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

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.1 by dl, Tue Jan 6 14:30:31 2009 UTC vs.
Revision 1.100 by dl, Fri Apr 1 20:20:37 2011 UTC

# Line 1 | Line 1
1   /*
2   * Written by Doug Lea with assistance from members of JCP JSR-166
3   * Expert Group and released to the public domain, as explained at
4 < * http://creativecommons.org/licenses/publicdomain
4 > * http://creativecommons.org/publicdomain/zero/1.0/
5   */
6  
7   package jsr166y;
8 < import java.util.*;
9 < import java.util.concurrent.*;
10 < import java.util.concurrent.locks.*;
11 < import java.util.concurrent.atomic.*;
12 < import sun.misc.Unsafe;
13 < import java.lang.reflect.*;
8 >
9 > import java.util.ArrayList;
10 > import java.util.Arrays;
11 > import java.util.Collection;
12 > import java.util.Collections;
13 > import java.util.List;
14 > import java.util.Random;
15 > import java.util.concurrent.AbstractExecutorService;
16 > import java.util.concurrent.Callable;
17 > import java.util.concurrent.ExecutorService;
18 > import java.util.concurrent.Future;
19 > import java.util.concurrent.RejectedExecutionException;
20 > import java.util.concurrent.RunnableFuture;
21 > import java.util.concurrent.TimeUnit;
22 > import java.util.concurrent.TimeoutException;
23 > import java.util.concurrent.atomic.AtomicInteger;
24 > import java.util.concurrent.locks.LockSupport;
25 > import java.util.concurrent.locks.ReentrantLock;
26 > import java.util.concurrent.locks.Condition;
27  
28   /**
29 < * Host for a group of ForkJoinWorkerThreads.  A ForkJoinPool provides
30 < * the entry point for tasks submitted from non-ForkJoinTasks, as well
31 < * as management and monitoring operations.  Normally a single
32 < * ForkJoinPool is used for a large number of submitted
20 < * tasks. Otherwise, use would not usually outweigh the construction
21 < * and bookkeeping overhead of creating a large set of threads.
22 < *
23 < * <p>ForkJoinPools differ from other kinds of Executor mainly in that
24 < * they provide <em>work-stealing</em>: all threads in the pool
25 < * attempt to find and execute subtasks created by other active tasks
26 < * (eventually blocking if none exist). This makes them efficient when
27 < * most tasks spawn other subtasks (as do most ForkJoinTasks) but
28 < * possibly less so otherwise. It is however fine to combine execution
29 < * of some plain Runnable- or Callable- based activities with that of
30 < * ForkJoinTasks.
29 > * An {@link ExecutorService} for running {@link ForkJoinTask}s.
30 > * A {@code ForkJoinPool} provides the entry point for submissions
31 > * from non-{@code ForkJoinTask} clients, as well as management and
32 > * monitoring operations.
33   *
34 < * <p>A ForkJoinPool may be constructed with a given parallelism level
35 < * (target pool size), which it attempts to maintain by dynamically
36 < * adding, suspending, or resuming threads, even if some tasks have
37 < * blocked waiting to join others. However, no such adjustments are
38 < * performed in the face of blocked IO or other unmanaged
39 < * synchronization. The nested ManagedBlocker interface enables
40 < * extension of the kinds of synchronization accommodated.
34 > * <p>A {@code ForkJoinPool} differs from other kinds of {@link
35 > * ExecutorService} mainly by virtue of employing
36 > * <em>work-stealing</em>: all threads in the pool attempt to find and
37 > * execute subtasks created by other active tasks (eventually blocking
38 > * waiting for work if none exist). This enables efficient processing
39 > * when most tasks spawn other subtasks (as do most {@code
40 > * ForkJoinTask}s). When setting <em>asyncMode</em> to true in
41 > * constructors, {@code ForkJoinPool}s may also be appropriate for use
42 > * with event-style tasks that are never joined.
43   *
44 < * <p>The target parallelism level may also be set dynamically. You
45 < * can limit the number of threads dynamically constructed using
46 < * method <tt>setMaximumPoolSize</tt> and/or
47 < * <tt>setMaintainParallelism</tt>.
44 > * <p>A {@code ForkJoinPool} is constructed with a given target
45 > * parallelism level; by default, equal to the number of available
46 > * processors. The pool attempts to maintain enough active (or
47 > * available) threads by dynamically adding, suspending, or resuming
48 > * internal worker threads, even if some tasks are stalled waiting to
49 > * join others. However, no such adjustments are guaranteed in the
50 > * face of blocked IO or other unmanaged synchronization. The nested
51 > * {@link ManagedBlocker} interface enables extension of the kinds of
52 > * synchronization accommodated.
53   *
54   * <p>In addition to execution and lifecycle control methods, this
55   * class provides status check methods (for example
56 < * <tt>getStealCount</tt>) that are intended to aid in developing,
56 > * {@link #getStealCount}) that are intended to aid in developing,
57   * tuning, and monitoring fork/join applications. Also, method
58 < * <tt>toString</tt> returns indications of pool state in a convenient
59 < * form for informal monitoring.
58 > * {@link #toString} returns indications of pool state in a
59 > * convenient form for informal monitoring.
60 > *
61 > * <p> As is the case with other ExecutorServices, there are three
62 > * main task execution methods summarized in the following
63 > * table. These are designed to be used by clients not already engaged
64 > * in fork/join computations in the current pool.  The main forms of
65 > * these methods accept instances of {@code ForkJoinTask}, but
66 > * overloaded forms also allow mixed execution of plain {@code
67 > * Runnable}- or {@code Callable}- based activities as well.  However,
68 > * tasks that are already executing in a pool should normally
69 > * <em>NOT</em> use these pool execution methods, but instead use the
70 > * within-computation forms listed in the table.
71 > *
72 > * <table BORDER CELLPADDING=3 CELLSPACING=1>
73 > *  <tr>
74 > *    <td></td>
75 > *    <td ALIGN=CENTER> <b>Call from non-fork/join clients</b></td>
76 > *    <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td>
77 > *  </tr>
78 > *  <tr>
79 > *    <td> <b>Arrange async execution</td>
80 > *    <td> {@link #execute(ForkJoinTask)}</td>
81 > *    <td> {@link ForkJoinTask#fork}</td>
82 > *  </tr>
83 > *  <tr>
84 > *    <td> <b>Await and obtain result</td>
85 > *    <td> {@link #invoke(ForkJoinTask)}</td>
86 > *    <td> {@link ForkJoinTask#invoke}</td>
87 > *  </tr>
88 > *  <tr>
89 > *    <td> <b>Arrange exec and obtain Future</td>
90 > *    <td> {@link #submit(ForkJoinTask)}</td>
91 > *    <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td>
92 > *  </tr>
93 > * </table>
94 > *
95 > * <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is
96 > * used for all parallel task execution in a program or subsystem.
97 > * Otherwise, use would not usually outweigh the construction and
98 > * bookkeeping overhead of creating a large set of threads. For
99 > * example, a common pool could be used for the {@code SortTasks}
100 > * illustrated in {@link RecursiveAction}. Because {@code
101 > * ForkJoinPool} uses threads in {@linkplain java.lang.Thread#isDaemon
102 > * daemon} mode, there is typically no need to explicitly {@link
103 > * #shutdown} such a pool upon program exit.
104 > *
105 > * <pre>
106 > * static final ForkJoinPool mainPool = new ForkJoinPool();
107 > * ...
108 > * public void sort(long[] array) {
109 > *   mainPool.invoke(new SortTask(array, 0, array.length));
110 > * }
111 > * </pre>
112   *
113   * <p><b>Implementation notes</b>: This implementation restricts the
114 < * maximum parallelism to 32767. Attempts to create pools with greater
115 < * than the maximum result in IllegalArgumentExceptions.
114 > * maximum number of running threads to 32767. Attempts to create
115 > * pools with greater than the maximum number result in
116 > * {@code IllegalArgumentException}.
117 > *
118 > * <p>This implementation rejects submitted tasks (that is, by throwing
119 > * {@link RejectedExecutionException}) only when the pool is shut down
120 > * or internal resources have been exhausted.
121 > *
122 > * @since 1.7
123 > * @author Doug Lea
124   */
125 < public class ForkJoinPool extends AbstractExecutorService
57 <    implements ExecutorService {
125 > public class ForkJoinPool extends AbstractExecutorService {
126  
127      /*
128 <     * See the extended comments interspersed below for design,
129 <     * rationale, and walkthroughs.
128 >     * Implementation Overview
129 >     *
130 >     * This class provides the central bookkeeping and control for a
131 >     * set of worker threads: Submissions from non-FJ threads enter
132 >     * into a submission queue. Workers take these tasks and typically
133 >     * split them into subtasks that may be stolen by other workers.
134 >     * Preference rules give first priority to processing tasks from
135 >     * their own queues (LIFO or FIFO, depending on mode), then to
136 >     * randomized FIFO steals of tasks in other worker queues, and
137 >     * lastly to new submissions.
138 >     *
139 >     * The main throughput advantages of work-stealing stem from
140 >     * decentralized control -- workers mostly take tasks from
141 >     * themselves or each other. We cannot negate this in the
142 >     * implementation of other management responsibilities. The main
143 >     * tactic for avoiding bottlenecks is packing nearly all
144 >     * essentially atomic control state into a single 64bit volatile
145 >     * variable ("ctl"). This variable is read on the order of 10-100
146 >     * times as often as it is modified (always via CAS). (There is
147 >     * some additional control state, for example variable "shutdown"
148 >     * for which we can cope with uncoordinated updates.)  This
149 >     * streamlines synchronization and control at the expense of messy
150 >     * constructions needed to repack status bits upon updates.
151 >     * Updates tend not to contend with each other except during
152 >     * bursts while submitted tasks begin or end.  In some cases when
153 >     * they do contend, threads can instead do something else
154 >     * (usually, scan for tasks) until contention subsides.
155 >     *
156 >     * To enable packing, we restrict maximum parallelism to (1<<15)-1
157 >     * (which is far in excess of normal operating range) to allow
158 >     * ids, counts, and their negations (used for thresholding) to fit
159 >     * into 16bit fields.
160 >     *
161 >     * Recording Workers.  Workers are recorded in the "workers" array
162 >     * that is created upon pool construction and expanded if (rarely)
163 >     * necessary.  This is an array as opposed to some other data
164 >     * structure to support index-based random steals by workers.
165 >     * Updates to the array recording new workers and unrecording
166 >     * terminated ones are protected from each other by a seqLock
167 >     * (scanGuard) but the array is otherwise concurrently readable,
168 >     * and accessed directly by workers. To simplify index-based
169 >     * operations, the array size is always a power of two, and all
170 >     * readers must tolerate null slots. To avoid flailing during
171 >     * start-up, the array is presized to hold twice #parallelism
172 >     * workers (which is unlikely to need further resizing during
173 >     * execution). But to avoid dealing with so many null slots,
174 >     * variable scanGuard includes a mask for the nearest power of two
175 >     * that contains all current workers.  All worker thread creation
176 >     * is on-demand, triggered by task submissions, replacement of
177 >     * terminated workers, and/or compensation for blocked
178 >     * workers. However, all other support code is set up to work with
179 >     * other policies.  To ensure that we do not hold on to worker
180 >     * references that would prevent GC, ALL accesses to workers are
181 >     * via indices into the workers array (which is one source of some
182 >     * of the messy code constructions here). In essence, the workers
183 >     * array serves as a weak reference mechanism. Thus for example
184 >     * the wait queue field of ctl stores worker indices, not worker
185 >     * references.  Access to the workers in associated methods (for
186 >     * example signalWork) must both index-check and null-check the
187 >     * IDs. All such accesses ignore bad IDs by returning out early
188 >     * from what they are doing, since this can only be associated
189 >     * with termination, in which case it is OK to give up.
190 >     *
191 >     * All uses of the workers array, as well as queue arrays, check
192 >     * that the array is non-null (even if previously non-null). This
193 >     * allows nulling during termination, which is currently not
194 >     * necessary, but remains an option for resource-revocation-based
195 >     * shutdown schemes.
196 >     *
197 >     * Wait Queuing. Unlike HPC work-stealing frameworks, we cannot
198 >     * let workers spin indefinitely scanning for tasks when none can
199 >     * be found immediately, and we cannot start/resume workers unless
200 >     * there appear to be tasks available.  On the other hand, we must
201 >     * quickly prod them into action when new tasks are submitted or
202 >     * generated.  We park/unpark workers after placing in an event
203 >     * wait queue when they cannot find work. This "queue" is actually
204 >     * a simple Treiber stack, headed by the "id" field of ctl, plus a
205 >     * 15bit counter value to both wake up waiters (by advancing their
206 >     * count) and avoid ABA effects. Successors are held in worker
207 >     * field "nextWait".  Queuing deals with several intrinsic races,
208 >     * mainly that a task-producing thread can miss seeing (and
209 >     * signalling) another thread that gave up looking for work but
210 >     * has not yet entered the wait queue. We solve this by requiring
211 >     * a full sweep of all workers both before (in scan()) and after
212 >     * (in tryAwaitWork()) a newly waiting worker is added to the wait
213 >     * queue. During a rescan, the worker might release some other
214 >     * queued worker rather than itself, which has the same net
215 >     * effect. Because enqueued workers may actually be rescanning
216 >     * rather than waiting, we set and clear the "parked" field of
217 >     * ForkJoinWorkerThread to reduce unnecessary calls to unpark.
218 >     * (Use of the parked field requires a secondary recheck to avoid
219 >     * missed signals.)
220 >     *
221 >     * Signalling.  We create or wake up workers only when there
222 >     * appears to be at least one task they might be able to find and
223 >     * execute.  When a submission is added or another worker adds a
224 >     * task to a queue that previously had two or fewer tasks, they
225 >     * signal waiting workers (or trigger creation of new ones if
226 >     * fewer than the given parallelism level -- see signalWork).
227 >     * These primary signals are buttressed by signals during rescans
228 >     * as well as those performed when a worker steals a task and
229 >     * notices that there are more tasks too; together these cover the
230 >     * signals needed in cases when more than two tasks are pushed
231 >     * but untaken.
232 >     *
233 >     * Trimming workers. To release resources after periods of lack of
234 >     * use, a worker starting to wait when the pool is quiescent will
235 >     * time out and terminate if the pool has remained quiescent for
236 >     * SHRINK_RATE nanosecs. This will slowly propagate, eventually
237 >     * terminating all workers after long periods of non-use.
238 >     *
239 >     * Submissions. External submissions are maintained in an
240 >     * array-based queue that is structured identically to
241 >     * ForkJoinWorkerThread queues except for the use of
242 >     * submissionLock in method addSubmission. Unlike the case for
243 >     * worker queues, multiple external threads can add new
244 >     * submissions, so adding requires a lock.
245 >     *
246 >     * Compensation. Beyond work-stealing support and lifecycle
247 >     * control, the main responsibility of this framework is to take
248 >     * actions when one worker is waiting to join a task stolen (or
249 >     * always held by) another.  Because we are multiplexing many
250 >     * tasks on to a pool of workers, we can't just let them block (as
251 >     * in Thread.join).  We also cannot just reassign the joiner's
252 >     * run-time stack with another and replace it later, which would
253 >     * be a form of "continuation", that even if possible is not
254 >     * necessarily a good idea since we sometimes need both an
255 >     * unblocked task and its continuation to progress. Instead we
256 >     * combine two tactics:
257 >     *
258 >     *   Helping: Arranging for the joiner to execute some task that it
259 >     *      would be running if the steal had not occurred.  Method
260 >     *      ForkJoinWorkerThread.joinTask tracks joining->stealing
261 >     *      links to try to find such a task.
262 >     *
263 >     *   Compensating: Unless there are already enough live threads,
264 >     *      method tryPreBlock() may create or re-activate a spare
265 >     *      thread to compensate for blocked joiners until they
266 >     *      unblock.
267 >     *
268 >     * The ManagedBlocker extension API can't use helping so relies
269 >     * only on compensation in method awaitBlocker.
270 >     *
271 >     * It is impossible to keep exactly the target parallelism number
272 >     * of threads running at any given time.  Determining the
273 >     * existence of conservatively safe helping targets, the
274 >     * availability of already-created spares, and the apparent need
275 >     * to create new spares are all racy and require heuristic
276 >     * guidance, so we rely on multiple retries of each.  Currently,
277 >     * in keeping with on-demand signalling policy, we compensate only
278 >     * if blocking would leave less than one active (non-waiting,
279 >     * non-blocked) worker. Additionally, to avoid some false alarms
280 >     * due to GC, lagging counters, system activity, etc, compensated
281 >     * blocking for joins is only attempted after rechecks stabilize
282 >     * (retries are interspersed with Thread.yield, for good
283 >     * citizenship).  The variable blockedCount, incremented before
284 >     * blocking and decremented after, is sometimes needed to
285 >     * distinguish cases of waiting for work vs blocking on joins or
286 >     * other managed sync. Both cases are equivalent for most pool
287 >     * control, so we can update non-atomically. (Additionally,
288 >     * contention on blockedCount alleviates some contention on ctl).
289 >     *
290 >     * Shutdown and Termination. A call to shutdownNow atomically sets
291 >     * the ctl stop bit and then (non-atomically) sets each workers
292 >     * "terminate" status, cancels all unprocessed tasks, and wakes up
293 >     * all waiting workers.  Detecting whether termination should
294 >     * commence after a non-abrupt shutdown() call requires more work
295 >     * and bookkeeping. We need consensus about quiesence (i.e., that
296 >     * there is no more work) which is reflected in active counts so
297 >     * long as there are no current blockers, as well as possible
298 >     * re-evaluations during independent changes in blocking or
299 >     * quiescing workers.
300 >     *
301 >     * Style notes: There is a lot of representation-level coupling
302 >     * among classes ForkJoinPool, ForkJoinWorkerThread, and
303 >     * ForkJoinTask.  Most fields of ForkJoinWorkerThread maintain
304 >     * data structures managed by ForkJoinPool, so are directly
305 >     * accessed.  Conversely we allow access to "workers" array by
306 >     * workers, and direct access to ForkJoinTask.status by both
307 >     * ForkJoinPool and ForkJoinWorkerThread.  There is little point
308 >     * trying to reduce this, since any associated future changes in
309 >     * representations will need to be accompanied by algorithmic
310 >     * changes anyway. All together, these low-level implementation
311 >     * choices produce as much as a factor of 4 performance
312 >     * improvement compared to naive implementations, and enable the
313 >     * processing of billions of tasks per second, at the expense of
314 >     * some ugliness.
