ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/jsr166/jsr166/src/main/java/util/concurrent/SynchronousQueue.java
Revision: 1.78
Committed: Wed Jun 8 00:50:35 2011 UTC (13 years ago) by jsr166
Branch: MAIN
Changes since 1.77: +0 -1 lines
Log Message:
clean up imports

File Contents

# Content
1 /*
2 * Written by Doug Lea, Bill Scherer, and Michael Scott with
3 * assistance from members of JCP JSR-166 Expert Group and released to
4 * the public domain, as explained at
5 * http://creativecommons.org/publicdomain/zero/1.0/
6 */
7
8 package java.util.concurrent;
9 import java.util.concurrent.locks.*;
10 import java.util.*;
11
12 /**
13 * A {@linkplain BlockingQueue blocking queue} in which each insert
14 * operation must wait for a corresponding remove operation by another
15 * thread, and vice versa. A synchronous queue does not have any
16 * internal capacity, not even a capacity of one. You cannot
17 * <tt>peek</tt> at a synchronous queue because an element is only
18 * present when you try to remove it; you cannot insert an element
19 * (using any method) unless another thread is trying to remove it;
20 * you cannot iterate as there is nothing to iterate. The
21 * <em>head</em> of the queue is the element that the first queued
22 * inserting thread is trying to add to the queue; if there is no such
23 * queued thread then no element is available for removal and
24 * <tt>poll()</tt> will return <tt>null</tt>. For purposes of other
25 * <tt>Collection</tt> methods (for example <tt>contains</tt>), a
26 * <tt>SynchronousQueue</tt> acts as an empty collection. This queue
27 * does not permit <tt>null</tt> elements.
28 *
29 * <p>Synchronous queues are similar to rendezvous channels used in
30 * CSP and Ada. They are well suited for handoff designs, in which an
31 * object running in one thread must sync up with an object running
32 * in another thread in order to hand it some information, event, or
33 * task.
34 *
35 * <p> This class supports an optional fairness policy for ordering
36 * waiting producer and consumer threads. By default, this ordering
37 * is not guaranteed. However, a queue constructed with fairness set
38 * to <tt>true</tt> grants threads access in FIFO order.
39 *
40 * <p>This class and its iterator implement all of the
41 * <em>optional</em> methods of the {@link Collection} and {@link
42 * Iterator} interfaces.
43 *
44 * <p>This class is a member of the
45 * <a href="{@docRoot}/../technotes/guides/collections/index.html">
46 * Java Collections Framework</a>.
47 *
48 * @since 1.5
49 * @author Doug Lea and Bill Scherer and Michael Scott
50 * @param <E> the type of elements held in this collection
51 */
52 public class SynchronousQueue<E> extends AbstractQueue<E>
53 implements BlockingQueue<E>, java.io.Serializable {
54 private static final long serialVersionUID = -3223113410248163686L;
55
56 /*
57 * This class implements extensions of the dual stack and dual
58 * queue algorithms described in "Nonblocking Concurrent Objects
59 * with Condition Synchronization", by W. N. Scherer III and
60 * M. L. Scott. 18th Annual Conf. on Distributed Computing,
61 * Oct. 2004 (see also
62 * http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/duals.html).
63 * The (Lifo) stack is used for non-fair mode, and the (Fifo)
64 * queue for fair mode. The performance of the two is generally
65 * similar. Fifo usually supports higher throughput under
66 * contention but Lifo maintains higher thread locality in common
67 * applications.
68 *
69 * A dual queue (and similarly stack) is one that at any given
70 * time either holds "data" -- items provided by put operations,
71 * or "requests" -- slots representing take operations, or is
72 * empty. A call to "fulfill" (i.e., a call requesting an item
73 * from a queue holding data or vice versa) dequeues a
74 * complementary node. The most interesting feature of these
75 * queues is that any operation can figure out which mode the
76 * queue is in, and act accordingly without needing locks.
77 *
78 * Both the queue and stack extend abstract class Transferer
79 * defining the single method transfer that does a put or a
80 * take. These are unified into a single method because in dual
81 * data structures, the put and take operations are symmetrical,
82 * so nearly all code can be combined. The resulting transfer
83 * methods are on the long side, but are easier to follow than
84 * they would be if broken up into nearly-duplicated parts.
85 *
86 * The queue and stack data structures share many conceptual
87 * similarities but very few concrete details. For simplicity,
88 * they are kept distinct so that they can later evolve
89 * separately.
90 *
91 * The algorithms here differ from the versions in the above paper
92 * in extending them for use in synchronous queues, as well as
93 * dealing with cancellation. The main differences include:
94 *
95 * 1. The original algorithms used bit-marked pointers, but
96 * the ones here use mode bits in nodes, leading to a number
97 * of further adaptations.
98 * 2. SynchronousQueues must block threads waiting to become
99 * fulfilled.
100 * 3. Support for cancellation via timeout and interrupts,
101 * including cleaning out cancelled nodes/threads
102 * from lists to avoid garbage retention and memory depletion.
