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 8package java.util.concurrent; 9 10import java.util.AbstractQueue; 11import java.util.Collection; 12import java.util.Collections; 13import java.util.Iterator; 14import java.util.Spliterator; 15import java.util.Spliterators; 16import java.util.concurrent.locks.LockSupport; 17import java.util.concurrent.locks.ReentrantLock; 18 19// BEGIN android-note 20// removed link to collections framework docs 21// END android-note 22 23/** 24 * A {@linkplain BlockingQueue blocking queue} in which each insert 25 * operation must wait for a corresponding remove operation by another 26 * thread, and vice versa. A synchronous queue does not have any 27 * internal capacity, not even a capacity of one. You cannot 28 * {@code peek} at a synchronous queue because an element is only 29 * present when you try to remove it; you cannot insert an element 30 * (using any method) unless another thread is trying to remove it; 31 * you cannot iterate as there is nothing to iterate. The 32 * <em>head</em> of the queue is the element that the first queued 33 * inserting thread is trying to add to the queue; if there is no such 34 * queued thread then no element is available for removal and 35 * {@code poll()} will return {@code null}. For purposes of other 36 * {@code Collection} methods (for example {@code contains}), a 37 * {@code SynchronousQueue} acts as an empty collection. This queue 38 * does not permit {@code null} elements. 39 * 40 * <p>Synchronous queues are similar to rendezvous channels used in 41 * CSP and Ada. They are well suited for handoff designs, in which an 42 * object running in one thread must sync up with an object running 43 * in another thread in order to hand it some information, event, or 44 * task. 45 * 46 * <p>This class supports an optional fairness policy for ordering 47 * waiting producer and consumer threads. By default, this ordering 48 * is not guaranteed. However, a queue constructed with fairness set 49 * to {@code true} grants threads access in FIFO order. 50 * 51 * <p>This class and its iterator implement all of the 52 * <em>optional</em> methods of the {@link Collection} and {@link 53 * Iterator} interfaces. 54 * 55 * @since 1.5 56 * @author Doug Lea and Bill Scherer and Michael Scott 57 * @param <E> the type of elements held in this queue 58 */ 59public class SynchronousQueue<E> extends AbstractQueue<E> 60 implements BlockingQueue<E>, java.io.Serializable { 61 private static final long serialVersionUID = -3223113410248163686L; 62 63 /* 64 * This class implements extensions of the dual stack and dual 65 * queue algorithms described in "Nonblocking Concurrent Objects 66 * with Condition Synchronization", by W. N. Scherer III and 67 * M. L. Scott. 18th Annual Conf. on Distributed Computing, 68 * Oct. 2004 (see also 69 * http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/duals.html). 70 * The (Lifo) stack is used for non-fair mode, and the (Fifo) 71 * queue for fair mode. The performance of the two is generally 72 * similar. Fifo usually supports higher throughput under 73 * contention but Lifo maintains higher thread locality in common 74 * applications. 75 * 76 * A dual queue (and similarly stack) is one that at any given 77 * time either holds "data" -- items provided by put operations, 78 * or "requests" -- slots representing take operations, or is 79 * empty. A call to "fulfill" (i.e., a call requesting an item 80 * from a queue holding data or vice versa) dequeues a 81 * complementary node. The most interesting feature of these 82 * queues is that any operation can figure out which mode the 83 * queue is in, and act accordingly without needing locks. 84 * 85 * Both the queue and stack extend abstract class Transferer 86 * defining the single method transfer that does a put or a 87 * take. These are unified into a single method because in dual 88 * data structures, the put and take operations are symmetrical, 89 * so nearly all code can be combined. The resulting transfer 90 * methods are on the long side, but are easier to follow than 91 * they would be if broken up into nearly-duplicated parts. 92 * 93 * The queue and stack data structures share many conceptual 94 * similarities but very few concrete details. For simplicity, 95 * they are kept distinct so that they can later evolve 96 * separately. 97 * 98 * The algorithms here differ from the versions in the above paper 99 * in extending them for use in synchronous queues, as well as 100 * dealing with cancellation. The main differences include: 101 * 102 * 1. The original algorithms used bit-marked pointers, but 103 * the ones here use mode bits in nodes, leading to a number 104 * of further adaptations. 