315 >     *
316 >     * Methods signalWork() and scan() are the main bottlenecks so are
317 >     * especially heavily micro-optimized/mangled.  There are lots of
318 >     * inline assignments (of form "while ((local = field) != 0)")
319 >     * which are usually the simplest way to ensure the required read
320 >     * orderings (which are sometimes critical). This leads to a
321 >     * "C"-like style of listing declarations of these locals at the
322 >     * heads of methods or blocks.  There are several occurrences of
323 >     * the unusual "do {} while (!cas...)"  which is the simplest way
324 >     * to force an update of a CAS'ed variable. There are also other
325 >     * coding oddities that help some methods perform reasonably even
326 >     * when interpreted (not compiled).
327 >     *
328 >     * The order of declarations in this file is: (1) declarations of
329 >     * statics (2) fields (along with constants used when unpacking
330 >     * some of them), listed in an order that tends to reduce
331 >     * contention among them a bit under most JVMs.  (3) internal
332 >     * control methods (4) callbacks and other support for
333 >     * ForkJoinTask and ForkJoinWorkerThread classes, (5) exported
334 >     * methods (plus a few little helpers). (6) static block
335 >     * initializing all statics in a minimally dependent order.
336       */
337  
64    /** Mask for packing and unpacking shorts */
65    private static final int  shortMask = 0xffff;
66
67    /** Max pool size -- must be a power of two minus 1 */
68    private static final int MAX_THREADS =  0x7FFF;
69
338      /**
339 <     * Factory for creating new ForkJoinWorkerThreads.  A
340 <     * ForkJoinWorkerThreadFactory must be defined and used for
341 <     * ForkJoinWorkerThread subclasses that extend base functionality
342 <     * or initialize threads with different contexts.
339 >     * Factory for creating new {@link ForkJoinWorkerThread}s.
340 >     * A {@code ForkJoinWorkerThreadFactory} must be defined and used
341 >     * for {@code ForkJoinWorkerThread} subclasses that extend base
342 >     * functionality or initialize threads with different contexts.
343       */
344      public static interface ForkJoinWorkerThreadFactory {
345          /**
346           * Returns a new worker thread operating in the given pool.
347           *
348           * @param pool the pool this thread works in
349 <         * @throws NullPointerException if pool is null;
349 >         * @throws NullPointerException if the pool is null
350           */
351          public ForkJoinWorkerThread newThread(ForkJoinPool pool);
352      }
353  
354      /**
355 <     * Default ForkJoinWorkerThreadFactory implementation, creates a
355 >     * Default ForkJoinWorkerThreadFactory implementation; creates a
356       * new ForkJoinWorkerThread.
357       */
358 <    public static class  DefaultForkJoinWorkerThreadFactory
358 >    static class DefaultForkJoinWorkerThreadFactory
359          implements ForkJoinWorkerThreadFactory {
360          public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
361 <            try {
94 <                return new ForkJoinWorkerThread(pool);
95 <            } catch (OutOfMemoryError oom)  {
96 <                return null;
97 <            }
361 >            return new ForkJoinWorkerThread(pool);
362          }
363      }
364  
365      /**
366 <     * The default ForkJoinWorkerThreadFactory, used unless overridden
367 <     * in ForkJoinPool constructors.
366 >     * Creates a new ForkJoinWorkerThread. This factory is used unless
367 >     * overridden in ForkJoinPool constructors.
368       */
369 <    private static final DefaultForkJoinWorkerThreadFactory
370 <        defaultForkJoinWorkerThreadFactory =
107 <        new DefaultForkJoinWorkerThreadFactory();
108 <
369 >    public static final ForkJoinWorkerThreadFactory
370 >        defaultForkJoinWorkerThreadFactory;
371  
372      /**
373       * Permission required for callers of methods that may start or
374       * kill threads.
375       */
376 <    private static final RuntimePermission modifyThreadPermission =
115 <        new RuntimePermission("modifyThread");
376 >    private static final RuntimePermission modifyThreadPermission;
377  
378      /**
379       * If there is a security manager, makes sure caller has
# Line 127 | Line 388 | public class ForkJoinPool extends Abstra
388      /**
389       * Generator for assigning sequence numbers as pool names.
390       */
391 <    private static final AtomicInteger poolNumberGenerator =
131 <        new AtomicInteger();
391 >    private static final AtomicInteger poolNumberGenerator;
392  
393      /**
394 <     * Array holding all worker threads in the pool. Array size must
395 <     * be a power of two.  Updates and replacements are protected by
396 <     * workerLock, but it is always kept in a consistent enough state
397 <     * to be randomly accessed without locking by workers performing
398 <     * work-stealing.
394 >     * Generator for initial random seeds for worker victim
395 >     * selection. This is used only to create initial seeds. Random
396 >     * steals use a cheaper xorshift generator per steal attempt. We
397 >     * don't expect much contention on seedGenerator, so just use a
398 >     * plain Random.
399       */
400 <    volatile ForkJoinWorkerThread[] workers;
400 >    static final Random workerSeedGenerator;
401  
402      /**
403 <     * Lock protecting access to workers.
403 >     * Array holding all worker threads in the pool.  Initialized upon
404 >     * construction. Array size must be a power of two.  Updates and
405 >     * replacements are protected by scanGuard, but the array is
406 >     * always kept in a consistent enough state to be randomly
407 >     * accessed without locking by workers performing work-stealing,
408 >     * as well as other traversal-based methods in this class, so long
409 >     * as reads memory-acquire by first reading ctl. All readers must
410 >     * tolerate that some array slots may be null.
411       */
412 <    private final ReentrantLock workerLock;
412 >    ForkJoinWorkerThread[] workers;
413  
414      /**
415 <     * Condition for awaitTermination.
415 >     * Initial size for submission queue array. Must be a power of
416 >     * two.  In many applications, these always stay small so we use a
417 >     * small initial cap.
418       */
419 <    private final Condition termination;
419 >    private static final int INITIAL_QUEUE_CAPACITY = 8;
420 >
421 >    /**
422 >     * Maximum size for submission queue array. Must be a power of two
423 >     * less than or equal to 1 << (31 - width of array entry) to
424 >     * ensure lack of index wraparound, but is capped at a lower
425 >     * value to help users trap runaway computations.
426 >     */
427 >    private static final int MAXIMUM_QUEUE_CAPACITY = 1 << 24; // 16M
428  
429      /**
430 <     * The uncaught exception handler used when any worker
154 <     * abrupty terminates
430 >     * Array serving as submission queue. Initialized upon construction.
431       */
432 <    private Thread.UncaughtExceptionHandler ueh;
432 >    private ForkJoinTask<?>[] submissionQueue;
433 >
434 >    /**
435 >     * Lock protecting submissions array for addSubmission
436 >     */
437 >    private final ReentrantLock submissionLock;
438 >
439 >    /**
440 >     * Condition for awaitTermination, using submissionLock for
441 >     * convenience.
442 >     */
443 >    private final Condition termination;
444  
445      /**
446       * Creation factory for worker threads.
# Line 161 | Line 448 | public class ForkJoinPool extends Abstra
448      private final ForkJoinWorkerThreadFactory factory;
449  
450      /**
451 <     * Head of stack of threads that were created to maintain
452 <     * parallelism when other threads blocked, but have since
453 <     * suspended when the parallelism level rose.
451 >     * The uncaught exception handler used when any worker abruptly
452 >     * terminates.
453 >     */
454 >    final Thread.UncaughtExceptionHandler ueh;
455 >
456 >    /**
457 >     * Prefix for assigning names to worker threads
458       */
459 <    private volatile WaitQueueNode spareStack;
459 >    private final String workerNamePrefix;
460  
461      /**
462       * Sum of per-thread steal counts, updated only when threads are
463       * idle or terminating.
464       */
465 <    private final AtomicLong stealCount;
465 >    private volatile long stealCount;
466  
467      /**
468 <     * Queue for external submissions.
468 >     * Main pool control -- a long packed with:
469 >     * AC: Number of active running workers minus target parallelism (16 bits)
470 >     * TC: Number of total workers minus target parallelism (16bits)
471 >     * ST: true if pool is terminating (1 bit)
472 >     * EC: the wait count of top waiting thread (15 bits)
473 >     * ID: ~poolIndex of top of Treiber stack of waiting threads (16 bits)
474 >     *
475 >     * When convenient, we can extract the upper 32 bits of counts and
476 >     * the lower 32 bits of queue state, u = (int)(ctl >>> 32) and e =
477 >     * (int)ctl.  The ec field is never accessed alone, but always
478 >     * together with id and st. The offsets of counts by the target
479 >     * parallelism and the positionings of fields makes it possible to
480 >     * perform the most common checks via sign tests of fields: When
481 >     * ac is negative, there are not enough active workers, when tc is
482 >     * negative, there are not enough total workers, when id is
483 >     * negative, there is at least one waiting worker, and when e is
484 >     * negative, the pool is terminating.  To deal with these possibly
485 >     * negative fields, we use casts in and out of "short" and/or
486 >     * signed shifts to maintain signedness.
487       */
488 <    private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue;
488 >    volatile long ctl;
489 >
490 >    // bit positions/shifts for fields
491 >    private static final int  AC_SHIFT   = 48;
492 >    private static final int  TC_SHIFT   = 32;
493 >    private static final int  ST_SHIFT   = 31;
494 >    private static final int  EC_SHIFT   = 16;
495 >
496 >    // bounds
497 >    private static final int  MAX_ID     = 0x7fff;  // max poolIndex
498 >    private static final int  SMASK      = 0xffff;  // mask short bits
499 >    private static final int  SHORT_SIGN = 1 << 15;
500 >    private static final int  INT_SIGN   = 1 << 31;
501 >
502 >    // masks
503 >    private static final long STOP_BIT   = 0x0001L << ST_SHIFT;
504 >    private static final long AC_MASK    = ((long)SMASK) << AC_SHIFT;
505 >    private static final long TC_MASK    = ((long)SMASK) << TC_SHIFT;
506 >
507 >    // units for incrementing and decrementing
508 >    private static final long TC_UNIT    = 1L << TC_SHIFT;
509 >    private static final long AC_UNIT    = 1L << AC_SHIFT;
510 >
511 >    // masks and units for dealing with u = (int)(ctl >>> 32)
512 >    private static final int  UAC_SHIFT  = AC_SHIFT - 32;
513 >    private static final int  UTC_SHIFT  = TC_SHIFT - 32;
514 >    private static final int  UAC_MASK   = SMASK << UAC_SHIFT;
515 >    private static final int  UTC_MASK   = SMASK << UTC_SHIFT;
516 >    private static final int  UAC_UNIT   = 1 << UAC_SHIFT;
517 >    private static final int  UTC_UNIT   = 1 << UTC_SHIFT;
518 >
519 >    // masks and units for dealing with e = (int)ctl
520 >    private static final int  E_MASK     = 0x7fffffff; // no STOP_BIT
521 >    private static final int  EC_UNIT    = 1 << EC_SHIFT;
522  
523      /**
524 <     * Head of Treiber stack for barrier sync. See below for explanation
524 >     * The target parallelism level.
525       */
526 <    private volatile WaitQueueNode barrierStack;
526 >    final int parallelism;
527  
528      /**
529 <     * The count for event barrier
529 >     * Index (mod submission queue length) of next element to take
530 >     * from submission queue. Usage is identical to that for
531 >     * per-worker queues -- see ForkJoinWorkerThread internal
532 >     * documentation.
533       */
534 <    private volatile long eventCount;
534 >    volatile int queueBase;
535  
536      /**
537 <     * Pool number, just for assigning useful names to worker threads
537 >     * Index (mod submission queue length) of next element to add
538 >     * in submission queue. Usage is identical to that for
539 >     * per-worker queues -- see ForkJoinWorkerThread internal
540 >     * documentation.
541       */
542 <    private final int poolNumber;
542 >    int queueTop;
543  
544      /**
545 <     * The maximum allowed pool size
545 >     * True when shutdown() has been called.
546       */
547 <    private volatile int maxPoolSize;
547 >    volatile boolean shutdown;
548  
549      /**
550 <     * The desired parallelism level, updated only under workerLock.
550 >     * True if use local fifo, not default lifo, for local polling
551 >     * Read by, and replicated by ForkJoinWorkerThreads
552       */
553 <    private volatile int parallelism;
553 >    final boolean locallyFifo;
554  
555      /**
556 <     * Holds number of total (i.e., created and not yet terminated)
557 <     * and running (i.e., not blocked on joins or other managed sync)
558 <     * threads, packed into one int to ensure consistent snapshot when
210 <     * making decisions about creating and suspending spare
211 <     * threads. Updated only by CAS.  Note: CASes in
212 <     * updateRunningCount and preJoin running active count is in low
213 <     * word, so need to be modified if this changes
556 >     * The number of threads in ForkJoinWorkerThreads.helpQuiescePool.
557 >     * When non-zero, suppresses automatic shutdown when active
558 >     * counts become zero.
559       */
560 <    private volatile int workerCounts;
216 <
217 <    private static int totalCountOf(int s)           { return s >>> 16;  }
218 <    private static int runningCountOf(int s)         { return s & shortMask; }
219 <    private static int workerCountsFor(int t, int r) { return (t << 16) + r; }
560 >    volatile int quiescerCount;
561  
562      /**
563 <     * Add delta (which may be negative) to running count.  This must
223 <     * be called before (with negative arg) and after (with positive)
224 <     * any managed synchronization (i.e., mainly, joins)
225 <     * @param delta the number to add
563 >     * The number of threads blocked in join.
564       */
565 <    final void updateRunningCount(int delta) {
228 <        int s;
229 <        do;while (!casWorkerCounts(s = workerCounts, s + delta));
230 <    }
565 >    volatile int blockedCount;
566  
567      /**
568 <     * Add delta (which may be negative) to both total and running
234 <     * count.  This must be called upon creation and termination of
235 <     * worker threads.
236 <     * @param delta the number to add
568 >     * Counter for worker Thread names (unrelated to their poolIndex)
569       */
570 <    private void updateWorkerCount(int delta) {
239 <        int d = delta + (delta << 16); // add to both lo and hi parts
240 <        int s;
241 <        do;while (!casWorkerCounts(s = workerCounts, s + d));
242 <    }
570 >    private volatile int nextWorkerNumber;
571  
572      /**
573 <     * Lifecycle control. High word contains runState, low word
246 <     * contains the number of workers that are (probably) executing
247 <     * tasks. This value is atomically incremented before a worker
248 <     * gets a task to run, and decremented when worker has no tasks
249 <     * and cannot find any. These two fields are bundled together to
250 <     * support correct termination triggering.  Note: activeCount
251 <     * CAS'es cheat by assuming active count is in low word, so need
252 <     * to be modified if this changes
573 >     * The index for the next created worker. Accessed under scanGuard.
574       */
575 <    private volatile int runControl;
575 >    private int nextWorkerIndex;
576  
577 <    // RunState values. Order among values matters
578 <    private static final int RUNNING     = 0;
579 <    private static final int SHUTDOWN    = 1;
580 <    private static final int TERMINATING = 2;
581 <    private static final int TERMINATED  = 3;
577 >    /**
578 >     * SeqLock and index masking for updates to workers array.  Locked
579 >     * when SG_UNIT is set. Unlocking clears bit by adding
580 >     * SG_UNIT. Staleness of read-only operations can be checked by
581 >     * comparing scanGuard to value before the reads. The low 16 bits
582 >     * (i.e, anding with SMASK) hold (the smallest power of two
583 >     * covering all worker indices, minus one, and is used to avoid
584 >     * dealing with large numbers of null slots when the workers array
585 >     * is overallocated.
586 >     */
587 >    volatile int scanGuard;
588  
589 <    private static int runStateOf(int c)             { return c >>> 16; }
263 <    private static int activeCountOf(int c)          { return c & shortMask; }
264 <    private static int runControlFor(int r, int a)   { return (r << 16) + a; }
589 >    private static final int SG_UNIT = 1 << 16;
590  
591      /**
592 <     * Increment active count. Called by workers before/during
593 <     * executing tasks.
592 >     * The wakeup interval (in nanoseconds) for a worker waiting for a
593 >     * task when the pool is quiescent to instead try to shrink the
594 >     * number of workers.  The exact value does not matter too
595 >     * much. It must be short enough to release resources during
596 >     * sustained periods of idleness, but not so short that threads
597 >     * are continually re-created.
598       */
599 <    final void incrementActiveCount() {
600 <        int c;
272 <        do;while (!casRunControl(c = runControl, c+1));
273 <    }
599 >    private static final long SHRINK_RATE =
600 >        4L * 1000L * 1000L * 1000L; // 4 seconds
601  
602      /**
603 <     * Decrement active count; possibly trigger termination.
604 <     * Called by workers when they can't find tasks.
603 >     * Top-level loop for worker threads: On each step: if the
604 >     * previous step swept through all queues and found no tasks, or
605 >     * there are excess threads, then possibly blocks. Otherwise,
606 >     * scans for and, if found, executes a task. Returns when pool
607 >     * and/or worker terminate.
608 >     *
609 >     * @param w the worker
610       */
611 <    final void decrementActiveCount() {
612 <        int c, nextc;
613 <        do;while (!casRunControl(c = runControl, nextc = c-1));
614 <        if (canTerminateOnShutdown(nextc))
615 <            terminateOnShutdown();
611 >    final void work(ForkJoinWorkerThread w) {
612 >        boolean swept = false;                // true on empty scans
613 >        long c;
614 >        while (!w.terminate && (int)(c = ctl) >= 0) {
615 >            int a;                            // active count
616 >            if (!swept && (a = (int)(c >> AC_SHIFT)) <= 0)
617 >                swept = scan(w, a);
618 >            else if (tryAwaitWork(w, c))
619 >                swept = false;
620 >        }
621      }
622  
623 <    /**
624 <     * Return true if argument represents zero active count and
625 <     * nonzero runstate, which is the triggering condition for
626 <     * terminating on shutdown.