103 *
104 * Blocking is mainly accomplished using LockSupport park/unpark,
105 * except that nodes that appear to be the next ones to become
106 * fulfilled first spin a bit (on multiprocessors only). On very
107 * busy synchronous queues, spinning can dramatically improve
108 * throughput. And on less busy ones, the amount of spinning is
109 * small enough not to be noticeable.
110 *
111 * Cleaning is done in different ways in queues vs stacks. For
112 * queues, we can almost always remove a node immediately in O(1)
113 * time (modulo retries for consistency checks) when it is
114 * cancelled. But if it may be pinned as the current tail, it must
115 * wait until some subsequent cancellation. For stacks, we need a
116 * potentially O(n) traversal to be sure that we can remove the
117 * node, but this can run concurrently with other threads
118 * accessing the stack.
119 *
120 * While garbage collection takes care of most node reclamation
121 * issues that otherwise complicate nonblocking algorithms, care
122 * is taken to "forget" references to data, other nodes, and
123 * threads that might be held on to long-term by blocked
124 * threads. In cases where setting to null would otherwise
125 * conflict with main algorithms, this is done by changing a
126 * node's link to now point to the node itself. This doesn't arise
127 * much for Stack nodes (because blocked threads do not hang on to
128 * old head pointers), but references in Queue nodes must be
129 * aggressively forgotten to avoid reachability of everything any
130 * node has ever referred to since arrival.
131 */
132
133 /**
134 * Shared internal API for dual stacks and queues.
135 */
136 abstract static class Transferer {
137 /**
138 * Performs a put or take.
139 *
140 * @param e if non-null, the item to be handed to a consumer;
141 * if null, requests that transfer return an item
142 * offered by producer.
143 * @param timed if this operation should timeout
144 * @param nanos the timeout, in nanoseconds
145 * @return if non-null, the item provided or received; if null,
146 * the operation failed due to timeout or interrupt --
147 * the caller can distinguish which of these occurred
148 * by checking Thread.interrupted.
149 */
150 abstract Object transfer(Object e, boolean timed, long nanos);
151 }
152
153 /** The number of CPUs, for spin control */
154 static final int NCPUS = Runtime.getRuntime().availableProcessors();
155
156 /**
157 * The number of times to spin before blocking in timed waits.
158 * The value is empirically derived -- it works well across a
159 * variety of processors and OSes. Empirically, the best value
160 * seems not to vary with number of CPUs (beyond 2) so is just
161 * a constant.
162 */
163 static final int maxTimedSpins = (NCPUS < 2) ? 0 : 32;
164
165 /**
166 * The number of times to spin before blocking in untimed waits.
167 * This is greater than timed value because untimed waits spin
168 * faster since they don't need to check times on each spin.
169 */
170 static final int maxUntimedSpins = maxTimedSpins * 16;
171
172 /**
173 * The number of nanoseconds for which it is faster to spin
174 * rather than to use timed park. A rough estimate suffices.
175 */
176 static final long spinForTimeoutThreshold = 1000L;
177
178 /** Dual stack */
179 static final class TransferStack extends Transferer {
180 /*
181 * This extends Scherer-Scott dual stack algorithm, differing,
182 * among other ways, by using "covering" nodes rather than
183 * bit-marked pointers: Fulfilling operations push on marker
184 * nodes (with FULFILLING bit set in mode) to reserve a spot
185 * to match a waiting node.
186 */
187
188 /* Modes for SNodes, ORed together in node fields */
189 /** Node represents an unfulfilled consumer */
190 static final int REQUEST = 0;
191 /** Node represents an unfulfilled producer */
192 static final int DATA = 1;
193 /** Node is fulfilling another unfulfilled DATA or REQUEST */
194 static final int FULFILLING = 2;
195
196 /** Return true if m has fulfilling bit set */
197 static boolean isFulfilling(int m) { return (m & FULFILLING) != 0; }
198
199 /** Node class for TransferStacks. */
200 static final class SNode {
201 volatile SNode next; // next node in stack
202 volatile SNode match; // the node matched to this
203 volatile Thread waiter; // to control park/unpark
204 Object item; // data; or null for REQUESTs
205 int mode;
206 // Note: item and mode fields don't need to be volatile
207 // since they are always written before, and read after,
208 // other volatile/atomic operations.
209
210 SNode(Object item) {
211 this.item = item;
212 }
213
214 boolean casNext(SNode cmp, SNode val) {
215 return cmp == next &&
216 UNSAFE.compareAndSwapObject(this, nextOffset, cmp, val);
217 }
218
219 /**
220 * Tries to match node s to this node, if so, waking up thread.
221 * Fulfillers call tryMatch to identify their waiters.
222 * Waiters block until they have been matched.
223 *
224 * @param s the node to match
225 * @return true if successfully matched to s
226 */
227 boolean tryMatch(SNode s) {
228 if (match == null &&
229 UNSAFE.compareAndSwapObject(this, matchOffset, null, s)) {
230 Thread w = waiter;
231 if (w != null) { // waiters need at most one unpark
232 waiter = null;
233 LockSupport.unpark(w);
234 }
235 return true;
236 }
237 return match == s;
238 }
239
240 /**
241 * Tries to cancel a wait by matching node to itself.