105 * 2. SynchronousQueues must block threads waiting to become 106 * fulfilled. 107 * 3. Support for cancellation via timeout and interrupts, 108 * including cleaning out cancelled nodes/threads 109 * from lists to avoid garbage retention and memory depletion. 110 * 111 * Blocking is mainly accomplished using LockSupport park/unpark, 112 * except that nodes that appear to be the next ones to become 113 * fulfilled first spin a bit (on multiprocessors only). On very 114 * busy synchronous queues, spinning can dramatically improve 115 * throughput. And on less busy ones, the amount of spinning is 116 * small enough not to be noticeable. 117 * 118 * Cleaning is done in different ways in queues vs stacks. For 119 * queues, we can almost always remove a node immediately in O(1) 120 * time (modulo retries for consistency checks) when it is 121 * cancelled. But if it may be pinned as the current tail, it must 122 * wait until some subsequent cancellation. For stacks, we need a 123 * potentially O(n) traversal to be sure that we can remove the 124 * node, but this can run concurrently with other threads 125 * accessing the stack. 126 * 127 * While garbage collection takes care of most node reclamation 128 * issues that otherwise complicate nonblocking algorithms, care 129 * is taken to "forget" references to data, other nodes, and 130 * threads that might be held on to long-term by blocked 131 * threads. In cases where setting to null would otherwise 132 * conflict with main algorithms, this is done by changing a 133 * node's link to now point to the node itself. This doesn't arise 134 * much for Stack nodes (because blocked threads do not hang on to 135 * old head pointers), but references in Queue nodes must be 136 * aggressively forgotten to avoid reachability of everything any 137 * node has ever referred to since arrival. 138 */ 139 140 /** 141 * Shared internal API for dual stacks and queues. 142 */ 143 abstract static class Transferer<E> { 144 /** 145 * Performs a put or take. 146 * 147 * @param e if non-null, the item to be handed to a consumer; 148 * if null, requests that transfer return an item 149 * offered by producer. 150 * @param timed if this operation should timeout 151 * @param nanos the timeout, in nanoseconds 152 * @return if non-null, the item provided or received; if null, 153 * the operation failed due to timeout or interrupt -- 154 * the caller can distinguish which of these occurred 155 * by checking Thread.interrupted. 156 */ 157 abstract E transfer(E e, boolean timed, long nanos); 158 } 159 160 /** 161 * The number of times to spin before blocking in timed waits. 162 * The value is empirically derived -- it works well across a 163 * variety of processors and OSes. Empirically, the best value 164 * seems not to vary with number of CPUs (beyond 2) so is just 165 * a constant. 166 */ 167 static final int MAX_TIMED_SPINS = 168 (Runtime.getRuntime().availableProcessors() < 2) ? 0 : 32; 169 170 /** 171 * The number of times to spin before blocking in untimed waits. 172 * This is greater than timed value because untimed waits spin 173 * faster since they don't need to check times on each spin. 174 */ 175 static final int MAX_UNTIMED_SPINS = MAX_TIMED_SPINS * 16; 176 177 /** 178 * The number of nanoseconds for which it is faster to spin 179 * rather than to use timed park. A rough estimate suffices. 180 */ 181 static final long SPIN_FOR_TIMEOUT_THRESHOLD = 1000L; 182 183 /** Dual stack */ 184 static final class TransferStack<E> extends Transferer<E> { 185 /* 186 * This extends Scherer-Scott dual stack algorithm, differing, 187 * among other ways, by using "covering" nodes rather than 188 * bit-marked pointers: Fulfilling operations push on marker 189 * nodes (with FULFILLING bit set in mode) to reserve a spot 190 * to match a waiting node. 191 */ 192 193 /* Modes for SNodes, ORed together in node fields */ 194 /** Node represents an unfulfilled consumer */ 195 static final int REQUEST = 0; 196 /** Node represents an unfulfilled producer */ 197 static final int DATA = 1; 198 /** Node is fulfilling another unfulfilled DATA or REQUEST */ 199 static final int FULFILLING = 2; 200 201 /** Returns true if m has fulfilling bit set. */ 202 static boolean isFulfilling(int m) { return (m & FULFILLING) != 0; } 203 204 /** Node class for TransferStacks. */ 205 static final class SNode { 206 volatile SNode next; // next node in stack 207 volatile SNode match; // the node matched to this 208 volatile Thread waiter; // to control park/unpark 209 Object item; // data; or null for REQUESTs 210 int mode; 211 // Note: item and mode fields don't need to be volatile 212 // since they are always written before, and read after, 213 // other volatile/atomic operations. 