627 <     */
628 <    private static boolean canTerminateOnShutdown(int c) {
629 <        return ((c & -c) >>> 16) != 0; // i.e. least bit is nonzero runState bit
623 >    // Signalling
624 >
625 >    /**
626 >     * Wakes up or creates a worker.
627 >     */
628 >    final void signalWork() {
629 >        /*
630 >         * The while condition is true if: (there is are too few total
631 >         * workers OR there is at least one waiter) AND (there are too
632 >         * few active workers OR the pool is terminating).  The value
633 >         * of e distinguishes the remaining cases: zero (no waiters)
634 >         * for create, negative if terminating (in which case do
635 >         * nothing), else release a waiter. The secondary checks for
636 >         * release (non-null array etc) can fail if the pool begins
637 >         * terminating after the test, and don't impose any added cost
638 >         * because JVMs must perform null and bounds checks anyway.
639 >         */
640 >        long c; int e, u;
641 >        while ((((e = (int)(c = ctl)) | (u = (int)(c >>> 32))) &
642 >                (INT_SIGN|SHORT_SIGN)) == (INT_SIGN|SHORT_SIGN) && e >= 0) {
643 >            if (e > 0) {                         // release a waiting worker
644 >                int i; ForkJoinWorkerThread w; ForkJoinWorkerThread[] ws;
645 >                if ((ws = workers) == null ||
646 >                    (i = ~e & SMASK) >= ws.length ||
647 >                    (w = ws[i]) == null)
648 >                    break;
649 >                long nc = (((long)(w.nextWait & E_MASK)) |
650 >                           ((long)(u + UAC_UNIT) << 32));
651 >                if (w.eventCount == e &&
652 >                    UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc)) {
653 >                    w.eventCount = (e + EC_UNIT) & E_MASK;
654 >                    if (w.parked)
655 >                        UNSAFE.unpark(w);
656 >                    break;
657 >                }
658 >            }
659 >            else if (UNSAFE.compareAndSwapLong
660 >                     (this, ctlOffset, c,
661 >                      (long)(((u + UTC_UNIT) & UTC_MASK) |
662 >                             ((u + UAC_UNIT) & UAC_MASK)) << 32)) {
663 >                addWorker();
664 >                break;
665 >            }
666 >        }
667      }
668  
669      /**
670 <     * Transition run state to at least the given state. Return true
671 <     * if not already at least given state.
670 >     * Variant of signalWork to help release waiters on rescans.
671 >     * Tries once to release a waiter if active count < 0.
672 >     *
673 >     * @return false if failed due to contention, else true
674 >     */
675 >    private boolean tryReleaseWaiter() {
676 >        long c; int e, i; ForkJoinWorkerThread w; ForkJoinWorkerThread[] ws;
677 >        if ((e = (int)(c = ctl)) > 0 &&
678 >            (int)(c >> AC_SHIFT) < 0 &&
679 >            (ws = workers) != null &&
680 >            (i = ~e & SMASK) < ws.length &&
681 >            (w = ws[i]) != null) {
682 >            long nc = ((long)(w.nextWait & E_MASK) |
683 >                       ((c + AC_UNIT) & (AC_MASK|TC_MASK)));
684 >            if (w.eventCount != e ||
685 >                !UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc))
686 >                return false;
687 >            w.eventCount = (e + EC_UNIT) & E_MASK;
688 >            if (w.parked)
689 >                UNSAFE.unpark(w);
690 >        }
691 >        return true;
692 >    }
693 >
694 >    // Scanning for tasks
695 >
696 >    /**
697 >     * Scans for and, if found, executes one task. Scans start at a
698 >     * random index of workers array, and randomly select the first
699 >     * (2*#workers)-1 probes, and then, if all empty, resort to 2
700 >     * circular sweeps, which is necessary to check quiescence. and
701 >     * taking a submission only if no stealable tasks were found.  The
702 >     * steal code inside the loop is a specialized form of
703 >     * ForkJoinWorkerThread.deqTask, followed bookkeeping to support
704 >     * helpJoinTask and signal propagation. The code for submission
705 >     * queues is almost identical. On each steal, the worker completes
706 >     * not only the task, but also all local tasks that this task may
707 >     * have generated. On detecting staleness or contention when
708 >     * trying to take a task, this method returns without finishing
709 >     * sweep, which allows global state rechecks before retry.
710 >     *
711 >     * @param w the worker
712 >     * @param a the number of active workers
713 >     * @return true if swept all queues without finding a task
714       */
715 <    private boolean transitionRunStateTo(int state) {
716 <        for (;;) {
717 <            int c = runControl;
718 <            if (runStateOf(c) >= state)
715 >    private boolean scan(ForkJoinWorkerThread w, int a) {
716 >        int g = scanGuard; // mask 0 avoids useless scans if only one active
717 >        int m = (parallelism == 1 - a && blockedCount == 0) ? 0 : g & SMASK;
718 >        ForkJoinWorkerThread[] ws = workers;
719 >        if (ws == null || ws.length <= m)         // staleness check
720 >            return false;
721 >        for (int r = w.seed, k = r, j = -(m + m); j <= m + m; ++j) {
722 >            ForkJoinTask<?> t; ForkJoinTask<?>[] q; int b, i;
723 >            ForkJoinWorkerThread v = ws[k & m];
724 >            if (v != null && (b = v.queueBase) != v.queueTop &&
725 >                (q = v.queue) != null && (i = (q.length - 1) & b) >= 0) {
726 >                long u = (i << ASHIFT) + ABASE;
727 >                if ((t = q[i]) != null && v.queueBase == b &&
728 >                    UNSAFE.compareAndSwapObject(q, u, t, null)) {
729 >                    int d = (v.queueBase = b + 1) - v.queueTop;
730 >                    v.stealHint = w.poolIndex;
731 >                    if (d != 0)
732 >                        signalWork();             // propagate if nonempty
733 >                    w.execTask(t);
734 >                }
735 >                r ^= r << 13; r ^= r >>> 17; w.seed = r ^ (r << 5);
736 >                return false;                     // store next seed
737 >            }
738 >            else if (j < 0) {                     // xorshift
739 >                r ^= r << 13; r ^= r >>> 17; k = r ^= r << 5;
740 >            }
741 >            else
742 >                ++k;
743 >        }
744 >        if (scanGuard != g)                       // staleness check
745 >            return false;
746 >        else {                                    // try to take submission
747 >            ForkJoinTask<?> t; ForkJoinTask<?>[] q; int b, i;
748 >            if ((b = queueBase) != queueTop &&
749 >                (q = submissionQueue) != null &&
750 >                (i = (q.length - 1) & b) >= 0) {
751 >                long u = (i << ASHIFT) + ABASE;
752 >                if ((t = q[i]) != null && queueBase == b &&
753 >                    UNSAFE.compareAndSwapObject(q, u, t, null)) {
754 >                    queueBase = b + 1;
755 >                    w.execTask(t);
756 >                }
757                  return false;
758 <            if (casRunControl(c, runControlFor(state, activeCountOf(c))))
758 >            }
759 >            return true;                         // all queues empty
760 >        }
761 >    }
762 >
763 >    /**
764 >     * Tries to enqueue worker w in wait queue and await change in
765 >     * worker's eventCount.  If the pool is quiescent and there is
766 >     * more than one worker, possibly terminates worker upon exit.
767 >     * Otherwise, before blocking, rescans queues to avoid missed
768 >     * signals.  Upon finding work, releases at least one worker
769 >     * (which may be the current worker). Rescans restart upon
770 >     * detected staleness or failure to release due to
771 >     * contention. Note the unusual conventions about Thread.interrupt
772 >     * here and elsewhere: Because interrupts are used solely to alert
773 >     * threads to check termination, which is checked here anyway, we
774 >     * clear status (using Thread.interrupted) before any call to
775 >     * park, so that park does not immediately return due to status
776 >     * being set via some other unrelated call to interrupt in user
777 >     * code.
778 >     *
779 >     * @param w the calling worker
780 >     * @param c the ctl value on entry
781 >     * @return true if waited or another thread was released upon enq
782 >     */
783 >    private boolean tryAwaitWork(ForkJoinWorkerThread w, long c) {
784 >        int v = w.eventCount;
785 >        w.nextWait = (int)c;                      // w's successor record
786 >        long nc = (long)(v & E_MASK) | ((c - AC_UNIT) & (AC_MASK|TC_MASK));
787 >        if (ctl != c || !UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc)) {
788 >            long d = ctl; // return true if lost to a deq, to force scan
789 >            return (int)d != (int)c && ((d - c) & AC_MASK) >= 0L;
790 >        }
791 >        for (int sc = w.stealCount; sc != 0;) {   // accumulate stealCount
792 >            long s = stealCount;
793 >            if (UNSAFE.compareAndSwapLong(this, stealCountOffset, s, s + sc))
794 >                sc = w.stealCount = 0;
795 >            else if (w.eventCount != v)
796 >                return true;                      // update next time
797 >        }
798 >        if ((int)c != 0 && parallelism + (int)(nc >> AC_SHIFT) == 0 &&
799 >            blockedCount == 0 && quiescerCount == 0)
800 >            idleAwaitWork(w, nc, c, v);           // quiescent
801 >        for (boolean rescanned = false;;) {
802 >            if (w.eventCount != v)
803                  return true;
804 +            if (!rescanned) {
805 +                int g = scanGuard, m = g & SMASK;
806 +                ForkJoinWorkerThread[] ws = workers;
807 +                if (ws != null && m < ws.length) {
808 +                    rescanned = true;
809 +                    for (int i = 0; i <= m; ++i) {
810 +                        ForkJoinWorkerThread u = ws[i];
811 +                        if (u != null) {
812 +                            if (u.queueBase != u.queueTop &&
813 +                                !tryReleaseWaiter())
814 +                                rescanned = false; // contended
815 +                            if (w.eventCount != v)
816 +                                return true;
817 +                        }
818 +                    }
819 +                }
820 +                if (scanGuard != g ||              // stale
821 +                    (queueBase != queueTop && !tryReleaseWaiter()))
822 +                    rescanned = false;
823 +                if (!rescanned)
824 +                    Thread.yield();                // reduce contention
825 +                else
826 +                    Thread.interrupted();          // clear before park
827 +            }
828 +            else {
829 +                w.parked = true;                   // must recheck
830 +                if (w.eventCount != v) {
831 +                    w.parked = false;
832 +                    return true;
833 +                }
834 +                LockSupport.park(this);
835 +                rescanned = w.parked = false;
836 +            }
837 +        }
838 +    }
839 +
840 +    /**
841 +     * If inactivating worker w has caused pool to become
842 +     * quiescent, check for pool termination, and wait for event
843 +     * for up to SHRINK_RATE nanosecs (rescans are unnecessary in
844 +     * this case because quiescence reflects consensus about lack
845 +     * of work). On timeout, if ctl has not changed, terminate the
846 +     * worker. Upon its termination (see deregisterWorker), it may
847 +     * wake up another worker to possibly repeat this process.
848 +     *
849 +     * @param w the calling worker
850 +     * @param currentCtl the ctl value after enqueuing w
851 +     * @param prevCtl the ctl value if w terminated
852 +     * @param v the eventCount w awaits change
853 +     */
854 +    private void idleAwaitWork(ForkJoinWorkerThread w, long currentCtl,
855 +                               long prevCtl, int v) {
856 +        if (w.eventCount == v) {
857 +            if (shutdown)
858 +                tryTerminate(false);
859 +            ForkJoinTask.helpExpungeStaleExceptions(); // help clean weak refs
860 +            while (ctl == currentCtl) {
861 +                long startTime = System.nanoTime();
862 +                w.parked = true;
863 +                if (w.eventCount == v)             // must recheck
864 +                    LockSupport.parkNanos(this, SHRINK_RATE);
865 +                w.parked = false;
866 +                if (w.eventCount != v)
867 +                    break;
868 +                else if (System.nanoTime() - startTime <
869 +                         SHRINK_RATE - (SHRINK_RATE / 10)) // timing slop
870 +                    Thread.interrupted();          // spurious wakeup
871 +                else if (UNSAFE.compareAndSwapLong(this, ctlOffset,
872 +                                                   currentCtl, prevCtl)) {
873 +                    w.terminate = true;            // restore previous
874 +                    w.eventCount = ((int)currentCtl + EC_UNIT) & E_MASK;
875 +                    break;
876 +                }
877 +            }
878          }
879      }
880  
881 +    // Submissions
882 +
883      /**
884 <     * Controls whether to add spares to maintain parallelism
884 >     * Enqueues the given task in the submissionQueue.  Same idea as
885 >     * ForkJoinWorkerThread.pushTask except for use of submissionLock.
886 >     *
887 >     * @param t the task
888       */
889 <    private volatile boolean maintainsParallelism;
889 >    private void addSubmission(ForkJoinTask<?> t) {
890 >        final ReentrantLock lock = this.submissionLock;
891 >        lock.lock();
892 >        try {
893 >            ForkJoinTask<?>[] q; int s, m;
894 >            if ((q = submissionQueue) != null) {    // ignore if queue removed
895 >                long u = (((s = queueTop) & (m = q.length-1)) << ASHIFT)+ABASE;
896 >                UNSAFE.putOrderedObject(q, u, t);
897 >                queueTop = s + 1;
898 >                if (s - queueBase == m)
899 >                    growSubmissionQueue();
900 >            }
901 >        } finally {
902 >            lock.unlock();
903 >        }
904 >        signalWork();
905 >    }
906  
907 <    // Constructors
907 >    //  (pollSubmission is defined below with exported methods)
908  
909      /**
910 <     * Creates a ForkJoinPool with a pool size equal to the number of
911 <     * processors available on the system and using the default
319 <     * ForkJoinWorkerThreadFactory,
320 <     * @throws SecurityException if a security manager exists and
321 <     *         the caller is not permitted to modify threads
322 <     *         because it does not hold {@link
323 <     *         java.lang.RuntimePermission}<tt>("modifyThread")</tt>,
910 >     * Creates or doubles submissionQueue array.
911 >     * Basically identical to ForkJoinWorkerThread version.
912       */
913 <    public ForkJoinPool() {
914 <        this(Runtime.getRuntime().availableProcessors(),
915 <             defaultForkJoinWorkerThreadFactory);
913 >    private void growSubmissionQueue() {
914 >        ForkJoinTask<?>[] oldQ = submissionQueue;
915 >        int size = oldQ != null ? oldQ.length << 1 : INITIAL_QUEUE_CAPACITY;
916 >        if (size > MAXIMUM_QUEUE_CAPACITY)
917 >            throw new RejectedExecutionException("Queue capacity exceeded");
918 >        if (size < INITIAL_QUEUE_CAPACITY)
919 >            size = INITIAL_QUEUE_CAPACITY;
920 >        ForkJoinTask<?>[] q = submissionQueue = new ForkJoinTask<?>[size];
921 >        int mask = size - 1;
922 >        int top = queueTop;
923 >        int oldMask;
924 >        if (oldQ != null && (oldMask = oldQ.length - 1) >= 0) {
925 >            for (int b = queueBase; b != top; ++b) {
926 >                long u = ((b & oldMask) << ASHIFT) + ABASE;
927 >                Object x = UNSAFE.getObjectVolatile(oldQ, u);
928 >                if (x != null && UNSAFE.compareAndSwapObject(oldQ, u, x, null))
929 >                    UNSAFE.putObjectVolatile
930 >                        (q, ((b & mask) << ASHIFT) + ABASE, x);
931 >            }
932 >        }
933      }
934  
935 +    // Blocking support
936 +
937      /**
938 <     * Creates a ForkJoinPool with the indicated parellelism level
939 <     * threads, and using the default ForkJoinWorkerThreadFactory,
940 <     * @param parallelism the number of worker threads
941 <     * @throws IllegalArgumentException if parallelism less than or
942 <     * equal to zero
943 <     * @throws SecurityException if a security manager exists and
337 <     *         the caller is not permitted to modify threads
338 <     *         because it does not hold {@link
339 <     *         java.lang.RuntimePermission}<tt>("modifyThread")</tt>,
938 >     * Tries to increment blockedCount, decrement active count
939 >     * (sometimes implicitly) and possibly release or create a
940 >     * compensating worker in preparation for blocking. Fails
941 >     * on contention or termination.
942 >     *
943 >     * @return true if the caller can block, else should recheck and retry
944       */
945 <    public ForkJoinPool(int parallelism) {
946 <        this(parallelism, defaultForkJoinWorkerThreadFactory);
945 >    private boolean tryPreBlock() {
946 >        int b = blockedCount;
947 >        if (UNSAFE.compareAndSwapInt(this, blockedCountOffset, b, b + 1)) {
948 >            int pc = parallelism;
949 >            do {
950 >                ForkJoinWorkerThread[] ws; ForkJoinWorkerThread w;
951 >                int e, ac, tc, rc, i;
952 >                long c = ctl;
953 >                int u = (int)(c >>> 32);
954 >                if ((e = (int)c) < 0) {
955 >                                                 // skip -- terminating
956 >                }
957 >                else if ((ac = (u >> UAC_SHIFT)) <= 0 && e != 0 &&
958 >                         (ws = workers) != null &&
959 >                         (i = ~e & SMASK) < ws.length &&
960 >                         (w = ws[i]) != null) {
961 >                    long nc = ((long)(w.nextWait & E_MASK) |
962 >                               (c & (AC_MASK|TC_MASK)));
963 >                    if (w.eventCount == e &&
964 >                        UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc)) {
965 >                        w.eventCount = (e + EC_UNIT) & E_MASK;
966 >                        if (w.parked)
967 >                            UNSAFE.unpark(w);
968 >                        return true;             // release an idle worker
969 >                    }
970 >                }
971 >                else if ((tc = (short)(u >>> UTC_SHIFT)) >= 0 && ac + pc > 1) {
972 >                    long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK);
973 >                    if (UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc))
974 >                        return true;             // no compensation needed
975 >                }
976 >                else if (tc + pc < MAX_ID) {
977 >                    long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK);
978 >                    if (UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc)) {
979 >                        addWorker();
980 >                        return true;            // create a replacement
981 >                    }
982 >                }
983 >                // try to back out on any failure and let caller retry
984 >            } while (!UNSAFE.compareAndSwapInt(this, blockedCountOffset,
985 >                                               b = blockedCount, b - 1));
986 >        }
987 >        return false;
988      }
989  
990      /**
991 <     * Creates a ForkJoinPool with a pool size equal to the number of
347 <     * processors available on the system and using the given
348 <     * ForkJoinWorkerThreadFactory,
349 <     * @param factory the factory for creating new threads
350 <     * @throws NullPointerException if factory is null
351 <     * @throws SecurityException if a security manager exists and
352 <     *         the caller is not permitted to modify threads
353 <     *         because it does not hold {@link
354 <     *         java.lang.RuntimePermission}<tt>("modifyThread")</tt>,
991 >     * Decrements blockedCount and increments active count
992       */
993 <    public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
994 <        this(Runtime.getRuntime().availableProcessors(), factory);
993 >    private void postBlock() {
994 >        long c;
995 >        do {} while (!UNSAFE.compareAndSwapLong(this, ctlOffset,  // no mask
996 >                                                c = ctl, c + AC_UNIT));
997 >        int b;
998 >        do {} while(!UNSAFE.compareAndSwapInt(this, blockedCountOffset,
999 >                                              b = blockedCount, b - 1));
1000      }
1001  
1002      /**
1003 <     * Creates a ForkJoinPool with the indicated target number of
1004 <     * worker threads and the given factory.
