bits)
     * SIGNAL: set when threads may be waiting on the lock (1 bit)
     *
     * The sequence number enables simple consistency checks:
     * Staleness of read-only operations on the workQueues array can
     * be checked by comparing plock before vs after the reads.
1195 1196 1197 1198 1199 1200 1201 1202 1203
     */

    // bit positions/shifts for fields
    private static final int  AC_SHIFT   = 48;
    private static final int  TC_SHIFT   = 32;
    private static final int  ST_SHIFT   = 31;
    private static final int  EC_SHIFT   = 16;

    // bounds
1204 1205 1206 1207
    private static final int  SMASK      = 0xffff;  // short bits
    private static final int  MAX_CAP    = 0x7fff;  // max #workers - 1
    private static final int  EVENMASK   = 0xfffe;  // even short bits
    private static final int  SQMASK     = 0x007e;  // max 64 (even) slots
1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228
    private static final int  SHORT_SIGN = 1 << 15;
    private static final int  INT_SIGN   = 1 << 31;

    // masks
    private static final long STOP_BIT   = 0x0001L << ST_SHIFT;
    private static final long AC_MASK    = ((long)SMASK) << AC_SHIFT;
    private static final long TC_MASK    = ((long)SMASK) << TC_SHIFT;

    // units for incrementing and decrementing
    private static final long TC_UNIT    = 1L << TC_SHIFT;
    private static final long AC_UNIT    = 1L << AC_SHIFT;

    // masks and units for dealing with u = (int)(ctl >>> 32)
    private static final int  UAC_SHIFT  = AC_SHIFT - 32;
    private static final int  UTC_SHIFT  = TC_SHIFT - 32;
    private static final int  UAC_MASK   = SMASK << UAC_SHIFT;
    private static final int  UTC_MASK   = SMASK << UTC_SHIFT;
    private static final int  UAC_UNIT   = 1 << UAC_SHIFT;
    private static final int  UTC_UNIT   = 1 << UTC_SHIFT;

    // masks and units for dealing with e = (int)ctl
1229 1230
    private static final int E_MASK      = 0x7fffffff; // no STOP_BIT
    private static final int E_SEQ       = 1 << EC_SHIFT;
D
dl 已提交
1231

1232 1233 1234 1235 1236
    // plock bits
    private static final int SHUTDOWN    = 1 << 31;
    private static final int PL_LOCK     = 2;
    private static final int PL_SIGNAL   = 1;
    private static final int PL_SPINS    = 1 << 8;
D
dl 已提交
1237

1238 1239 1240 1241
    // access mode for WorkQueue
    static final int LIFO_QUEUE          =  0;
    static final int FIFO_QUEUE          =  1;
    static final int SHARED_QUEUE        = -1;
D
dl 已提交
1242

1243 1244 1245 1246 1247
    // Instance fields
    volatile long stealCount;                  // collects worker counts
    volatile long ctl;                         // main pool control
    volatile int plock;                        // shutdown status and seqLock
    volatile int indexSeed;                    // worker/submitter index seed
1248 1249
    final short parallelism;                   // parallelism level
    final short mode;                          // LIFO/FIFO
1250 1251
    WorkQueue[] workQueues;                    // main registry
    final ForkJoinWorkerThreadFactory factory;
1252
    final UncaughtExceptionHandler ueh;        // per-worker UEH
1253
    final String workerNamePrefix;             // to create worker name string
1254

1255
    /**
1256 1257
     * Acquires the plock lock to protect worker array and related
     * updates. This method is called only if an initial CAS on plock
1258
     * fails. This acts as a spinlock for normal cases, but falls back
1259 1260 1261 1262 1263
     * to builtin monitor to block when (rarely) needed. This would be
     * a terrible idea for a highly contended lock, but works fine as
     * a more conservative alternative to a pure spinlock.
     */
    private int acquirePlock() {
1264
        int spins = PL_SPINS, ps, nps;
1265 1266 1267 1268 1269
        for (;;) {
            if (((ps = plock) & PL_LOCK) == 0 &&
                U.compareAndSwapInt(this, PLOCK, ps, nps = ps + PL_LOCK))
                return nps;
            else if (spins >= 0) {
1270
                if (ThreadLocalRandom.nextSecondarySeed() >= 0)
1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290
                    --spins;
            }
            else if (U.compareAndSwapInt(this, PLOCK, ps, ps | PL_SIGNAL)) {
                synchronized (this) {
                    if ((plock & PL_SIGNAL) != 0) {
                        try {
                            wait();
                        } catch (InterruptedException ie) {
                            try {
                                Thread.currentThread().interrupt();
                            } catch (SecurityException ignore) {
                            }
                        }
                    }
                    else
                        notifyAll();
                }
            }
        }
    }
D
dl 已提交
1291 1292

    /**
1293 1294
     * Unlocks and signals any thread waiting for plock. Called only
     * when CAS of seq value for unlock fails.
D
dl 已提交
1295
     */
1296 1297 1298 1299
    private void releasePlock(int ps) {
        plock = ps;
        synchronized (this) { notifyAll(); }
    }
D
dl 已提交
1300

1301 1302 1303 1304 1305
    /**
     * Tries to create and start one worker if fewer than target
     * parallelism level exist. Adjusts counts etc on failure.
     */
    private void tryAddWorker() {
1306
        long c; int u, e;
1307
        while ((u = (int)((c = ctl) >>> 32)) < 0 &&
1308 1309 1310
               (u & SHORT_SIGN) != 0 && (e = (int)c) >= 0) {
            long nc = ((long)(((u + UTC_UNIT) & UTC_MASK) |
                              ((u + UAC_UNIT) & UAC_MASK)) << 32) | (long)e;
1311 1312 1313 1314 1315 1316 1317 1318 1319 1320
            if (U.compareAndSwapLong(this, CTL, c, nc)) {
                ForkJoinWorkerThreadFactory fac;
                Throwable ex = null;
                ForkJoinWorkerThread wt = null;
                try {
                    if ((fac = factory) != null &&
                        (wt = fac.newThread(this)) != null) {
                        wt.start();
                        break;
                    }
1321 1322
                } catch (Throwable rex) {
                    ex = rex;
1323 1324 1325 1326
                }
                deregisterWorker(wt, ex);
                break;
            }
D
dl 已提交
1327 1328 1329
        }
    }

1330
    //  Registering and deregistering workers
D
dl 已提交
1331

D
dl 已提交
1332
    /**
1333 1334 1335 1336 1337 1338 1339 1340
     * Callback from ForkJoinWorkerThread to establish and record its
     * WorkQueue. To avoid scanning bias due to packing entries in
     * front of the workQueues array, we treat the array as a simple
     * power-of-two hash table using per-thread seed as hash,
     * expanding as needed.
     *
     * @param wt the worker thread
     * @return the worker's queue
D
dl 已提交
1341
     */
1342
    final WorkQueue registerWorker(ForkJoinWorkerThread wt) {
1343
        UncaughtExceptionHandler handler; WorkQueue[] ws; int s, ps;
1344 1345 1346 1347 1348 1349
        wt.setDaemon(true);
        if ((handler = ueh) != null)
            wt.setUncaughtExceptionHandler(handler);
        do {} while (!U.compareAndSwapInt(this, INDEXSEED, s = indexSeed,
                                          s += SEED_INCREMENT) ||
                     s == 0); // skip 0
1350
        WorkQueue w = new WorkQueue(this, wt, mode, s);
1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368
        if (((ps = plock) & PL_LOCK) != 0 ||
            !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
            ps = acquirePlock();
        int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
        try {
            if ((ws = workQueues) != null) {    // skip if shutting down
                int n = ws.length, m = n - 1;
                int r = (s << 1) | 1;           // use odd-numbered indices
                if (ws[r &= m] != null) {       // collision
                    int probes = 0;             // step by approx half size
                    int step = (n <= 4) ? 2 : ((n >>> 1) & EVENMASK) + 2;
                    while (ws[r = (r + step) & m] != null) {
                        if (++probes >= n) {
                            workQueues = ws = Arrays.copyOf(ws, n <<= 1);
                            m = n - 1;
                            probes = 0;
                        }
                    }
1369
                }
1370 1371
                w.poolIndex = (short)r;
                w.eventCount = r; // volatile write orders
1372
                ws[r] = w;
1373
            }
1374 1375 1376 1377
        } finally {
            if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
                releasePlock(nps);
        }
1378
        wt.setName(workerNamePrefix.concat(Integer.toString(w.poolIndex >>> 1)));
1379 1380 1381 1382 1383 1384 1385 1386 1387
        return w;
    }

    /**
     * Final callback from terminating worker, as well as upon failure
     * to construct or start a worker.  Removes record of worker from
     * array, and adjusts counts. If pool is shutting down, tries to
     * complete termination.
     *
1388
     * @param wt the worker thread, or null if construction failed
1389 1390 1391 1392 1393 1394 1395
     * @param ex the exception causing failure, or null if none
     */
    final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
        WorkQueue w = null;
        if (wt != null && (w = wt.workQueue) != null) {
            int ps;
            w.qlock = -1;                // ensure set
1396
            U.getAndAddLong(this, STEALCOUNT, w.nsteals); // collect steals
1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408
            if (((ps = plock) & PL_LOCK) != 0 ||
                !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
                ps = acquirePlock();
            int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
            try {
                int idx = w.poolIndex;
                WorkQueue[] ws = workQueues;
                if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w)
                    ws[idx] = null;
            } finally {
                if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
                    releasePlock(nps);
D
dl 已提交
1409 1410
            }
        }
D
dl 已提交
1411

1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424
        long c;                          // adjust ctl counts
        do {} while (!U.compareAndSwapLong
                     (this, CTL, c = ctl, (((c - AC_UNIT) & AC_MASK) |
                                           ((c - TC_UNIT) & TC_MASK) |
                                           (c & ~(AC_MASK|TC_MASK)))));

        if (!tryTerminate(false, false) && w != null && w.array != null) {
            w.cancelAll();               // cancel remaining tasks
            WorkQueue[] ws; WorkQueue v; Thread p; int u, i, e;
            while ((u = (int)((c = ctl) >>> 32)) < 0 && (e = (int)c) >= 0) {
                if (e > 0) {             // activate or create replacement
                    if ((ws = workQueues) == null ||
                        (i = e & SMASK) >= ws.length ||
1425
                        (v = ws[i]) == null)
1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443
                        break;
                    long nc = (((long)(v.nextWait & E_MASK)) |
                               ((long)(u + UAC_UNIT) << 32));
                    if (v.eventCount != (e | INT_SIGN))
                        break;
                    if (U.compareAndSwapLong(this, CTL, c, nc)) {
                        v.eventCount = (e + E_SEQ) & E_MASK;
                        if ((p = v.parker) != null)
                            U.unpark(p);
                        break;
                    }
                }
                else {
                    if ((short)u < 0)
                        tryAddWorker();
                    break;
                }
            }
D
dl 已提交
1444
        }
1445 1446 1447 1448
        if (ex == null)                     // help clean refs on way out
            ForkJoinTask.helpExpungeStaleExceptions();
        else                                // rethrow
            ForkJoinTask.rethrow(ex);
D
dl 已提交
1449 1450
    }