214 215 SNode(Object item) { 216 this.item = item; 217 } 218 219 boolean casNext(SNode cmp, SNode val) { 220 return cmp == next && 221 U.compareAndSwapObject(this, NEXT, cmp, val); 222 } 223 224 /** 225 * Tries to match node s to this node, if so, waking up thread. 226 * Fulfillers call tryMatch to identify their waiters. 227 * Waiters block until they have been matched. 228 * 229 * @param s the node to match 230 * @return true if successfully matched to s 231 */ 232 boolean tryMatch(SNode s) { 233 if (match == null && 234 U.compareAndSwapObject(this, MATCH, null, s)) { 235 Thread w = waiter; 236 if (w != null) { // waiters need at most one unpark 237 waiter = null; 238 LockSupport.unpark(w); 239 } 240 return true; 241 } 242 return match == s; 243 } 244 245 /** 246 * Tries to cancel a wait by matching node to itself. 247 */ 248 void tryCancel() { 249 U.compareAndSwapObject(this, MATCH, null, this); 250 } 251 252 boolean isCancelled() { 253 return match == this; 254 } 255 256 // Unsafe mechanics 257 private static final sun.misc.Unsafe U = sun.misc.Unsafe.getUnsafe(); 258 private static final long MATCH; 259 private static final long NEXT; 260 261 static { 262 try { 263 MATCH = U.objectFieldOffset 264 (SNode.class.getDeclaredField("match")); 265 NEXT = U.objectFieldOffset 266 (SNode.class.getDeclaredField("next")); 267 } catch (ReflectiveOperationException e) { 268 throw new Error(e); 269 } 270 } 271 } 272 273 /** The head (top) of the stack */ 274 volatile SNode head; 275 276 boolean casHead(SNode h, SNode nh) { 277 return h == head && 278 U.compareAndSwapObject(this, HEAD, h, nh); 279 } 280 281 /** 282 * Creates or resets fields of a node. Called only from transfer 283 * where the node to push on stack is lazily created and 284 * reused when possible to help reduce intervals between reads 285 * and CASes of head and to avoid surges of garbage when CASes 286 * to push nodes fail due to contention. 287 */ 288 static SNode snode(SNode s, Object e, SNode next, int mode) { 289 if (s == null) s = new SNode(e); 290 s.mode = mode; 291 s.next = next; 292 return s; 293 } 294 295 /** 296 * Puts or takes an item. 297 */ 298 @SuppressWarnings("unchecked") 299 E transfer(E e, boolean timed, long nanos) { 300 /* 301 * Basic algorithm is to loop trying one of three actions: 302 * 303 * 1. If apparently empty or already containing nodes of same 304 * mode, try to push node on stack and wait for a match, 305 * returning it, or null if cancelled. 306 * 307 * 2. If apparently containing node of complementary mode, 308 * try to push a fulfilling node on to stack, match 309 * with corresponding waiting node, pop both from 310 * stack, and return matched item. The matching or 311 * unlinking might not actually be necessary because of 312 * other threads performing action 3: 313 * 314 * 3. If top of stack already holds another fulfilling node, 315 * help it out by doing its match and/or pop 316 * operations, and then continue. The code for helping 317 * is essentially the same as for fulfilling, except 318 * that it doesn't return the item. 319 */ 320 321 SNode s = null; // constructed/reused as needed 322 int mode = (e == null) ? REQUEST : DATA; 323 324 for (;;) { 325 SNode h = head; 326 if (h == null || h.mode == mode) { // empty or same-mode 327 if (timed && nanos <= 0L) { // can't wait 328 if (h != null && h.isCancelled()) 329 casHead(h, h.next); // pop cancelled node 330 else 331 return null; 332 } else if (casHead(h, s = snode(s, e, h, mode))) { 333 SNode m = awaitFulfill(s, timed, nanos); 334 if (m == s) { // wait was cancelled 335 clean(s); 336 return null; 337 } 338 if ((h = head) != null && h.next == s) 339 casHead(h, s.next); // help s's fulfiller 340 return (E) ((mode == REQUEST) ? m.item : s.item); 341 } 342 } else if (!isFulfilling(h.mode)) { // try to fulfill 343 if (h.isCancelled()) // already cancelled 344 casHead(h, h.next); // pop and retry 345 else if (casHead(h, s=snode(s, e, h, FULFILLING|mode))) { 346 for (;;) { // loop until matched or waiters disappear 347 SNode m = s.next; // m is s's match 348 if (m == null) { // all waiters are gone 349 casHead(s, null); // pop fulfill node 350 s = null; // use new node next time 351 break; // restart main loop 352 } 353 SNode mn = m.next; 354 if (m.tryMatch(s)) { 355 casHead(s, mn); // pop both s and m 356 return (E) ((mode == REQUEST) ? m.item : s.item); 357 } else // lost match 358 s.casNext(m, mn); // help unlink 359 } 360 } 361 } else { // help a fulfiller 362 SNode m = h.next; // m is h's match 363 if (m == null) // waiter is gone 364 casHead(h, null); // pop fulfilling node 365 else { 366 SNode mn = m.next; 367 if (m.tryMatch(h)) // help match 368 casHead(h, mn); // pop both h and m 369 else // lost match 370 h.casNext(m, mn); // help unlink 371 } 372 } 373 } 374 } 375 376 /** 377 * Spins/blocks until node s is matched by a fulfill operation. 