1003 >     * Possibly blocks waiting for the given task to complete, or
1004 >     * cancels the task if terminating.  Fails to wait if contended.
1005       *
1006 <     * @param parallelism the targeted number of worker threads
365 <     * @param factory the factory for creating new threads
366 <     * @throws IllegalArgumentException if parallelism less than or
367 <     * equal to zero, or greater than implementation limit.
368 <     * @throws NullPointerException if factory is null
369 <     * @throws SecurityException if a security manager exists and
370 <     *         the caller is not permitted to modify threads
371 <     *         because it does not hold {@link
372 <     *         java.lang.RuntimePermission}<tt>("modifyThread")</tt>,
1006 >     * @param joinMe the task
1007       */
1008 <    public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
1009 <        if (parallelism <= 0 || parallelism > MAX_THREADS)
1010 <            throw new IllegalArgumentException();
1011 <        if (factory == null)
1012 <            throw new NullPointerException();
1013 <        checkPermission();
1014 <        this.factory = factory;
1015 <        this.parallelism = parallelism;
1016 <        this.maxPoolSize = MAX_THREADS;
1017 <        this.maintainsParallelism = true;
1018 <        this.poolNumber = poolNumberGenerator.incrementAndGet();
385 <        this.workerLock = new ReentrantLock();
386 <        this.termination = workerLock.newCondition();
387 <        this.stealCount = new AtomicLong();
388 <        this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>();
389 <        createAndStartInitialWorkers(parallelism);
1008 >    final void tryAwaitJoin(ForkJoinTask<?> joinMe) {
1009 >        int s;
1010 >        Thread.interrupted(); // clear interrupts before checking termination
1011 >        if (joinMe.status >= 0) {
1012 >            if (tryPreBlock()) {
1013 >                joinMe.tryAwaitDone(0L);
1014 >                postBlock();
1015 >            }
1016 >            else if ((ctl & STOP_BIT) != 0L)
1017 >                joinMe.cancelIgnoringExceptions();
1018 >        }
1019      }
1020  
1021      /**
1022 <     * Create new worker using factory.
1023 <     * @param index the index to assign worker
1024 <     * @return new worker, or null of factory failed
1022 >     * Possibly blocks the given worker waiting for joinMe to
1023 >     * complete or timeout
1024 >     *
1025 >     * @param joinMe the task
1026 >     * @param millis the wait time for underlying Object.wait
1027       */
1028 <    private ForkJoinWorkerThread createWorker(int index) {
1029 <        Thread.UncaughtExceptionHandler h = ueh;
1030 <        ForkJoinWorkerThread w = factory.newThread(this);
1031 <        if (w != null) {
1032 <            w.poolIndex = index;
1033 <            w.setDaemon(true);
1034 <            w.setName("ForkJoinPool-" + poolNumber + "-worker-" + index);
1035 <            if (h != null)
1036 <                w.setUncaughtExceptionHandler(h);
1028 >    final void timedAwaitJoin(ForkJoinTask<?> joinMe, long nanos) {
1029 >        while (joinMe.status >= 0) {
1030 >            Thread.interrupted();
1031 >            if ((ctl & STOP_BIT) != 0L) {
1032 >                joinMe.cancelIgnoringExceptions();
1033 >                break;
1034 >            }
1035 >            if (tryPreBlock()) {
1036 >                long last = System.nanoTime();
1037 >                while (joinMe.status >= 0) {
1038 >                    long millis = TimeUnit.NANOSECONDS.toMillis(nanos);
1039 >                    if (millis <= 0)
1040 >                        break;
1041 >                    joinMe.tryAwaitDone(millis);
1042 >                    if (joinMe.status < 0)
1043 >                        break;
1044 >                    if ((ctl & STOP_BIT) != 0L) {
1045 >                        joinMe.cancelIgnoringExceptions();
1046 >                        break;
1047 >                    }
1048 >                    long now = System.nanoTime();
1049 >                    nanos -= now - last;
1050 >                    last = now;
1051 >                }
1052 >                postBlock();
1053 >                break;
1054 >            }
1055          }
407        return w;
1056      }
1057  
1058      /**
1059 <     * Return a good size for worker array given pool size.
412 <     * Currently requires size to be a power of two.
1059 >     * If necessary, compensates for blocker, and blocks
1060       */
1061 <    private static int arraySizeFor(int ps) {
1062 <        return ps <= 1? 1 : (1 << (32 - Integer.numberOfLeadingZeros(ps-1)));
1061 >    private void awaitBlocker(ManagedBlocker blocker)
1062 >        throws InterruptedException {
1063 >        while (!blocker.isReleasable()) {
1064 >            if (tryPreBlock()) {
1065 >                try {
1066 >                    do {} while (!blocker.isReleasable() && !blocker.block());
1067 >                } finally {
1068 >                    postBlock();
1069 >                }
1070 >                break;
1071 >            }
1072 >        }
1073      }
1074  
1075 +    // Creating, registering and deregistring workers
1076 +
1077      /**
1078 <     * Create or resize array if necessary to hold newLength
1079 <     * @return the array
1078 >     * Tries to create and start a worker; minimally rolls back counts
1079 >     * on failure.
1080       */
1081 <    private ForkJoinWorkerThread[] ensureWorkerArrayCapacity(int newLength) {
1082 <        ForkJoinWorkerThread[] ws = workers;
1083 <        if (ws == null)
1084 <            return workers = new ForkJoinWorkerThread[arraySizeFor(newLength)];
1085 <        else if (newLength > ws.length)
1086 <            return workers = Arrays.copyOf(ws, arraySizeFor(newLength));
1081 >    private void addWorker() {
1082 >        Throwable ex = null;
1083 >        ForkJoinWorkerThread t = null;
1084 >        try {
1085 >            t = factory.newThread(this);
1086 >        } catch (Throwable e) {
1087 >            ex = e;
1088 >        }
1089 >        if (t == null) {  // null or exceptional factory return
1090 >            long c;       // adjust counts
1091 >            do {} while (!UNSAFE.compareAndSwapLong
1092 >                         (this, ctlOffset, c = ctl,
1093 >                          (((c - AC_UNIT) & AC_MASK) |
1094 >                           ((c - TC_UNIT) & TC_MASK) |
1095 >                           (c & ~(AC_MASK|TC_MASK)))));
1096 >            // Propagate exception if originating from an external caller
1097 >            if (!tryTerminate(false) && ex != null &&
1098 >                !(Thread.currentThread() instanceof ForkJoinWorkerThread))
1099 >                UNSAFE.throwException(ex);
1100 >        }
1101          else
1102 <            return ws;
1102 >            t.start();
1103      }
1104  
1105      /**
1106 <     * Try to shrink workers into smaller array after one or more terminate
1106 >     * Callback from ForkJoinWorkerThread constructor to assign a
1107 >     * public name
1108       */
1109 <    private void tryShrinkWorkerArray() {
1110 <        ForkJoinWorkerThread[] ws = workers;
1111 <        int len = ws.length;
1112 <        int last = len - 1;
1113 <        while (last >= 0 && ws[last] == null)
1114 <            --last;
441 <        int newLength = arraySizeFor(last+1);
442 <        if (newLength < len)
443 <            workers = Arrays.copyOf(ws, newLength);
1109 >    final String nextWorkerName() {
1110 >        for (int n;;) {
1111 >            if (UNSAFE.compareAndSwapInt(this, nextWorkerNumberOffset,
1112 >                                         n = nextWorkerNumber, ++n))
1113 >                return workerNamePrefix + n;
1114 >        }
1115      }
1116  
1117      /**
1118 <     * Initial worker array and worker creation and startup. (This
1119 <     * must be done under lock to avoid interference by some of the
1120 <     * newly started threads while creating others.)
1118 >     * Callback from ForkJoinWorkerThread constructor to
1119 >     * determine its poolIndex and record in workers array.
1120 >     *
1121 >     * @param w the worker
1122 >     * @return the worker's pool index
1123       */
1124 <    private void createAndStartInitialWorkers(int ps) {
1125 <        final ReentrantLock lock = this.workerLock;
1126 <        lock.lock();
1127 <        try {
1128 <            ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(ps);
1129 <            for (int i = 0; i < ps; ++i) {
1130 <                ForkJoinWorkerThread w = createWorker(i);
1131 <                if (w != null) {
1132 <                    ws[i] = w;
1133 <                    w.start();
1134 <                    updateWorkerCount(1);
1124 >    final int registerWorker(ForkJoinWorkerThread w) {
1125 >        /*
1126 >         * In the typical case, a new worker acquires the lock, uses
1127 >         * next available index and returns quickly.  Since we should
1128 >         * not block callers (ultimately from signalWork or
1129 >         * tryPreBlock) waiting for the lock needed to do this, we
1130 >         * instead help release other workers while waiting for the
1131 >         * lock.
1132 >         */
1133 >        for (int g;;) {
1134 >            ForkJoinWorkerThread[] ws;
1135 >            if (((g = scanGuard) & SG_UNIT) == 0 &&
1136 >                UNSAFE.compareAndSwapInt(this, scanGuardOffset,
1137 >                                         g, g | SG_UNIT)) {
1138 >                int k = nextWorkerIndex;
1139 >                try {
1140 >                    if ((ws = workers) != null) { // ignore on shutdown
1141 >                        int n = ws.length;
1142 >                        if (k < 0 || k >= n || ws[k] != null) {
1143 >                            for (k = 0; k < n && ws[k] != null; ++k)
1144 >                                ;
1145 >                            if (k == n)
1146 >                                ws = workers = Arrays.copyOf(ws, n << 1);
1147 >                        }
1148 >                        ws[k] = w;
1149 >                        nextWorkerIndex = k + 1;
1150 >                        int m = g & SMASK;
1151 >                        g = k > m? ((m << 1) + 1) & SMASK : g + (SG_UNIT<<1);
1152 >                    }
1153 >                } finally {
1154 >                    scanGuard = g;
1155 >                }
1156 >                return k;
1157 >            }
1158 >            else if ((ws = workers) != null) { // help release others
1159 >                for (ForkJoinWorkerThread u : ws) {
1160 >                    if (u != null && u.queueBase != u.queueTop) {
1161 >                        if (tryReleaseWaiter())
1162 >                            break;
1163 >                    }
1164                  }
1165              }
464        } finally {
465            lock.unlock();
1166          }
1167      }
1168  
1169      /**
1170 <     * Worker creation and startup for threads added via setParallelism.
1170 >     * Final callback from terminating worker.  Removes record of
1171 >     * worker from array, and adjusts counts. If pool is shutting
1172 >     * down, tries to complete termination.
1173 >     *
1174 >     * @param w the worker
1175       */
1176 <    private void createAndStartAddedWorkers() {
1177 <        resumeAllSpares();  // Allow spares to convert to nonspare
1178 <        int ps = parallelism;
1179 <        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(ps);
1180 <        int len = ws.length;
1181 <        // Sweep through slots, to keep lowest indices most populated
1182 <        int k = 0;
1183 <        while (k < len) {
1184 <            if (ws[k] != null) {
1185 <                ++k;
1186 <                continue;
1176 >    final void deregisterWorker(ForkJoinWorkerThread w, Throwable ex) {
1177 >        int idx = w.poolIndex;
1178 >        int sc = w.stealCount;
1179 >        int steps = 0;
1180 >        // Remove from array, adjust worker counts and collect steal count.
1181 >        // We can intermix failed removes or adjusts with steal updates
1182 >        do {
1183 >            long s, c;
1184 >            int g;
1185 >            if (steps == 0 && ((g = scanGuard) & SG_UNIT) == 0 &&
1186 >                UNSAFE.compareAndSwapInt(this, scanGuardOffset,
1187 >                                         g, g |= SG_UNIT)) {
1188 >                ForkJoinWorkerThread[] ws = workers;
1189 >                if (ws != null && idx >= 0 &&
1190 >                    idx < ws.length && ws[idx] == w)
1191 >                    ws[idx] = null;    // verify
1192 >                nextWorkerIndex = idx;
1193 >                scanGuard = g + SG_UNIT;
1194 >                steps = 1;
1195 >            }
1196 >            if (steps == 1 &&
1197 >                UNSAFE.compareAndSwapLong(this, ctlOffset, c = ctl,
1198 >                                          (((c - AC_UNIT) & AC_MASK) |
1199 >                                           ((c - TC_UNIT) & TC_MASK) |
1200 >                                           (c & ~(AC_MASK|TC_MASK)))))
1201 >                steps = 2;
1202 >            if (sc != 0 &&
1203 >                UNSAFE.compareAndSwapLong(this, stealCountOffset,
1204 >                                          s = stealCount, s + sc))
1205 >                sc = 0;
1206 >        } while (steps != 2 || sc != 0);
1207 >        if (!tryTerminate(false)) {
1208 >            if (ex != null)   // possibly replace if died abnormally
1209 >                signalWork();
1210 >            else
1211 >                tryReleaseWaiter();
1212 >        }
1213 >    }
1214 >
1215 >    // Shutdown and termination
1216 >
1217 >    /**
1218 >     * Possibly initiates and/or completes termination.
1219 >     *
1220 >     * @param now if true, unconditionally terminate, else only
1221 >     * if shutdown and empty queue and no active workers
1222 >     * @return true if now terminating or terminated
1223 >     */
1224 >    private boolean tryTerminate(boolean now) {
1225 >        long c;
1226 >        while (((c = ctl) & STOP_BIT) == 0) {
1227 >            if (!now) {
1228 >                if ((int)(c >> AC_SHIFT) != -parallelism)
1229 >                    return false;
1230 >                if (!shutdown || blockedCount != 0 || quiescerCount != 0 ||
1231 >                    queueBase != queueTop) {
1232 >                    if (ctl == c) // staleness check
1233 >                        return false;
1234 >                    continue;
1235 >                }
1236              }
1237 <            int s = workerCounts;
1238 <            int tc = totalCountOf(s);
1239 <            int rc = runningCountOf(s);
1240 <            if (rc >= ps || tc >= ps)
1241 <                break;
1242 <            if (casWorkerCounts (s, workerCountsFor(tc+1, rc+1))) {
1243 <                ForkJoinWorkerThread w = createWorker(k);
1244 <                if (w != null) {
1245 <                    ws[k++] = w;
1246 <                    w.start();
1237 >            if (UNSAFE.compareAndSwapLong(this, ctlOffset, c, c | STOP_BIT))
1238 >                startTerminating();
1239 >        }
1240 >        if ((short)(c >>> TC_SHIFT) == -parallelism) { // signal when 0 workers
1241 >            final ReentrantLock lock = this.submissionLock;
1242 >            lock.lock();
1243 >            try {
1244 >                termination.signalAll();
1245 >            } finally {
1246 >                lock.unlock();
1247 >            }
1248 >        }
1249 >        return true;
1250 >    }
1251 >
1252 >    /**
1253 >     * Runs up to three passes through workers: (0) Setting
1254 >     * termination status for each worker, followed by wakeups up to
1255 >     * queued workers; (1) helping cancel tasks; (2) interrupting
1256 >     * lagging threads (likely in external tasks, but possibly also
1257 >     * blocked in joins).  Each pass repeats previous steps because of
1258 >     * potential lagging thread creation.
1259 >     */
1260 >    private void startTerminating() {
1261 >        cancelSubmissions();
1262 >        for (int pass = 0; pass < 3; ++pass) {
1263 >            ForkJoinWorkerThread[] ws = workers;
1264 >            if (ws != null) {
1265 >                for (ForkJoinWorkerThread w : ws) {
1266 >                    if (w != null) {
1267 >                        w.terminate = true;
1268 >                        if (pass > 0) {
1269 >                            w.cancelTasks();
1270 >                            if (pass > 1 && !w.isInterrupted()) {
1271 >                                try {
1272 >                                    w.interrupt();
1273 >                                } catch (SecurityException ignore) {
1274 >                                }
1275 >                            }
1276 >                        }
1277 >                    }
1278                  }
1279 <                else {
1280 <                    updateWorkerCount(-1); // back out on failed creation
1281 <                    break;
1279 >                terminateWaiters();
1280 >            }
1281 >        }
1282 >    }
1283 >
1284 >    /**
1285 >     * Polls and cancels all submissions. Called only during termination.
1286 >     */
1287 >    private void cancelSubmissions() {
1288 >        while (queueBase != queueTop) {
1289 >            ForkJoinTask<?> task = pollSubmission();
1290 >            if (task != null) {
1291 >                try {
1292 >                    task.cancel(false);
1293 >                } catch (Throwable ignore) {
1294 >                }
1295 >            }
1296 >        }
1297 >    }
1298 >
1299 >    /**
1300 >     * Tries to set the termination status of waiting workers, and
1301 >     * then wakes them up (after which they will terminate).