1451
    // Submissions
1452

D
dl 已提交
1453
    /**
1454 1455 1456 1457 1458 1459 1460 1461
     * Unless shutting down, adds the given task to a submission queue
     * at submitter's current queue index (modulo submission
     * range). Only the most common path is directly handled in this
     * method. All others are relayed to fullExternalPush.
     *
     * @param task the task. Caller must ensure non-null.
     */
    final void externalPush(ForkJoinTask<?> task) {
1462 1463 1464 1465 1466 1467
        WorkQueue q; int m, s, n, am; ForkJoinTask<?>[] a;
        int r = ThreadLocalRandom.getProbe();
        int ps = plock;
        WorkQueue[] ws = workQueues;
        if (ps > 0 && ws != null && (m = (ws.length - 1)) >= 0 &&
            (q = ws[m & r & SQMASK]) != null && r != 0 &&
1468
            U.compareAndSwapInt(q, QLOCK, 0, 1)) { // lock
1469 1470 1471
            if ((a = q.array) != null &&
                (am = a.length - 1) > (n = (s = q.top) - q.base)) {
                int j = ((am & s) << ASHIFT) + ABASE;
1472 1473 1474
                U.putOrderedObject(a, j, task);
                q.top = s + 1;                     // push on to deque
                q.qlock = 0;
1475 1476
                if (n <= 1)
                    signalWork(ws, q);
1477 1478 1479 1480 1481 1482 1483 1484 1485 1486
                return;
            }
            q.qlock = 0;
        }
        fullExternalPush(task);
    }

    /**
     * Full version of externalPush. This method is called, among
     * other times, upon the first submission of the first task to the
1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499
     * pool, so must perform secondary initialization.  It also
     * detects first submission by an external thread by looking up
     * its ThreadLocal, and creates a new shared queue if the one at
     * index if empty or contended. The plock lock body must be
     * exception-free (so no try/finally) so we optimistically
     * allocate new queues outside the lock and throw them away if
     * (very rarely) not needed.
     *
     * Secondary initialization occurs when plock is zero, to create
     * workQueue array and set plock to a valid value.  This lock body
     * must also be exception-free. Because the plock seq value can
     * eventually wrap around zero, this method harmlessly fails to
     * reinitialize if workQueues exists, while still advancing plock.
1500 1501
     */
    private void fullExternalPush(ForkJoinTask<?> task) {
1502 1503 1504 1505 1506 1507
        int r;
        if ((r = ThreadLocalRandom.getProbe()) == 0) {
            ThreadLocalRandom.localInit();
            r = ThreadLocalRandom.getProbe();
        }
        for (;;) {
1508
            WorkQueue[] ws; WorkQueue q; int ps, m, k;
1509 1510
            boolean move = false;
            if ((ps = plock) < 0)
1511 1512
                throw new RejectedExecutionException();
            else if (ps == 0 || (ws = workQueues) == null ||
1513
                     (m = ws.length - 1) < 0) { // initialize workQueues
1514
                int p = parallelism;            // find power of two table size
1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528
                int n = (p > 1) ? p - 1 : 1;    // ensure at least 2 slots
                n |= n >>> 1; n |= n >>> 2;  n |= n >>> 4;
                n |= n >>> 8; n |= n >>> 16; n = (n + 1) << 1;
                WorkQueue[] nws = ((ws = workQueues) == null || ws.length == 0 ?
                                   new WorkQueue[n] : null);
                if (((ps = plock) & PL_LOCK) != 0 ||
                    !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
                    ps = acquirePlock();
                if (((ws = workQueues) == null || ws.length == 0) && nws != null)
                    workQueues = nws;
                int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
                if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
                    releasePlock(nps);
            }
1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545
            else if ((q = ws[k = r & m & SQMASK]) != null) {
                if (q.qlock == 0 && U.compareAndSwapInt(q, QLOCK, 0, 1)) {
                    ForkJoinTask<?>[] a = q.array;
                    int s = q.top;
                    boolean submitted = false;
                    try {                      // locked version of push
                        if ((a != null && a.length > s + 1 - q.base) ||
                            (a = q.growArray()) != null) {   // must presize
                            int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
                            U.putOrderedObject(a, j, task);
                            q.top = s + 1;
                            submitted = true;
                        }
                    } finally {
                        q.qlock = 0;  // unlock
                    }
                    if (submitted) {
1546
                        signalWork(ws, q);
1547 1548
                        return;
                    }
1549
                }
1550
                move = true; // move on failure
D
dl 已提交
1551
            }
1552 1553
            else if (((ps = plock) & PL_LOCK) == 0) { // create new queue
                q = new WorkQueue(this, null, SHARED_QUEUE, r);
1554
                q.poolIndex = (short)k;
1555 1556 1557 1558 1559 1560 1561 1562
                if (((ps = plock) & PL_LOCK) != 0 ||
                    !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
                    ps = acquirePlock();
                if ((ws = workQueues) != null && k < ws.length && ws[k] == null)
                    ws[k] = q;
                int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN);
                if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
                    releasePlock(nps);
1563 1564
            }
            else
1565 1566 1567
                move = true; // move if busy
            if (move)
                r = ThreadLocalRandom.advanceProbe(r);
1568
        }
1569 1570 1571 1572 1573 1574 1575 1576 1577
    }

    // Maintaining ctl counts

    /**
     * Increments active count; mainly called upon return from blocking.
     */
    final void incrementActiveCount() {
        long c;
1578 1579 1580
        do {} while (!U.compareAndSwapLong
                     (this, CTL, c = ctl, ((c & ~AC_MASK) |
                                           ((c & AC_MASK) + AC_UNIT))));
1581 1582 1583 1584 1585
    }

    /**
     * Tries to create or activate a worker if too few are active.
     *
1586 1587
     * @param ws the worker array to use to find signallees
     * @param q if non-null, the queue holding tasks to be processed
1588
     */
1589 1590 1591 1592 1593 1594
    final void signalWork(WorkQueue[] ws, WorkQueue q) {
        for (;;) {
            long c; int e, u, i; WorkQueue w; Thread p;
            if ((u = (int)((c = ctl) >>> 32)) >= 0)
                break;
            if ((e = (int)c) <= 0) {
1595 1596 1597
                if ((short)u < 0)
                    tryAddWorker();
                break;
1598
            }
1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613
            if (ws == null || ws.length <= (i = e & SMASK) ||
                (w = ws[i]) == null)
                break;
            long nc = (((long)(w.nextWait & E_MASK)) |
                       ((long)(u + UAC_UNIT)) << 32);
            int ne = (e + E_SEQ) & E_MASK;
            if (w.eventCount == (e | INT_SIGN) &&
                U.compareAndSwapLong(this, CTL, c, nc)) {
                w.eventCount = ne;
                if ((p = w.parker) != null)
                    U.unpark(p);
                break;
            }
            if (q != null && q.base >= q.top)
                break;
D
dl 已提交
1614
        }
D
dl 已提交
1615 1616
    }

1617 1618 1619 1620 1621 1622 1623
    // Scanning for tasks

    /**
     * Top-level runloop for workers, called by ForkJoinWorkerThread.run.
     */
    final void runWorker(WorkQueue w) {
        w.growArray(); // allocate queue
1624 1625 1626
        for (int r = w.hint; scan(w, r) == 0; ) {
            r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // xorshift
        }
1627 1628
    }

D
dl 已提交
1629
    /**
1630
     * Scans for and, if found, runs one task, else possibly
1631 1632 1633 1634 1635 1636
     * inactivates the worker. This method operates on single reads of
     * volatile state and is designed to be re-invoked continuously,
     * in part because it returns upon detecting inconsistencies,
     * contention, or state changes that indicate possible success on
     * re-invocation.
     *
1637 1638 1639 1640 1641 1642 1643 1644 1645
     * The scan searches for tasks across queues starting at a random
     * index, checking each at least twice.  The scan terminates upon
     * either finding a non-empty queue, or completing the sweep. If
     * the worker is not inactivated, it takes and runs a task from
     * this queue. Otherwise, if not activated, it tries to activate
     * itself or some other worker by signalling. On failure to find a
     * task, returns (for retry) if pool state may have changed during
     * an empty scan, or tries to inactivate if active, else possibly
     * blocks or terminates via method awaitWork.
1646 1647
     *
     * @param w the worker (via its WorkQueue)
1648 1649
     * @param r a random seed
     * @return worker qlock status if would have waited, else 0
1650
     */
1651
    private final int scan(WorkQueue w, int r) {
1652
        WorkQueue[] ws; int m;
1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670
        long c = ctl;                            // for consistency check
        if ((ws = workQueues) != null && (m = ws.length - 1) >= 0 && w != null) {
            for (int j = m + m + 1, ec = w.eventCount;;) {
                WorkQueue q; int b, e; ForkJoinTask<?>[] a; ForkJoinTask<?> t;
                if ((q = ws[(r - j) & m]) != null &&
                    (b = q.base) - q.top < 0 && (a = q.array) != null) {
                    long i = (((a.length - 1) & b) << ASHIFT) + ABASE;
                    if ((t = ((ForkJoinTask<?>)
                              U.getObjectVolatile(a, i))) != null) {
                        if (ec < 0)
                            helpRelease(c, ws, w, q, b);
                        else if (q.base == b &&
                                 U.compareAndSwapObject(a, i, t, null)) {
                            U.putOrderedInt(q, QBASE, b + 1);
                            if ((b + 1) - q.top < 0)
                                signalWork(ws, q);
                            w.runTask(t);
                        }
1671
                    }
1672
                    break;
1673
                }
1674 1675 1676 1677 1678 1679
                else if (--j < 0) {
                    if ((ec | (e = (int)c)) < 0) // inactive or terminating
                        return awaitWork(w, c, ec);
                    else if (ctl == c) {         // try to inactivate and enqueue
                        long nc = (long)ec | ((c - AC_UNIT) & (AC_MASK|TC_MASK));
                        w.nextWait = e;
1680
                        w.eventCount = ec | INT_SIGN;
1681 1682
                        if (!U.compareAndSwapLong(this, CTL, c, nc))
                            w.eventCount = ec;   // back out
1683
                    }
1684
                    break;
1685
                }
D
dl 已提交
1686
            }
D
dl 已提交
1687
        }
1688
        return 0;
D
dl 已提交
1689 1690 1691
    }

    /**
1692 1693 1694 1695 1696 1697 1698 1699
     * A continuation of scan(), possibly blocking or terminating
     * worker w. Returns without blocking if pool state has apparently
     * changed since last invocation.  Also, if inactivating w has
     * caused the pool to become quiescent, checks for pool
     * termination, and, so long as this is not the only worker, waits
     * for event for up to a given duration.  On timeout, if ctl has
     * not changed, terminates the worker, which will in turn wake up
     * another worker to possibly repeat this process.
1700 1701
     *
     * @param w the calling worker
1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742
     * @param c the ctl value on entry to scan
     * @param ec the worker's eventCount on entry to scan
     */
    private final int awaitWork(WorkQueue w, long c, int ec) {
        int stat, ns; long parkTime, deadline;
        if ((stat = w.qlock) >= 0 && w.eventCount == ec && ctl == c &&
            !Thread.interrupted()) {
            int e = (int)c;
            int u = (int)(c >>> 32);
            int d = (u >> UAC_SHIFT) + parallelism; // active count

            if (e < 0 || (d <= 0 && tryTerminate(false, false)))
                stat = w.qlock = -1;          // pool is terminating
            else if ((ns = w.nsteals) != 0) { // collect steals and retry
                w.nsteals = 0;
                U.getAndAddLong(this, STEALCOUNT, (long)ns);
            }
            else {
                long pc = ((d > 0 || ec != (e | INT_SIGN)) ? 0L :
                           ((long)(w.nextWait & E_MASK)) | // ctl to restore
                           ((long)(u + UAC_UNIT)) << 32);
                if (pc != 0L) {               // timed wait if last waiter
                    int dc = -(short)(c >>> TC_SHIFT);
                    parkTime = (dc < 0 ? FAST_IDLE_TIMEOUT:
                                (dc + 1) * IDLE_TIMEOUT);
                    deadline = System.nanoTime() + parkTime - TIMEOUT_SLOP;
                }
                else
                    parkTime = deadline = 0L;
                if (w.eventCount == ec && ctl == c) {
                    Thread wt = Thread.currentThread();
                    U.putObject(wt, PARKBLOCKER, this);
                    w.parker = wt;            // emulate LockSupport.park
                    if (w.eventCount == ec && ctl == c)
                        U.park(false, parkTime);  // must recheck before park
                    w.parker = null;
                    U.putObject(wt, PARKBLOCKER, null);
                    if (parkTime != 0L && ctl == c &&
                        deadline - System.nanoTime() <= 0L &&
                        U.compareAndSwapLong(this, CTL, c, pc))
                        stat = w.qlock = -1;  // shrink pool
D
dl 已提交
1743 1744 1745
                }
            }
        }
1746
        return stat;
D
dl 已提交
1747 1748 1749
    }

    /**
1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769
     * Possibly releases (signals) a worker. Called only from scan()
     * when a worker with apparently inactive status finds a non-empty
     * queue. This requires revalidating all of the associated state
     * from caller.
     */
    private final void helpRelease(long c, WorkQueue[] ws, WorkQueue w,
                                   WorkQueue q, int b) {
        WorkQueue v; int e, i; Thread p;
        if (w != null && w.eventCount < 0 && (e = (int)c) > 0 &&
            ws != null && ws.length > (i = e & SMASK) &&
            (v = ws[i]) != null && ctl == c) {
            long nc = (((long)(v.nextWait & E_MASK)) |
                       ((long)((int)(c >>> 32) + UAC_UNIT)) << 32);
            int ne = (e + E_SEQ) & E_MASK;
            if (q != null && q.base == b && w.eventCount < 0 &&
                v.eventCount == (e | INT_SIGN) &&
                U.compareAndSwapLong(this, CTL, c, nc)) {
                v.eventCount = ne;
                if ((p = v.parker) != null)
                    U.unpark(p);
1770 1771
            }
        }
D
dl 已提交
1772 1773 1774
    }

    /**
1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793
     * Tries to locate and execute tasks for a stealer of the given
     * task, or in turn one of its stealers, Traces currentSteal ->
     * currentJoin links looking for a thread working on a descendant
     * of the given task and with a non-empty queue to steal back and
     * execute tasks from. The first call to this method upon a
     * waiting join will often entail scanning/search, (which is OK
     * because the joiner has nothing better to do), but this method
     * leaves hints in workers to speed up subsequent calls. The
     * implementation is very branchy to cope with potential
     * inconsistencies or loops encountering chains that are stale,
     * unknown, or so long that they are likely cyclic.
     *
     * @param joiner the joining worker
     * @param task the task to join
     * @return 0 if no progress can be made, negative if task
     * known complete, else positive
     */
    private int tryHelpStealer(WorkQueue joiner, ForkJoinTask<?> task) {
        int stat = 0, steps = 0;                    // bound to avoid cycles
1794 1795
        if (task != null && joiner != null &&
            joiner.base - joiner.top >= 0) {        // hoist checks
1796 1797 1798 1799 1800 1801 1802
            restart: for (;;) {
                ForkJoinTask<?> subtask = task;     // current target
                for (WorkQueue j = joiner, v;;) {   // v is stealer of subtask
                    WorkQueue[] ws; int m, s, h;
                    if ((s = task.status) < 0) {
                        stat = s;
                        break restart;
1803
                    }
1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821
                    if ((ws = workQueues) == null || (m = ws.length - 1) <= 0)
                        break restart;              // shutting down
                    if ((v = ws[h = (j.hint | 1) & m]) == null ||
                        v.currentSteal != subtask) {
                        for (int origin = h;;) {    // find stealer
                            if (((h = (h + 2) & m) & 15) == 1 &&
                                (subtask.status < 0 || j.currentJoin != subtask))
                                continue restart;   // occasional staleness check
                            if ((v = ws[h]) != null &&
                                v.currentSteal == subtask) {
                                j.hint = h;        // save hint
                                break;
                            }
                            if (h == origin)
                                break restart;      // cannot find stealer
                        }
                    }
                    for (;;) { // help stealer or descend to its stealer
1822
                        ForkJoinTask[] a; int b;
1823 1824 1825 1826 1827 1828 1829 1830 1831 1832
                        if (subtask.status < 0)     // surround probes with
                            continue restart;       //   consistency checks
                        if ((b = v.base) - v.top < 0 && (a = v.array) != null) {
                            int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
                            ForkJoinTask<?> t =
                                (ForkJoinTask<?>)U.getObjectVolatile(a, i);
                            if (subtask.status < 0 || j.currentJoin != subtask ||
                                v.currentSteal != subtask)
                                continue restart;   // stale
                            stat = 1;               // apparent progress
1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848
                            if (v.base == b) {
                                if (t == null)
                                    break restart;
                                if (U.compareAndSwapObject(a, i, t, null)) {
                                    U.putOrderedInt(v, QBASE, b + 1);
                                    ForkJoinTask<?> ps = joiner.currentSteal;
                                    int jt = joiner.top;
                                    do {
                                        joiner.currentSteal = t;
                                        t.doExec(); // clear local tasks too
                                    } while (task.status >= 0 &&
                                             joiner.top != jt &&
                                             (t = joiner.pop()) != null);
                                    joiner.currentSteal = ps;
                                    break restart;
                                }
1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863
                            }
                        }
                        else {                      // empty -- try to descend
                            ForkJoinTask<?> next = v.currentJoin;
                            if (subtask.status < 0 || j.currentJoin != subtask ||
                                v.currentSteal != subtask)
                                continue restart;   // stale
                            else if (next == null || ++steps == MAX_HELP)
                                break restart;      // dead-end or maybe cyclic
                            else {
                                subtask = next;
                                j = v;
                                break;
                            }
                        }
1864 1865
                    }
                }
1866
            }
D
dl 已提交
1867
        }
1868
        return stat;
D
dl 已提交
1869 1870 1871
    }

    /**
1872 1873 1874 1875
     * Analog of tryHelpStealer for CountedCompleters. Tries to steal
     * and run tasks within the target's computation.
     *
     * @param task the task to join
1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888
     * @param maxTasks the maximum number of other tasks to run
     */
    final int helpComplete(WorkQueue joiner, CountedCompleter<?> task,
                           int maxTasks) {
        WorkQueue[] ws; int m;
        int s = 0;
        if ((ws = workQueues) != null && (m = ws.length - 1) >= 0 &&
            joiner != null && task != null) {
            int j = joiner.poolIndex;
            int scans = m + m + 1;
            long c = 0L;              // for stability check
            for (int k = scans; ; j += 2) {
                WorkQueue q;
1889
                if ((s = task.status) < 0)
1890 1891 1892 1893
                    break;
                else if (joiner.internalPopAndExecCC(task)) {
                    if (--maxTasks <= 0) {
                        s = task.status;
1894
                        break;
1895 1896
                    }
                    k = scans;
1897
                }
1898
                else if ((s = task.status) < 0)
1899
                    break;
1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911
                else if ((q = ws[j & m]) != null && q.pollAndExecCC(task)) {
                    if (--maxTasks <= 0) {
                        s = task.status;
                        break;
                    }
                    k = scans;
                }
                else if (--k < 0) {
                    if (c == (c = ctl))
                        break;
                    k = scans;
                }
1912 1913
            }
        }
1914
        return s;
1915 1916 1917 1918 1919 1920 1921 1922
    }

    /**
     * Tries to decrement active count (sometimes implicitly) and
     * possibly release or create a compensating worker in preparation
     * for blocking. Fails on contention or termination. Otherwise,
     * adds a new thread if no idle workers are available and pool
     * may become starved.
1923 1924
     *
     * @param c the assumed ctl value
1925
     */
1926 1927 1928 1929 1930 1931 1932
    final boolean tryCompensate(long c) {
        WorkQueue[] ws = workQueues;
        int pc = parallelism, e = (int)c, m, tc;
        if (ws != null && (m = ws.length - 1) >= 0 && e >= 0 && ctl == c) {
            WorkQueue w = ws[e & m];
            if (e != 0 && w != null) {
                Thread p;
1933 1934
                long nc = ((long)(w.nextWait & E_MASK) |
                           (c & (AC_MASK|TC_MASK)));
1935 1936 1937 1938
                int ne = (e + E_SEQ) & E_MASK;
                if (w.eventCount == (e | INT_SIGN) &&
                    U.compareAndSwapLong(this, CTL, c, nc)) {
                    w.eventCount = ne;
1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966
                    if ((p = w.parker) != null)
                        U.unpark(p);
                    return true;   // replace with idle worker
                }
            }
            else if ((tc = (short)(c >>> TC_SHIFT)) >= 0 &&
                     (int)(c >> AC_SHIFT) + pc > 1) {
                long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK);
                if (U.compareAndSwapLong(this, CTL, c, nc))
                    return true;   // no compensation
            }
            else if (tc + pc < MAX_CAP) {
                long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
                if (U.compareAndSwapLong(this, CTL, c, nc)) {
                    ForkJoinWorkerThreadFactory fac;
                    Throwable ex = null;
                    ForkJoinWorkerThread wt = null;
                    try {
                        if ((fac = factory) != null &&
                            (wt = fac.newThread(this)) != null) {
                            wt.start();
                            return true;
                        }
                    } catch (Throwable rex) {
                        ex = rex;
                    }
                    deregisterWorker(wt, ex); // clean up and return false
                }
D
dl 已提交
1967 1968
            }
        }
1969
        return false;
D
dl 已提交
1970 1971 1972
    }

    /**
1973
     * Helps and/or blocks until the given task is done.
D
dl 已提交
1974
     *
1975 1976 1977 1978 1979 1980
     * @param joiner the joining worker
     * @param task the task
     * @return task status on exit
     */
    final int awaitJoin(WorkQueue joiner, ForkJoinTask<?> task) {
        int s = 0;
1981
        if (task != null && (s = task.status) >= 0 && joiner != null) {
1982 1983
            ForkJoinTask<?> prevJoin = joiner.currentJoin;
            joiner.currentJoin = task;
1984 1985 1986 1987 1988
            do {} while (joiner.tryRemoveAndExec(task) && // process local tasks
                         (s = task.status) >= 0);
            if (s >= 0 && (task instanceof CountedCompleter))
                s = helpComplete(joiner, (CountedCompleter<?>)task, Integer.MAX_VALUE);
            long cc = 0;        // for stability checks
1989
            while (s >= 0 && (s = task.status) >= 0) {
1990
                if ((s = tryHelpStealer(joiner, task)) == 0 &&
1991
                    (s = task.status) >= 0) {
1992 1993 1994
                    if (!tryCompensate(cc))
                        cc = ctl;
                    else {
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006
                        if (task.trySetSignal() && (s = task.status) >= 0) {
                            synchronized (task) {
                                if (task.status >= 0) {
                                    try {                // see ForkJoinTask
                                        task.wait();     //  for explanation
                                    } catch (InterruptedException ie) {
                                    }
                                }
                                else
                                    task.notifyAll();
                            }
                        }
2007
                        long c; // reactivate
2008
                        do {} while (!U.compareAndSwapLong
2009 2010 2011
                                     (this, CTL, c = ctl,
                                      ((c & ~AC_MASK) |
                                       ((c & AC_MASK) + AC_UNIT))));
2012
                    }
2013
                }
D
dl 已提交
2014
            }
2015
            joiner.currentJoin = prevJoin;
D
dl 已提交
2016
        }
2017
        return s;
D
dl 已提交
2018 2019
    }

D
dl 已提交
2020
    /**
2021 2022 2023 2024 2025 2026
     * Stripped-down variant of awaitJoin used by timed joins. Tries
     * to help join only while there is continuous progress. (Caller
     * will then enter a timed wait.)
     *
     * @param joiner the joining worker
     * @param task the task
D
dl 已提交
2027
     */
2028 2029 2030 2031 2032
    final void helpJoinOnce(WorkQueue joiner, ForkJoinTask<?> task) {
        int s;
        if (joiner != null && task != null && (s = task.status) >= 0) {
            ForkJoinTask<?> prevJoin = joiner.currentJoin;
            joiner.currentJoin = task;
2033 2034 2035 2036 2037
            do {} while (joiner.tryRemoveAndExec(task) && // process local tasks
                         (s = task.status) >= 0);
            if (s >= 0) {
                if (task instanceof CountedCompleter)
                    helpComplete(joiner, (CountedCompleter<?>)task, Integer.MAX_VALUE);
2038 2039 2040 2041
                do {} while (task.status >= 0 &&
                             tryHelpStealer(joiner, task) > 0);
            }
            joiner.currentJoin = prevJoin;
2042
        }
2043 2044 2045
    }

    /**
2046
     * Returns a (probably) non-empty steal queue, if one is found
2047 2048
     * during a scan, else null.  This method must be retried by
     * caller if, by the time it tries to use the queue, it is empty.
2049
     */
2050 2051
    private WorkQueue findNonEmptyStealQueue() {
        int r = ThreadLocalRandom.nextSecondarySeed();
2052 2053 2054 2055
        for (;;) {
            int ps = plock, m; WorkQueue[] ws; WorkQueue q;
            if ((ws = workQueues) != null && (m = ws.length - 1) >= 0) {
                for (int j = (m + 1) << 2; j >= 0; --j) {
2056
                    if ((q = ws[(((r - j) << 1) | 1) & m]) != null &&
2057 2058
                        q.base - q.top < 0)
                        return q;
2059 2060
                }
            }
2061 2062
            if (plock == ps)
                return null;
D
dl 已提交
2063 2064
        }
    }
D
dl 已提交
2065 2066

    /**
2067 2068 2069 2070 2071 2072
     * Runs tasks until {@code isQuiescent()}. We piggyback on
     * active count ctl maintenance, but rather than blocking
     * when tasks cannot be found, we rescan until all others cannot
     * find tasks either.
     */
    final void helpQuiescePool(WorkQueue w) {
2073
        ForkJoinTask<?> ps = w.currentSteal;
2074
        for (boolean active = true;;) {
2075
            long c; WorkQueue q; ForkJoinTask<?> t; int b;
2076
            while ((t = w.nextLocalTask()) != null)
2077
                t.doExec();
2078
            if ((q = findNonEmptyStealQueue()) != null) {
2079 2080 2081
                if (!active) {      // re-establish active count
                    active = true;
                    do {} while (!U.compareAndSwapLong
2082 2083 2084
                                 (this, CTL, c = ctl,
                                  ((c & ~AC_MASK) |
                                   ((c & AC_MASK) + AC_UNIT))));
2085
                }
2086
                if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null) {
2087 2088
                    (w.currentSteal = t).doExec();
                    w.currentSteal = ps;
2089
                }
2090
            }
2091
            else if (active) {       // decrement active count without queuing
2092 2093 2094
                long nc = ((c = ctl) & ~AC_MASK) | ((c & AC_MASK) - AC_UNIT);
                if ((int)(nc >> AC_SHIFT) + parallelism == 0)
                    break;          // bypass decrement-then-increment
2095
                if (U.compareAndSwapLong(this, CTL, c, nc))
2096 2097
                    active = false;
            }
2098 2099 2100 2101 2102
            else if ((int)((c = ctl) >> AC_SHIFT) + parallelism <= 0 &&
                     U.compareAndSwapLong
                     (this, CTL, c, ((c & ~AC_MASK) |
                                     ((c & AC_MASK) + AC_UNIT))))
                break;
2103 2104 2105 2106
        }
    }

    /**
2107
     * Gets and removes a local or stolen task for the given worker.
2108
     *
2109
     * @return a task, if available
2110
     */
2111 2112 2113 2114 2115
    final ForkJoinTask<?> nextTaskFor(WorkQueue w) {
        for (ForkJoinTask<?> t;;) {
            WorkQueue q; int b;
            if ((t = w.nextLocalTask()) != null)
                return t;
2116
            if ((q = findNonEmptyStealQueue()) == null)
2117
                return null;
2118
            if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
2119
                return t;
2120 2121 2122 2123
        }
    }

    /**
2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147
     * Returns a cheap heuristic guide for task partitioning when
     * programmers, frameworks, tools, or languages have little or no
     * idea about task granularity.  In essence by offering this
     * method, we ask users only about tradeoffs in overhead vs
     * expected throughput and its variance, rather than how finely to
     * partition tasks.
     *
     * In a steady state strict (tree-structured) computation, each
     * thread makes available for stealing enough tasks for other
     * threads to remain active. Inductively, if all threads play by
     * the same rules, each thread should make available only a
     * constant number of tasks.
     *
     * The minimum useful constant is just 1. But using a value of 1
     * would require immediate replenishment upon each steal to
     * maintain enough tasks, which is infeasible.  Further,
     * partitionings/granularities of offered tasks should minimize
     * steal rates, which in general means that threads nearer the top
     * of computation tree should generate more than those nearer the
     * bottom. In perfect steady state, each thread is at
     * approximately the same level of computation tree. However,
     * producing extra tasks amortizes the uncertainty of progress and
     * diffusion assumptions.
     *
2148
     * So, users will want to use values larger (but not much larger)
2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171
     * than 1 to both smooth over transient shortages and hedge
     * against uneven progress; as traded off against the cost of
     * extra task overhead. We leave the user to pick a threshold
     * value to compare with the results of this call to guide
     * decisions, but recommend values such as 3.
     *
     * When all threads are active, it is on average OK to estimate
     * surplus strictly locally. In steady-state, if one thread is
     * maintaining say 2 surplus tasks, then so are others. So we can
     * just use estimated queue length.  However, this strategy alone
     * leads to serious mis-estimates in some non-steady-state
     * conditions (ramp-up, ramp-down, other stalls). We can detect
     * many of these by further considering the number of "idle"
     * threads, that are known to have zero queued tasks, so
     * compensate by a factor of (#idle/#active) threads.
     *
     * Note: The approximation of #busy workers as #active workers is
     * not very good under current signalling scheme, and should be
     * improved.
     */
    static int getSurplusQueuedTaskCount() {
        Thread t; ForkJoinWorkerThread wt; ForkJoinPool pool; WorkQueue q;
        if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread)) {
2172
            int p = (pool = (wt = (ForkJoinWorkerThread)t).pool).parallelism;
2173 2174 2175 2176 2177 2178 2179
            int n = (q = wt.workQueue).top - q.base;
            int a = (int)(pool.ctl >> AC_SHIFT) + p;
            return n - (a > (p >>>= 1) ? 0 :
                        a > (p >>>= 1) ? 1 :
                        a > (p >>>= 1) ? 2 :
                        a > (p >>>= 1) ? 4 :
                        8);
D
dl 已提交
2180
        }
2181
        return 0;
D
dl 已提交
2182 2183
    }