378 * 379 * @param s the waiting node 380 * @param timed true if timed wait 381 * @param nanos timeout value 382 * @return matched node, or s if cancelled 383 */ 384 SNode awaitFulfill(SNode s, boolean timed, long nanos) { 385 /* 386 * When a node/thread is about to block, it sets its waiter 387 * field and then rechecks state at least one more time 388 * before actually parking, thus covering race vs 389 * fulfiller noticing that waiter is non-null so should be 390 * woken. 391 * 392 * When invoked by nodes that appear at the point of call 393 * to be at the head of the stack, calls to park are 394 * preceded by spins to avoid blocking when producers and 395 * consumers are arriving very close in time. This can 396 * happen enough to bother only on multiprocessors. 397 * 398 * The order of checks for returning out of main loop 399 * reflects fact that interrupts have precedence over 400 * normal returns, which have precedence over 401 * timeouts. (So, on timeout, one last check for match is 402 * done before giving up.) Except that calls from untimed 403 * SynchronousQueue.{poll/offer} don't check interrupts 404 * and don't wait at all, so are trapped in transfer 405 * method rather than calling awaitFulfill. 406 */ 407 final long deadline = timed ? System.nanoTime() + nanos : 0L; 408 Thread w = Thread.currentThread(); 409 int spins = shouldSpin(s) 410 ? (timed ? MAX_TIMED_SPINS : MAX_UNTIMED_SPINS) 411 : 0; 412 for (;;) { 413 if (w.isInterrupted()) 414 s.tryCancel(); 415 SNode m = s.match; 416 if (m != null) 417 return m; 418 if (timed) { 419 nanos = deadline - System.nanoTime(); 420 if (nanos <= 0L) { 421 s.tryCancel(); 422 continue; 423 } 424 } 425 if (spins > 0) 426 spins = shouldSpin(s) ? (spins - 1) : 0; 427 else if (s.waiter == null) 428 s.waiter = w; // establish waiter so can park next iter 429 else if (!timed) 430 LockSupport.park(this); 431 else if (nanos > SPIN_FOR_TIMEOUT_THRESHOLD) 432 LockSupport.parkNanos(this, nanos); 433 } 434 } 435 436 /** 437 * Returns true if node s is at head or there is an active 438 * fulfiller. 439 */ 440 boolean shouldSpin(SNode s) { 441 SNode h = head; 442 return (h == s || h == null || isFulfilling(h.mode)); 443 } 444 445 /** 446 * Unlinks s from the stack. 447 */ 448 void clean(SNode s) { 449 s.item = null; // forget item 450 s.waiter = null; // forget thread 451 452 /* 453 * At worst we may need to traverse entire stack to unlink 454 * s. If there are multiple concurrent calls to clean, we 455 * might not see s if another thread has already removed 456 * it. But we can stop when we see any node known to 457 * follow s. We use s.next unless it too is cancelled, in 458 * which case we try the node one past. We don't check any 459 * further because we don't want to doubly traverse just to 460 * find sentinel. 461 */ 462 463 SNode past = s.next; 464 if (past != null && past.isCancelled()) 465 past = past.next; 466 467 // Absorb cancelled nodes at head 468 SNode p; 469 while ((p = head) != null && p != past && p.isCancelled()) 470 casHead(p, p.next); 471 472 // Unsplice embedded nodes 473 while (p != null && p != past) { 474 SNode n = p.next; 475 if (n != null && n.isCancelled()) 476 p.casNext(n, n.next); 477 else 478 p = n; 479 } 480 } 481 482 // Unsafe mechanics 483 private static final sun.misc.Unsafe U = sun.misc.Unsafe.getUnsafe(); 484 private static final long HEAD; 485 static { 486 try { 487 HEAD = U.objectFieldOffset 488 (TransferStack.class.getDeclaredField("head")); 489 } catch (ReflectiveOperationException e) { 490 throw new Error(e); 491 } 492 } 493 } 494 495 /** Dual Queue */ 496 static final class TransferQueue<E> extends Transferer<E> { 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 U.compareAndSwapObject(this, NEXT, cmp, val); 521 } 522 523 boolean casItem(Object cmp, Object val) { 524 return item == cmp && 525 U.compareAndSwapObject(this, ITEM, cmp, val); 526 } 527 528 /** 529 * Tries to cancel by CAS'ing ref to this as item. 530 */ 531 void tryCancel(Object cmp) { 532 U.compareAndSwapObject(this, ITEM, 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 U = sun.misc.Unsafe.getUnsafe(); 550 private static final long ITEM; 551 private static final long NEXT; 552 553 static { 554 try { 555 ITEM = U.objectFieldOffset 556 (QNode.class.getDeclaredField("item")); 557 NEXT = U.objectFieldOffset 558 (QNode.class.getDeclaredField("next")); 559 } catch (ReflectiveOperationException e) { 560 throw new Error(e); 561 } 562 } 563 } 564 565 /** Head of queue */ 566 transient volatile QNode head; 567 /** Tail of queue */ 568 transient volatile QNode tail; 569 /** 570 * Reference to a cancelled node that might not yet have been 571 * unlinked from queue because it was the last inserted node 572 * when it was cancelled. 573 */ 574 transient volatile QNode cleanMe; 575 576 TransferQueue() { 577 QNode h = new QNode(null, false); // initialize to dummy node. 578 head = h; 579 tail = h; 580 } 581 582 /** 583 * Tries to cas nh as new head; if successful, unlink 584 * old head's next node to avoid garbage retention. 585 */ 586 void advanceHead(QNode h, QNode nh) { 587 if (h == head && 588 U.compareAndSwapObject(this, HEAD, h, nh)) 589 h.next = h; // forget old next 590 } 591 592 /** 593 * Tries to cas nt as new tail. 594 */ 595 void advanceTail(QNode t, QNode nt) { 596 if (tail == t) 597 U.compareAndSwapObject(this, TAIL, t, nt); 598 } 599 600 /** 601 * Tries to CAS cleanMe slot. 602 */ 603 boolean casCleanMe(QNode cmp, QNode val) { 604 return cleanMe == cmp && 605 U.compareAndSwapObject(this, CLEANME, cmp, val); 606 } 607 608 /** 609 * Puts or takes an item. 610 */ 611 @SuppressWarnings("unchecked") 612 E transfer(E e, boolean timed, long nanos) { 613 /* Basic algorithm is to loop trying to take either of 614 * two actions: 615 * 616 * 1. If queue apparently empty or holding same-mode nodes, 617 * try to add node to queue of waiters, wait to be 618 * fulfilled (or cancelled) and return matching item. 619 * 620 * 2. If queue apparently contains waiting items, and this 621 * call is of complementary mode, try to fulfill by CAS'ing 622 * item field of waiting node and dequeuing it, and then 623 * returning matching item. 624 * 625 * In each case, along the way, check for and try to help 626 * advance head and tail on behalf of other stalled/slow 627 * threads. 628 * 629 * The loop starts off with a null check guarding against 630 * seeing uninitialized head or tail values. This never 631 * happens in current SynchronousQueue, but could if 632 * callers held non-volatile/final ref to the 633 * transferer. The check is here anyway because it places 634 * null checks at top of loop, which is usually faster 635 * than having them implicitly interspersed. 636 */ 637 638 QNode s = null; // constructed/reused as needed 639 boolean isData = (e != null); 640 641 for (;;) { 642 QNode t = tail; 643 QNode h = head; 644 if (t == null || h == null) // saw uninitialized value 645 continue; // spin 646 647 if (h == t || t.isData == isData) { // empty or same-mode 648 QNode tn = t.next; 649 if (t != tail) // inconsistent read 650 continue; 651 if (tn != null) { // lagging tail 652 advanceTail(t, tn); 653 continue; 654 } 655 if (timed && nanos <= 0L) // can't wait 656 return null; 657 if (s == null) 658 s = new QNode(e, isData); 659 if (!t.casNext(null, s)) // failed to link in 660 continue; 661 662 advanceTail(t, s); // swing tail and wait 663 Object x = awaitFulfill(s, e, timed, nanos); 664 if (x == s) { // wait was cancelled 665 clean(t, s); 666 return null; 667 } 668 669 if (!s.isOffList()) { // not already unlinked 670 advanceHead(t, s); // unlink if head 671 if (x != null) // and forget fields 672 s.item = s; 673 s.waiter = null; 674 } 675 return (x != null) ? (E)x : e; 676 677 } else { // complementary-mode 678 QNode m = h.next; // node to fulfill 679 if (t != tail || m == null || h != head) 680 continue; // inconsistent read 681 682 Object x = m.item; 683 if (isData == (x != null) || // m already fulfilled 684 x == m || // m cancelled 685 !m.casItem(x, e)) { // lost CAS 686 advanceHead(h, m); // dequeue and retry 687 continue; 688 } 689 690 advanceHead(h, m); // successfully fulfilled 691 LockSupport.unpark(m.waiter); 692 return (x != null) ? (E)x : e; 693 } 694 } 695 } 696 697 /** 698 * Spins/blocks until node s is fulfilled. 