1302 >     */
1303 >    private void terminateWaiters() {
1304 >        ForkJoinWorkerThread[] ws = workers;
1305 >        if (ws != null) {
1306 >            ForkJoinWorkerThread w; long c; int i, e;
1307 >            int n = ws.length;
1308 >            while ((i = ~(e = (int)(c = ctl)) & SMASK) < n &&
1309 >                   (w = ws[i]) != null && w.eventCount == (e & E_MASK)) {
1310 >                if (UNSAFE.compareAndSwapLong(this, ctlOffset, c,
1311 >                                              (long)(w.nextWait & E_MASK) |
1312 >                                              ((c + AC_UNIT) & AC_MASK) |
1313 >                                              (c & (TC_MASK|STOP_BIT)))) {
1314 >                    w.terminate = true;
1315 >                    w.eventCount = e + EC_UNIT;
1316 >                    if (w.parked)
1317 >                        UNSAFE.unpark(w);
1318                  }
1319              }
1320          }
1321      }
1322  
1323 +    // misc ForkJoinWorkerThread support
1324 +
1325      /**
1326 <     * Sets the handler for internal worker threads that terminate due
1327 <     * to unrecoverable errors encountered while executing tasks.
1328 <     * Unless set, the current default or ThreadGroup handler is used
1329 <     * as handler.
1326 >     * Increment or decrement quiescerCount. Needed only to prevent
1327 >     * triggering shutdown if a worker is transiently inactive while
1328 >     * checking quiescence.
1329 >     *
1330 >     * @param delta 1 for increment, -1 for decrement
1331 >     */
1332 >    final void addQuiescerCount(int delta) {
1333 >        int c;
1334 >        do {} while(!UNSAFE.compareAndSwapInt(this, quiescerCountOffset,
1335 >                                              c = quiescerCount, c + delta));
1336 >    }
1337 >
1338 >    /**
1339 >     * Directly increment or decrement active count without
1340 >     * queuing. This method is used to transiently assert inactivation
1341 >     * while checking quiescence.
1342 >     *
1343 >     * @param delta 1 for increment, -1 for decrement
1344 >     */
1345 >    final void addActiveCount(int delta) {
1346 >        long d = delta < 0 ? -AC_UNIT : AC_UNIT;
1347 >        long c;
1348 >        do {} while (!UNSAFE.compareAndSwapLong(this, ctlOffset, c = ctl,
1349 >                                                ((c + d) & AC_MASK) |
1350 >                                                (c & ~AC_MASK)));
1351 >    }
1352 >
1353 >    /**
1354 >     * Returns the approximate (non-atomic) number of idle threads per
1355 >     * active thread.
1356 >     */
1357 >    final int idlePerActive() {
1358 >        // Approximate at powers of two for small values, saturate past 4
1359 >        int p = parallelism;
1360 >        int a = p + (int)(ctl >> AC_SHIFT);
1361 >        return (a > (p >>>= 1) ? 0 :
1362 >                a > (p >>>= 1) ? 1 :
1363 >                a > (p >>>= 1) ? 2 :
1364 >                a > (p >>>= 1) ? 4 :
1365 >                8);
1366 >    }
1367 >
1368 >    // Exported methods
1369 >
1370 >    // Constructors
1371 >
1372 >    /**
1373 >     * Creates a {@code ForkJoinPool} with parallelism equal to {@link
1374 >     * java.lang.Runtime#availableProcessors}, using the {@linkplain
1375 >     * #defaultForkJoinWorkerThreadFactory default thread factory},
1376 >     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
1377       *
509     * @param h the new handler
510     * @return the old handler, or null if none
1378       * @throws SecurityException if a security manager exists and
1379       *         the caller is not permitted to modify threads
1380       *         because it does not hold {@link
1381 <     *         java.lang.RuntimePermission}<tt>("modifyThread")</tt>,
1381 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1382       */
1383 <    public Thread.UncaughtExceptionHandler
1384 <        setUncaughtExceptionHandler(Thread.UncaughtExceptionHandler h) {
1385 <        checkPermission();
519 <        Thread.UncaughtExceptionHandler old = null;
520 <        final ReentrantLock lock = this.workerLock;
521 <        lock.lock();
522 <        try {
523 <            old = ueh;
524 <            ueh = h;
525 <            ForkJoinWorkerThread[] ws = workers;
526 <            for (int i = 0; i < ws.length; ++i) {
527 <                ForkJoinWorkerThread w = ws[i];
528 <                if (w != null)
529 <                    w.setUncaughtExceptionHandler(h);
530 <            }
531 <        } finally {
532 <            lock.unlock();
533 <        }
534 <        return old;
1383 >    public ForkJoinPool() {
1384 >        this(Runtime.getRuntime().availableProcessors(),
1385 >             defaultForkJoinWorkerThreadFactory, null, false);
1386      }
1387  
1388      /**
1389 <     * Returns the handler for internal worker threads that terminate
1390 <     * due to unrecoverable errors encountered while executing tasks.
1391 <     * @return the handler, or null if none
1389 >     * Creates a {@code ForkJoinPool} with the indicated parallelism
1390 >     * level, the {@linkplain
1391 >     * #defaultForkJoinWorkerThreadFactory default thread factory},
1392 >     * no UncaughtExceptionHandler, and non-async LIFO processing mode.
1393 >     *
1394 >     * @param parallelism the parallelism level
1395 >     * @throws IllegalArgumentException if parallelism less than or
1396 >     *         equal to zero, or greater than implementation limit
1397 >     * @throws SecurityException if a security manager exists and
1398 >     *         the caller is not permitted to modify threads
1399 >     *         because it does not hold {@link
1400 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1401       */
1402 <    public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
1403 <        Thread.UncaughtExceptionHandler h;
544 <        final ReentrantLock lock = this.workerLock;
545 <        lock.lock();
546 <        try {
547 <            h = ueh;
548 <        } finally {
549 <            lock.unlock();
550 <        }
551 <        return h;
1402 >    public ForkJoinPool(int parallelism) {
1403 >        this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
1404      }
1405  
554    // Execution methods
555
1406      /**
1407 <     * Common code for execute, invoke and submit
1407 >     * Creates a {@code ForkJoinPool} with the given parameters.
1408 >     *
1409 >     * @param parallelism the parallelism level. For default value,
1410 >     * use {@link java.lang.Runtime#availableProcessors}.
1411 >     * @param factory the factory for creating new threads. For default value,
1412 >     * use {@link #defaultForkJoinWorkerThreadFactory}.
1413 >     * @param handler the handler for internal worker threads that
1414 >     * terminate due to unrecoverable errors encountered while executing
1415 >     * tasks. For default value, use {@code null}.
1416 >     * @param asyncMode if true,
1417 >     * establishes local first-in-first-out scheduling mode for forked
1418 >     * tasks that are never joined. This mode may be more appropriate
1419 >     * than default locally stack-based mode in applications in which
1420 >     * worker threads only process event-style asynchronous tasks.
1421 >     * For default value, use {@code false}.
1422 >     * @throws IllegalArgumentException if parallelism less than or
1423 >     *         equal to zero, or greater than implementation limit
1424 >     * @throws NullPointerException if the factory is null
1425 >     * @throws SecurityException if a security manager exists and
1426 >     *         the caller is not permitted to modify threads
1427 >     *         because it does not hold {@link
1428 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1429       */
1430 <    private <T> void doSubmit(ForkJoinTask<T> task) {
1431 <        if (isShutdown())
1432 <            throw new RejectedExecutionException();
1433 <        submissionQueue.offer(task);
1434 <        signalIdleWorkers(true);
1430 >    public ForkJoinPool(int parallelism,
1431 >                        ForkJoinWorkerThreadFactory factory,
1432 >                        Thread.UncaughtExceptionHandler handler,
1433 >                        boolean asyncMode) {
1434 >        checkPermission();
1435 >        if (factory == null)
1436 >            throw new NullPointerException();
1437 >        if (parallelism <= 0 || parallelism > MAX_ID)
1438 >            throw new IllegalArgumentException();
1439 >        this.parallelism = parallelism;
1440 >        this.factory = factory;
1441 >        this.ueh = handler;
1442 >        this.locallyFifo = asyncMode;
1443 >        long np = (long)(-parallelism); // offset ctl counts
1444 >        this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK);
1445 >        this.submissionQueue = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY];
1446 >        // initialize workers array with room for 2*parallelism if possible
1447 >        int n = parallelism << 1;
1448 >        if (n >= MAX_ID)
1449 >            n = MAX_ID;
1450 >        else { // See Hackers Delight, sec 3.2, where n < (1 << 16)
1451 >            n |= n >>> 1; n |= n >>> 2; n |= n >>> 4; n |= n >>> 8;
1452 >        }
1453 >        workers = new ForkJoinWorkerThread[n + 1];
1454 >        this.submissionLock = new ReentrantLock();
1455 >        this.termination = submissionLock.newCondition();
1456 >        StringBuilder sb = new StringBuilder("ForkJoinPool-");
1457 >        sb.append(poolNumberGenerator.incrementAndGet());
1458 >        sb.append("-worker-");
1459 >        this.workerNamePrefix = sb.toString();
1460      }
1461  
1462 +    // Execution methods
1463 +
1464      /**
1465 <     * Performs the given task; returning its result upon completion
1465 >     * Performs the given task, returning its result upon completion.
1466 >     * If the computation encounters an unchecked Exception or Error,
1467 >     * it is rethrown as the outcome of this invocation.  Rethrown
1468 >     * exceptions behave in the same way as regular exceptions, but,
1469 >     * when possible, contain stack traces (as displayed for example
1470 >     * using {@code ex.printStackTrace()}) of both the current thread
1471 >     * as well as the thread actually encountering the exception;
1472 >     * minimally only the latter.
1473 >     *
1474       * @param task the task
1475       * @return the task's result
1476 <     * @throws NullPointerException if task is null
1477 <     * @throws RejectedExecutionException if pool is shut down
1476 >     * @throws NullPointerException if the task is null
1477 >     * @throws RejectedExecutionException if the task cannot be
1478 >     *         scheduled for execution
1479       */
1480      public <T> T invoke(ForkJoinTask<T> task) {
1481 <        doSubmit(task);
1482 <        return task.join();
1481 >        Thread t = Thread.currentThread();
1482 >        if (task == null)
1483 >            throw new NullPointerException();
1484 >        if (shutdown)
1485 >            throw new RejectedExecutionException();
1486 >        if ((t instanceof ForkJoinWorkerThread) &&
1487 >            ((ForkJoinWorkerThread)t).pool == this)
1488 >            return task.invoke();  // bypass submit if in same pool
1489 >        else {
1490 >            addSubmission(task);
1491 >            return task.join();
1492 >        }
1493 >    }
1494 >
1495 >    /**
1496 >     * Unless terminating, forks task if within an ongoing FJ
1497 >     * computation in the current pool, else submits as external task.
1498 >     */
1499 >    private <T> void forkOrSubmit(ForkJoinTask<T> task) {
1500 >        ForkJoinWorkerThread w;
1501 >        Thread t = Thread.currentThread();
1502 >        if (shutdown)
1503 >            throw new RejectedExecutionException();
1504 >        if ((t instanceof ForkJoinWorkerThread) &&
1505 >            (w = (ForkJoinWorkerThread)t).pool == this)
1506 >            w.pushTask(task);
1507 >        else
1508 >            addSubmission(task);
1509      }
1510  
1511      /**
1512       * Arranges for (asynchronous) execution of the given task.
1513 +     *
1514       * @param task the task
1515 <     * @throws NullPointerException if task is null
1516 <     * @throws RejectedExecutionException if pool is shut down
1515 >     * @throws NullPointerException if the task is null
1516 >     * @throws RejectedExecutionException if the task cannot be
1517 >     *         scheduled for execution
1518       */
1519 <    public <T> void execute(ForkJoinTask<T> task) {
1520 <        doSubmit(task);
1519 >    public void execute(ForkJoinTask<?> task) {
1520 >        if (task == null)
1521 >            throw new NullPointerException();
1522 >        forkOrSubmit(task);
1523      }
1524  
1525      // AbstractExecutorService methods
1526  
1527 +    /**
1528 +     * @throws NullPointerException if the task is null
1529 +     * @throws RejectedExecutionException if the task cannot be
1530 +     *         scheduled for execution
1531 +     */
1532      public void execute(Runnable task) {
1533 <        doSubmit(new AdaptedRunnable<Void>(task, null));
1533 >        if (task == null)
1534 >            throw new NullPointerException();
1535 >        ForkJoinTask<?> job;
1536 >        if (task instanceof ForkJoinTask<?>) // avoid re-wrap
1537 >            job = (ForkJoinTask<?>) task;
1538 >        else
1539 >            job = ForkJoinTask.adapt(task, null);
1540 >        forkOrSubmit(job);
1541      }
1542  
1543 <    public <T> ForkJoinTask<T> submit(Callable<T> task) {
1544 <        ForkJoinTask<T> job = new AdaptedCallable<T>(task);
1545 <        doSubmit(job);
1546 <        return job;
1543 >    /**
1544 >     * Submits a ForkJoinTask for execution.
1545 >     *
1546 >     * @param task the task to submit
1547 >     * @return the task
1548 >     * @throws NullPointerException if the task is null
1549 >     * @throws RejectedExecutionException if the task cannot be
1550 >     *         scheduled for execution
1551 >     */
1552 >    public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
1553 >        if (task == null)
1554 >            throw new NullPointerException();
1555 >        forkOrSubmit(task);
1556 >        return task;
1557      }
1558  
1559 <    public <T> ForkJoinTask<T> submit(Runnable task, T result) {
1560 <        ForkJoinTask<T> job = new AdaptedRunnable<T>(task, result);
1561 <        doSubmit(job);
1559 >    /**
1560 >     * @throws NullPointerException if the task is null
1561 >     * @throws RejectedExecutionException if the task cannot be
1562 >     *         scheduled for execution
1563 >     */
1564 >    public <T> ForkJoinTask<T> submit(Callable<T> task) {
1565 >        if (task == null)
1566 >            throw new NullPointerException();
1567 >        ForkJoinTask<T> job = ForkJoinTask.adapt(task);
1568 >        forkOrSubmit(job);
1569          return job;
1570      }
1571  
1572 <    public ForkJoinTask<?> submit(Runnable task) {
1573 <        ForkJoinTask<Void> job = new AdaptedRunnable<Void>(task, null);
1574 <        doSubmit(job);
1572 >    /**
1573 >     * @throws NullPointerException if the task is null
1574 >     * @throws RejectedExecutionException if the task cannot be
1575 >     *         scheduled for execution
1576 >     */
1577 >    public <T> ForkJoinTask<T> submit(Runnable task, T result) {
1578 >        if (task == null)
1579 >            throw new NullPointerException();
1580 >        ForkJoinTask<T> job = ForkJoinTask.adapt(task, result);
1581 >        forkOrSubmit(job);
1582          return job;
1583      }
1584  
612    protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
613        return new AdaptedRunnable(runnable, value);
614    }
615
616    protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
617        return new AdaptedCallable(callable);
618    }
619
1585      /**
1586 <     * Adaptor for Runnables. This implements RunnableFuture
1587 <     * to be compliant with AbstractExecutorService constraints
1586 >     * @throws NullPointerException if the task is null
1587 >     * @throws RejectedExecutionException if the task cannot be
1588 >     *         scheduled for execution
1589       */
1590 <    static final class AdaptedRunnable<T> extends ForkJoinTask<T>
1591 <        implements RunnableFuture<T> {
1592 <        final Runnable runnable;
1593 <        final T resultOnCompletion;
1594 <        T result;
1595 <        AdaptedRunnable(Runnable runnable, T result) {
1596 <            if (runnable == null) throw new NullPointerException();
1597 <            this.runnable = runnable;
1598 <            this.resultOnCompletion = result;
1599 <        }
634 <        public T getRawResult() { return result; }
635 <        public void setRawResult(T v) { result = v; }
636 <        public boolean exec() {
637 <            runnable.run();
638 <            result = resultOnCompletion;
639 <            return true;
640 <        }
641 <        public void run() { invoke(); }
1590 >    public ForkJoinTask<?> submit(Runnable task) {
1591 >        if (task == null)
1592 >            throw new NullPointerException();
1593 >        ForkJoinTask<?> job;
1594 >        if (task instanceof ForkJoinTask<?>) // avoid re-wrap
1595 >            job = (ForkJoinTask<?>) task;
1596 >        else
1597 >            job = ForkJoinTask.adapt(task, null);
1598 >        forkOrSubmit(job);
1599 >        return job;
1600      }
1601  
1602      /**
1603 <     * Adaptor for Callables
1603 >     * @throws NullPointerException       {@inheritDoc}
1604 >     * @throws RejectedExecutionException {@inheritDoc}
1605       */
647    static final class AdaptedCallable<T> extends ForkJoinTask<T>
648        implements RunnableFuture<T> {
649        final Callable<T> callable;
650        T result;
651        AdaptedCallable(Callable<T> callable) {
652            if (callable == null) throw new NullPointerException();
653            this.callable = callable;
654        }
655        public T getRawResult() { return result; }
656        public void setRawResult(T v) { result = v; }
657        public boolean exec() {
658            try {
659                result = callable.call();
660                return true;
661            } catch (Error err) {
662                throw err;
663            } catch (RuntimeException rex) {
664                throw rex;
665            } catch (Exception ex) {
666                throw new RuntimeException(ex);
667            }
668        }
669        public void run() { invoke(); }
670    }
671
1606      public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
1607 <        ArrayList<ForkJoinTask<T>> ts =
1607 >        ArrayList<ForkJoinTask<T>> forkJoinTasks =
1608              new ArrayList<ForkJoinTask<T>>(tasks.size());
1609 <        for (Callable<T> c : tasks)
1610 <            ts.add(new AdaptedCallable<T>(c));
1611 <        invoke(new InvokeAll<T>(ts));
1612 <        return (List<Future<T>>)(List)ts;
1609 >        for (Callable<T> task : tasks)
1610 >            forkJoinTasks.add(ForkJoinTask.adapt(task));
1611 >        invoke(new InvokeAll<T>(forkJoinTasks));
1612 >
1613 >        @SuppressWarnings({"unchecked", "rawtypes"})
1614 >            List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks;
1615 >        return futures;
1616      }
1617  
1618      static final class InvokeAll<T> extends RecursiveAction {
1619          final ArrayList<ForkJoinTask<T>> tasks;
1620          InvokeAll(ArrayList<ForkJoinTask<T>> tasks) { this.tasks = tasks; }
1621          public void compute() {
1622 <            try { invokeAll(tasks); } catch(Exception ignore) {}
1622 >            try { invokeAll(tasks); }
1623 >            catch (Exception ignore) {}
1624          }
1625 +        private static final long serialVersionUID = -7914297376763021607L;
1626      }
1627  
689    // Configuration and status settings and queries
690
1628      /**
1629 <     * Returns the factory used for constructing new workers
1629 >     * Returns the factory used for constructing new workers.