2184
    //  Termination
2185

D
dl 已提交
2186
    /**
2187 2188 2189 2190 2191 2192 2193
     * Possibly initiates and/or completes termination.  The caller
     * triggering termination runs three passes through workQueues:
     * (0) Setting termination status, followed by wakeups of queued
     * workers; (1) cancelling all tasks; (2) interrupting lagging
     * threads (likely in external tasks, but possibly also blocked in
     * joins).  Each pass repeats previous steps because of potential
     * lagging thread creation.
D
dl 已提交
2194
     *
D
dl 已提交
2195
     * @param now if true, unconditionally terminate, else only
2196 2197
     * if no work and no active workers
     * @param enable if true, enable shutdown when next possible
D
dl 已提交
2198 2199
     * @return true if now terminating or terminated
     */
2200
    private boolean tryTerminate(boolean now, boolean enable) {
2201
        int ps;
2202
        if (this == common)                        // cannot shut down
2203
            return false;
2204 2205 2206 2207 2208 2209 2210 2211 2212 2213
        if ((ps = plock) >= 0) {                   // enable by setting plock
            if (!enable)
                return false;
            if ((ps & PL_LOCK) != 0 ||
                !U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK))
                ps = acquirePlock();
            int nps = ((ps + PL_LOCK) & ~SHUTDOWN) | SHUTDOWN;
            if (!U.compareAndSwapInt(this, PLOCK, ps, nps))
                releasePlock(nps);
        }
2214
        for (long c;;) {
2215
            if (((c = ctl) & STOP_BIT) != 0) {     // already terminating
2216
                if ((short)(c >>> TC_SHIFT) + parallelism <= 0) {
2217
                    synchronized (this) {
2218
                        notifyAll();               // signal when 0 workers
2219
                    }
2220
                }
2221
                return true;
2222
            }
2223 2224
            if (!now) {                            // check if idle & no tasks
                WorkQueue[] ws; WorkQueue w;
2225
                if ((int)(c >> AC_SHIFT) + parallelism > 0)
2226
                    return false;
2227 2228
                if ((ws = workQueues) != null) {
                    for (int i = 0; i < ws.length; ++i) {
2229 2230 2231 2232 2233
                        if ((w = ws[i]) != null &&
                            (!w.isEmpty() ||
                             ((i & 1) != 0 && w.eventCount >= 0))) {
                            signalWork(ws, w);
                            return false;
2234
                        }
2235 2236 2237 2238 2239
                    }
                }
            }
            if (U.compareAndSwapLong(this, CTL, c, c | STOP_BIT)) {
                for (int pass = 0; pass < 3; ++pass) {
2240 2241
                    WorkQueue[] ws; WorkQueue w; Thread wt;
                    if ((ws = workQueues) != null) {
2242 2243 2244 2245 2246 2247 2248 2249 2250 2251
                        int n = ws.length;
                        for (int i = 0; i < n; ++i) {
                            if ((w = ws[i]) != null) {
                                w.qlock = -1;
                                if (pass > 0) {
                                    w.cancelAll();
                                    if (pass > 1 && (wt = w.owner) != null) {
                                        if (!wt.isInterrupted()) {
                                            try {
                                                wt.interrupt();
2252
                                            } catch (Throwable ignore) {
2253 2254 2255 2256
                                            }
                                        }
                                        U.unpark(wt);
                                    }
2257
                                }
D
dl 已提交
2258 2259
                            }
                        }
2260 2261 2262
                        // Wake up workers parked on event queue
                        int i, e; long cc; Thread p;
                        while ((e = (int)(cc = ctl) & E_MASK) != 0 &&
2263
                               (i = e & SMASK) < n && i >= 0 &&
2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275
                               (w = ws[i]) != null) {
                            long nc = ((long)(w.nextWait & E_MASK) |
                                       ((cc + AC_UNIT) & AC_MASK) |
                                       (cc & (TC_MASK|STOP_BIT)));
                            if (w.eventCount == (e | INT_SIGN) &&
                                U.compareAndSwapLong(this, CTL, cc, nc)) {
                                w.eventCount = (e + E_SEQ) & E_MASK;
                                w.qlock = -1;
                                if ((p = w.parker) != null)
                                    U.unpark(p);
                            }
                        }
D
dl 已提交
2276 2277 2278 2279 2280 2281
                    }
                }
            }
        }
    }

2282 2283
    // external operations on common pool

D
dl 已提交
2284
    /**
2285 2286
     * Returns common pool queue for a thread that has submitted at
     * least one task.
D
dl 已提交
2287
     */
2288
    static WorkQueue commonSubmitterQueue() {
2289 2290 2291
        ForkJoinPool p; WorkQueue[] ws; int m, z;
        return ((z = ThreadLocalRandom.getProbe()) != 0 &&
                (p = common) != null &&
2292 2293
                (ws = p.workQueues) != null &&
                (m = ws.length - 1) >= 0) ?
2294
            ws[m & z & SQMASK] : null;
2295 2296 2297
    }

    /**
2298 2299
     * Tries to pop the given task from submitter's queue in common pool.
     */
2300 2301 2302 2303 2304 2305 2306 2307 2308
    final boolean tryExternalUnpush(ForkJoinTask<?> task) {
        WorkQueue joiner; ForkJoinTask<?>[] a; int m, s;
        WorkQueue[] ws = workQueues;
        int z = ThreadLocalRandom.getProbe();
        boolean popped = false;
        if (ws != null && (m = ws.length - 1) >= 0 &&
            (joiner = ws[z & m & SQMASK]) != null &&
            joiner.base != (s = joiner.top) &&
            (a = joiner.array) != null) {
2309
            long j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE;
2310 2311 2312 2313 2314 2315
            if (U.getObject(a, j) == task &&
                U.compareAndSwapInt(joiner, QLOCK, 0, 1)) {
                if (joiner.top == s && joiner.array == a &&
                    U.compareAndSwapObject(a, j, task, null)) {
                    joiner.top = s - 1;
                    popped = true;
2316
                }
2317
                joiner.qlock = 0;
D
dl 已提交
2318 2319
            }
        }
2320
        return popped;
D
dl 已提交
2321 2322
    }

2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340
    final int externalHelpComplete(CountedCompleter<?> task, int maxTasks) {
        WorkQueue joiner; int m;
        WorkQueue[] ws = workQueues;
        int j = ThreadLocalRandom.getProbe();
        int s = 0;
        if (ws != null && (m = ws.length - 1) >= 0 &&
            (joiner = ws[j & m & SQMASK]) != null && task != null) {
            int scans = m + m + 1;
            long c = 0L;             // for stability check
            j |= 1;                  // poll odd queues
            for (int k = scans; ; j += 2) {
                WorkQueue q;
                if ((s = task.status) < 0)
                    break;
                else if (joiner.externalPopAndExecCC(task)) {
                    if (--maxTasks <= 0) {
                        s = task.status;
                        break;
2341
                    }
2342
                    k = scans;
2343
                }
2344
                else if ((s = task.status) < 0)
2345
                    break;
2346 2347 2348 2349
                else if ((q = ws[j & m]) != null && q.pollAndExecCC(task)) {
                    if (--maxTasks <= 0) {
                        s = task.status;
                        break;
2350
                    }
2351 2352 2353 2354 2355 2356
                    k = scans;
                }
                else if (--k < 0) {
                    if (c == (c = ctl))
                        break;
                    k = scans;
2357 2358 2359
                }
            }
        }
2360
        return s;
D
dl 已提交
2361 2362
    }

2363
    // Exported methods
D
dl 已提交
2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378

    // Constructors

    /**
     * Creates a {@code ForkJoinPool} with parallelism equal to {@link
     * java.lang.Runtime#availableProcessors}, using the {@linkplain
     * #defaultForkJoinWorkerThreadFactory default thread factory},
     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
     *
     * @throws SecurityException if a security manager exists and
     *         the caller is not permitted to modify threads
     *         because it does not hold {@link
     *         java.lang.RuntimePermission}{@code ("modifyThread")}
     */
    public ForkJoinPool() {
2379
        this(Math.min(MAX_CAP, Runtime.getRuntime().availableProcessors()),
D
dl 已提交
2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426
             defaultForkJoinWorkerThreadFactory, null, false);
    }

    /**
     * Creates a {@code ForkJoinPool} with the indicated parallelism
     * level, the {@linkplain
     * #defaultForkJoinWorkerThreadFactory default thread factory},
     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
     *
     * @param parallelism the parallelism level
     * @throws IllegalArgumentException if parallelism less than or
     *         equal to zero, or greater than implementation limit
     * @throws SecurityException if a security manager exists and
     *         the caller is not permitted to modify threads
     *         because it does not hold {@link
     *         java.lang.RuntimePermission}{@code ("modifyThread")}
     */
    public ForkJoinPool(int parallelism) {
        this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
    }

    /**
     * Creates a {@code ForkJoinPool} with the given parameters.
     *
     * @param parallelism the parallelism level. For default value,
     * use {@link java.lang.Runtime#availableProcessors}.
     * @param factory the factory for creating new threads. For default value,
     * use {@link #defaultForkJoinWorkerThreadFactory}.
     * @param handler the handler for internal worker threads that
     * terminate due to unrecoverable errors encountered while executing
     * tasks. For default value, use {@code null}.
     * @param asyncMode if true,
     * establishes local first-in-first-out scheduling mode for forked
     * tasks that are never joined. This mode may be more appropriate
     * than default locally stack-based mode in applications in which
     * worker threads only process event-style asynchronous tasks.
     * For default value, use {@code false}.
     * @throws IllegalArgumentException if parallelism less than or
     *         equal to zero, or greater than implementation limit
     * @throws NullPointerException if the factory is null
     * @throws SecurityException if a security manager exists and
     *         the caller is not permitted to modify threads
     *         because it does not hold {@link
     *         java.lang.RuntimePermission}{@code ("modifyThread")}
     */
    public ForkJoinPool(int parallelism,
                        ForkJoinWorkerThreadFactory factory,
2427
                        UncaughtExceptionHandler handler,
D
dl 已提交
2428
                        boolean asyncMode) {
2429 2430 2431
        this(checkParallelism(parallelism),
             checkFactory(factory),
             handler,
2432
             (asyncMode ? FIFO_QUEUE : LIFO_QUEUE),
2433
             "ForkJoinPool-" + nextPoolId() + "-worker-");
D
dl 已提交
2434
        checkPermission();
2435 2436 2437
    }

    private static int checkParallelism(int parallelism) {
2438
        if (parallelism <= 0 || parallelism > MAX_CAP)
D
dl 已提交
2439
            throw new IllegalArgumentException();
2440 2441 2442 2443 2444 2445 2446 2447
        return parallelism;
    }

    private static ForkJoinWorkerThreadFactory checkFactory
        (ForkJoinWorkerThreadFactory factory) {
        if (factory == null)
            throw new NullPointerException();
        return factory;
D
dl 已提交
2448 2449
    }

2450
    /**
2451 2452 2453
     * Creates a {@code ForkJoinPool} with the given parameters, without
     * any security checks or parameter validation.  Invoked directly by
     * makeCommonPool.
2454
     */
2455 2456 2457
    private ForkJoinPool(int parallelism,
                         ForkJoinWorkerThreadFactory factory,
                         UncaughtExceptionHandler handler,
2458
                         int mode,
2459 2460
                         String workerNamePrefix) {
        this.workerNamePrefix = workerNamePrefix;
2461 2462
        this.factory = factory;
        this.ueh = handler;
2463 2464
        this.mode = (short)mode;
        this.parallelism = (short)parallelism;
2465 2466
        long np = (long)(-parallelism); // offset ctl counts
        this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
2467 2468 2469 2470
    }

    /**
     * Returns the common pool instance. This pool is statically
2471 2472 2473 2474 2475 2476 2477
     * constructed; its run state is unaffected by attempts to {@link
     * #shutdown} or {@link #shutdownNow}. However this pool and any
     * ongoing processing are automatically terminated upon program
     * {@link System#exit}.  Any program that relies on asynchronous
     * task processing to complete before program termination should
     * invoke {@code commonPool().}{@link #awaitQuiescence awaitQuiescence},
     * before exit.
2478 2479
     *
     * @return the common pool instance
2480
     * @since 1.8
2481 2482
     */
    public static ForkJoinPool commonPool() {
2483 2484
        // assert common != null : "static init error";
        return common;
2485 2486
    }

D
dl 已提交
2487 2488 2489 2490
    // Execution methods

    /**
     * Performs the given task, returning its result upon completion.
2491 2492 2493 2494 2495 2496 2497
     * If the computation encounters an unchecked Exception or Error,
     * it is rethrown as the outcome of this invocation.  Rethrown
     * exceptions behave in the same way as regular exceptions, but,
     * when possible, contain stack traces (as displayed for example
     * using {@code ex.printStackTrace()}) of both the current thread
     * as well as the thread actually encountering the exception;
     * minimally only the latter.
D
dl 已提交
2498 2499
     *
     * @param task the task
2500
     * @param <T> the type of the task's result
D
dl 已提交
2501
     * @return the task's result
D
dl 已提交
2502 2503 2504 2505
     * @throws NullPointerException if the task is null
     * @throws RejectedExecutionException if the task cannot be
     *         scheduled for execution
     */
D
dl 已提交
2506
    public <T> T invoke(ForkJoinTask<T> task) {
2507 2508
        if (task == null)
            throw new NullPointerException();
2509 2510
        externalPush(task);
        return task.join();
D
dl 已提交
2511 2512 2513
    }

    /**
D
dl 已提交
2514 2515 2516
     * Arranges for (asynchronous) execution of the given task.
     *
     * @param task the task
D
dl 已提交
2517 2518 2519 2520
     * @throws NullPointerException if the task is null
     * @throws RejectedExecutionException if the task cannot be
     *         scheduled for execution
     */
D
dl 已提交
2521
    public void execute(ForkJoinTask<?> task) {
2522 2523
        if (task == null)
            throw new NullPointerException();
2524
        externalPush(task);
D
dl 已提交
2525 2526
    }

D
dl 已提交
2527 2528
    // AbstractExecutorService methods

D
dl 已提交
2529 2530 2531 2532 2533
    /**
     * @throws NullPointerException if the task is null
     * @throws RejectedExecutionException if the task cannot be
     *         scheduled for execution
     */
D
dl 已提交
2534
    public void execute(Runnable task) {
2535 2536
        if (task == null)
            throw new NullPointerException();
D
dl 已提交
2537 2538 2539 2540
        ForkJoinTask<?> job;
        if (task instanceof ForkJoinTask<?>) // avoid re-wrap
            job = (ForkJoinTask<?>) task;
        else
2541
            job = new ForkJoinTask.RunnableExecuteAction(task);
2542
        externalPush(job);
D
dl 已提交
2543 2544 2545 2546 2547 2548
    }

    /**
     * Submits a ForkJoinTask for execution.
     *
     * @param task the task to submit
2549
     * @param <T> the type of the task's result
D
dl 已提交
2550 2551 2552 2553 2554 2555
     * @return the task
     * @throws NullPointerException if the task is null
     * @throws RejectedExecutionException if the task cannot be
     *         scheduled for execution
     */
    public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
2556 2557
        if (task == null)
            throw new NullPointerException();
2558
        externalPush(task);
D
dl 已提交
2559 2560 2561
        return task;
    }

D
dl 已提交
2562 2563 2564 2565 2566 2567
    /**
     * @throws NullPointerException if the task is null
     * @throws RejectedExecutionException if the task cannot be
     *         scheduled for execution
     */
    public <T> ForkJoinTask<T> submit(Callable<T> task) {
2568 2569
        ForkJoinTask<T> job = new ForkJoinTask.AdaptedCallable<T>(task);
        externalPush(job);
D
dl 已提交
2570 2571 2572 2573 2574 2575 2576 2577 2578
        return job;
    }

    /**
     * @throws NullPointerException if the task is null
     * @throws RejectedExecutionException if the task cannot be
     *         scheduled for execution
     */
    public <T> ForkJoinTask<T> submit(Runnable task, T result) {
2579 2580
        ForkJoinTask<T> job = new ForkJoinTask.AdaptedRunnable<T>(task, result);
        externalPush(job);
D
dl 已提交
2581 2582 2583 2584 2585 2586 2587 2588 2589
        return job;
    }

    /**
     * @throws NullPointerException if the task is null
     * @throws RejectedExecutionException if the task cannot be
     *         scheduled for execution
     */
    public ForkJoinTask<?> submit(Runnable task) {
2590 2591
        if (task == null)
            throw new NullPointerException();
D
dl 已提交
2592 2593 2594 2595
        ForkJoinTask<?> job;
        if (task instanceof ForkJoinTask<?>) // avoid re-wrap
            job = (ForkJoinTask<?>) task;
        else
2596 2597
            job = new ForkJoinTask.AdaptedRunnableAction(task);
        externalPush(job);
D
dl 已提交
2598 2599
        return job;
    }
D
dl 已提交
2600 2601 2602 2603 2604 2605

    /**
     * @throws NullPointerException       {@inheritDoc}
     * @throws RejectedExecutionException {@inheritDoc}
     */
    public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
2606 2607 2608
        // In previous versions of this class, this method constructed
        // a task to run ForkJoinTask.invokeAll, but now external
        // invocation of multiple tasks is at least as efficient.
2609
        ArrayList<Future<T>> futures = new ArrayList<Future<T>>(tasks.size());
D
dl 已提交
2610

2611 2612 2613 2614
        boolean done = false;
        try {
            for (Callable<T> t : tasks) {
                ForkJoinTask<T> f = new ForkJoinTask.AdaptedCallable<T>(t);
2615
                futures.add(f);
2616 2617
                externalPush(f);
            }
2618 2619
            for (int i = 0, size = futures.size(); i < size; i++)
                ((ForkJoinTask<?>)futures.get(i)).quietlyJoin();
2620 2621 2622 2623
            done = true;
            return futures;
        } finally {
            if (!done)
2624 2625
                for (int i = 0, size = futures.size(); i < size; i++)
                    futures.get(i).cancel(false);
D
dl 已提交
2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643
        }
    }

    /**
     * Returns the factory used for constructing new workers.
     *
     * @return the factory used for constructing new workers
     */
    public ForkJoinWorkerThreadFactory getFactory() {
        return factory;
    }

    /**
     * Returns the handler for internal worker threads that terminate
     * due to unrecoverable errors encountered while executing tasks.
     *
     * @return the handler, or {@code null} if none
     */
2644
    public UncaughtExceptionHandler getUncaughtExceptionHandler() {
D
dl 已提交
2645
        return ueh;
D
dl 已提交
2646 2647 2648 2649 2650 2651 2652 2653
    }

    /**
     * Returns the targeted parallelism level of this pool.
     *
     * @return the targeted parallelism level of this pool
     */
    public int getParallelism() {
2654 2655
        int par;
        return ((par = parallelism) > 0) ? par : 1;
2656 2657 2658 2659 2660 2661
    }

    /**
     * Returns the targeted parallelism level of the common pool.
     *
     * @return the targeted parallelism level of the common pool
2662
     * @since 1.8
2663 2664
     */
    public static int getCommonPoolParallelism() {
2665
        return commonParallelism;
D
dl 已提交
2666 2667 2668 2669
    }

    /**
     * Returns the number of worker threads that have started but not
D
dl 已提交
2670
     * yet terminated.  The result returned by this method may differ
D
dl 已提交
2671 2672 2673 2674 2675 2676
     * from {@link #getParallelism} when threads are created to
     * maintain parallelism when others are cooperatively blocked.
     *
     * @return the number of worker threads
     */
    public int getPoolSize() {
2677
        return parallelism + (short)(ctl >>> TC_SHIFT);
D
dl 已提交
2678 2679 2680 2681 2682 2683 2684 2685 2686
    }

    /**
     * Returns {@code true} if this pool uses local first-in-first-out
     * scheduling mode for forked tasks that are never joined.
     *
     * @return {@code true} if this pool uses async mode
     */
    public boolean getAsyncMode() {
2687
        return mode == FIFO_QUEUE;
D
dl 已提交
2688 2689 2690 2691 2692
    }

    /**
     * Returns an estimate of the number of worker threads that are
     * not blocked waiting to join tasks or for other managed
D
dl 已提交
2693 2694
     * synchronization. This method may overestimate the
     * number of running threads.
D
dl 已提交
2695 2696 2697 2698
     *
     * @return the number of worker threads
     */
    public int getRunningThreadCount() {
2699 2700 2701 2702 2703 2704 2705 2706 2707
        int rc = 0;
        WorkQueue[] ws; WorkQueue w;
        if ((ws = workQueues) != null) {
            for (int i = 1; i < ws.length; i += 2) {
                if ((w = ws[i]) != null && w.isApparentlyUnblocked())
                    ++rc;
            }
        }
        return rc;
D
dl 已提交
2708 2709 2710 2711 2712 2713 2714 2715 2716 2717
    }

    /**
     * Returns an estimate of the number of threads that are currently
     * stealing or executing tasks. This method may overestimate the
     * number of active threads.
     *
     * @return the number of active threads
     */
    public int getActiveThreadCount() {
2718
        int r = parallelism + (int)(ctl >> AC_SHIFT);
D
dl 已提交
2719
        return (r <= 0) ? 0 : r; // suppress momentarily negative values
D
dl 已提交
2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733
    }

    /**
     * Returns {@code true} if all worker threads are currently idle.
     * An idle worker is one that cannot obtain a task to execute
     * because none are available to steal from other threads, and
     * there are no pending submissions to the pool. This method is
     * conservative; it might not return {@code true} immediately upon
     * idleness of all threads, but will eventually become true if
     * threads remain inactive.
     *
     * @return {@code true} if all threads are currently idle
     */
    public boolean isQuiescent() {
2734
        return parallelism + (int)(ctl >> AC_SHIFT) <= 0;
D
dl 已提交
2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748
    }

    /**
     * Returns an estimate of the total number of tasks stolen from
     * one thread's work queue by another. The reported value
     * underestimates the actual total number of steals when the pool
     * is not quiescent. This value may be useful for monitoring and
     * tuning fork/join programs: in general, steal counts should be
     * high enough to keep threads busy, but low enough to avoid
     * overhead and contention across threads.
     *
     * @return the number of steals
     */
    public long getStealCount() {
2749 2750 2751 2752 2753 2754 2755 2756 2757
        long count = stealCount;
        WorkQueue[] ws; WorkQueue w;
        if ((ws = workQueues) != null) {
            for (int i = 1; i < ws.length; i += 2) {
                if ((w = ws[i]) != null)
                    count += w.nsteals;
            }
        }
        return count;
D
dl 已提交
2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771
    }

    /**
     * Returns an estimate of the total number of tasks currently held
     * in queues by worker threads (but not including tasks submitted
     * to the pool that have not begun executing). This value is only
     * an approximation, obtained by iterating across all threads in
     * the pool. This method may be useful for tuning task
     * granularities.
     *
     * @return the number of queued tasks
     */
    public long getQueuedTaskCount() {
        long count = 0;
2772 2773 2774 2775 2776 2777
        WorkQueue[] ws; WorkQueue w;
        if ((ws = workQueues) != null) {
            for (int i = 1; i < ws.length; i += 2) {
                if ((w = ws[i]) != null)
                    count += w.queueSize();
            }
2778
        }
D
dl 已提交
2779 2780 2781 2782 2783
        return count;
    }