699 * 700 * @param s the waiting node 701 * @param e the comparison value for checking match 702 * @param timed true if timed wait 703 * @param nanos timeout value 704 * @return matched item, or s if cancelled 705 */ 706 Object awaitFulfill(QNode s, E e, boolean timed, long nanos) { 707 /* Same idea as TransferStack.awaitFulfill */ 708 final long deadline = timed ? System.nanoTime() + nanos : 0L; 709 Thread w = Thread.currentThread(); 710 int spins = (head.next == s) 711 ? (timed ? MAX_TIMED_SPINS : MAX_UNTIMED_SPINS) 712 : 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 nanos = deadline - System.nanoTime(); 721 if (nanos <= 0L) { 722 s.tryCancel(e); 723 continue; 724 } 725 } 726 if (spins > 0) 727 --spins; 728 else if (s.waiter == null) 729 s.waiter = w; 730 else if (!timed) 731 LockSupport.park(this); 732 else if (nanos > SPIN_FOR_TIMEOUT_THRESHOLD) 733 LockSupport.parkNanos(this, nanos); 734 } 735 } 736 737 /** 738 * Gets rid of cancelled node s with original predecessor pred. 739 */ 740 void clean(QNode pred, QNode s) { 741 s.waiter = null; // forget thread 742 /* 743 * At any given time, exactly one node on list cannot be 744 * deleted -- the last inserted node. To accommodate this, 745 * if we cannot delete s, we save its predecessor as 746 * "cleanMe", deleting the previously saved version 747 * first. At least one of node s or the node previously 748 * saved can always be deleted, so this always terminates. 749 */ 750 while (pred.next == s) { // Return early if already unlinked 751 QNode h = head; 752 QNode hn = h.next; // Absorb cancelled first node as head 753 if (hn != null && hn.isCancelled()) { 754 advanceHead(h, hn); 755 continue; 756 } 757 QNode t = tail; // Ensure consistent read for tail 758 if (t == h) 759 return; 760 QNode tn = t.next; 761 if (t != tail) 762 continue; 763 if (tn != null) { 764 advanceTail(t, tn); 765 continue; 766 } 767 if (s != t) { // If not tail, try to unsplice 768 QNode sn = s.next; 769 if (sn == s || pred.casNext(s, sn)) 770 return; 771 } 772 QNode dp = cleanMe; 773 if (dp != null) { // Try unlinking previous cancelled node 774 QNode d = dp.next; 775 QNode dn; 776 if (d == null || // d is gone or 777 d == dp || // d is off list or 778 !d.isCancelled() || // d not cancelled or 779 (d != t && // d not tail and 780 (dn = d.next) != null && // has successor 781 dn != d && // that is on list 782 dp.casNext(d, dn))) // d unspliced 783 casCleanMe(dp, null); 784 if (dp == pred) 785 return; // s is already saved node 786 } else if (casCleanMe(null, pred)) 787 return; // Postpone cleaning s 788 } 789 } 790 791 private static final sun.misc.Unsafe U = sun.misc.Unsafe.getUnsafe(); 792 private static final long HEAD; 793 private static final long TAIL; 794 private static final long CLEANME; 795 static { 796 try { 797 HEAD = U.objectFieldOffset 798 (TransferQueue.class.getDeclaredField("head")); 799 TAIL = U.objectFieldOffset 800 (TransferQueue.class.getDeclaredField("tail")); 801 CLEANME = U.objectFieldOffset 802 (TransferQueue.class.getDeclaredField("cleanMe")); 803 } catch (ReflectiveOperationException e) { 804 throw new Error(e); 805 } 806 } 807 } 808 809 /** 810 * The transferer. Set only in constructor, but cannot be declared 811 * as final without further complicating serialization. Since 812 * this is accessed only at most once per public method, there 813 * isn't a noticeable performance penalty for using volatile 814 * instead of final here. 815 */ 816 private transient volatile Transferer<E> transferer; 817 818 /** 819 * Creates a {@code SynchronousQueue} with nonfair access policy. 820 */ 821 public SynchronousQueue() { 822 this(false); 823 } 824 825 /** 826 * Creates a {@code SynchronousQueue} with the specified fairness policy. 827 * 828 * @param fair if true, waiting threads contend in FIFO order for 829 * access; otherwise the order is unspecified. 830 */ 831 public SynchronousQueue(boolean fair) { 832 transferer = fair ? new TransferQueue<E>() : new TransferStack<E>(); 833 } 834 835 /** 836 * Adds the specified element to this queue, waiting if necessary for 837 * another thread to receive it. 838 * 839 * @throws InterruptedException {@inheritDoc} 840 * @throws NullPointerException {@inheritDoc} 841 */ 842 public void put(E e) throws InterruptedException { 843 if (e == null) throw new NullPointerException(); 844 if (transferer.transfer(e, false, 0) == null) { 845 Thread.interrupted(); 846 throw new InterruptedException(); 847 } 848 } 849 850 /** 851 * Inserts the specified element into this queue, waiting if necessary 852 * up to the specified wait time for another thread to receive it. 853 * 854 * @return {@code true} if successful, or {@code false} if the 