1630       *
1631       * @return the factory used for constructing new workers
1632       */
# Line 698 | Line 1635 | public class ForkJoinPool extends Abstra
1635      }
1636  
1637      /**
1638 <     * Sets the target paralleism level of this pool.
1639 <     * @param parallelism the target parallelism
1640 <     * @throws IllegalArgumentException if parallelism less than or
1641 <     * equal to zero or greater than maximum size bounds.
705 <     * @throws SecurityException if a security manager exists and
706 <     *         the caller is not permitted to modify threads
707 <     *         because it does not hold {@link
708 <     *         java.lang.RuntimePermission}<tt>("modifyThread")</tt>,
1638 >     * Returns the handler for internal worker threads that terminate
1639 >     * due to unrecoverable errors encountered while executing tasks.
1640 >     *
1641 >     * @return the handler, or {@code null} if none
1642       */
1643 <    public void setParallelism(int parallelism) {
1644 <        checkPermission();
712 <        if (parallelism <= 0 || parallelism > maxPoolSize)
713 <            throw new IllegalArgumentException();
714 <        final ReentrantLock lock = this.workerLock;
715 <        lock.lock();
716 <        try {
717 <            if (!isTerminating()) {
718 <                int p = this.parallelism;
719 <                this.parallelism = parallelism;
720 <                if (parallelism > p)
721 <                    createAndStartAddedWorkers();
722 <                else
723 <                    trimSpares();
724 <            }
725 <        } finally {
726 <            lock.unlock();
727 <        }
728 <        signalIdleWorkers(false);
1643 >    public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
1644 >        return ueh;
1645      }
1646  
1647      /**
1648 <     * Returns the targeted number of worker threads in this pool.
733 <     * This value does not necessarily reflect transient changes as
734 <     * threads are added, removed, or abruptly terminate.
1648 >     * Returns the targeted parallelism level of this pool.
1649       *
1650 <     * @return the targeted number of worker threads in this pool
1650 >     * @return the targeted parallelism level of this pool
1651       */
1652      public int getParallelism() {
1653          return parallelism;
# Line 741 | Line 1655 | public class ForkJoinPool extends Abstra
1655  
1656      /**
1657       * Returns the number of worker threads that have started but not
1658 <     * yet terminated.  This result returned by this method may differ
1659 <     * from <tt>getParallelism</tt> when threads are created to
1658 >     * yet terminated.  The result returned by this method may differ
1659 >     * from {@link #getParallelism} when threads are created to
1660       * maintain parallelism when others are cooperatively blocked.
1661       *
1662       * @return the number of worker threads
1663       */
1664      public int getPoolSize() {
1665 <        return totalCountOf(workerCounts);
752 <    }
753 <
754 <    /**
755 <     * Returns the maximum number of threads allowed to exist in the
756 <     * pool, even if there are insufficient unblocked running threads.
757 <     * @return the maximum
758 <     */
759 <    public int getMaximumPoolSize() {
760 <        return maxPoolSize;
761 <    }
762 <
763 <    /**
764 <     * Sets the maximum number of threads allowed to exist in the
765 <     * pool, even if there are insufficient unblocked running threads.
766 <     * Setting this value has no effect on current pool size. It
767 <     * controls construction of new threads.
768 <     * @throws IllegalArgumentException if negative or greater then
769 <     * internal implementation limit.
770 <     */
771 <    public void setMaximumPoolSize(int newMax) {
772 <        if (newMax < 0 || newMax > MAX_THREADS)
773 <            throw new IllegalArgumentException();
774 <        maxPoolSize = newMax;
775 <    }
776 <
777 <
778 <    /**
779 <     * Returns true if this pool dynamically maintains its target
780 <     * parallelism level. If false, new threads are added only to
781 <     * avoid possible starvation.
782 <     * This setting is by default true;
783 <     * @return true if maintains parallelism
784 <     */
785 <    public boolean getMaintainsParallelism() {
786 <        return maintainsParallelism;
1665 >        return parallelism + (short)(ctl >>> TC_SHIFT);
1666      }
1667  
1668      /**
1669 <     * Sets whether this pool dynamically maintains its target
1670 <     * parallelism level. If false, new threads are added only to
1671 <     * avoid possible starvation.
1672 <     * @param enable true to maintains parallelism
1669 >     * Returns {@code true} if this pool uses local first-in-first-out
1670 >     * scheduling mode for forked tasks that are never joined.
1671 >     *
1672 >     * @return {@code true} if this pool uses async mode
1673       */
1674 <    public void setMaintainsParallelism(boolean enable) {
1675 <        maintainsParallelism = enable;
1674 >    public boolean getAsyncMode() {
1675 >        return locallyFifo;
1676      }
1677  
1678      /**
1679 <     * Returns the approximate number of worker threads that are not
1680 <     * blocked waiting to join tasks or for other managed
1681 <     * synchronization.
1679 >     * Returns an estimate of the number of worker threads that are
1680 >     * not blocked waiting to join tasks or for other managed
1681 >     * synchronization. This method may overestimate the
1682 >     * number of running threads.
1683       *
1684       * @return the number of worker threads
1685       */
1686      public int getRunningThreadCount() {
1687 <        return runningCountOf(workerCounts);
1687 >        int r = parallelism + (int)(ctl >> AC_SHIFT);
1688 >        return r <= 0? 0 : r; // suppress momentarily negative values
1689      }
1690  
1691      /**
1692 <     * Returns the approximate number of threads that are currently
1692 >     * Returns an estimate of the number of threads that are currently
1693       * stealing or executing tasks. This method may overestimate the
1694       * number of active threads.
1695 <     * @return the number of active threads.
1695 >     *
1696 >     * @return the number of active threads
1697       */
1698      public int getActiveThreadCount() {
1699 <        return activeCountOf(runControl);
1700 <    }
819 <
820 <    /**
821 <     * Returns the approximate number of threads that are currently
822 <     * idle waiting for tasks. This method may underestimate the
823 <     * number of idle threads.
824 <     * @return the number of idle threads.
825 <     */
826 <    final int getIdleThreadCount() {
827 <        int c = runningCountOf(workerCounts) - activeCountOf(runControl);
828 <        return (c <= 0)? 0 : c;
1699 >        int r = parallelism + (int)(ctl >> AC_SHIFT) + blockedCount;
1700 >        return r <= 0? 0 : r; // suppress momentarily negative values
1701      }
1702  
1703      /**
1704 <     * Returns true if all worker threads are currently idle. An idle
1705 <     * worker is one that cannot obtain a task to execute because none
1706 <     * are available to steal from other threads, and there are no
1707 <     * pending submissions to the pool. This method is conservative:
1708 <     * It might not return true immediately upon idleness of all
1709 <     * threads, but will eventually become true if threads remain
1710 <     * inactive.
1711 <     * @return true if all threads are currently idle
1704 >     * Returns {@code true} if all worker threads are currently idle.
1705 >     * An idle worker is one that cannot obtain a task to execute
1706 >     * because none are available to steal from other threads, and
1707 >     * there are no pending submissions to the pool. This method is
1708 >     * conservative; it might not return {@code true} immediately upon
1709 >     * idleness of all threads, but will eventually become true if
1710 >     * threads remain inactive.
1711 >     *
1712 >     * @return {@code true} if all threads are currently idle
1713       */
1714      public boolean isQuiescent() {
1715 <        return activeCountOf(runControl) == 0;
1715 >        return parallelism + (int)(ctl >> AC_SHIFT) + blockedCount == 0;
1716      }
1717  
1718      /**
# Line 847 | Line 1720 | public class ForkJoinPool extends Abstra
1720       * one thread's work queue by another. The reported value
1721       * underestimates the actual total number of steals when the pool
1722       * is not quiescent. This value may be useful for monitoring and
1723 <     * tuning fork/join programs: In general, steal counts should be
1723 >     * tuning fork/join programs: in general, steal counts should be
1724       * high enough to keep threads busy, but low enough to avoid
1725       * overhead and contention across threads.
1726 <     * @return the number of steals.
1726 >     *
1727 >     * @return the number of steals
1728       */
1729      public long getStealCount() {
1730 <        return stealCount.get();
857 <    }
858 <
859 <    /**
860 <     * Accumulate steal count from a worker. Call only
861 <     * when worker known to be idle.
862 <     */
863 <    private void updateStealCount(ForkJoinWorkerThread w) {
864 <        int sc = w.getAndClearStealCount();
865 <        if (sc != 0)
866 <            stealCount.addAndGet(sc);
1730 >        return stealCount;
1731      }
1732  
1733      /**
1734 <     * Returns the total number of tasks currently held in queues by
1735 <     * worker threads (but not including tasks submitted to the pool
1736 <     * that have not begun executing). This value is only an
1737 <     * approximation, obtained by iterating across all threads in the
1738 <     * pool. This method may be useful for tuning task granularities.
1739 <     * @return the number of queued tasks.
1734 >     * Returns an estimate of the total number of tasks currently held
1735 >     * in queues by worker threads (but not including tasks submitted
1736 >     * to the pool that have not begun executing). This value is only
1737 >     * an approximation, obtained by iterating across all threads in
1738 >     * the pool. This method may be useful for tuning task
1739 >     * granularities.
1740 >     *
1741 >     * @return the number of queued tasks
1742       */
1743      public long getQueuedTaskCount() {
1744          long count = 0;
1745 <        ForkJoinWorkerThread[] ws = workers;
1746 <        for (int i = 0; i < ws.length; ++i) {
1747 <            ForkJoinWorkerThread t = ws[i];
1748 <            if (t != null)
1749 <                count += t.getQueueSize();
1745 >        ForkJoinWorkerThread[] ws;
1746 >        if ((short)(ctl >>> TC_SHIFT) > -parallelism &&
1747 >            (ws = workers) != null) {
1748 >            for (ForkJoinWorkerThread w : ws)
1749 >                if (w != null)
1750 >                    count -= w.queueBase - w.queueTop; // must read base first
1751          }
1752          return count;
1753      }
1754  
1755      /**
1756 <     * Returns the approximate number tasks submitted to this pool
1757 <     * that have not yet begun executing. This method takes time
1758 <     * proportional to the number of submissions.
1759 <     * @return the number of queued submissions.
1756 >     * Returns an estimate of the number of tasks submitted to this
1757 >     * pool that have not yet begun executing.  This method may take
1758 >     * time proportional to the number of submissions.
1759 >     *
1760 >     * @return the number of queued submissions
1761       */
1762      public int getQueuedSubmissionCount() {
1763 <        return submissionQueue.size();
1763 >        return -queueBase + queueTop;
1764      }
1765  
1766      /**
1767 <     * Returns true if there are any tasks submitted to this pool
1768 <     * that have not yet begun executing.
1769 <     * @return <tt>true</tt> if there are any queued submissions.
1767 >     * Returns {@code true} if there are any tasks submitted to this
1768 >     * pool that have not yet begun executing.
1769 >     *
1770 >     * @return {@code true} if there are any queued submissions
1771       */
1772      public boolean hasQueuedSubmissions() {
1773 <        return !submissionQueue.isEmpty();
1773 >        return queueBase != queueTop;
1774      }
1775  
1776      /**
1777       * Removes and returns the next unexecuted submission if one is
1778       * available.  This method may be useful in extensions to this
1779       * class that re-assign work in systems with multiple pools.
1780 <     * @return the next submission, or null if none
1780 >     *
1781 >     * @return the next submission, or {@code null} if none
1782       */
1783      protected ForkJoinTask<?> pollSubmission() {
1784 <        return submissionQueue.poll();
1784 >        ForkJoinTask<?> t; ForkJoinTask<?>[] q; int b, i;
1785 >        while ((b = queueBase) != queueTop &&
1786 >               (q = submissionQueue) != null &&
1787 >               (i = (q.length - 1) & b) >= 0) {
1788 >            long u = (i << ASHIFT) + ABASE;
1789 >            if ((t = q[i]) != null &&
1790 >                queueBase == b &&
1791 >                UNSAFE.compareAndSwapObject(q, u, t, null)) {
1792 >                queueBase = b + 1;
1793 >                return t;
1794 >            }
1795 >        }
1796 >        return null;
1797 >    }
1798 >
1799 >    /**
1800 >     * Removes all available unexecuted submitted and forked tasks
1801 >     * from scheduling queues and adds them to the given collection,
1802 >     * without altering their execution status. These may include
1803 >     * artificially generated or wrapped tasks. This method is
1804 >     * designed to be invoked only when the pool is known to be
1805 >     * quiescent. Invocations at other times may not remove all
1806 >     * tasks. A failure encountered while attempting to add elements
1807 >     * to collection {@code c} may result in elements being in
1808 >     * neither, either or both collections when the associated
1809 >     * exception is thrown.  The behavior of this operation is
1810 >     * undefined if the specified collection is modified while the
1811 >     * operation is in progress.
1812 >     *
1813 >     * @param c the collection to transfer elements into
1814 >     * @return the number of elements transferred
1815 >     */
1816 >    protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
1817 >        int count = 0;
1818 >        while (queueBase != queueTop) {
1819 >            ForkJoinTask<?> t = pollSubmission();
1820 >            if (t != null) {
1821 >                c.add(t);
1822 >                ++count;
1823 >            }
1824 >        }
1825 >        ForkJoinWorkerThread[] ws;
1826 >        if ((short)(ctl >>> TC_SHIFT) > -parallelism &&
1827 >            (ws = workers) != null) {
1828 >            for (ForkJoinWorkerThread w : ws)
1829 >                if (w != null)
1830 >                    count += w.drainTasksTo(c);
1831 >        }
1832 >        return count;
1833      }
1834  
1835      /**
# Line 922 | Line 1840 | public class ForkJoinPool extends Abstra
1840       * @return a string identifying this pool, as well as its state
1841       */
1842      public String toString() {
925        int ps = parallelism;
926        int wc = workerCounts;
927        int rc = runControl;
1843          long st = getStealCount();
1844          long qt = getQueuedTaskCount();
1845          long qs = getQueuedSubmissionCount();
1846 +        int pc = parallelism;
1847 +        long c = ctl;
1848 +        int tc = pc + (short)(c >>> TC_SHIFT);
1849 +        int rc = pc + (int)(c >> AC_SHIFT);
1850 +        if (rc < 0) // ignore transient negative
1851 +            rc = 0;
1852 +        int ac = rc + blockedCount;
1853 +        String level;
1854 +        if ((c & STOP_BIT) != 0)
1855 +            level = (tc == 0)? "Terminated" : "Terminating";
1856 +        else
1857 +            level = shutdown? "Shutting down" : "Running";
1858          return super.toString() +
1859 <            "[" + runStateToString(runStateOf(rc)) +
1860 <            ", parallelism = " + ps +
1861 <            ", size = " + totalCountOf(wc) +
1862 <            ", active = " + activeCountOf(rc) +
1863 <            ", running = " + runningCountOf(wc) +
1859 >            "[" + level +
1860 >            ", parallelism = " + pc +
1861 >            ", size = " + tc +
1862 >            ", active = " + ac +
1863 >            ", running = " + rc +
1864              ", steals = " + st +
1865              ", tasks = " + qt +
1866              ", submissions = " + qs +
1867              "]";
1868      }
1869  
943    private static String runStateToString(int rs) {
944        switch(rs) {
945        case RUNNING: return "Running";
946        case SHUTDOWN: return "Shutting down";
947        case TERMINATING: return "Terminating";
948        case TERMINATED: return "Terminated";
949        default: throw new Error("Unknown run state");
950        }
951    }
952
953    // lifecycle control
954
1870      /**
1871       * Initiates an orderly shutdown in which previously submitted
1872       * tasks are executed, but no new tasks will be accepted.
1873       * Invocation has no additional effect if already shut down.
1874       * Tasks that are in the process of being submitted concurrently
1875       * during the course of this method may or may not be rejected.
1876 +     *
1877       * @throws SecurityException if a security manager exists and
1878       *         the caller is not permitted to modify threads
1879       *         because it does not hold {@link
1880 <     *         java.lang.RuntimePermission}<tt>("modifyThread")</tt>,
1880 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1881       */
1882      public void shutdown() {
1883          checkPermission();
1884 <        transitionRunStateTo(SHUTDOWN);
1885 <        if (canTerminateOnShutdown(runControl))
970 <            terminateOnShutdown();
1884 >        shutdown = true;
1885 >        tryTerminate(false);
1886      }
1887  
1888      /**
1889 <     * Attempts to stop all actively executing tasks, and cancels all
1890 <     * waiting tasks.  Tasks that are in the process of being
1891 <     * submitted or executed concurrently during the course of this
1892 <     * method may or may not be rejected. Unlike some other executors,
1893 <     * this method cancels rather than collects non-executed tasks,
1894 <     * so always returns an empty list.
1889 >     * Attempts to cancel and/or stop all tasks, and reject all
1890 >     * subsequently submitted tasks.  Tasks that are in the process of
1891 >     * being submitted or executed concurrently during the course of
1892 >     * this method may or may not be rejected. This method cancels
1893 >     * both existing and unexecuted tasks, in order to permit
1894 >     * termination in the presence of task dependencies. So the method
1895 >     * always returns an empty list (unlike the case for some other
1896 >     * Executors).
1897 >     *
1898       * @return an empty list
1899       * @throws SecurityException if a security manager exists and
1900       *         the caller is not permitted to modify threads
1901       *         because it does not hold {@link
1902 <     *         java.lang.RuntimePermission}<tt>("modifyThread")</tt>,
1902 >     *         java.lang.RuntimePermission}{@code ("modifyThread")}
1903       */
1904      public List<Runnable> shutdownNow() {
1905          checkPermission();
1906 <        terminate();
1906 >        shutdown = true;
1907 >        tryTerminate(true);
1908          return Collections.emptyList();
1909      }
1910  
1911      /**
1912 <     * Returns <tt>true</tt> if all tasks have completed following shut down.