242 */
243 void tryCancel() {
244 UNSAFE.compareAndSwapObject(this, matchOffset, null, this);
245 }
246
247 boolean isCancelled() {
248 return match == this;
249 }
250
251 // Unsafe mechanics
252 private static final sun.misc.Unsafe UNSAFE;
253 private static final long matchOffset;
254 private static final long nextOffset;
255
256 static {
257 try {
258 UNSAFE = sun.misc.Unsafe.getUnsafe();
259 Class<?> k = SNode.class;
260 matchOffset = UNSAFE.objectFieldOffset
261 (k.getDeclaredField("match"));
262 nextOffset = UNSAFE.objectFieldOffset
263 (k.getDeclaredField("next"));
264 } catch (Exception e) {
265 throw new Error(e);
266 }
267 }
268 }
269
270 /** The head (top) of the stack */
271 volatile SNode head;
272
273 boolean casHead(SNode h, SNode nh) {
274 return h == head &&
275 UNSAFE.compareAndSwapObject(this, headOffset, h, nh);
276 }
277
278 /**
279 * Creates or resets fields of a node. Called only from transfer
280 * where the node to push on stack is lazily created and
281 * reused when possible to help reduce intervals between reads
282 * and CASes of head and to avoid surges of garbage when CASes
283 * to push nodes fail due to contention.
284 */
285 static SNode snode(SNode s, Object e, SNode next, int mode) {
286 if (s == null) s = new SNode(e);
287 s.mode = mode;
288 s.next = next;
289 return s;
290 }
291
292 /**
293 * Puts or takes an item.
294 */
295 Object transfer(Object e, boolean timed, long nanos) {
296 /*
297 * Basic algorithm is to loop trying one of three actions:
298 *
299 * 1. If apparently empty or already containing nodes of same
300 * mode, try to push node on stack and wait for a match,
301 * returning it, or null if cancelled.
302 *
303 * 2. If apparently containing node of complementary mode,
304 * try to push a fulfilling node on to stack, match
305 * with corresponding waiting node, pop both from
306 * stack, and return matched item. The matching or
307 * unlinking might not actually be necessary because of
308 * other threads performing action 3:
309 *
310 * 3. If top of stack already holds another fulfilling node,
311 * help it out by doing its match and/or pop
312 * operations, and then continue. The code for helping
313 * is essentially the same as for fulfilling, except
314 * that it doesn't return the item.
315 */
316
317 SNode s = null; // constructed/reused as needed
318 int mode = (e == null) ? REQUEST : DATA;
319
320 for (;;) {
321 SNode h = head;
322 if (h == null || h.mode == mode) { // empty or same-mode
323 if (timed && nanos <= 0) { // can't wait
324 if (h != null && h.isCancelled())
325 casHead(h, h.next); // pop cancelled node
326 else
327 return null;
328 } else if (casHead(h, s = snode(s, e, h, mode))) {
329 SNode m = awaitFulfill(s, timed, nanos);
330 if (m == s) { // wait was cancelled
331 clean(s);
332 return null;
333 }
334 if ((h = head) != null && h.next == s)
335 casHead(h, s.next); // help s's fulfiller
336 return (mode == REQUEST) ? m.item : s.item;
337 }
338 } else if (!isFulfilling(h.mode)) { // try to fulfill
339 if (h.isCancelled()) // already cancelled
340 casHead(h, h.next); // pop and retry
341 else if (casHead(h, s=snode(s, e, h, FULFILLING|mode))) {
342 for (;;) { // loop until matched or waiters disappear
343 SNode m = s.next; // m is s's match
344 if (m == null) { // all waiters are gone
345 casHead(s, null); // pop fulfill node
346 s = null; // use new node next time
347 break; // restart main loop
348 }
349 SNode mn = m.next;
350 if (m.tryMatch(s)) {
351 casHead(s, mn); // pop both s and m
352 return (mode == REQUEST) ? m.item : s.item;
353 } else // lost match
354 s.casNext(m, mn); // help unlink
355 }
356 }
357 } else { // help a fulfiller
358 SNode m = h.next; // m is h's match
359 if (m == null) // waiter is gone
360 casHead(h, null); // pop fulfilling node
361 else {
362 SNode mn = m.next;
363 if (m.tryMatch(h)) // help match
364 casHead(h, mn); // pop both h and m
365 else // lost match
366 h.casNext(m, mn); // help unlink
367 }
368 }
369 }
370 }
371
372 /**
373 * Spins/blocks until node s is matched by a fulfill operation.
374 *
375 * @param s the waiting node
376 * @param timed true if timed wait
377 * @param nanos timeout value
378 * @return matched node, or s if cancelled
379 */
380 SNode awaitFulfill(SNode s, boolean timed, long nanos) {
381 /*
382 * When a node/thread is about to block, it sets its waiter
383 * field and then rechecks state at least one more time
384 * before actually parking, thus covering race vs
385 * fulfiller noticing that waiter is non-null so should be
386 * woken.