855 * specified waiting time elapses before a consumer appears 856 * @throws InterruptedException {@inheritDoc} 857 * @throws NullPointerException {@inheritDoc} 858 */ 859 public boolean offer(E e, long timeout, TimeUnit unit) 860 throws InterruptedException { 861 if (e == null) throw new NullPointerException(); 862 if (transferer.transfer(e, true, unit.toNanos(timeout)) != null) 863 return true; 864 if (!Thread.interrupted()) 865 return false; 866 throw new InterruptedException(); 867 } 868 869 /** 870 * Inserts the specified element into this queue, if another thread is 871 * waiting to receive it. 872 * 873 * @param e the element to add 874 * @return {@code true} if the element was added to this queue, else 875 * {@code false} 876 * @throws NullPointerException if the specified element is null 877 */ 878 public boolean offer(E e) { 879 if (e == null) throw new NullPointerException(); 880 return transferer.transfer(e, true, 0) != null; 881 } 882 883 /** 884 * Retrieves and removes the head of this queue, waiting if necessary 885 * for another thread to insert it. 886 * 887 * @return the head of this queue 888 * @throws InterruptedException {@inheritDoc} 889 */ 890 public E take() throws InterruptedException { 891 E e = transferer.transfer(null, false, 0); 892 if (e != null) 893 return e; 894 Thread.interrupted(); 895 throw new InterruptedException(); 896 } 897 898 /** 899 * Retrieves and removes the head of this queue, waiting 900 * if necessary up to the specified wait time, for another thread 901 * to insert it. 902 * 903 * @return the head of this queue, or {@code null} if the 904 * specified waiting time elapses before an element is present 905 * @throws InterruptedException {@inheritDoc} 906 */ 907 public E poll(long timeout, TimeUnit unit) throws InterruptedException { 908 E e = transferer.transfer(null, true, unit.toNanos(timeout)); 909 if (e != null || !Thread.interrupted()) 910 return e; 911 throw new InterruptedException(); 912 } 913 914 /** 915 * Retrieves and removes the head of this queue, if another thread 916 * is currently making an element available. 917 * 918 * @return the head of this queue, or {@code null} if no 919 * element is available 920 */ 921 public E poll() { 922 return transferer.transfer(null, true, 0); 923 } 924 925 /** 926 * Always returns {@code true}. 927 * A {@code SynchronousQueue} has no internal capacity. 928 * 929 * @return {@code true} 930 */ 931 public boolean isEmpty() { 932 return true; 933 } 934 935 /** 936 * Always returns zero. 937 * A {@code SynchronousQueue} has no internal capacity. 938 * 939 * @return zero 940 */ 941 public int size() { 942 return 0; 943 } 944 945 /** 946 * Always returns zero. 947 * A {@code SynchronousQueue} has no internal capacity. 948 * 949 * @return zero 950 */ 951 public int remainingCapacity() { 952 return 0; 953 } 954 955 /** 956 * Does nothing. 957 * A {@code SynchronousQueue} has no internal capacity. 958 */ 959 public void clear() { 960 } 961 962 /** 963 * Always returns {@code false}. 964 * A {@code SynchronousQueue} has no internal capacity. 965 * 966 * @param o the element 967 * @return {@code false} 968 */ 969 public boolean contains(Object o) { 970 return false; 971 } 972 973 /** 974 * Always returns {@code false}. 975 * A {@code SynchronousQueue} has no internal capacity. 976 * 977 * @param o the element to remove 978 * @return {@code false} 979 */ 980 public boolean remove(Object o) { 981 return false; 982 } 983 984 /** 985 * Returns {@code false} unless the given collection is empty. 986 * A {@code SynchronousQueue} has no internal capacity. 987 * 988 * @param c the collection 989 * @return {@code false} unless given collection is empty 990 */ 991 public boolean containsAll(Collection<?> c) { 992 return c.isEmpty(); 993 } 994 995 /** 996 * Always returns {@code false}. 997 * A {@code SynchronousQueue} has no internal capacity. 998 * 999 * @param c the collection 1000 * @return {@code false} 1001 */ 1002 public boolean removeAll(Collection<?> c) { 1003 return false; 1004 } 1005 1006 /** 1007 * Always returns {@code false}. 1008 * A {@code SynchronousQueue} has no internal capacity. 1009 * 1010 * @param c the collection 1011 * @return {@code false} 1012 */ 1013 public boolean retainAll(Collection<?> c) { 1014 return false; 1015 } 1016 1017 /** 1018 * Always returns {@code null}. 1019 * A {@code SynchronousQueue} does not return elements 1020 * unless actively waited on. 1021 * 1022 * @return {@code null} 1023 */ 1024 public E peek() { 1025 return null; 1026 } 1027 1028 /** 1029 * Returns an empty iterator in which {@code hasNext} always returns 1030 * {@code false}. 1031 * 1032 * @return an empty iterator 1033 */ 1034 public Iterator<E> iterator() { 1035 return Collections.emptyIterator(); 1036 } 1037 1038 /** 1039 * Returns an empty spliterator in which calls to 1040 * {@link java.util.Spliterator#trySplit()} always return {@code null}. 1041 * 1042 * @return an empty spliterator 1043 * @since 1.8 1044 */ 1045 public Spliterator<E> spliterator() { 1046 return Spliterators.emptySpliterator(); 1047 } 1048 1049 /** 1050 * Returns a zero-length array. 1051 * @return a zero-length array 1052 */ 1053 public Object[] toArray() { 1054 return new Object[0]; 1055 } 1056 1057 /** 1058 * Sets the zeroth element of the specified array to {@code null} 1059 * (if the array has non-zero length) and returns it. 1060 * 1061 * @param a the array 1062 * @return the specified array 1063 * @throws NullPointerException if the specified array is null 1064 */ 1065 public <T> T[] toArray(T[] a) { 1066 if (a.length > 0) 1067 a[0] = null; 1068 return a; 1069 } 1070 1071 /** 1072 * Always returns {@code "[]"}. 1073 * @return {@code "[]"} 1074 */ 1075 public String toString() { 1076 return "[]"; 1077 } 1078 1079 /** 1080 * @throws UnsupportedOperationException {@inheritDoc} 1081 * @throws ClassCastException {@inheritDoc} 1082 * @throws NullPointerException {@inheritDoc} 1083 * @throws IllegalArgumentException {@inheritDoc} 1084 */ 1085 public int drainTo(Collection<? super E> c) { 1086 if (c == null) 1087 throw new NullPointerException(); 1088 if (c == this) 1089 throw new IllegalArgumentException(); 1090 int n = 0; 1091 for (E e; (e = poll()) != null;) { 1092 c.add(e); 1093 ++n; 1094 } 1095 return n; 1096 } 1097 1098 /** 1099 * @throws UnsupportedOperationException {@inheritDoc} 1100 * @throws ClassCastException {@inheritDoc} 1101 * @throws NullPointerException {@inheritDoc} 1102 * @throws IllegalArgumentException {@inheritDoc} 1103 */ 1104 public int drainTo(Collection<? super E> c, int maxElements) { 1105 if (c == null) 1106 throw new NullPointerException(); 1107 if (c == this) 1108 throw new IllegalArgumentException(); 1109 int n = 0; 1110 for (E e; n < maxElements && (e = poll()) != null;) { 1111 c.add(e); 1112 ++n; 1113 } 1114 return n; 1115 } 1116 1117 /* 1118 * To cope with serialization strategy in the 1.5 version of 1119 * SynchronousQueue, we declare some unused classes and fields 1120 * that exist solely to enable serializability across versions. 1121 * These fields are never used, so are initialized only if this 1122 * object is ever serialized or deserialized. 1123 */ 1124 1125 @SuppressWarnings("serial") 1126 static class WaitQueue implements java.io.Serializable { } 1127 static class LifoWaitQueue extends WaitQueue { 1128 private static final long serialVersionUID = -3633113410248163686L; 1129 } 1130 static class FifoWaitQueue extends WaitQueue { 1131 private static final long serialVersionUID = -3623113410248163686L; 1132 } 1133 private ReentrantLock qlock; 1134 private WaitQueue waitingProducers; 1135 private WaitQueue waitingConsumers; 1136 1137 /** 1138 * Saves this queue to a stream (that is, serializes it). 1139 * @param s the stream 1140 * @throws java.io.IOException if an I/O error occurs 1141 */ 1142 private void writeObject(java.io.ObjectOutputStream s) 1143 throws java.io.IOException { 1144 boolean fair = transferer instanceof TransferQueue; 1145 if (fair) { 1146 qlock = new ReentrantLock(true); 1147 waitingProducers = new FifoWaitQueue(); 1148 waitingConsumers = new FifoWaitQueue(); 1149 } 1150 else { 1151 qlock = new ReentrantLock(); 1152 waitingProducers = new LifoWaitQueue(); 1153 waitingConsumers = new LifoWaitQueue(); 1154 } 1155 s.defaultWriteObject(); 1156 } 1157 1158 /** 1159 * Reconstitutes this queue from a stream (that is, deserializes it). 1160 * @param s the stream 1161 * @throws ClassNotFoundException if the class of a serialized object 1162 * could not be found 1163 * @throws java.io.IOException if an I/O error occurs 1164 */ 1165 private void readObject(java.io.ObjectInputStream s) 1166 throws java.io.IOException, ClassNotFoundException { 1167 s.defaultReadObject(); 1168 if (waitingProducers instanceof FifoWaitQueue) 1169 transferer = new TransferQueue<E>(); 1170 else 1171 transferer = new TransferStack<E>(); 1172 } 1173 1174 static { 1175 // Reduce the risk of rare disastrous classloading in first call to 1176 // LockSupport.park: https://bugs.openjdk.java.net/browse/JDK-8074773 1177 Class<?> ensureLoaded = LockSupport.class; 1178 } 1179} 1180