    /**
     * Returns an estimate of the number of tasks submitted to this
2784 2785
     * pool that have not yet begun executing.  This method may take
     * time proportional to the number of submissions.
D
dl 已提交
2786 2787 2788 2789
     *
     * @return the number of queued submissions
     */
    public int getQueuedSubmissionCount() {
2790 2791 2792 2793 2794 2795 2796 2797 2798
        int count = 0;
        WorkQueue[] ws; WorkQueue w;
        if ((ws = workQueues) != null) {
            for (int i = 0; i < ws.length; i += 2) {
                if ((w = ws[i]) != null)
                    count += w.queueSize();
            }
        }
        return count;
D
dl 已提交
2799 2800 2801 2802 2803 2804 2805 2806 2807
    }

    /**
     * Returns {@code true} if there are any tasks submitted to this
     * pool that have not yet begun executing.
     *
     * @return {@code true} if there are any queued submissions
     */
    public boolean hasQueuedSubmissions() {
2808 2809 2810 2811 2812 2813 2814 2815
        WorkQueue[] ws; WorkQueue w;
        if ((ws = workQueues) != null) {
            for (int i = 0; i < ws.length; i += 2) {
                if ((w = ws[i]) != null && !w.isEmpty())
                    return true;
            }
        }
        return false;
D
dl 已提交
2816 2817 2818 2819 2820 2821 2822 2823 2824 2825
    }

    /**
     * Removes and returns the next unexecuted submission if one is
     * available.  This method may be useful in extensions to this
     * class that re-assign work in systems with multiple pools.
     *
     * @return the next submission, or {@code null} if none
     */
    protected ForkJoinTask<?> pollSubmission() {
2826 2827 2828 2829 2830
        WorkQueue[] ws; WorkQueue w; ForkJoinTask<?> t;
        if ((ws = workQueues) != null) {
            for (int i = 0; i < ws.length; i += 2) {
                if ((w = ws[i]) != null && (t = w.poll()) != null)
                    return t;
2831 2832 2833
            }
        }
        return null;
D
dl 已提交
2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853
    }

    /**
     * Removes all available unexecuted submitted and forked tasks
     * from scheduling queues and adds them to the given collection,
     * without altering their execution status. These may include
     * artificially generated or wrapped tasks. This method is
     * designed to be invoked only when the pool is known to be
     * quiescent. Invocations at other times may not remove all
     * tasks. A failure encountered while attempting to add elements
     * to collection {@code c} may result in elements being in
     * neither, either or both collections when the associated
     * exception is thrown.  The behavior of this operation is
     * undefined if the specified collection is modified while the
     * operation is in progress.
     *
     * @param c the collection to transfer elements into
     * @return the number of elements transferred
     */
    protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
2854
        int count = 0;
2855 2856 2857 2858 2859 2860 2861 2862 2863
        WorkQueue[] ws; WorkQueue w; ForkJoinTask<?> t;
        if ((ws = workQueues) != null) {
            for (int i = 0; i < ws.length; ++i) {
                if ((w = ws[i]) != null) {
                    while ((t = w.poll()) != null) {
                        c.add(t);
                        ++count;
                    }
                }
2864 2865
            }
        }
D
dl 已提交
2866
        return count;
D
dl 已提交
2867 2868 2869 2870 2871 2872 2873 2874 2875 2876
    }

    /**
     * Returns a string identifying this pool, as well as its state,
     * including indications of run state, parallelism level, and
     * worker and task counts.
     *
     * @return a string identifying this pool, as well as its state
     */
    public String toString() {
2877 2878 2879
        // Use a single pass through workQueues to collect counts
        long qt = 0L, qs = 0L; int rc = 0;
        long st = stealCount;
2880
        long c = ctl;
2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896
        WorkQueue[] ws; WorkQueue w;
        if ((ws = workQueues) != null) {
            for (int i = 0; i < ws.length; ++i) {
                if ((w = ws[i]) != null) {
                    int size = w.queueSize();
                    if ((i & 1) == 0)
                        qs += size;
                    else {
                        qt += size;
                        st += w.nsteals;
                        if (w.isApparentlyUnblocked())
                            ++rc;
                    }
                }
            }
        }
2897
        int pc = parallelism;
2898
        int tc = pc + (short)(c >>> TC_SHIFT);
2899 2900 2901
        int ac = pc + (int)(c >> AC_SHIFT);
        if (ac < 0) // ignore transient negative
            ac = 0;
2902 2903
        String level;
        if ((c & STOP_BIT) != 0)
D
dl 已提交
2904
            level = (tc == 0) ? "Terminated" : "Terminating";
2905
        else
2906
            level = plock < 0 ? "Shutting down" : "Running";
D
dl 已提交
2907
        return super.toString() +
2908
            "[" + level +
D
dl 已提交
2909 2910 2911 2912
            ", parallelism = " + pc +
            ", size = " + tc +
            ", active = " + ac +
            ", running = " + rc +
D
dl 已提交
2913 2914 2915 2916 2917 2918 2919
            ", steals = " + st +
            ", tasks = " + qt +
            ", submissions = " + qs +
            "]";
    }

    /**
2920 2921 2922
     * Possibly initiates an orderly shutdown in which previously
     * submitted tasks are executed, but no new tasks will be
     * accepted. Invocation has no effect on execution state if this
2923
     * is the {@link #commonPool()}, and no additional effect if
2924 2925 2926
     * already shut down.  Tasks that are in the process of being
     * submitted concurrently during the course of this method may or
     * may not be rejected.
D
dl 已提交
2927 2928 2929 2930 2931 2932 2933 2934
     *
     * @throws SecurityException if a security manager exists and
     *         the caller is not permitted to modify threads
     *         because it does not hold {@link
     *         java.lang.RuntimePermission}{@code ("modifyThread")}
     */
    public void shutdown() {
        checkPermission();
2935
        tryTerminate(false, true);
D
dl 已提交
2936 2937 2938
    }

    /**
2939 2940
     * Possibly attempts to cancel and/or stop all tasks, and reject
     * all subsequently submitted tasks.  Invocation has no effect on
2941
     * execution state if this is the {@link #commonPool()}, and no
2942 2943 2944 2945 2946 2947 2948
     * additional effect if already shut down. Otherwise, tasks that
     * are in the process of being submitted or executed concurrently
     * during the course of this method may or may not be
     * rejected. This method cancels both existing and unexecuted
     * tasks, in order to permit termination in the presence of task
     * dependencies. So the method always returns an empty list
     * (unlike the case for some other Executors).
D
dl 已提交
2949 2950 2951 2952 2953 2954 2955 2956 2957
     *
     * @return an empty list
     * @throws SecurityException if a security manager exists and
     *         the caller is not permitted to modify threads
     *         because it does not hold {@link
     *         java.lang.RuntimePermission}{@code ("modifyThread")}
     */
    public List<Runnable> shutdownNow() {
        checkPermission();
2958
        tryTerminate(true, true);
D
dl 已提交
2959 2960 2961 2962 2963 2964 2965 2966 2967
        return Collections.emptyList();
    }

    /**
     * Returns {@code true} if all tasks have completed following shut down.
     *
     * @return {@code true} if all tasks have completed following shut down
     */
    public boolean isTerminated() {
2968 2969
        long c = ctl;
        return ((c & STOP_BIT) != 0L &&
2970
                (short)(c >>> TC_SHIFT) + parallelism <= 0);
D
dl 已提交
2971 2972 2973 2974 2975 2976 2977
    }

    /**
     * Returns {@code true} if the process of termination has
     * commenced but not yet completed.  This method may be useful for
     * debugging. A return of {@code true} reported a sufficient
     * period after shutdown may indicate that submitted tasks have
2978
     * ignored or suppressed interruption, or are waiting for I/O,
2979 2980 2981 2982
     * causing this executor not to properly terminate. (See the
     * advisory notes for class {@link ForkJoinTask} stating that
     * tasks should not normally entail blocking operations.  But if
     * they do, they must abort them on interrupt.)
D
dl 已提交
2983 2984 2985 2986
     *
     * @return {@code true} if terminating but not yet terminated
     */
    public boolean isTerminating() {
2987 2988
        long c = ctl;
        return ((c & STOP_BIT) != 0L &&
2989
                (short)(c >>> TC_SHIFT) + parallelism > 0);
2990 2991
    }

D
dl 已提交
2992 2993 2994 2995 2996 2997
    /**
     * Returns {@code true} if this pool has been shut down.
     *
     * @return {@code true} if this pool has been shut down
     */
    public boolean isShutdown() {
2998
        return plock < 0;
D
dl 已提交
2999 3000 3001
    }

    /**
3002 3003
     * Blocks until all tasks have completed execution after a
     * shutdown request, or the timeout occurs, or the current thread
3004 3005 3006 3007
     * is interrupted, whichever happens first. Because the {@link
     * #commonPool()} never terminates until program shutdown, when
     * applied to the common pool, this method is equivalent to {@link
     * #awaitQuiescence(long, TimeUnit)} but always returns {@code false}.
D
dl 已提交
3008 3009 3010 3011 3012 3013 3014 3015 3016
     *
     * @param timeout the maximum time to wait
     * @param unit the time unit of the timeout argument
     * @return {@code true} if this executor terminated and
     *         {@code false} if the timeout elapsed before termination
     * @throws InterruptedException if interrupted while waiting
     */
    public boolean awaitTermination(long timeout, TimeUnit unit)
        throws InterruptedException {
3017 3018 3019 3020 3021 3022
        if (Thread.interrupted())
            throw new InterruptedException();
        if (this == common) {
            awaitQuiescence(timeout, unit);
            return false;
        }
3023
        long nanos = unit.toNanos(timeout);
3024 3025
        if (isTerminated())
            return true;
3026 3027 3028
        if (nanos <= 0L)
            return false;
        long deadline = System.nanoTime() + nanos;
3029
        synchronized (this) {
3030 3031 3032 3033 3034 3035 3036 3037
            for (;;) {
                if (isTerminated())
                    return true;
                if (nanos <= 0L)
                    return false;
                long millis = TimeUnit.NANOSECONDS.toMillis(nanos);
                wait(millis > 0L ? millis : 1L);
                nanos = deadline - System.nanoTime();
3038
            }
D
dl 已提交
3039 3040 3041
        }
    }

3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077
    /**
     * If called by a ForkJoinTask operating in this pool, equivalent
     * in effect to {@link ForkJoinTask#helpQuiesce}. Otherwise,
     * waits and/or attempts to assist performing tasks until this
     * pool {@link #isQuiescent} or the indicated timeout elapses.
     *
     * @param timeout the maximum time to wait
     * @param unit the time unit of the timeout argument
     * @return {@code true} if quiescent; {@code false} if the
     * timeout elapsed.
     */
    public boolean awaitQuiescence(long timeout, TimeUnit unit) {
        long nanos = unit.toNanos(timeout);
        ForkJoinWorkerThread wt;
        Thread thread = Thread.currentThread();
        if ((thread instanceof ForkJoinWorkerThread) &&
            (wt = (ForkJoinWorkerThread)thread).pool == this) {
            helpQuiescePool(wt.workQueue);
            return true;
        }
        long startTime = System.nanoTime();
        WorkQueue[] ws;
        int r = 0, m;
        boolean found = true;
        while (!isQuiescent() && (ws = workQueues) != null &&
               (m = ws.length - 1) >= 0) {
            if (!found) {
                if ((System.nanoTime() - startTime) > nanos)
                    return false;
                Thread.yield(); // cannot block
            }
            found = false;
            for (int j = (m + 1) << 2; j >= 0; --j) {
                ForkJoinTask<?> t; WorkQueue q; int b;
                if ((q = ws[r++ & m]) != null && (b = q.base) - q.top < 0) {
                    found = true;
3078
                    if ((t = q.pollAt(b)) != null)
3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094
                        t.doExec();
                    break;
                }
            }
        }
        return true;
    }

    /**
     * Waits and/or attempts to assist performing tasks indefinitely
     * until the {@link #commonPool()} {@link #isQuiescent}.
     */
    static void quiesceCommonPool() {
        common.awaitQuiescence(Long.MAX_VALUE, TimeUnit.NANOSECONDS);
    }

D
dl 已提交
3095 3096 3097 3098
    /**
     * Interface for extending managed parallelism for tasks running
     * in {@link ForkJoinPool}s.
     *
D
dl 已提交
3099 3100 3101 3102
     * <p>A {@code ManagedBlocker} provides two methods.  Method
     * {@code isReleasable} must return {@code true} if blocking is
     * not necessary. Method {@code block} blocks the current thread
     * if necessary (perhaps internally invoking {@code isReleasable}
3103
     * before actually blocking). These actions are performed by any
3104 3105 3106
     * thread invoking {@link ForkJoinPool#managedBlock(ManagedBlocker)}.
     * The unusual methods in this API accommodate synchronizers that
     * may, but don't usually, block for long periods. Similarly, they
3107 3108 3109 3110 3111
     * allow more efficient internal handling of cases in which
     * additional workers may be, but usually are not, needed to
     * ensure sufficient parallelism.  Toward this end,
     * implementations of method {@code isReleasable} must be amenable
     * to repeated invocation.
D
dl 已提交
3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128
     *
     * <p>For example, here is a ManagedBlocker based on a
     * ReentrantLock:
     *  <pre> {@code
     * class ManagedLocker implements ManagedBlocker {
     *   final ReentrantLock lock;
     *   boolean hasLock = false;
     *   ManagedLocker(ReentrantLock lock) { this.lock = lock; }
     *   public boolean block() {
     *     if (!hasLock)
     *       lock.lock();
     *     return true;
     *   }
     *   public boolean isReleasable() {
     *     return hasLock || (hasLock = lock.tryLock());
     *   }
     * }}</pre>
D
dl 已提交
3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148
     *
     * <p>Here is a class that possibly blocks waiting for an
     * item on a given queue:
     *  <pre> {@code
     * class QueueTaker<E> implements ManagedBlocker {
     *   final BlockingQueue<E> queue;
     *   volatile E item = null;
     *   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
     *   public boolean block() throws InterruptedException {
     *     if (item == null)
     *       item = queue.take();
     *     return true;
     *   }
     *   public boolean isReleasable() {
     *     return item != null || (item = queue.poll()) != null;
     *   }
     *   public E getItem() { // call after pool.managedBlock completes
     *     return item;
     *   }
     * }}</pre>
D
dl 已提交
3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163
     */
    public static interface ManagedBlocker {
        /**
         * Possibly blocks the current thread, for example waiting for
         * a lock or condition.
         *
         * @return {@code true} if no additional blocking is necessary
         * (i.e., if isReleasable would return true)
         * @throws InterruptedException if interrupted while waiting
         * (the method is not required to do so, but is allowed to)
         */
        boolean block() throws InterruptedException;

        /**
         * Returns {@code true} if blocking is unnecessary.
3164
         * @return {@code true} if blocking is unnecessary
D
dl 已提交
3165 3166 3167 3168 3169 3170 3171 3172
         */
        boolean isReleasable();
    }

    /**
     * Blocks in accord with the given blocker.  If the current thread
     * is a {@link ForkJoinWorkerThread}, this method possibly
     * arranges for a spare thread to be activated if necessary to
D
dl 已提交
3173
     * ensure sufficient parallelism while the current thread is blocked.
D
dl 已提交
3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188
     *
     * <p>If the caller is not a {@link ForkJoinTask}, this method is
     * behaviorally equivalent to
     *  <pre> {@code
     * while (!blocker.isReleasable())
     *   if (blocker.block())
     *     return;
     * }</pre>
     *
     * If the caller is a {@code ForkJoinTask}, then the pool may
     * first be expanded to ensure parallelism, and later adjusted.
     *
     * @param blocker the blocker
     * @throws InterruptedException if blocker.block did so
     */
D
dl 已提交
3189
    public static void managedBlock(ManagedBlocker blocker)
D
dl 已提交
3190 3191
        throws InterruptedException {
        Thread t = Thread.currentThread();
D
dl 已提交
3192
        if (t instanceof ForkJoinWorkerThread) {
3193
            ForkJoinPool p = ((ForkJoinWorkerThread)t).pool;
3194 3195
            while (!blocker.isReleasable()) {
                if (p.tryCompensate(p.ctl)) {
3196 3197 3198 3199 3200 3201 3202 3203 3204
                    try {
                        do {} while (!blocker.isReleasable() &&
                                     !blocker.block());
                    } finally {
                        p.incrementActiveCount();
                    }
                    break;
                }
            }
D
dl 已提交
3205 3206
        }
        else {
3207 3208
            do {} while (!blocker.isReleasable() &&
                         !blocker.block());
D
dl 已提交
3209 3210 3211 3212 3213 3214 3215 3216
        }
    }

    // AbstractExecutorService overrides.  These rely on undocumented
    // fact that ForkJoinTask.adapt returns ForkJoinTasks that also
    // implement RunnableFuture.

    protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
3217
        return new ForkJoinTask.AdaptedRunnable<T>(runnable, value);
D
dl 已提交
3218 3219 3220
    }

    protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
3221
        return new ForkJoinTask.AdaptedCallable<T>(callable);
D
dl 已提交
3222 3223 3224
    }

    // Unsafe mechanics
3225 3226 3227 3228
    private static final sun.misc.Unsafe U;
    private static final long CTL;
    private static final long PARKBLOCKER;
    private static final int ABASE;
3229
    private static final int ASHIFT;
3230 3231 3232
    private static final long STEALCOUNT;
    private static final long PLOCK;
    private static final long INDEXSEED;
3233
    private static final long QBASE;
3234
    private static final long QLOCK;
3235 3236

    static {
3237
        // initialize field offsets for CAS etc
D
dl 已提交
3238
        try {
3239
            U = sun.misc.Unsafe.getUnsafe();
3240
            Class<?> k = ForkJoinPool.class;
3241
            CTL = U.objectFieldOffset
3242
                (k.getDeclaredField("ctl"));
3243
            STEALCOUNT = U.objectFieldOffset
3244
                (k.getDeclaredField("stealCount"));
3245 3246 3247 3248 3249 3250 3251 3252
            PLOCK = U.objectFieldOffset
                (k.getDeclaredField("plock"));
            INDEXSEED = U.objectFieldOffset
                (k.getDeclaredField("indexSeed"));
            Class<?> tk = Thread.class;
            PARKBLOCKER = U.objectFieldOffset
                (tk.getDeclaredField("parkBlocker"));
            Class<?> wk = WorkQueue.class;
3253 3254
            QBASE = U.objectFieldOffset
                (wk.getDeclaredField("base"));
3255 3256 3257 3258
            QLOCK = U.objectFieldOffset
                (wk.getDeclaredField("qlock"));
            Class<?> ak = ForkJoinTask[].class;
            ABASE = U.arrayBaseOffset(ak);
3259 3260 3261 3262
            int scale = U.arrayIndexScale(ak);
            if ((scale & (scale - 1)) != 0)
                throw new Error("data type scale not a power of two");
            ASHIFT = 31 - Integer.numberOfLeadingZeros(scale);
3263 3264
        } catch (Exception e) {
            throw new Error(e);
D
dl 已提交
3265
        }
3266

3267
        defaultForkJoinWorkerThreadFactory =
3268 3269 3270
            new DefaultForkJoinWorkerThreadFactory();
        modifyThreadPermission = new RuntimePermission("modifyThread");

3271 3272 3273
        common = java.security.AccessController.doPrivileged
            (new java.security.PrivilegedAction<ForkJoinPool>() {
                public ForkJoinPool run() { return makeCommonPool(); }});
3274
        int par = common.parallelism; // report 1 even if threads disabled
3275 3276
        commonParallelism = par > 0 ? par : 1;
    }
3277

3278 3279 3280 3281 3282 3283 3284 3285 3286
    /**
     * Creates and returns the common pool, respecting user settings
     * specified via system properties.
     */
    private static ForkJoinPool makeCommonPool() {
        int parallelism = -1;
        ForkJoinWorkerThreadFactory factory
            = defaultForkJoinWorkerThreadFactory;
        UncaughtExceptionHandler handler = null;
3287
        try {  // ignore exceptions in accessing/parsing properties
3288 3289 3290 3291
            String pp = System.getProperty
                ("java.util.concurrent.ForkJoinPool.common.parallelism");
            String fp = System.getProperty
                ("java.util.concurrent.ForkJoinPool.common.threadFactory");
3292 3293 3294 3295
            String hp = System.getProperty
                ("java.util.concurrent.ForkJoinPool.common.exceptionHandler");
            if (pp != null)
                parallelism = Integer.parseInt(pp);
3296
            if (fp != null)
3297 3298
                factory = ((ForkJoinWorkerThreadFactory)ClassLoader.
                           getSystemClassLoader().loadClass(fp).newInstance());
3299
            if (hp != null)
3300
                handler = ((UncaughtExceptionHandler)ClassLoader.
3301 3302 3303 3304
                           getSystemClassLoader().loadClass(hp).newInstance());
        } catch (Exception ignore) {
        }

3305 3306 3307
        if (parallelism < 0 && // default 1 less than #cores
            (parallelism = Runtime.getRuntime().availableProcessors() - 1) < 0)
            parallelism = 0;
3308 3309
        if (parallelism > MAX_CAP)
            parallelism = MAX_CAP;
3310
        return new ForkJoinPool(parallelism, factory, handler, LIFO_QUEUE,
3311
                                "ForkJoinPool.commonPool-worker-");
D
dl 已提交
3312
    }
3313

D
dl 已提交
3314
}