387 *
388 * When invoked by nodes that appear at the point of call
389 * to be at the head of the stack, calls to park are
390 * preceded by spins to avoid blocking when producers and
391 * consumers are arriving very close in time. This can
392 * happen enough to bother only on multiprocessors.
393 *
394 * The order of checks for returning out of main loop
395 * reflects fact that interrupts have precedence over
396 * normal returns, which have precedence over
397 * timeouts. (So, on timeout, one last check for match is
398 * done before giving up.) Except that calls from untimed
399 * SynchronousQueue.{poll/offer} don't check interrupts
400 * and don't wait at all, so are trapped in transfer
401 * method rather than calling awaitFulfill.
402 */
403 long lastTime = timed ? System.nanoTime() : 0;
404 Thread w = Thread.currentThread();
405 SNode h = head;
406 int spins = (shouldSpin(s) ?
407 (timed ? maxTimedSpins : maxUntimedSpins) : 0);
408 for (;;) {
409 if (w.isInterrupted())
410 s.tryCancel();
411 SNode m = s.match;
412 if (m != null)
413 return m;
414 if (timed) {
415 long now = System.nanoTime();
416 nanos -= now - lastTime;
417 lastTime = now;
418 if (nanos <= 0) {
419 s.tryCancel();
420 continue;
421 }
422 }
423 if (spins > 0)
424 spins = shouldSpin(s) ? (spins-1) : 0;
425 else if (s.waiter == null)
426 s.waiter = w; // establish waiter so can park next iter
427 else if (!timed)
428 LockSupport.park(this);
429 else if (nanos > spinForTimeoutThreshold)
430 LockSupport.parkNanos(this, nanos);
431 }
432 }
433
434 /**
435 * Returns true if node s is at head or there is an active
436 * fulfiller.
437 */
438 boolean shouldSpin(SNode s) {
439 SNode h = head;
440 return (h == s || h == null || isFulfilling(h.mode));
441 }
442
443 /**
444 * Unlinks s from the stack.
445 */
446 void clean(SNode s) {
447 s.item = null; // forget item
448 s.waiter = null; // forget thread
449
450 /*
451 * At worst we may need to traverse entire stack to unlink
452 * s. If there are multiple concurrent calls to clean, we
453 * might not see s if another thread has already removed
454 * it. But we can stop when we see any node known to
455 * follow s. We use s.next unless it too is cancelled, in
456 * which case we try the node one past. We don't check any
457 * further because we don't want to doubly traverse just to
458 * find sentinel.
459 */
460
461 SNode past = s.next;
462 if (past != null && past.isCancelled())
463 past = past.next;
464
465 // Absorb cancelled nodes at head
466 SNode p;
467 while ((p = head) != null && p != past && p.isCancelled())
468 casHead(p, p.next);
469
470 // Unsplice embedded nodes
471 while (p != null && p != past) {
472 SNode n = p.next;
473 if (n != null && n.isCancelled())
474 p.casNext(n, n.next);
475 else
476 p = n;
477 }
478 }
479
480 // Unsafe mechanics
481 private static final sun.misc.Unsafe UNSAFE;
482 private static final long headOffset;
483 static {
484 try {
485 UNSAFE = sun.misc.Unsafe.getUnsafe();
486 Class<?> k = TransferStack.class;
487 headOffset = UNSAFE.objectFieldOffset
488 (k.getDeclaredField("head"));
489 } catch (Exception e) {
490 throw new Error(e);
491 }
492 }
493 }
494
495 /** Dual Queue */
496 static final class TransferQueue extends Transferer {
497 /*
498 * This extends Scherer-Scott dual queue algorithm, differing,
499 * among other ways, by using modes within nodes rather than
500 * marked pointers. The algorithm is a little simpler than
501 * that for stacks because fulfillers do not need explicit
502 * nodes, and matching is done by CAS'ing QNode.item field
503 * from non-null to null (for put) or vice versa (for take).
504 */
505
506 /** Node class for TransferQueue. */
507 static final class QNode {
508 volatile QNode next; // next node in queue
509 volatile Object item; // CAS'ed to or from null
510 volatile Thread waiter; // to control park/unpark
511 final boolean isData;
512
513 QNode(Object item, boolean isData) {
514 this.item = item;
515 this.isData = isData;
516 }
517
518 boolean casNext(QNode cmp, QNode val) {
519 return next == cmp &&
520 UNSAFE.compareAndSwapObject(this, nextOffset, cmp, val);
521 }
522
523 boolean casItem(Object cmp, Object val) {
524 return item == cmp &&
525 UNSAFE.compareAndSwapObject(this, itemOffset, cmp, val);
526 }
527
528 /**
529 * Tries to cancel by CAS'ing ref to this as item.
530 */
531 void tryCancel(Object cmp) {
532 UNSAFE.compareAndSwapObject(this, itemOffset, cmp, this);
533 }
534
535 boolean isCancelled() {
536 return item == this;
537 }
538
539 /**
540 * Returns true if this node is known to be off the queue
541 * because its next pointer has been forgotten due to
542 * an advanceHead operation.
543 */
544 boolean isOffList() {
545 return next == this;
546 }
547
548 // Unsafe mechanics
549 private static final sun.misc.Unsafe UNSAFE;
550 private static final long itemOffset;
551 private static final long nextOffset;
552
553 static {
554 try {
555 UNSAFE = sun.misc.Unsafe.getUnsafe();
556 Class<?> k = QNode.class;
557 itemOffset = UNSAFE.objectFieldOffset
558 (k.getDeclaredField("item"));
559 nextOffset = UNSAFE.objectFieldOffset
560 (k.getDeclaredField("next"));
561 } catch (Exception e) {
562 throw new Error(e);
563 }
564 }
565 }
566
567 /** Head of queue */
568 transient volatile QNode head;
569 /** Tail of queue */
570 transient volatile QNode tail;
571 /**
572 * Reference to a cancelled node that might not yet have been
573 * unlinked from queue because it was the last inserted node
574 * when it cancelled.
575 */
576 transient volatile QNode cleanMe;
577
578 TransferQueue() {
579 QNode h = new QNode(null, false); // initialize to dummy node.
580 head = h;
581 tail = h;
582 }
583
584 /**
585 * Tries to cas nh as new head; if successful, unlink
586 * old head's next node to avoid garbage retention.
587 */
588 void advanceHead(QNode h, QNode nh) {
589 if (h == head &&
590 UNSAFE.compareAndSwapObject(this, headOffset, h, nh))
591 h.next = h; // forget old next
592 }
593
594 /**
595 * Tries to cas nt as new tail.
596 */
597 void advanceTail(QNode t, QNode nt) {
598 if (tail == t)
599 UNSAFE.compareAndSwapObject(this, tailOffset, t, nt);
600 }
601
602 /**
603 * Tries to CAS cleanMe slot.
604 */
605 boolean casCleanMe(QNode cmp, QNode val) {
606 return cleanMe == cmp &&
607 UNSAFE.compareAndSwapObject(this, cleanMeOffset, cmp, val);
608 }
609
610 /**
611 * Puts or takes an item.
612 */
613 Object transfer(Object e, boolean timed, long nanos) {
614 /* Basic algorithm is to loop trying to take either of
615 * two actions:
616 *
617 * 1. If queue apparently empty or holding same-mode nodes,
618 * try to add node to queue of waiters, wait to be
619 * fulfilled (or cancelled) and return matching item.
620 *
621 * 2. If queue apparently contains waiting items, and this
622 * call is of complementary mode, try to fulfill by CAS'ing
623 * item field of waiting node and dequeuing it, and then
624 * returning matching item.
625 *
626 * In each case, along the way, check for and try to help
627 * advance head and tail on behalf of other stalled/slow
628 * threads.
629 *
630 * The loop starts off with a null check guarding against
631 * seeing uninitialized head or tail values. This never
632 * happens in current SynchronousQueue, but could if
633 * callers held non-volatile/final ref to the
634 * transferer. The check is here anyway because it places
635 * null checks at top of loop, which is usually faster
636 * than having them implicitly interspersed.
637 */
638
639 QNode s = null; // constructed/reused as needed
640 boolean isData = (e != null);
641
642 for (;;) {
643 QNode t = tail;
644 QNode h = head;
645 if (t == null || h == null) // saw uninitialized value
646 continue; // spin
647
648 if (h == t || t.isData == isData) { // empty or same-mode
649 QNode tn = t.next;
650 if (t != tail) // inconsistent read
651 continue;
652 if (tn != null) { // lagging tail
653 advanceTail(t, tn);
654 continue;
655 }
656 if (timed && nanos <= 0) // can't wait
657 return null;
658 if (s == null)
659 s = new QNode(e, isData);
660 if (!t.casNext(null, s)) // failed to link in
661 continue;
662
663 advanceTail(t, s); // swing tail and wait
664 Object x = awaitFulfill(s, e, timed, nanos);
665 if (x == s) { // wait was cancelled
666 clean(t, s);
667 return null;
668 }
669
670 if (!s.isOffList()) { // not already unlinked
671 advanceHead(t, s); // unlink if head
672 if (x != null) // and forget fields
673 s.item = s;
674 s.waiter = null;
675 }
676 return (x != null) ? x : e;
677
678 } else { // complementary-mode
679 QNode m = h.next; // node to fulfill
680 if (t != tail || m == null || h != head)
681 continue; // inconsistent read
682
683 Object x = m.item;
684 if (isData == (x != null) || // m already fulfilled
685 x == m || // m cancelled
686 !m.casItem(x, e)) { // lost CAS
687 advanceHead(h, m); // dequeue and retry
688 continue;
689 }
690
691 advanceHead(h, m); // successfully fulfilled
692 LockSupport.unpark(m.waiter);
693 return (x != null) ? x : e;
694 }
695 }
696 }
697
698 /**
699 * Spins/blocks until node s is fulfilled.
700 *
701 * @param s the waiting node
702 * @param e the comparison value for checking match
703 * @param timed true if timed wait
704 * @param nanos timeout value
705 * @return matched item, or s if cancelled
706 */
707 Object awaitFulfill(QNode s, Object e, boolean timed, long nanos) {
708 /* Same idea as TransferStack.awaitFulfill */
709 long lastTime = timed ? System.nanoTime() : 0;
710 Thread w = Thread.currentThread();
711 int spins = ((head.next == s) ?
712 (timed ? maxTimedSpins : maxUntimedSpins) : 0);
713 for (;;) {
714 if (w.isInterrupted())
715 s.tryCancel(e);
716 Object x = s.item;
717 if (x != e)
718 return x;
719 if (timed) {
720 long now = System.nanoTime();
721 nanos -= now - lastTime;
722 lastTime = now;
723 if (nanos <= 0) {
724 s.tryCancel(e);
725 continue;
726 }
727 }
728 if (spins > 0)
729 --spins;
730 else if (s.waiter == null)
731 s.waiter = w;
732 else if (!timed)
733 LockSupport.park(this);
734 else if (nanos > spinForTimeoutThreshold)
735 LockSupport.parkNanos(this, nanos);
736 }
737 }
738
739 /**
740 * Gets rid of cancelled node s with original predecessor pred.
741 */
742 void clean(QNode pred, QNode s) {
743 s.waiter = null; // forget thread
744 /*
745 * At any given time, exactly one node on list cannot be
746 * deleted -- the last inserted node. To accommodate this,
747 * if we cannot delete s, we save its predecessor as
748 * "cleanMe", deleting the previously saved version
749 * first. At least one of node s or the node previously
750 * saved can always be deleted, so this always terminates.
751 */
752 while (pred.next == s) { // Return early if already unlinked
753 QNode h = head;
754 QNode hn = h.next; // Absorb cancelled first node as head
755 if (hn != null && hn.isCancelled()) {
756 advanceHead(h, hn);
757 continue;
758 }
759 QNode t = tail; // Ensure consistent read for tail
760 if (t == h)
761 return;
762 QNode tn = t.next;
763 if (t != tail)
764 continue;
765 if (tn != null) {
766 advanceTail(t, tn);
767 continue;
768 }
769 if (s != t) { // If not tail, try to unsplice
770 QNode sn = s.next;
771 if (sn == s || pred.casNext(s, sn))
772 return;
773 }
774 QNode dp = cleanMe;
775 if (dp != null) { // Try unlinking previous cancelled node
776 QNode d = dp.next;
777 QNode dn;
778 if (d == null || // d is gone or
779 d == dp || // d is off list or
780 !d.isCancelled() || // d not cancelled or
781 (d != t && // d not tail and
782 (dn = d.next) != null && // has successor
783 dn != d && // that is on list
784 dp.casNext(d, dn))) // d unspliced
785 casCleanMe(dp, null);
786 if (dp == pred)
787 return; // s is already saved node
788 } else if (casCleanMe(null, pred))
789 return; // Postpone cleaning s
790 }
791 }
792
793 private static final sun.misc.Unsafe UNSAFE;
794 private static final long headOffset;
795 private static final long tailOffset;
796 private static final long cleanMeOffset;
797 static {
798 try {
799 UNSAFE = sun.misc.Unsafe.getUnsafe();
800 Class<?> k = TransferQueue.class;
801 headOffset = UNSAFE.objectFieldOffset
802 (k.getDeclaredField("head"));
803 tailOffset = UNSAFE.objectFieldOffset
804 (k.getDeclaredField("tail"));
805 cleanMeOffset = UNSAFE.objectFieldOffset
806 (k.getDeclaredField("cleanMe"));
807 } catch (Exception e) {
808 throw new Error(e);
809 }
810 }
811 }
812
813 /**
814 * The transferer. Set only in constructor, but cannot be declared
815 * as final without further complicating serialization. Since
816 * this is accessed only at most once per public method, there
817 * isn't a noticeable performance penalty for using volatile
818 * instead of final here.
819 */
820 private transient volatile Transferer transferer;
821
822 /**
823 * Creates a <tt>SynchronousQueue</tt> with nonfair access policy.
824 */
825 public SynchronousQueue() {
826 this(false);
827 }
828
829 /**
830 * Creates a <tt>SynchronousQueue</tt> with the specified fairness policy.
831 *
832 * @param fair if true, waiting threads contend in FIFO order for
833 * access; otherwise the order is unspecified.
834 */
835 public SynchronousQueue(boolean fair) {
836 transferer = fair ? new TransferQueue() : new TransferStack();
837 }
838
839 /**
840 * Adds the specified element to this queue, waiting if necessary for
841 * another thread to receive it.
842 *
843 * @throws InterruptedException {@inheritDoc}
844 * @throws NullPointerException {@inheritDoc}
845 */
846 public void put(E o) throws InterruptedException {
847 if (o == null) throw new NullPointerException();
848 if (transferer.transfer(o, false, 0) == null) {
849 Thread.interrupted();
850 throw new InterruptedException();
851 }
852 }
853
854 /**
855 * Inserts the specified element into this queue, waiting if necessary
856 * up to the specified wait time for another thread to receive it.
857 *
858 * @return <tt>true</tt> if successful, or <tt>false</tt> if the
859 * specified waiting time elapses before a consumer appears.
860 * @throws InterruptedException {@inheritDoc}
861 * @throws NullPointerException {@inheritDoc}
862 */
863 public boolean offer(E o, long timeout, TimeUnit unit)
864 throws InterruptedException {
865 if (o == null) throw new NullPointerException();
866 if (transferer.transfer(o, true, unit.toNanos(timeout)) != null)
867 return true;
868 if (!Thread.interrupted())
869 return false;
870 throw new InterruptedException();
871 }
872
873 /**
874 * Inserts the specified element into this queue, if another thread is
875 * waiting to receive it.
876 *
877 * @param e the element to add
878 * @return <tt>true</tt> if the element was added to this queue, else
879 * <tt>false</tt>
880 * @throws NullPointerException if the specified element is null
881 */
882 public boolean offer(E e) {
883 if (e == null) throw new NullPointerException();
884 return transferer.transfer(e, true, 0) != null;
885 }
886
887 /**
888 * Retrieves and removes the head of this queue, waiting if necessary
889 * for another thread to insert it.
890 *
891 * @return the head of this queue
892 * @throws InterruptedException {@inheritDoc}
893 */
894 public E take() throws InterruptedException {
895 Object e = transferer.transfer(null, false, 0);
896 if (e != null)
897 return (E)e;
898 Thread.interrupted();
899 throw new InterruptedException();
900 }
901
902 /**
903 * Retrieves and removes the head of this queue, waiting
904 * if necessary up to the specified wait time, for another thread
905 * to insert it.
906 *
907 * @return the head of this queue, or <tt>null</tt> if the
908 * specified waiting time elapses before an element is present.
909 * @throws InterruptedException {@inheritDoc}
910 */
911 public E poll(long timeout, TimeUnit unit) throws InterruptedException {
912 Object e = transferer.transfer(null, true, unit.toNanos(timeout));
913 if (e != null || !Thread.interrupted())
914 return (E)e;
915 throw new InterruptedException();
916 }
917
918 /**
919 * Retrieves and removes the head of this queue, if another thread
920 * is currently making an element available.
921 *
922 * @return the head of this queue, or <tt>null</tt> if no
923 * element is available.
924 */
925 public E poll() {
926 return (E)transferer.transfer(null, true, 0);
927 }
928
929 /**
930 * Always returns <tt>true</tt>.
931 * A <tt>SynchronousQueue</tt> has no internal capacity.
932 *
933 * @return <tt>true</tt>
934 */
935 public boolean isEmpty() {
936 return true;
937 }
938
939 /**
940 * Always returns zero.
941 * A <tt>SynchronousQueue</tt> has no internal capacity.
942 *
943 * @return zero.
944 */
945 public int size() {
946 return 0;
947 }
948
949 /**
950 * Always returns zero.
951 * A <tt>SynchronousQueue</tt> has no internal capacity.
952 *
953 * @return zero.
954 */
955 public int remainingCapacity() {
956 return 0;
957 }
958
959 /**
960 * Does nothing.
961 * A <tt>SynchronousQueue</tt> has no internal capacity.
962 */
963 public void clear() {
964 }
965
966 /**
967 * Always returns <tt>false</tt>.
968 * A <tt>SynchronousQueue</tt> has no internal capacity.
969 *
970 * @param o the element
971 * @return <tt>false</tt>
972 */
973 public boolean contains(Object o) {
974 return false;
975 }
976
977 /**
978 * Always returns <tt>false</tt>.
979 * A <tt>SynchronousQueue</tt> has no internal capacity.
980 *
981 * @param o the element to remove
982 * @return <tt>false</tt>
983 */
984 public boolean remove(Object o) {
985 return false;
986 }
987
988 /**
989 * Returns <tt>false</tt> unless the given collection is empty.
990 * A <tt>SynchronousQueue</tt> has no internal capacity.
991 *
992 * @param c the collection
993 * @return <tt>false</tt> unless given collection is empty
994 */
995 public boolean containsAll(Collection<?> c) {
996 return c.isEmpty();
997 }
998
999 /**
1000 * Always returns <tt>false</tt>.
1001 * A <tt>SynchronousQueue</tt> has no internal capacity.
1002 *
1003 * @param c the collection
1004 * @return <tt>false</tt>
1005 */
1006 public boolean removeAll(Collection<?> c) {
1007 return false;
1008 }
1009
1010 /**
1011 * Always returns <tt>false</tt>.
1012 * A <tt>SynchronousQueue</tt> has no internal capacity.
1013 *
1014 * @param c the collection
1015 * @return <tt>false</tt>
1016 */
1017 public boolean retainAll(Collection<?> c) {
1018 return false;
1019 }
1020
1021 /**
1022 * Always returns <tt>null</tt>.
1023 * A <tt>SynchronousQueue</tt> does not return elements
1024 * unless actively waited on.
1025 *
1026 * @return <tt>null</tt>
1027 */
1028 public E peek() {
1029 return null;
1030 }
1031
1032 /**
1033 * Returns an empty iterator in which <tt>hasNext</tt> always returns
1034 * <tt>false</tt>.
1035 *
1036 * @return an empty iterator
1037 */
1038 @SuppressWarnings("unchecked")
1039 public Iterator<E> iterator() {
1040 return (Iterator<E>) EmptyIterator.EMPTY_ITERATOR;
1041 }
1042
1043 // Replicated from a previous version of Collections
1044 private static class EmptyIterator<E> implements Iterator<E> {
1045 static final EmptyIterator<Object> EMPTY_ITERATOR
1046 = new EmptyIterator<Object>();
1047
1048 public boolean hasNext() { return false; }
1049 public E next() { throw new NoSuchElementException(); }
1050 public void remove() { throw new IllegalStateException(); }
1051 }
1052
1053 /**
1054 * Returns a zero-length array.
1055 * @return a zero-length array
1056 */
1057 public Object[] toArray() {
1058 return new Object[0];
1059 }
1060
1061 /**
1062 * Sets the zeroeth element of the specified array to <tt>null</tt>
1063 * (if the array has non-zero length) and returns it.
1064 *
1065 * @param a the array
1066 * @return the specified array
1067 * @throws NullPointerException if the specified array is null
1068 */
1069 public <T> T[] toArray(T[] a) {
1070 if (a.length > 0)
1071 a[0] = null;
1072 return a;
1073 }
1074
1075 /**
1076 * @throws UnsupportedOperationException {@inheritDoc}
1077 * @throws ClassCastException {@inheritDoc}
1078 * @throws NullPointerException {@inheritDoc}
1079 * @throws IllegalArgumentException {@inheritDoc}
1080 */
1081 public int drainTo(Collection<? super E> c) {
1082 if (c == null)
1083 throw new NullPointerException();
1084 if (c == this)
1085 throw new IllegalArgumentException();
1086 int n = 0;
1087 E e;
1088 while ( (e = poll()) != null) {
1089 c.add(e);
1090 ++n;
1091 }
1092 return n;
1093 }
1094
1095 /**
1096 * @throws UnsupportedOperationException {@inheritDoc}
1097 * @throws ClassCastException {@inheritDoc}
1098 * @throws NullPointerException {@inheritDoc}
1099 * @throws IllegalArgumentException {@inheritDoc}
1100 */
1101 public int drainTo(Collection<? super E> c, int maxElements) {
1102 if (c == null)
1103 throw new NullPointerException();
1104 if (c == this)
1105 throw new IllegalArgumentException();
1106 int n = 0;
1107 E e;
1108 while (n < maxElements && (e = poll()) != null) {
1109 c.add(e);
1110 ++n;
1111 }
1112 return n;
1113 }
1114
1115 /*
1116 * To cope with serialization strategy in the 1.5 version of
1117 * SynchronousQueue, we declare some unused classes and fields
1118 * that exist solely to enable serializability across versions.
1119 * These fields are never used, so are initialized only if this
1120 * object is ever serialized or deserialized.
1121 */
1122
1123 static class WaitQueue implements java.io.Serializable { }
1124 static class LifoWaitQueue extends WaitQueue {
1125 private static final long serialVersionUID = -3633113410248163686L;
1126 }
1127 static class FifoWaitQueue extends WaitQueue {
1128 private static final long serialVersionUID = -3623113410248163686L;
1129 }
1130 private ReentrantLock qlock;
1131 private WaitQueue waitingProducers;
1132 private WaitQueue waitingConsumers;
1133
1134 /**
1135 * Save the state to a stream (that is, serialize it).
1136 *
1137 * @param s the stream
1138 */
1139 private void writeObject(java.io.ObjectOutputStream s)
1140 throws java.io.IOException {
1141 boolean fair = transferer instanceof TransferQueue;
1142 if (fair) {
1143 qlock = new ReentrantLock(true);
1144 waitingProducers = new FifoWaitQueue();
1145 waitingConsumers = new FifoWaitQueue();
1146 }
1147 else {
1148 qlock = new ReentrantLock();
1149 waitingProducers = new LifoWaitQueue();
1150 waitingConsumers = new LifoWaitQueue();
1151 }
1152 s.defaultWriteObject();
1153 }
1154
1155 private void readObject(final java.io.ObjectInputStream s)
1156 throws java.io.IOException, ClassNotFoundException {
1157 s.defaultReadObject();
1158 if (waitingProducers instanceof FifoWaitQueue)
1159 transferer = new TransferQueue();
1160 else
1161 transferer = new TransferStack();
1162 }
1163
1164 // Unsafe mechanics
1165 static long objectFieldOffset(sun.misc.Unsafe UNSAFE,
1166 String field, Class<?> klazz) {
1167 try {
1168 return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
1169 } catch (NoSuchFieldException e) {
1170 // Convert Exception to corresponding Error
1171 NoSuchFieldError error = new NoSuchFieldError(field);
1172 error.initCause(e);
1173 throw error;
1174 }
1175 }
1176
1177 }