1912 >     * Returns {@code true} if all tasks have completed following shut down.
1913       *
1914 <     * @return <tt>true</tt> if all tasks have completed following shut down
1914 >     * @return {@code true} if all tasks have completed following shut down
1915       */
1916      public boolean isTerminated() {
1917 <        return runStateOf(runControl) == TERMINATED;
1917 >        long c = ctl;
1918 >        return ((c & STOP_BIT) != 0L &&
1919 >                (short)(c >>> TC_SHIFT) == -parallelism);
1920      }
1921  
1922      /**
1923 <     * Returns <tt>true</tt> if the process of termination has
1924 <     * commenced but possibly not yet completed.
1923 >     * Returns {@code true} if the process of termination has
1924 >     * commenced but not yet completed.  This method may be useful for
1925 >     * debugging. A return of {@code true} reported a sufficient
1926 >     * period after shutdown may indicate that submitted tasks have
1927 >     * ignored or suppressed interruption, or are waiting for IO,
1928 >     * causing this executor not to properly terminate. (See the
1929 >     * advisory notes for class {@link ForkJoinTask} stating that
1930 >     * tasks should not normally entail blocking operations.  But if
1931 >     * they do, they must abort them on interrupt.)
1932       *
1933 <     * @return <tt>true</tt> if terminating
1933 >     * @return {@code true} if terminating but not yet terminated
1934       */
1935      public boolean isTerminating() {
1936 <        return runStateOf(runControl) >= TERMINATING;
1936 >        long c = ctl;
1937 >        return ((c & STOP_BIT) != 0L &&
1938 >                (short)(c >>> TC_SHIFT) != -parallelism);
1939 >    }
1940 >
1941 >    /**
1942 >     * Returns true if terminating or terminated. Used by ForkJoinWorkerThread.
1943 >     */
1944 >    final boolean isAtLeastTerminating() {
1945 >        return (ctl & STOP_BIT) != 0L;
1946      }
1947  
1948      /**
1949 <     * Returns <tt>true</tt> if this pool has been shut down.
1949 >     * Returns {@code true} if this pool has been shut down.
1950       *
1951 <     * @return <tt>true</tt> if this pool has been shut down
1951 >     * @return {@code true} if this pool has been shut down
1952       */
1953      public boolean isShutdown() {
1954 <        return runStateOf(runControl) >= SHUTDOWN;
1954 >        return shutdown;
1955      }
1956  
1957      /**
# Line 1024 | Line 1961 | public class ForkJoinPool extends Abstra
1961       *
1962       * @param timeout the maximum time to wait
1963       * @param unit the time unit of the timeout argument
1964 <     * @return <tt>true</tt> if this executor terminated and
1965 <     *         <tt>false</tt> if the timeout elapsed before termination
1964 >     * @return {@code true} if this executor terminated and
1965 >     *         {@code false} if the timeout elapsed before termination
1966       * @throws InterruptedException if interrupted while waiting
1967       */
1968      public boolean awaitTermination(long timeout, TimeUnit unit)
1969          throws InterruptedException {
1970          long nanos = unit.toNanos(timeout);
1971 <        final ReentrantLock lock = this.workerLock;
1971 >        final ReentrantLock lock = this.submissionLock;
1972          lock.lock();
1973          try {
1974              for (;;) {
# Line 1046 | Line 1983 | public class ForkJoinPool extends Abstra
1983          }
1984      }
1985  
1049    // Shutdown and termination support
1050
1051    /**
1052     * Callback from terminating worker. Null out the corresponding
1053     * workers slot, and if terminating, try to terminate, else try to
1054     * shrink workers array.
1055     * @param w the worker
1056     */
1057    final void workerTerminated(ForkJoinWorkerThread w) {
1058        updateStealCount(w);
1059        updateWorkerCount(-1);
1060        final ReentrantLock lock = this.workerLock;
1061        lock.lock();
1062        try {
1063            ForkJoinWorkerThread[] ws = workers;
1064            int idx = w.poolIndex;
1065            if (idx >= 0 && idx < ws.length && ws[idx] == w)
1066                ws[idx] = null;
1067            if (totalCountOf(workerCounts) == 0) {
1068                terminate(); // no-op if already terminating
1069                transitionRunStateTo(TERMINATED);
1070                termination.signalAll();
1071            }
1072            else if (!isTerminating()) {
1073                tryShrinkWorkerArray();
1074                tryResumeSpare(true); // allow replacement
1075            }
1076        } finally {
1077            lock.unlock();
1078        }
1079        signalIdleWorkers(false);
1080    }
1081
1082    /**
1083     * Initiate termination.
1084     */
1085    private void terminate() {
1086        if (transitionRunStateTo(TERMINATING)) {
1087            stopAllWorkers();
1088            resumeAllSpares();
1089            signalIdleWorkers(true);
1090            cancelQueuedSubmissions();
1091            cancelQueuedWorkerTasks();
1092            interruptUnterminatedWorkers();
1093            signalIdleWorkers(true); // resignal after interrupt
1094        }
1095    }
1096
1097    /**
1098     * Possibly terminate when on shutdown state
1099     */
1100    private void terminateOnShutdown() {
1101        if (!hasQueuedSubmissions() && canTerminateOnShutdown(runControl))
1102            terminate();
1103    }
1104
1105    /**
1106     * Clear out and cancel submissions
1107     */
1108    private void cancelQueuedSubmissions() {
1109        ForkJoinTask<?> task;
1110        while ((task = pollSubmission()) != null)
1111            task.cancel(false);
1112    }
1113
1114    /**
1115     * Clean out worker queues.
1116     */
1117    private void cancelQueuedWorkerTasks() {
1118        final ReentrantLock lock = this.workerLock;
1119        lock.lock();
1120        try {
1121            ForkJoinWorkerThread[] ws = workers;
1122            for (int i = 0; i < ws.length; ++i) {
1123                ForkJoinWorkerThread t = ws[i];
1124                if (t != null)
1125                    t.cancelTasks();
1126            }
1127        } finally {
1128            lock.unlock();
1129        }
1130    }
1131
1132    /**
1133     * Set each worker's status to terminating. Requires lock to avoid
1134     * conflicts with add/remove
1135     */
1136    private void stopAllWorkers() {
1137        final ReentrantLock lock = this.workerLock;
1138        lock.lock();
1139        try {
1140            ForkJoinWorkerThread[] ws = workers;
1141            for (int i = 0; i < ws.length; ++i) {
1142                ForkJoinWorkerThread t = ws[i];
1143                if (t != null)
1144                    t.shutdownNow();
1145            }
1146        } finally {
1147            lock.unlock();
1148        }
1149    }
1150
1151    /**
1152     * Interrupt all unterminated workers.  This is not required for
1153     * sake of internal control, but may help unstick user code during
1154     * shutdown.
1155     */
1156    private void interruptUnterminatedWorkers() {
1157        final ReentrantLock lock = this.workerLock;
1158        lock.lock();
1159        try {
1160            ForkJoinWorkerThread[] ws = workers;
1161            for (int i = 0; i < ws.length; ++i) {
1162                ForkJoinWorkerThread t = ws[i];
1163                if (t != null && !t.isTerminated()) {
1164                    try {
1165                        t.interrupt();
1166                    } catch (SecurityException ignore) {
1167                    }
1168                }
1169            }
1170        } finally {
1171            lock.unlock();
1172        }
1173    }
1174
1175
1176    /*
1177     * Nodes for event barrier to manage idle threads.
1178     *
1179     * The event barrier has an event count and a wait queue (actually
1180     * a Treiber stack).  Workers are enabled to look for work when
1181     * the eventCount is incremented. If they fail to find some,
1182     * they may wait for next count. Synchronization events occur only
1183     * in enough contexts to maintain overall liveness:
1184     *
1185     *   - Submission of a new task to the pool
1186     *   - Creation or termination of a worker
1187     *   - pool termination
1188     *   - A worker pushing a task on an empty queue
1189     *
1190     * The last case (pushing a task) occurs often enough, and is
1191     * heavy enough compared to simple stack pushes to require some
1192     * special handling: Method signalNonEmptyWorkerQueue returns
1193     * without advancing count if the queue appears to be empty.  This
1194     * would ordinarily result in races causing some queued waiters
1195     * not to be woken up. To avoid this, a worker in sync
1196     * rescans for tasks after being enqueued if it was the first to
1197     * enqueue, and aborts the wait if finding one, also helping to
1198     * signal others. This works well because the worker has nothing
1199     * better to do anyway, and so might as well help alleviate the
1200     * overhead and contention on the threads actually doing work.
1201     *
1202     * Queue nodes are basic Treiber stack nodes, also used for spare
1203     * stack.
1204     */
1205    static final class WaitQueueNode {
1206        WaitQueueNode next; // only written before enqueued
1207        volatile ForkJoinWorkerThread thread; // nulled to cancel wait
1208        final long count; // unused for spare stack
1209        WaitQueueNode(ForkJoinWorkerThread w, long c) {
1210            count = c;
1211            thread = w;
1212        }
1213        final boolean signal() {
1214            ForkJoinWorkerThread t = thread;
1215            thread = null;
1216            if (t != null) {
1217                LockSupport.unpark(t);
1218                return true;
1219            }
1220            return false;
1221        }
1222    }
1223
1224    /**
1225     * Release at least one thread waiting for event count to advance,
1226     * if one exists. If initial attempt fails, release all threads.
1227     * @param all if false, at first try to only release one thread
1228     * @return current event
1229     */
1230    private long releaseIdleWorkers(boolean all) {
1231        long c;
1232        for (;;) {
1233            WaitQueueNode q = barrierStack;
1234            c = eventCount;
1235            long qc;
1236            if (q == null || (qc = q.count) >= c)
1237                break;
1238            if (!all) {
1239                if (casBarrierStack(q, q.next) && q.signal())
1240                    break;
1241                all = true;
1242            }
1243            else if (casBarrierStack(q, null)) {
1244                do {
1245                 q.signal();
1246                } while ((q = q.next) != null);
1247                break;
1248            }
1249        }
1250        return c;
1251    }
1252
1253    /**
1254     * Returns current barrier event count
1255     * @return current barrier event count
1256     */
1257    final long getEventCount() {
1258        long ec = eventCount;
1259        releaseIdleWorkers(true); // release to ensure accurate result
1260        return ec;
1261    }
1262
1263    /**
1264     * Increment event count and release at least one waiting thread,
1265     * if one exists (released threads will in turn wake up others).
1266     * @param all if true, try to wake up all
1267     */
1268    final void signalIdleWorkers(boolean all) {
1269        long c;
1270        do;while (!casEventCount(c = eventCount, c+1));
1271        releaseIdleWorkers(all);
1272    }
1273
1274    /**
1275     * Wake up threads waiting to steal a task. Because method
1276     * sync rechecks availability, it is OK to only proceed if
1277     * queue appears to be non-empty.
1278     */
1279    final void signalNonEmptyWorkerQueue() {
1280        // If CAS fails another signaller must have succeeded
1281        long c;
1282        if (barrierStack != null && casEventCount(c = eventCount, c+1))
1283            releaseIdleWorkers(false);
1284    }
1285
1286    /**
1287     * Waits until event count advances from count, or some thread is
1288     * waiting on a previous count, or there is stealable work
1289     * available. Help wake up others on release.
1290     * @param w the calling worker thread
1291     * @param prev previous value returned by sync (or 0)
1292     * @return current event count
1293     */
1294    final long sync(ForkJoinWorkerThread w, long prev) {
1295        updateStealCount(w);
1296
1297        while (!w.isShutdown() && !isTerminating() &&
1298               (parallelism >= runningCountOf(workerCounts) ||
1299                !suspendIfSpare(w))) { // prefer suspend to waiting here
1300            WaitQueueNode node = null;
1301            boolean queued = false;
1302            for (;;) {
1303                if (!queued) {
1304                    if (eventCount != prev)
1305                        break;
1306                    WaitQueueNode h = barrierStack;
1307                    if (h != null && h.count != prev)
1308                        break; // release below and maybe retry
1309                    if (node == null)
1310                        node = new WaitQueueNode(w, prev);
1311                    queued = casBarrierStack(node.next = h, node);
1312                }
1313                else if (Thread.interrupted() ||
1314                         node.thread == null ||
1315                         (node.next == null && w.prescan()) ||
1316                         eventCount != prev) {
1317                    node.thread = null;
1318                    if (eventCount == prev) // help trigger
1319                        casEventCount(prev, prev+1);
1320                    break;
1321                }
1322                else
1323                    LockSupport.park(this);
1324            }
1325            long ec = eventCount;
1326            if (releaseIdleWorkers(false) != prev)
1327                return ec;
1328        }
1329        return prev; // return old count if aborted
1330    }
1331
1332    //  Parallelism maintenance
1333
1334    /**
1335     * Decrement running count; if too low, add spare.
1336     *
1337     * Conceptually, all we need to do here is add or resume a
1338     * spare thread when one is about to block (and remove or
1339     * suspend it later when unblocked -- see suspendIfSpare).
1340     * However, implementing this idea requires coping with
1341     * several problems: We have imperfect information about the
1342     * states of threads. Some count updates can and usually do
1343     * lag run state changes, despite arrangements to keep them
1344     * accurate (for example, when possible, updating counts
1345     * before signalling or resuming), especially when running on
1346     * dynamic JVMs that don't optimize the infrequent paths that
1347     * update counts. Generating too many threads can make these
1348     * problems become worse, because excess threads are more
1349     * likely to be context-switched with others, slowing them all
1350     * down, especially if there is no work available, so all are
1351     * busy scanning or idling.  Also, excess spare threads can
1352     * only be suspended or removed when they are idle, not
1353     * immediately when they aren't needed. So adding threads will
1354     * raise parallelism level for longer than necessary.  Also,
1355     * FJ applications often enounter highly transient peaks when
1356     * many threads are blocked joining, but for less time than it
1357     * takes to create or resume spares.
1358     *
1359     * @param joinMe if non-null, return early if done
1360     * @param maintainParallelism if true, try to stay within
1361     * target counts, else create only to avoid starvation
1362     * @return true if joinMe known to be done
1363     */
1364    final boolean preJoin(ForkJoinTask<?> joinMe, boolean maintainParallelism) {
1365        maintainParallelism &= maintainsParallelism; // overrride
1366        boolean dec = false;  // true when running count decremented
1367        while (spareStack == null || !tryResumeSpare(dec)) {
1368            int counts = workerCounts;
1369            if (dec || (dec = casWorkerCounts(counts, --counts))) { // CAS cheat
1370                if (!needSpare(counts, maintainParallelism))
1371                    break;
1372                if (joinMe.status < 0)
1373                    return true;
1374                if (tryAddSpare(counts))
1375                    break;
1376            }
1377        }
1378        return false;
1379    }
1380
1381    /**
1382     * Same idea as preJoin
1383     */
1384    final boolean preBlock(ManagedBlocker blocker, boolean maintainParallelism){
1385        maintainParallelism &= maintainsParallelism;
1386        boolean dec = false;
1387        while (spareStack == null || !tryResumeSpare(dec)) {
1388            int counts = workerCounts;
1389            if (dec || (dec = casWorkerCounts(counts, --counts))) {
1390                if (!needSpare(counts, maintainParallelism))
1391                    break;
1392                if (blocker.isReleasable())
1393                    return true;
1394                if (tryAddSpare(counts))
1395                    break;
1396            }
1397        }
1398        return false;
1399    }
1400
1401    /**
1402     * Returns true if a spare thread appears to be needed.  If
1403     * maintaining parallelism, returns true when the deficit in
1404     * running threads is more than the surplus of total threads, and
1405     * there is apparently some work to do.  This self-limiting rule
1406     * means that the more threads that have already been added, the
1407     * less parallelism we will tolerate before adding another.
1408     * @param counts current worker counts
1409     * @param maintainParallelism try to maintain parallelism
1410     */
1411    private boolean needSpare(int counts, boolean maintainParallelism) {
1412        int ps = parallelism;
1413        int rc = runningCountOf(counts);
1414        int tc = totalCountOf(counts);
1415        int runningDeficit = ps - rc;
1416        int totalSurplus = tc - ps;
1417        return (tc < maxPoolSize &&
1418                (rc == 0 || totalSurplus < 0 ||
1419                 (maintainParallelism &&
1420                  runningDeficit > totalSurplus && mayHaveQueuedWork())));
1421    }
1422
1423    /**
1424     * Returns true if at least one worker queue appears to be
1425     * nonempty. This is expensive but not often called. It is not
1426     * critical that this be accurate, but if not, more or fewer
1427     * running threads than desired might be maintained.
1428     */
1429    private boolean mayHaveQueuedWork() {
1430        ForkJoinWorkerThread[] ws = workers;
1431        int len = ws.length;
1432        ForkJoinWorkerThread v;
1433        for (int i = 0; i < len; ++i) {
1434            if ((v = ws[i]) != null && v.getRawQueueSize() > 0) {
1435                releaseIdleWorkers(false); // help wake up stragglers
1436                return true;
1437            }
1438        }
1439        return false;
1440    }
1441
1442    /**
1443     * Add a spare worker if lock available and no more than the
1444     * expected numbers of threads exist
1445     * @return true if successful
1446     */
1447    private boolean tryAddSpare(int expectedCounts) {
1448        final ReentrantLock lock = this.workerLock;
1449        int expectedRunning = runningCountOf(expectedCounts);
1450        int expectedTotal = totalCountOf(expectedCounts);
1451        boolean success = false;
1452        boolean locked = false;
1453        // confirm counts while locking; CAS after obtaining lock
1454        try {
1455            for (;;) {
1456                int s = workerCounts;
1457                int tc = totalCountOf(s);
1458                int rc = runningCountOf(s);
1459                if (rc > expectedRunning || tc > expectedTotal)
1460                    break;
1461                if (!locked && !(locked = lock.tryLock()))
1462                    break;
1463                if (casWorkerCounts(s, workerCountsFor(tc+1, rc+1))) {
1464                    createAndStartSpare(tc);
1465                    success = true;
1466                    break;
1467                }
1468            }
1469        } finally {
1470            if (locked)
1471                lock.unlock();
1472        }
1473        return success;
1474    }
1475
1476    /**
1477     * Add the kth spare worker. On entry, pool coounts are already
1478     * adjusted to reflect addition.
1479     */
1480    private void createAndStartSpare(int k) {
1481        ForkJoinWorkerThread w = null;
1482        ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(k + 1);
1483        int len = ws.length;
1484        // Probably, we can place at slot k. If not, find empty slot
1485        if (k < len && ws[k] != null) {
1486            for (k = 0; k < len && ws[k] != null; ++k)
1487                ;
1488        }
1489        if (k < len && (w = createWorker(k)) != null) {
1490            ws[k] = w;
1491            w.start();
1492        }
1493        else
1494            updateWorkerCount(-1); // adjust on failure
1495        signalIdleWorkers(false);
1496    }
1497
1498    /**
1499     * Suspend calling thread w if there are excess threads.  Called
1500     * only from sync.  Spares are enqueued in a Treiber stack
1501     * using the same WaitQueueNodes as barriers.  They are resumed
1502     * mainly in preJoin, but are also woken on pool events that
1503     * require all threads to check run state.
1504     * @param w the caller
1505     */
1506    private boolean suspendIfSpare(ForkJoinWorkerThread w) {
1507        WaitQueueNode node = null;
1508        int s;
1509        while (parallelism < runningCountOf(s = workerCounts)) {
1510            if (node == null)
1511                node = new WaitQueueNode(w, 0);
1512            if (casWorkerCounts(s, s-1)) { // representation-dependent
1513                // push onto stack
1514                do;while (!casSpareStack(node.next = spareStack, node));
1515
1516                // block until released by resumeSpare
1517                while (node.thread != null) {
1518                    if (!Thread.interrupted())
1519                        LockSupport.park(this);
1520                }
1521                w.activate(); // help warm up
1522                return true;
1523            }
1524        }
1525        return false;
1526    }
1527
1528    /**
1529     * Try to pop and resume a spare thread.
1530     * @param updateCount if true, increment running count on success
1531     * @return true if successful
1532     */
1533    private boolean tryResumeSpare(boolean updateCount) {
1534        WaitQueueNode q;
1535        while ((q = spareStack) != null) {
1536            if (casSpareStack(q, q.next)) {
1537                if (updateCount)
1538                    updateRunningCount(1);
1539                q.signal();
1540                return true;
1541            }
1542        }
1543        return false;
1544    }
1545
1546    /**
1547     * Pop and resume all spare threads. Same idea as
1548     * releaseIdleWorkers.
1549     * @return true if any spares released
1550     */
1551    private boolean resumeAllSpares() {
1552        WaitQueueNode q;
1553        while ( (q = spareStack) != null) {
1554            if (casSpareStack(q, null)) {
1555                do {
1556                    updateRunningCount(1);
1557                    q.signal();
1558                } while ((q = q.next) != null);
1559                return true;
1560            }
1561        }
1562        return false;
1563    }
1564
1565    /**
1566     * Pop and shutdown excessive spare threads. Call only while
1567     * holding lock. This is not guaranteed to eliminate all excess
1568     * threads, only those suspended as spares, which are the ones
1569     * unlikely to be needed in the future.
1570     */
1571    private void trimSpares() {
1572        int surplus = totalCountOf(workerCounts) - parallelism;
1573        WaitQueueNode q;
1574        while (surplus > 0 && (q = spareStack) != null) {
1575            if (casSpareStack(q, null)) {
1576                do {
1577                    updateRunningCount(1);
1578                    ForkJoinWorkerThread w = q.thread;
1579                    if (w != null && surplus > 0 &&
1580                        runningCountOf(workerCounts) > 0 && w.shutdown())
1581                        --surplus;
1582                    q.signal();
1583                } while ((q = q.next) != null);
1584            }
1585        }
1586    }
1587
1588    /**
1589     * Returns approximate number of spares, just for diagnostics.
1590     */
1591    private int countSpares() {
1592        int sum = 0;
1593        for (WaitQueueNode q = spareStack; q != null; q = q.next)
1594            ++sum;
1595        return sum;
1596    }
1597
1986      /**
1987       * Interface for extending managed parallelism for tasks running
1988 <     * in ForkJoinPools. A ManagedBlocker provides two methods.
1989 <     * Method <tt>isReleasable</tt> must return true if blocking is not
1990 <     * necessary. Method <tt>block</tt> blocks the current thread
1991 <     * if necessary (perhaps internally invoking isReleasable before
1992 <     * actually blocking.).
1988 >     * in {@link ForkJoinPool}s.
1989 >     *
1990 >     * <p>A {@code ManagedBlocker} provides two methods.  Method
1991 >     * {@code isReleasable} must return {@code true} if blocking is
1992 >     * not necessary. Method {@code block} blocks the current thread
1993 >     * if necessary (perhaps internally invoking {@code isReleasable}
1994 >     * before actually blocking). These actions are performed by any
1995 >     * thread invoking {@link ForkJoinPool#managedBlock}.  The
1996 >     * unusual methods in this API accommodate synchronizers that may,
1997 >     * but don't usually, block for long periods. Similarly, they
1998 >     * allow more efficient internal handling of cases in which
1999 >     * additional workers may be, but usually are not, needed to
2000 >     * ensure sufficient parallelism.  Toward this end,
2001 >     * implementations of method {@code isReleasable} must be amenable
2002 >     * to repeated invocation.
2003 >     *
2004       * <p>For example, here is a ManagedBlocker based on a
2005       * ReentrantLock:
2006 <     * <pre>
2007 <     *   class ManagedLocker implements ManagedBlocker {
2008 <     *     final ReentrantLock lock;
2009 <     *     boolean hasLock = false;
2010 <     *     ManagedLocker(ReentrantLock lock) { this.lock = lock; }
2011 <     *     public boolean block() {
2012 <     *        if (!hasLock)
2013 <     *           lock.lock();
2014 <     *        return true;
2015 <     *     }
2016 <     *     public boolean isReleasable() {
2017 <     *        return hasLock || (hasLock = lock.tryLock());
2018 <     *     }
2006 >     *  <pre> {@code
2007 >     * class ManagedLocker implements ManagedBlocker {
2008 >     *   final ReentrantLock lock;
2009 >     *   boolean hasLock = false;
2010 >     *   ManagedLocker(ReentrantLock lock) { this.lock = lock; }
2011 >     *   public boolean block() {
2012 >     *     if (!hasLock)
2013 >     *       lock.lock();
2014 >     *     return true;
2015 >     *   }
2016 >     *   public boolean isReleasable() {
2017 >     *     return hasLock || (hasLock = lock.tryLock());
2018 >     *   }
2019 >     * }}</pre>
2020 >     *
2021 >     * <p>Here is a class that possibly blocks waiting for an
2022 >     * item on a given queue:
2023 >     *  <pre> {@code
2024 >     * class QueueTaker<E> implements ManagedBlocker {
2025 >     *   final BlockingQueue<E> queue;
2026 >     *   volatile E item = null;
2027 >     *   QueueTaker(BlockingQueue<E> q) { this.queue = q; }
2028 >     *   public boolean block() throws InterruptedException {
2029 >     *     if (item == null)
2030 >     *       item = queue.take();
2031 >     *     return true;
2032       *   }
2033 <     * </pre>
2033 >     *   public boolean isReleasable() {
2034 >     *     return item != null || (item = queue.poll()) != null;
2035 >     *   }
2036 >     *   public E getItem() { // call after pool.managedBlock completes
2037 >     *     return item;
2038 >     *   }
2039 >     * }}</pre>
2040       */
2041      public static interface ManagedBlocker {
2042          /**
2043           * Possibly blocks the current thread, for example waiting for
2044           * a lock or condition.
2045 <         * @return true if no additional blocking is necessary (i.e.,
2046 <         * if isReleasable would return true).
2045 >         *
2046 >         * @return {@code true} if no additional blocking is necessary
2047 >         * (i.e., if isReleasable would return true)
2048           * @throws InterruptedException if interrupted while waiting
2049 <         * (the method is not required to do so, but is allowe to).
2049 >         * (the method is not required to do so, but is allowed to)
2050           */
2051          boolean block() throws InterruptedException;
2052  
2053          /**
2054 <         * Returns true if blocking is unnecessary.
2054 >         * Returns {@code true} if blocking is unnecessary.
2055           */
2056          boolean isReleasable();
2057      }
2058  
2059      /**
2060       * Blocks in accord with the given blocker.  If the current thread
2061 <     * is a ForkJoinWorkerThread, this method possibly arranges for a
2062 <     * spare thread to be activated if necessary to ensure parallelism
2063 <     * while the current thread is blocked.  If
2064 <     * <tt>maintainParallelism</tt> is true and the pool supports it
2065 <     * (see <tt>getMaintainsParallelism</tt>), this method attempts to
2066 <     * maintain the pool's nominal parallelism. Otherwise if activates
2067 <     * a thread only if necessary to avoid complete starvation. This
2068 <     * option may be preferable when blockages use timeouts, or are
2069 <     * almost always brief.
2070 <     *
2071 <     * <p> If the caller is not a ForkJoinTask, this method is behaviorally
2072 <     * equivalent to
2073 <     * <pre>
2074 <     *   while (!blocker.isReleasable())
1656 <     *      if (blocker.block())
1657 <     *         return;
1658 <     * </pre>
1659 <     * If the caller is a ForkJoinTask, then the pool may first
1660 <     * be expanded to ensure parallelism, and later adjusted.
2061 >     * is a {@link ForkJoinWorkerThread}, this method possibly
2062 >     * arranges for a spare thread to be activated if necessary to
2063 >     * ensure sufficient parallelism while the current thread is blocked.
2064 >     *
2065 >     * <p>If the caller is not a {@link ForkJoinTask}, this method is
2066 >     * behaviorally equivalent to
2067 >     *  <pre> {@code
2068 >     * while (!blocker.isReleasable())
2069 >     *   if (blocker.block())
2070 >     *     return;
2071 >     * }</pre>
2072 >     *
2073 >     * If the caller is a {@code ForkJoinTask}, then the pool may
2074 >     * first be expanded to ensure parallelism, and later adjusted.
2075       *
2076       * @param blocker the blocker
2077 <     * @param maintainParallelism if true and supported by this pool,
1664 <     * attempt to maintain the pool's nominal parallelism; otherwise
1665 <     * activate a thread only if necessary to avoid complete
1666 <     * starvation.
1667 <     * @throws InterruptedException if blocker.block did so.
2077 >     * @throws InterruptedException if blocker.block did so
2078       */
2079 <    public static void managedBlock(ManagedBlocker blocker,
1670 <                                    boolean maintainParallelism)
2079 >    public static void managedBlock(ManagedBlocker blocker)
2080          throws InterruptedException {
2081          Thread t = Thread.currentThread();
2082 <        ForkJoinPool pool = (t instanceof ForkJoinWorkerThread?
2083 <                             ((ForkJoinWorkerThread)t).pool : null);
2084 <        if (!blocker.isReleasable()) {
2085 <            try {
2086 <                if (pool == null ||
2087 <                    !pool.preBlock(blocker, maintainParallelism))
1679 <                    awaitBlocker(blocker);
1680 <            } finally {
1681 <                if (pool != null)
1682 <                    pool.updateRunningCount(1);
1683 <            }
2082 >        if (t instanceof ForkJoinWorkerThread) {
2083 >            ForkJoinWorkerThread w = (ForkJoinWorkerThread) t;
2084 >            w.pool.awaitBlocker(blocker);
2085 >        }
2086 >        else {
2087 >            do {} while (!blocker.isReleasable() && !blocker.block());
2088          }
2089      }
2090  
2091 <    private static void awaitBlocker(ManagedBlocker blocker)
2092 <        throws InterruptedException {
2093 <        do;while (!blocker.isReleasable() && !blocker.block());
1690 <    }
2091 >    // AbstractExecutorService overrides.  These rely on undocumented
2092 >    // fact that ForkJoinTask.adapt returns ForkJoinTasks that also
2093 >    // implement RunnableFuture.
2094  
2095 +    protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
2096 +        return (RunnableFuture<T>) ForkJoinTask.adapt(runnable, value);
2097 +    }
2098  
2099 <    // Temporary Unsafe mechanics for preliminary release
2099 >    protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
2100 >        return (RunnableFuture<T>) ForkJoinTask.adapt(callable);
2101 >    }
2102  
2103 <    static final Unsafe _unsafe;
2104 <    static final long eventCountOffset;
2105 <    static final long workerCountsOffset;
2106 <    static final long runControlOffset;
2107 <    static final long barrierStackOffset;
2108 <    static final long spareStackOffset;
2103 >    // Unsafe mechanics
2104 >    private static final sun.misc.Unsafe UNSAFE;
2105 >    private static final long ctlOffset;
2106 >    private static final long stealCountOffset;
2107 >    private static final long blockedCountOffset;
2108 >    private static final long quiescerCountOffset;
2109 >    private static final long scanGuardOffset;
2110 >    private static final long nextWorkerNumberOffset;
2111 >    private static final long ABASE;
2112 >    private static final int ASHIFT;
2113  
2114      static {
2115 +        poolNumberGenerator = new AtomicInteger();
2116 +        workerSeedGenerator = new Random();
2117 +        modifyThreadPermission = new RuntimePermission("modifyThread");
2118 +        defaultForkJoinWorkerThreadFactory =
2119 +            new DefaultForkJoinWorkerThreadFactory();
2120 +        int s;
2121          try {
2122 <            if (ForkJoinPool.class.getClassLoader() != null) {
2123 <                Field f = Unsafe.class.getDeclaredField("theUnsafe");
2124 <                f.setAccessible(true);
2125 <                _unsafe = (Unsafe)f.get(null);
2126 <            }
2127 <            else
2128 <                _unsafe = Unsafe.getUnsafe();
2129 <            eventCountOffset = _unsafe.objectFieldOffset
2130 <                (ForkJoinPool.class.getDeclaredField("eventCount"));
2131 <            workerCountsOffset = _unsafe.objectFieldOffset
2132 <                (ForkJoinPool.class.getDeclaredField("workerCounts"));
2133 <            runControlOffset = _unsafe.objectFieldOffset
2134 <                (ForkJoinPool.class.getDeclaredField("runControl"));
2135 <            barrierStackOffset = _unsafe.objectFieldOffset
2136 <                (ForkJoinPool.class.getDeclaredField("barrierStack"));
2137 <            spareStackOffset = _unsafe.objectFieldOffset
2138 <                (ForkJoinPool.class.getDeclaredField("spareStack"));
2122 >            UNSAFE = getUnsafe();
2123 >            Class k = ForkJoinPool.class;
2124 >            ctlOffset = UNSAFE.objectFieldOffset
2125 >                (k.getDeclaredField("ctl"));
2126 >            stealCountOffset = UNSAFE.objectFieldOffset
2127 >                (k.getDeclaredField("stealCount"));
2128 >            blockedCountOffset = UNSAFE.objectFieldOffset
2129 >                (k.getDeclaredField("blockedCount"));
2130 >            quiescerCountOffset = UNSAFE.objectFieldOffset
2131 >                (k.getDeclaredField("quiescerCount"));
2132 >            scanGuardOffset = UNSAFE.objectFieldOffset
2133 >                (k.getDeclaredField("scanGuard"));
2134 >            nextWorkerNumberOffset = UNSAFE.objectFieldOffset
2135 >                (k.getDeclaredField("nextWorkerNumber"));
2136 >            Class a = ForkJoinTask[].class;
2137 >            ABASE = UNSAFE.arrayBaseOffset(a);
2138 >            s = UNSAFE.arrayIndexScale(a);
2139          } catch (Exception e) {
2140 <            throw new RuntimeException("Could not initialize intrinsics", e);
2140 >            throw new Error(e);
2141          }
2142 +        if ((s & (s-1)) != 0)
2143 +            throw new Error("data type scale not a power of two");
2144 +        ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
2145      }
2146  
2147 <    private boolean casEventCount(long cmp, long val) {
2148 <        return _unsafe.compareAndSwapLong(this, eventCountOffset, cmp, val);
2149 <    }
2150 <    private boolean casWorkerCounts(int cmp, int val) {
2151 <        return _unsafe.compareAndSwapInt(this, workerCountsOffset, cmp, val);
2152 <    }
2153 <    private boolean casRunControl(int cmp, int val) {
2154 <        return _unsafe.compareAndSwapInt(this, runControlOffset, cmp, val);
2155 <    }
2156 <    private boolean casSpareStack(WaitQueueNode cmp, WaitQueueNode val) {
2157 <        return _unsafe.compareAndSwapObject(this, spareStackOffset, cmp, val);
2158 <    }
2159 <    private boolean casBarrierStack(WaitQueueNode cmp, WaitQueueNode val) {
2160 <        return _unsafe.compareAndSwapObject(this, barrierStackOffset, cmp, val);
2147 >    /**
2148 >     * Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
2149 >     * Replace with a simple call to Unsafe.getUnsafe when integrating
2150 >     * into a jdk.
2151 >     *
2152 >     * @return a sun.misc.Unsafe
2153 >     */
2154 >    private static sun.misc.Unsafe getUnsafe() {
2155 >        try {
2156 >            return sun.misc.Unsafe.getUnsafe();
2157 >        } catch (SecurityException se) {
2158 >            try {
2159 >                return java.security.AccessController.doPrivileged
2160 >                    (new java.security
2161 >                     .PrivilegedExceptionAction<sun.misc.Unsafe>() {
2162 >                        public sun.misc.Unsafe run() throws Exception {
2163 >                            java.lang.reflect.Field f = sun.misc
2164 >                                .Unsafe.class.getDeclaredField("theUnsafe");
2165 >                            f.setAccessible(true);
2166 >                            return (sun.misc.Unsafe) f.get(null);
2167 >                        }});
2168 >            } catch (java.security.PrivilegedActionException e) {
2169 >                throw new RuntimeException("Could not initialize intrinsics",
2170 >                                           e.getCause());
2171 >            }
2172 >        }
2173      }
2174   }

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines