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

Comparing jsr166/src/jsr166y/LinkedTransferQueue.java (file contents):
Revision 1.41 by jsr166, Sat Aug 1 20:44:05 2009 UTC vs.
Revision 1.71 by jsr166, Mon Nov 16 01:02:49 2009 UTC

# Line 15 | Line 15 | import java.util.Iterator;
15   import java.util.NoSuchElementException;
16   import java.util.Queue;
17   import java.util.concurrent.locks.LockSupport;
18 import java.util.concurrent.atomic.AtomicReference;
19
18   /**
19 < * An unbounded {@linkplain TransferQueue} based on linked nodes.
19 > * An unbounded {@link TransferQueue} based on linked nodes.
20   * This queue orders elements FIFO (first-in-first-out) with respect
21   * to any given producer.  The <em>head</em> of the queue is that
22   * element that has been on the queue the longest time for some
# Line 54 | Line 52 | public class LinkedTransferQueue<E> exte
52      private static final long serialVersionUID = -3223113410248163686L;
53  
54      /*
55 <     * This class extends the approach used in FIFO-mode
58 <     * SynchronousQueues. See the internal documentation, as well as
59 <     * the PPoPP 2006 paper "Scalable Synchronous Queues" by Scherer,
60 <     * Lea & Scott
61 <     * (http://www.cs.rice.edu/~wns1/papers/2006-PPoPP-SQ.pdf)
55 >     * *** Overview of Dual Queues with Slack ***
56       *
57 <     * The main extension is to provide different Wait modes for the
58 <     * main "xfer" method that puts or takes items.  These don't
59 <     * impact the basic dual-queue logic, but instead control whether
60 <     * or how threads block upon insertion of request or data nodes
61 <     * into the dual queue. It also uses slightly different
62 <     * conventions for tracking whether nodes are off-list or
63 <     * cancelled.
64 <     */
65 <
66 <    // Wait modes for xfer method
67 <    static final int NOWAIT  = 0;
68 <    static final int TIMEOUT = 1;
69 <    static final int WAIT    = 2;
70 <
71 <    /** The number of CPUs, for spin control */
72 <    static final int NCPUS = Runtime.getRuntime().availableProcessors();
73 <
74 <    /**
75 <     * The number of times to spin before blocking in timed waits.
76 <     * The value is empirically derived -- it works well across a
77 <     * variety of processors and OSes. Empirically, the best value
78 <     * seems not to vary with number of CPUs (beyond 2) so is just
79 <     * a constant.
80 <     */
81 <    static final int maxTimedSpins = (NCPUS < 2) ? 0 : 32;
57 >     * Dual Queues, introduced by Scherer and Scott
58 >     * (http://www.cs.rice.edu/~wns1/papers/2004-DISC-DDS.pdf) are
59 >     * (linked) queues in which nodes may represent either data or
60 >     * requests.  When a thread tries to enqueue a data node, but
61 >     * encounters a request node, it instead "matches" and removes it;
62 >     * and vice versa for enqueuing requests. Blocking Dual Queues
63 >     * arrange that threads enqueuing unmatched requests block until
64 >     * other threads provide the match. Dual Synchronous Queues (see
65 >     * Scherer, Lea, & Scott
66 >     * http://www.cs.rochester.edu/u/scott/papers/2009_Scherer_CACM_SSQ.pdf)
67 >     * additionally arrange that threads enqueuing unmatched data also
68 >     * block.  Dual Transfer Queues support all of these modes, as
69 >     * dictated by callers.
70 >     *
71 >     * A FIFO dual queue may be implemented using a variation of the
72 >     * Michael & Scott (M&S) lock-free queue algorithm
73 >     * (http://www.cs.rochester.edu/u/scott/papers/1996_PODC_queues.pdf).
74 >     * It maintains two pointer fields, "head", pointing to a
75 >     * (matched) node that in turn points to the first actual
76 >     * (unmatched) queue node (or null if empty); and "tail" that
77 >     * points to the last node on the queue (or again null if
78 >     * empty). For example, here is a possible queue with four data
79 >     * elements:
80 >     *
81 >     *  head                tail
82 >     *    |                   |
83 >     *    v                   v
84 >     *    M -> U -> U -> U -> U
85 >     *
86 >     * The M&S queue algorithm is known to be prone to scalability and
87 >     * overhead limitations when maintaining (via CAS) these head and
88 >     * tail pointers. This has led to the development of
89 >     * contention-reducing variants such as elimination arrays (see
90 >     * Moir et al http://portal.acm.org/citation.cfm?id=1074013) and
91 >     * optimistic back pointers (see Ladan-Mozes & Shavit
92 >     * http://people.csail.mit.edu/edya/publications/OptimisticFIFOQueue-journal.pdf).
93 >     * However, the nature of dual queues enables a simpler tactic for
94 >     * improving M&S-style implementations when dual-ness is needed.
95 >     *
96 >     * In a dual queue, each node must atomically maintain its match
97 >     * status. While there are other possible variants, we implement
98 >     * this here as: for a data-mode node, matching entails CASing an
99 >     * "item" field from a non-null data value to null upon match, and
100 >     * vice-versa for request nodes, CASing from null to a data
101 >     * value. (Note that the linearization properties of this style of
102 >     * queue are easy to verify -- elements are made available by
103 >     * linking, and unavailable by matching.) Compared to plain M&S
104 >     * queues, this property of dual queues requires one additional
105 >     * successful atomic operation per enq/deq pair. But it also
106 >     * enables lower cost variants of queue maintenance mechanics. (A
107 >     * variation of this idea applies even for non-dual queues that
108 >     * support deletion of interior elements, such as
109 >     * j.u.c.ConcurrentLinkedQueue.)
110 >     *
111 >     * Once a node is matched, its match status can never again
112 >     * change.  We may thus arrange that the linked list of them
113 >     * contain a prefix of zero or more matched nodes, followed by a
114 >     * suffix of zero or more unmatched nodes. (Note that we allow
115 >     * both the prefix and suffix to be zero length, which in turn
116 >     * means that we do not use a dummy header.)  If we were not
117 >     * concerned with either time or space efficiency, we could
118 >     * correctly perform enqueue and dequeue operations by traversing
119 >     * from a pointer to the initial node; CASing the item of the
120 >     * first unmatched node on match and CASing the next field of the
121 >     * trailing node on appends. (Plus some special-casing when
122 >     * initially empty).  While this would be a terrible idea in
123 >     * itself, it does have the benefit of not requiring ANY atomic
124 >     * updates on head/tail fields.
125 >     *
126 >     * We introduce here an approach that lies between the extremes of
127 >     * never versus always updating queue (head and tail) pointers.
128 >     * This offers a tradeoff between sometimes requiring extra
129 >     * traversal steps to locate the first and/or last unmatched
130 >     * nodes, versus the reduced overhead and contention of fewer
131 >     * updates to queue pointers. For example, a possible snapshot of
132 >     * a queue is:
133 >     *
134 >     *  head           tail
135 >     *    |              |
136 >     *    v              v
137 >     *    M -> M -> U -> U -> U -> U
138 >     *
139 >     * The best value for this "slack" (the targeted maximum distance
140 >     * between the value of "head" and the first unmatched node, and
141 >     * similarly for "tail") is an empirical matter. We have found
142 >     * that using very small constants in the range of 1-3 work best
143 >     * over a range of platforms. Larger values introduce increasing
144 >     * costs of cache misses and risks of long traversal chains, while
145 >     * smaller values increase CAS contention and overhead.
146 >     *
147 >     * Dual queues with slack differ from plain M&S dual queues by
148 >     * virtue of only sometimes updating head or tail pointers when
149 >     * matching, appending, or even traversing nodes; in order to
150 >     * maintain a targeted slack.  The idea of "sometimes" may be
151 >     * operationalized in several ways. The simplest is to use a
152 >     * per-operation counter incremented on each traversal step, and
153 >     * to try (via CAS) to update the associated queue pointer
154 >     * whenever the count exceeds a threshold. Another, that requires
155 >     * more overhead, is to use random number generators to update
156 >     * with a given probability per traversal step.
157 >     *
158 >     * In any strategy along these lines, because CASes updating
159 >     * fields may fail, the actual slack may exceed targeted
160 >     * slack. However, they may be retried at any time to maintain
161 >     * targets.  Even when using very small slack values, this
162 >     * approach works well for dual queues because it allows all
163 >     * operations up to the point of matching or appending an item
164 >     * (hence potentially allowing progress by another thread) to be
165 >     * read-only, thus not introducing any further contention. As
166 >     * described below, we implement this by performing slack
167 >     * maintenance retries only after these points.
168 >     *
169 >     * As an accompaniment to such techniques, traversal overhead can
170 >     * be further reduced without increasing contention of head
171 >     * pointer updates: Threads may sometimes shortcut the "next" link
172 >     * path from the current "head" node to be closer to the currently
173 >     * known first unmatched node, and similarly for tail. Again, this
174 >     * may be triggered with using thresholds or randomization.
175 >     *
176 >     * These ideas must be further extended to avoid unbounded amounts
177 >     * of costly-to-reclaim garbage caused by the sequential "next"
178 >     * links of nodes starting at old forgotten head nodes: As first
179 >     * described in detail by Boehm
180 >     * (http://portal.acm.org/citation.cfm?doid=503272.503282) if a GC
181 >     * delays noticing that any arbitrarily old node has become
182 >     * garbage, all newer dead nodes will also be unreclaimed.
183 >     * (Similar issues arise in non-GC environments.)  To cope with
184 >     * this in our implementation, upon CASing to advance the head
185 >     * pointer, we set the "next" link of the previous head to point
186 >     * only to itself; thus limiting the length of connected dead lists.
187 >     * (We also take similar care to wipe out possibly garbage
188 >     * retaining values held in other Node fields.)  However, doing so
189 >     * adds some further complexity to traversal: If any "next"
190 >     * pointer links to itself, it indicates that the current thread
191 >     * has lagged behind a head-update, and so the traversal must
192 >     * continue from the "head".  Traversals trying to find the
193 >     * current tail starting from "tail" may also encounter
194 >     * self-links, in which case they also continue at "head".
195 >     *
196 >     * It is tempting in slack-based scheme to not even use CAS for
197 >     * updates (similarly to Ladan-Mozes & Shavit). However, this
198 >     * cannot be done for head updates under the above link-forgetting
199 >     * mechanics because an update may leave head at a detached node.
200 >     * And while direct writes are possible for tail updates, they
201 >     * increase the risk of long retraversals, and hence long garbage
202 >     * chains, which can be much more costly than is worthwhile
203 >     * considering that the cost difference of performing a CAS vs
204 >     * write is smaller when they are not triggered on each operation
205 >     * (especially considering that writes and CASes equally require
206 >     * additional GC bookkeeping ("write barriers") that are sometimes
207 >     * more costly than the writes themselves because of contention).
208 >     *
209 >     * *** Overview of implementation ***
210 >     *
211 >     * We use a threshold-based approach to updates, with a slack
212 >     * threshold of two -- that is, we update head/tail when the
213 >     * current pointer appears to be two or more steps away from the
214 >     * first/last node. The slack value is hard-wired: a path greater
215 >     * than one is naturally implemented by checking equality of
216 >     * traversal pointers except when the list has only one element,
217 >     * in which case we keep slack threshold at one. Avoiding tracking
218 >     * explicit counts across method calls slightly simplifies an
219 >     * already-messy implementation. Using randomization would
220 >     * probably work better if there were a low-quality dirt-cheap
221 >     * per-thread one available, but even ThreadLocalRandom is too
222 >     * heavy for these purposes.
223 >     *
224 >     * With such a small slack threshold value, it is not worthwhile
225 >     * to augment this with path short-circuiting (i.e., unsplicing
226 >     * interior nodes) except in the case of cancellation/removal (see
227 >     * below).
228 >     *
229 >     * We allow both the head and tail fields to be null before any
230 >     * nodes are enqueued; initializing upon first append.  This
231 >     * simplifies some other logic, as well as providing more
232 >     * efficient explicit control paths instead of letting JVMs insert
233 >     * implicit NullPointerExceptions when they are null.  While not
234 >     * currently fully implemented, we also leave open the possibility
235 >     * of re-nulling these fields when empty (which is complicated to
236 >     * arrange, for little benefit.)
237 >     *
238 >     * All enqueue/dequeue operations are handled by the single method
239 >     * "xfer" with parameters indicating whether to act as some form
240 >     * of offer, put, poll, take, or transfer (each possibly with
241 >     * timeout). The relative complexity of using one monolithic
242 >     * method outweighs the code bulk and maintenance problems of
243 >     * using separate methods for each case.
244 >     *
245 >     * Operation consists of up to three phases. The first is
246 >     * implemented within method xfer, the second in tryAppend, and
247 >     * the third in method awaitMatch.
248 >     *
249 >     * 1. Try to match an existing node
250 >     *
251 >     *    Starting at head, skip already-matched nodes until finding
252 >     *    an unmatched node of opposite mode, if one exists, in which
253 >     *    case matching it and returning, also if necessary updating
254 >     *    head to one past the matched node (or the node itself if the
255 >     *    list has no other unmatched nodes). If the CAS misses, then
256 >     *    a loop retries advancing head by two steps until either
257 >     *    success or the slack is at most two. By requiring that each
258 >     *    attempt advances head by two (if applicable), we ensure that
259 >     *    the slack does not grow without bound. Traversals also check
260 >     *    if the initial head is now off-list, in which case they
261 >     *    start at the new head.
262 >     *
263 >     *    If no candidates are found and the call was untimed
264 >     *    poll/offer, (argument "how" is NOW) return.
265 >     *
266 >     * 2. Try to append a new node (method tryAppend)
267 >     *
268 >     *    Starting at current tail pointer, find the actual last node
269 >     *    and try to append a new node (or if head was null, establish
270 >     *    the first node). Nodes can be appended only if their
271 >     *    predecessors are either already matched or are of the same
272 >     *    mode. If we detect otherwise, then a new node with opposite
273 >     *    mode must have been appended during traversal, so we must
274 >     *    restart at phase 1. The traversal and update steps are
275 >     *    otherwise similar to phase 1: Retrying upon CAS misses and
276 >     *    checking for staleness.  In particular, if a self-link is
277 >     *    encountered, then we can safely jump to a node on the list
278 >     *    by continuing the traversal at current head.
279 >     *
280 >     *    On successful append, if the call was ASYNC, return.
281 >     *
282 >     * 3. Await match or cancellation (method awaitMatch)
283 >     *
284 >     *    Wait for another thread to match node; instead cancelling if
285 >     *    the current thread was interrupted or the wait timed out. On
286 >     *    multiprocessors, we use front-of-queue spinning: If a node
287 >     *    appears to be the first unmatched node in the queue, it
288 >     *    spins a bit before blocking. In either case, before blocking
289 >     *    it tries to unsplice any nodes between the current "head"
290 >     *    and the first unmatched node.
291 >     *
292 >     *    Front-of-queue spinning vastly improves performance of
293 >     *    heavily contended queues. And so long as it is relatively
294 >     *    brief and "quiet", spinning does not much impact performance
295 >     *    of less-contended queues.  During spins threads check their
296 >     *    interrupt status and generate a thread-local random number
297 >     *    to decide to occasionally perform a Thread.yield. While
298 >     *    yield has underdefined specs, we assume that might it help,
299 >     *    and will not hurt in limiting impact of spinning on busy
300 >     *    systems.  We also use smaller (1/2) spins for nodes that are
301 >     *    not known to be front but whose predecessors have not
302 >     *    blocked -- these "chained" spins avoid artifacts of
303 >     *    front-of-queue rules which otherwise lead to alternating
304 >     *    nodes spinning vs blocking. Further, front threads that
305 >     *    represent phase changes (from data to request node or vice
306 >     *    versa) compared to their predecessors receive additional
307 >     *    chained spins, reflecting longer paths typically required to
308 >     *    unblock threads during phase changes.
309 >     *
310 >     *
311 >     * ** Unlinking removed interior nodes **
312 >     *
313 >     * In addition to minimizing garbage retention via self-linking
314 >     * described above, we also unlink removed interior nodes. These
315 >     * may arise due to timed out or interrupted waits, or calls to
316 >     * remove(x) or Iterator.remove.  Normally, given a node that was
317 >     * at one time known to be the predecessor of some node s that is
318 >     * to be removed, we can unsplice s by CASing the next field of
319 >     * its predecessor if it still points to s (otherwise s must
320 >     * already have been removed or is now offlist). But there are two
321 >     * situations in which we cannot guarantee to make node s
322 >     * unreachable in this way: (1) If s is the trailing node of list
323 >     * (i.e., with null next), then it is pinned as the target node
324 >     * for appends, so can only be removed later when other nodes are
325 >     * appended. (2) We cannot necessarily unlink s given a
326 >     * predecessor node that is matched (including the case of being
327 >     * cancelled): the predecessor may already be unspliced, in which
328 >     * case some previous reachable node may still point to s.
329 >     * (For further explanation see Herlihy & Shavit "The Art of
330 >     * Multiprocessor Programming" chapter 9).  Although, in both
331 >     * cases, we can rule out the need for further action if either s
332 >     * or its predecessor are (or can be made to be) at, or fall off
333 >     * from, the head of list.
334 >     *
335 >     * Without taking these into account, it would be possible for an
336 >     * unbounded number of supposedly removed nodes to remain
337 >     * reachable.  Situations leading to such buildup are uncommon but
338 >     * can occur in practice; for example when a series of short timed
339 >     * calls to poll repeatedly time out but never otherwise fall off
340 >     * the list because of an untimed call to take at the front of the
341 >     * queue.
342 >     *
343 >     * When these cases arise, rather than always retraversing the
344 >     * entire list to find an actual predecessor to unlink (which
345 >     * won't help for case (1) anyway), we record a conservative
346 >     * estimate of possible unsplice failures (in "sweepVotes").  We
347 >     * trigger a full sweep when the estimate exceeds a threshold
348 >     * indicating the maximum number of estimated removal failures to
349 >     * tolerate before sweeping through, unlinking cancelled nodes
350 >     * that were not unlinked upon initial removal. We perform sweeps
351 >     * by the thread hitting threshold (rather than background threads
352 >     * or by spreading work to other threads) because in the main
353 >     * contexts in which removal occurs, the caller is already
354 >     * timed-out, cancelled, or performing a potentially O(n)
355 >     * operation (i.e., remove(x)), none of which are time-critical
356 >     * enough to warrant the overhead that alternatives would impose
357 >     * on other threads.
358 >     *
359 >     * Because the sweepVotes estimate is conservative, and because
360 >     * nodes become unlinked "naturally" as they fall off the head of
361 >     * the queue, and because we allow votes to accumulate even while
362 >     * sweeps are in progress, there are typically significantly fewer
363 >     * such nodes than estimated.  Choice of a threshold value
364 >     * balances the likelihood of wasted effort and contention, versus
365 >     * providing a worst-case bound on retention of interior nodes in
366 >     * quiescent queues. The value defined below was chosen
367 >     * empirically to balance these under various timeout scenarios.
368 >     *
369 >     * Note that we cannot self-link unlinked interior nodes during
370 >     * sweeps. However, the associated garbage chains terminate when
371 >     * some successor ultimately falls off the head of the list and is
372 >     * self-linked.
373 >     */
374 >
375 >    /** True if on multiprocessor */
376 >    private static final boolean MP =
377 >        Runtime.getRuntime().availableProcessors() > 1;
378 >
379 >    /**
380 >     * The number of times to spin (with randomly interspersed calls
381 >     * to Thread.yield) on multiprocessor before blocking when a node
382 >     * is apparently the first waiter in the queue.  See above for
383 >     * explanation. Must be a power of two. The value is empirically
384 >     * derived -- it works pretty well across a variety of processors,
385 >     * numbers of CPUs, and OSes.
386 >     */
387 >    private static final int FRONT_SPINS   = 1 << 7;
388 >
389 >    /**
390 >     * The number of times to spin before blocking when a node is
391 >     * preceded by another node that is apparently spinning.  Also
392 >     * serves as an increment to FRONT_SPINS on phase changes, and as
393 >     * base average frequency for yielding during spins. Must be a
394 >     * power of two.
395 >     */
396 >    private static final int CHAINED_SPINS = FRONT_SPINS >>> 1;
397 >
398 >    /**
399 >     * The maximum number of estimated removal failures (sweepVotes)
400 >     * to tolerate before sweeping through the queue unlinking
401 >     * cancelled nodes that were not unlinked upon initial
402 >     * removal. See above for explanation. The value must be at least
403 >     * two to avoid useless sweeps when removing trailing nodes.
404 >     */
405 >    static final int SWEEP_THRESHOLD = 32;
406 >
407 >    /**
408 >     * Queue nodes. Uses Object, not E, for items to allow forgetting
409 >     * them after use.  Relies heavily on Unsafe mechanics to minimize
410 >     * unnecessary ordering constraints: Writes that are intrinsically
411 >     * ordered wrt other accesses or CASes use simple relaxed forms.
412 >     */
413 >    static final class Node {
414 >        final boolean isData;   // false if this is a request node
415 >        volatile Object item;   // initially non-null if isData; CASed to match
416 >        volatile Node next;
417 >        volatile Thread waiter; // null until waiting
418  
419 <    /**
420 <     * The number of times to spin before blocking in untimed waits.
421 <     * This is greater than timed value because untimed waits spin
422 <     * faster since they don't need to check times on each spin.
93 <     */
94 <    static final int maxUntimedSpins = maxTimedSpins * 16;
95 <
96 <    /**
97 <     * The number of nanoseconds for which it is faster to spin
98 <     * rather than to use timed park. A rough estimate suffices.
99 <     */
100 <    static final long spinForTimeoutThreshold = 1000L;
419 >        // CAS methods for fields
420 >        final boolean casNext(Node cmp, Node val) {
421 >            return UNSAFE.compareAndSwapObject(this, nextOffset, cmp, val);
422 >        }
423  
424 <    /**
425 <     * Node class for LinkedTransferQueue. Opportunistically
426 <     * subclasses from AtomicReference to represent item. Uses Object,
427 <     * not E, to allow setting item to "this" after use, to avoid
106 <     * garbage retention. Similarly, setting the next field to this is
107 <     * used as sentinel that node is off list.
108 <     */
109 <    static final class Node<E> extends AtomicReference<Object> {
110 <        volatile Node<E> next;
111 <        volatile Thread waiter;       // to control park/unpark
112 <        final boolean isData;
424 >        final boolean casItem(Object cmp, Object val) {
425 >            assert cmp == null || cmp.getClass() != Node.class;
426 >            return UNSAFE.compareAndSwapObject(this, itemOffset, cmp, val);
427 >        }
428  
429 <        Node(E item, boolean isData) {
430 <            super(item);
429 >        /**
430 >         * Creates a new node. Uses relaxed write because item can only
431 >         * be seen if followed by CAS.
432 >         */
433 >        Node(Object item, boolean isData) {
434 >            UNSAFE.putObject(this, itemOffset, item); // relaxed write
435              this.isData = isData;
436          }
437  
438 <        // Unsafe mechanics
438 >        /**
439 >         * Links node to itself to avoid garbage retention.  Called
440 >         * only after CASing head field, so uses relaxed write.
441 >         */
442 >        final void forgetNext() {
443 >            UNSAFE.putObject(this, nextOffset, this);
444 >        }
445  
446 <        private static final sun.misc.Unsafe UNSAFE = getUnsafe();
447 <        private static final long nextOffset =
448 <            objectFieldOffset(UNSAFE, "next", Node.class);
446 >        /**
447 >         * Sets item to self and waiter to null, to avoid garbage
448 >         * retention after matching or cancelling. Uses relaxed writes
449 >         * bacause order is already constrained in the only calling
450 >         * contexts: item is forgotten only after volatile/atomic
451 >         * mechanics that extract items.  Similarly, clearing waiter
452 >         * follows either CAS or return from park (if ever parked;
453 >         * else we don't care).
454 >         */
455 >        final void forgetContents() {
456 >            UNSAFE.putObject(this, itemOffset, this);
457 >            UNSAFE.putObject(this, waiterOffset, null);
458 >        }
459  
460 <        final boolean casNext(Node<E> cmp, Node<E> val) {
461 <            return UNSAFE.compareAndSwapObject(this, nextOffset, cmp, val);
460 >        /**
461 >         * Returns true if this node has been matched, including the
462 >         * case of artificial matches due to cancellation.
463 >         */
464 >        final boolean isMatched() {
465 >            Object x = item;
466 >            return (x == this) || ((x == null) == isData);
467 >        }
468 >
469 >        /**
470 >         * Returns true if this is an unmatched request node.
471 >         */
472 >        final boolean isUnmatchedRequest() {
473 >            return !isData && item == null;
474          }
475  
476 <        final void clearNext() {
477 <            UNSAFE.putOrderedObject(this, nextOffset, this);
476 >        /**
477 >         * Returns true if a node with the given mode cannot be
478 >         * appended to this node because this node is unmatched and
479 >         * has opposite data mode.
480 >         */
481 >        final boolean cannotPrecede(boolean haveData) {
482 >            boolean d = isData;
483 >            Object x;
484 >            return d != haveData && (x = item) != this && (x != null) == d;
485          }
486  
487          /**
488 <         * Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
135 <         * Replace with a simple call to Unsafe.getUnsafe when integrating
136 <         * into a jdk.
137 <         *
138 <         * @return a sun.misc.Unsafe
488 >         * Tries to artificially match a data node -- used by remove.
489           */
490 <        private static sun.misc.Unsafe getUnsafe() {
491 <            try {
492 <                return sun.misc.Unsafe.getUnsafe();
493 <            } catch (SecurityException se) {
494 <                try {
495 <                    return java.security.AccessController.doPrivileged
146 <                        (new java.security
147 <                         .PrivilegedExceptionAction<sun.misc.Unsafe>() {
148 <                            public sun.misc.Unsafe run() throws Exception {
149 <                                java.lang.reflect.Field f = sun.misc
150 <                                    .Unsafe.class.getDeclaredField("theUnsafe");
151 <                                f.setAccessible(true);
152 <                                return (sun.misc.Unsafe) f.get(null);
153 <                            }});
154 <                } catch (java.security.PrivilegedActionException e) {
155 <                    throw new RuntimeException("Could not initialize intrinsics",
156 <                                               e.getCause());
157 <                }
490 >        final boolean tryMatchData() {
491 >            assert isData;
492 >            Object x = item;
493 >            if (x != null && x != this && casItem(x, null)) {
494 >                LockSupport.unpark(waiter);
495 >                return true;
496              }
497 +            return false;
498          }
499  
500 +        // Unsafe mechanics
501 +        private static final sun.misc.Unsafe UNSAFE = getUnsafe();
502 +        private static final long nextOffset =
503 +            objectFieldOffset(UNSAFE, "next", Node.class);
504 +        private static final long itemOffset =
505 +            objectFieldOffset(UNSAFE, "item", Node.class);
506 +        private static final long waiterOffset =
507 +            objectFieldOffset(UNSAFE, "waiter", Node.class);
508 +
509          private static final long serialVersionUID = -3375979862319811754L;
510      }
511  
512 <    /**
513 <     * Padded version of AtomicReference used for head, tail and
166 <     * cleanMe, to alleviate contention across threads CASing one vs
167 <     * the other.
168 <     */
169 <    static final class PaddedAtomicReference<T> extends AtomicReference<T> {
170 <        // enough padding for 64bytes with 4byte refs
171 <        Object p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pa, pb, pc, pd, pe;
172 <        PaddedAtomicReference(T r) { super(r); }
173 <        private static final long serialVersionUID = 8170090609809740854L;
174 <    }
512 >    /** head of the queue; null until first enqueue */
513 >    transient volatile Node head;
514  
515 +    /** tail of the queue; null until first append */
516 +    private transient volatile Node tail;
517  
518 <    /** head of the queue */
519 <    private transient final PaddedAtomicReference<Node<E>> head;
518 >    /** The number of apparent failures to unsplice removed nodes */
519 >    private transient volatile int sweepVotes;
520  
521 <    /** tail of the queue */
522 <    private transient final PaddedAtomicReference<Node<E>> tail;
521 >    // CAS methods for fields
522 >    private boolean casTail(Node cmp, Node val) {
523 >        return UNSAFE.compareAndSwapObject(this, tailOffset, cmp, val);
524 >    }
525  
526 <    /**
527 <     * Reference to a cancelled node that might not yet have been
528 <     * unlinked from queue because it was the last inserted node
186 <     * when it cancelled.
187 <     */
188 <    private transient final PaddedAtomicReference<Node<E>> cleanMe;
526 >    private boolean casHead(Node cmp, Node val) {
527 >        return UNSAFE.compareAndSwapObject(this, headOffset, cmp, val);
528 >    }
529  
530 <    /**
531 <     * Tries to cas nh as new head; if successful, unlink
532 <     * old head's next node to avoid garbage retention.
530 >    private boolean casSweepVotes(int cmp, int val) {
531 >        return UNSAFE.compareAndSwapInt(this, sweepVotesOffset, cmp, val);
532 >    }
533 >
534 >    /*
535 >     * Possible values for "how" argument in xfer method.
536       */
537 <    private boolean advanceHead(Node<E> h, Node<E> nh) {
538 <        if (h == head.get() && head.compareAndSet(h, nh)) {
539 <            h.clearNext(); // forget old next
540 <            return true;
541 <        }
542 <        return false;
537 >    private static final int NOW   = 0; // for untimed poll, tryTransfer
538 >    private static final int ASYNC = 1; // for offer, put, add
539 >    private static final int SYNC  = 2; // for transfer, take
540 >    private static final int TIMED = 3; // for timed poll, tryTransfer
541 >
542 >    @SuppressWarnings("unchecked")
543 >    static <E> E cast(Object item) {
544 >        assert item == null || item.getClass() != Node.class;
545 >        return (E) item;
546      }
547  
548      /**
549 <     * Puts or takes an item. Used for most queue operations (except
550 <     * poll() and tryTransfer()). See the similar code in
551 <     * SynchronousQueue for detailed explanation.
552 <     *
553 <     * @param e the item or if null, signifies that this is a take
554 <     * @param mode the wait mode: NOWAIT, TIMEOUT, WAIT
555 <     * @param nanos timeout in nanosecs, used only if mode is TIMEOUT
556 <     * @return an item, or null on failure
557 <     */
558 <    private E xfer(E e, int mode, long nanos) {
559 <        boolean isData = (e != null);
560 <        Node<E> s = null;
561 <        final PaddedAtomicReference<Node<E>> head = this.head;
216 <        final PaddedAtomicReference<Node<E>> tail = this.tail;
549 >     * Implements all queuing methods. See above for explanation.
550 >     *
551 >     * @param e the item or null for take
552 >     * @param haveData true if this is a put, else a take
553 >     * @param how NOW, ASYNC, SYNC, or TIMED
554 >     * @param nanos timeout in nanosecs, used only if mode is TIMED
555 >     * @return an item if matched, else e
556 >     * @throws NullPointerException if haveData mode but e is null
557 >     */
558 >    private E xfer(E e, boolean haveData, int how, long nanos) {
559 >        if (haveData && (e == null))
560 >            throw new NullPointerException();
561 >        Node s = null;                        // the node to append, if needed
562  
563 <        for (;;) {
219 <            Node<E> t = tail.get();
220 <            Node<E> h = head.get();
563 >        retry: for (;;) {                     // restart on append race
564  
565 <            if (t != null && (t == h || t.isData == isData)) {
566 <                if (s == null)
567 <                    s = new Node<E>(e, isData);
568 <                Node<E> last = t.next;
569 <                if (last != null) {
570 <                    if (t == tail.get())
571 <                        tail.compareAndSet(t, last);
572 <                }
573 <                else if (t.casNext(null, s)) {
574 <                    tail.compareAndSet(t, s);
575 <                    return awaitFulfill(t, s, e, mode, nanos);
576 <                }
577 <            }
578 <
579 <            else if (h != null) {
580 <                Node<E> first = h.next;
581 <                if (t == tail.get() && first != null &&
582 <                    advanceHead(h, first)) {
583 <                    Object x = first.get();
241 <                    if (x != first && first.compareAndSet(x, e)) {
242 <                        LockSupport.unpark(first.waiter);
243 <                        return isData ? e : (E) x;
565 >            for (Node h = head, p = h; p != null;) { // find & match first node
566 >                boolean isData = p.isData;
567 >                Object item = p.item;
568 >                if (item != p && (item != null) == isData) { // unmatched
569 >                    if (isData == haveData)   // can't match
570 >                        break;
571 >                    if (p.casItem(item, e)) { // match
572 >                        for (Node q = p; q != h;) {
573 >                            Node n = q.next;  // update by 2 unless singleton
574 >                            if (head == h && casHead(h, n == null? q : n)) {
575 >                                h.forgetNext();
576 >                                break;
577 >                            }                 // advance and retry
578 >                            if ((h = head)   == null ||
579 >                                (q = h.next) == null || !q.isMatched())
580 >                                break;        // unless slack < 2
581 >                        }
582 >                        LockSupport.unpark(p.waiter);
583 >                        return this.<E>cast(item);
584                      }
585                  }
586 +                Node n = p.next;
587 +                p = (p != n) ? n : (h = head); // Use head if p offlist
588              }
589 +
590 +            if (how != NOW) {                 // No matches available
591 +                if (s == null)
592 +                    s = new Node(e, haveData);
593 +                Node pred = tryAppend(s, haveData);
594 +                if (pred == null)
595 +                    continue retry;           // lost race vs opposite mode
596 +                if (how != ASYNC)
597 +                    return awaitMatch(s, pred, e, (how == TIMED), nanos);
598 +            }
599 +            return e; // not waiting
600          }
601      }
602  
250
603      /**
604 <     * Version of xfer for poll() and tryTransfer, which
605 <     * simplifies control paths both here and in xfer.
606 <     */
607 <    private E fulfill(E e) {
608 <        boolean isData = (e != null);
609 <        final PaddedAtomicReference<Node<E>> head = this.head;
610 <        final PaddedAtomicReference<Node<E>> tail = this.tail;
611 <
612 <        for (;;) {
613 <            Node<E> t = tail.get();
614 <            Node<E> h = head.get();
615 <
616 <            if (t != null && (t == h || t.isData == isData)) {
617 <                Node<E> last = t.next;
618 <                if (t == tail.get()) {
619 <                    if (last != null)
620 <                        tail.compareAndSet(t, last);
621 <                    else
622 <                        return null;
623 <                }
624 <            }
625 <            else if (h != null) {
626 <                Node<E> first = h.next;
627 <                if (t == tail.get() &&
628 <                    first != null &&
629 <                    advanceHead(h, first)) {
630 <                    Object x = first.get();
631 <                    if (x != first && first.compareAndSet(x, e)) {
280 <                        LockSupport.unpark(first.waiter);
281 <                        return isData ? e : (E) x;
282 <                    }
604 >     * Tries to append node s as tail.
605 >     *
606 >     * @param s the node to append
607 >     * @param haveData true if appending in data mode
608 >     * @return null on failure due to losing race with append in
609 >     * different mode, else s's predecessor, or s itself if no
610 >     * predecessor
611 >     */
612 >    private Node tryAppend(Node s, boolean haveData) {
613 >        for (Node t = tail, p = t;;) {        // move p to last node and append
614 >            Node n, u;                        // temps for reads of next & tail
615 >            if (p == null && (p = head) == null) {
616 >                if (casHead(null, s))
617 >                    return s;                 // initialize
618 >            }
619 >            else if (p.cannotPrecede(haveData))
620 >                return null;                  // lost race vs opposite mode
621 >            else if ((n = p.next) != null)    // not last; keep traversing
622 >                p = p != t && t != (u = tail) ? (t = u) : // stale tail
623 >                    (p != n) ? n : null;      // restart if off list
624 >            else if (!p.casNext(null, s))
625 >                p = p.next;                   // re-read on CAS failure
626 >            else {
627 >                if (p != t) {                 // update if slack now >= 2
628 >                    while ((tail != t || !casTail(t, s)) &&
629 >                           (t = tail)   != null &&
630 >                           (s = t.next) != null && // advance and retry
631 >                           (s = s.next) != null && s != t);
632                  }
633 +                return p;
634              }
635          }
636      }
637  
638      /**
639 <     * Spins/blocks until node s is fulfilled or caller gives up,
290 <     * depending on wait mode.
639 >     * Spins/yields/blocks until node s is matched or caller gives up.
640       *
292     * @param pred the predecessor of waiting node
641       * @param s the waiting node
642 +     * @param pred the predecessor of s, or s itself if it has no
643 +     * predecessor, or null if unknown (the null case does not occur
644 +     * in any current calls but may in possible future extensions)
645       * @param e the comparison value for checking match
646 <     * @param mode mode
647 <     * @param nanos timeout value
648 <     * @return matched item, or null if cancelled
649 <     */
650 <    private E awaitFulfill(Node<E> pred, Node<E> s, E e,
651 <                           int mode, long nanos) {
301 <        if (mode == NOWAIT)
302 <            return null;
303 <
304 <        long lastTime = (mode == TIMEOUT) ? System.nanoTime() : 0;
646 >     * @param timed if true, wait only until timeout elapses
647 >     * @param nanos timeout in nanosecs, used only if timed is true
648 >     * @return matched item, or e if unmatched on interrupt or timeout
649 >     */
650 >    private E awaitMatch(Node s, Node pred, E e, boolean timed, long nanos) {
651 >        long lastTime = timed ? System.nanoTime() : 0L;
652          Thread w = Thread.currentThread();
653 <        int spins = -1; // set to desired spin count below
653 >        int spins = -1; // initialized after first item and cancel checks
654 >        ThreadLocalRandom randomYields = null; // bound if needed
655 >
656          for (;;) {
657 <            if (w.isInterrupted())
658 <                s.compareAndSet(e, s);
659 <            Object x = s.get();
660 <            if (x != e) {                 // Node was matched or cancelled
661 <                advanceHead(pred, s);     // unlink if head
662 <                if (x == s) {             // was cancelled
663 <                    clean(pred, s);
664 <                    return null;
665 <                }
666 <                else if (x != null) {
667 <                    s.set(s);             // avoid garbage retention
668 <                    return (E) x;
669 <                }
670 <                else
671 <                    return e;
657 >            Object item = s.item;
658 >            if (item != e) {                  // matched
659 >                assert item != s;
660 >                s.forgetContents();           // avoid garbage
661 >                return this.<E>cast(item);
662 >            }
663 >            if ((w.isInterrupted() || (timed && nanos <= 0)) &&
664 >                    s.casItem(e, s)) {        // cancel
665 >                unsplice(pred, s);
666 >                return e;
667 >            }
668 >
669 >            if (spins < 0) {                  // establish spins at/near front
670 >                if ((spins = spinsFor(pred, s.isData)) > 0)
671 >                    randomYields = ThreadLocalRandom.current();
672 >            }
673 >            else if (spins > 0) {             // spin
674 >                --spins;
675 >                if (randomYields.nextInt(CHAINED_SPINS) == 0)
676 >                    Thread.yield();           // occasionally yield
677              }
678 <            if (mode == TIMEOUT) {
678 >            else if (s.waiter == null) {
679 >                s.waiter = w;                 // request unpark then recheck
680 >            }
681 >            else if (timed) {
682                  long now = System.nanoTime();
683 <                nanos -= now - lastTime;
683 >                if ((nanos -= now - lastTime) > 0)
684 >                    LockSupport.parkNanos(this, nanos);
685                  lastTime = now;
328                if (nanos <= 0) {
329                    s.compareAndSet(e, s); // try to cancel
330                    continue;
331                }
686              }
687 <            if (spins < 0) {
334 <                Node<E> h = head.get(); // only spin if at head
335 <                spins = ((h != null && h.next == s) ?
336 <                         ((mode == TIMEOUT) ?
337 <                          maxTimedSpins : maxUntimedSpins) : 0);
338 <            }
339 <            if (spins > 0)
340 <                --spins;
341 <            else if (s.waiter == null)
342 <                s.waiter = w;
343 <            else if (mode != TIMEOUT) {
687 >            else {
688                  LockSupport.park(this);
345                s.waiter = null;
346                spins = -1;
347            }
348            else if (nanos > spinForTimeoutThreshold) {
349                LockSupport.parkNanos(this, nanos);
350                s.waiter = null;
351                spins = -1;
689              }
690          }
691      }
692  
693      /**
694 <     * Returns validated tail for use in cleaning methods.
694 >     * Returns spin/yield value for a node with given predecessor and
695 >     * data mode. See above for explanation.
696       */
697 <    private Node<E> getValidatedTail() {
698 <        for (;;) {
699 <            Node<E> h = head.get();
700 <            Node<E> first = h.next;
701 <            if (first != null && first.get() == first) { // help advance
702 <                advanceHead(h, first);
703 <                continue;
704 <            }
367 <            Node<E> t = tail.get();
368 <            Node<E> last = t.next;
369 <            if (t == tail.get()) {
370 <                if (last != null)
371 <                    tail.compareAndSet(t, last); // help advance
372 <                else
373 <                    return t;
374 <            }
697 >    private static int spinsFor(Node pred, boolean haveData) {
698 >        if (MP && pred != null) {
699 >            if (pred.isData != haveData)      // phase change
700 >                return FRONT_SPINS + CHAINED_SPINS;
701 >            if (pred.isMatched())             // probably at front
702 >                return FRONT_SPINS;
703 >            if (pred.waiter == null)          // pred apparently spinning
704 >                return CHAINED_SPINS;
705          }
706 +        return 0;
707      }
708  
709 +    /* -------------- Traversal methods -------------- */
710 +
711      /**
712 <     * Gets rid of cancelled node s with original predecessor pred.
713 <     *
714 <     * @param pred predecessor of cancelled node
715 <     * @param s the cancelled node
712 >     * Returns the successor of p, or the head node if p.next has been
713 >     * linked to self, which will only be true if traversing with a
714 >     * stale pointer that is now off the list.
715 >     */
716 >    final Node succ(Node p) {
717 >        Node next = p.next;
718 >        return (p == next) ? head : next;
719 >    }
720 >
721 >    /**
722 >     * Returns the first unmatched node of the given mode, or null if
723 >     * none.  Used by methods isEmpty, hasWaitingConsumer.
724       */
725 <    private void clean(Node<E> pred, Node<E> s) {
726 <        Thread w = s.waiter;
727 <        if (w != null) {             // Wake up thread
728 <            s.waiter = null;
729 <            if (w != Thread.currentThread())
730 <                LockSupport.unpark(w);
725 >    private Node firstOfMode(boolean isData) {
726 >        for (Node p = head; p != null; p = succ(p)) {
727 >            if (!p.isMatched())
728 >                return (p.isData == isData) ? p : null;
729 >        }
730 >        return null;
731 >    }
732 >
733 >    /**
734 >     * Returns the item in the first unmatched node with isData; or
735 >     * null if none.  Used by peek.
736 >     */
737 >    private E firstDataItem() {
738 >        for (Node p = head; p != null; p = succ(p)) {
739 >            Object item = p.item;
740 >            if (p.isData) {
741 >                if (item != null && item != p)
742 >                    return this.<E>cast(item);
743 >            }
744 >            else if (item == null)
745 >                return null;
746          }
747 +        return null;
748 +    }
749  
750 <        if (pred == null)
751 <            return;
750 >    /**
751 >     * Traverses and counts unmatched nodes of the given mode.
752 >     * Used by methods size and getWaitingConsumerCount.
753 >     */
754 >    private int countOfMode(boolean data) {
755 >        int count = 0;
756 >        for (Node p = head; p != null; ) {
757 >            if (!p.isMatched()) {
758 >                if (p.isData != data)
759 >                    return 0;
760 >                if (++count == Integer.MAX_VALUE) // saturated
761 >                    break;
762 >            }
763 >            Node n = p.next;
764 >            if (n != p)
765 >                p = n;
766 >            else {
767 >                count = 0;
768 >                p = head;
769 >            }
770 >        }
771 >        return count;
772 >    }
773 >
774 >    final class Itr implements Iterator<E> {
775 >        private Node nextNode;   // next node to return item for
776 >        private E nextItem;      // the corresponding item
777 >        private Node lastRet;    // last returned node, to support remove
778 >        private Node lastPred;   // predecessor to unlink lastRet
779  
780 <        /*
781 <         * At any given time, exactly one node on list cannot be
397 <         * deleted -- the last inserted node. To accommodate this, if
398 <         * we cannot delete s, we save its predecessor as "cleanMe",
399 <         * processing the previously saved version first. At least one
400 <         * of node s or the node previously saved can always be
401 <         * processed, so this always terminates.
780 >        /**
781 >         * Moves to next node after prev, or first node if prev null.
782           */
783 <        while (pred.next == s) {
784 <            Node<E> oldpred = reclean();  // First, help get rid of cleanMe
785 <            Node<E> t = getValidatedTail();
786 <            if (s != t) {               // If not tail, try to unsplice
787 <                Node<E> sn = s.next;      // s.next == s means s already off list
788 <                if (sn == s || pred.casNext(s, sn))
783 >        private void advance(Node prev) {
784 >            lastPred = lastRet;
785 >            lastRet = prev;
786 >            for (Node p = (prev == null) ? head : succ(prev);
787 >                 p != null; p = succ(p)) {
788 >                Object item = p.item;
789 >                if (p.isData) {
790 >                    if (item != null && item != p) {
791 >                        nextItem = LinkedTransferQueue.this.<E>cast(item);
792 >                        nextNode = p;
793 >                        return;
794 >                    }
795 >                }
796 >                else if (item == null)
797                      break;
798              }
799 <            else if (oldpred == pred || // Already saved
800 <                     (oldpred == null && cleanMe.compareAndSet(null, pred)))
801 <                break;                  // Postpone cleaning
799 >            nextNode = null;
800 >        }
801 >
802 >        Itr() {
803 >            advance(null);
804 >        }
805 >
806 >        public final boolean hasNext() {
807 >            return nextNode != null;
808 >        }
809 >
810 >        public final E next() {
811 >            Node p = nextNode;
812 >            if (p == null) throw new NoSuchElementException();
813 >            E e = nextItem;
814 >            advance(p);
815 >            return e;
816 >        }
817 >
818 >        public final void remove() {
819 >            Node p = lastRet;
820 >            if (p == null) throw new IllegalStateException();
821 >            if (p.tryMatchData())
822 >                unsplice(lastPred, p);
823          }
824      }
825  
826 +    /* -------------- Removal methods -------------- */
827 +
828      /**
829 <     * Tries to unsplice the cancelled node held in cleanMe that was
830 <     * previously uncleanable because it was at tail.
829 >     * Unsplices (now or later) the given deleted/cancelled node with
830 >     * the given predecessor.
831       *
832 <     * @return current cleanMe node (or null)
832 >     * @param pred a node that was at one time known to be the
833 >     * predecessor of s, or null or s itself if s is/was at head
834 >     * @param s the node to be unspliced
835       */
836 <    private Node<E> reclean() {
836 >    final void unsplice(Node pred, Node s) {
837 >        s.forgetContents(); // forget unneeded fields
838          /*
839 <         * cleanMe is, or at one time was, predecessor of cancelled
840 <         * node s that was the tail so could not be unspliced.  If s
841 <         * is no longer the tail, try to unsplice if necessary and
842 <         * make cleanMe slot available.  This differs from similar
843 <         * code in clean() because we must check that pred still
430 <         * points to a cancelled node that must be unspliced -- if
431 <         * not, we can (must) clear cleanMe without unsplicing.
432 <         * This can loop only due to contention on casNext or
433 <         * clearing cleanMe.
839 >         * See above for rationale. Briefly: if pred still points to
840 >         * s, try to unlink s.  If s cannot be unlinked, because it is
841 >         * trailing node or pred might be unlinked, and neither pred
842 >         * nor s are head or offlist, add to sweepVotes, and if enough
843 >         * votes have accumulated, sweep.
844           */
845 <        Node<E> pred;
846 <        while ((pred = cleanMe.get()) != null) {
847 <            Node<E> t = getValidatedTail();
848 <            Node<E> s = pred.next;
849 <            if (s != t) {
850 <                Node<E> sn;
851 <                if (s == null || s == pred || s.get() != s ||
852 <                    (sn = s.next) == s || pred.casNext(s, sn))
853 <                    cleanMe.compareAndSet(pred, null);
845 >        if (pred != null && pred != s && pred.next == s) {
846 >            Node n = s.next;
847 >            if (n == null ||
848 >                (n != s && pred.casNext(s, n) && pred.isMatched())) {
849 >                for (;;) {               // check if at, or could be, head
850 >                    Node h = head;
851 >                    if (h == pred || h == s || h == null)
852 >                        return;          // at head or list empty
853 >                    if (!h.isMatched())
854 >                        break;
855 >                    Node hn = h.next;
856 >                    if (hn == null)
857 >                        return;          // now empty
858 >                    if (hn != h && casHead(h, hn))
859 >                        h.forgetNext();  // advance head
860 >                }
861 >                if (pred.next != pred && s.next != s) { // recheck if offlist
862 >                    for (;;) {           // sweep now if enough votes
863 >                        int v = sweepVotes;
864 >                        if (v < SWEEP_THRESHOLD) {
865 >                            if (casSweepVotes(v, v + 1))
866 >                                break;
867 >                        }
868 >                        else if (casSweepVotes(v, 0)) {
869 >                            sweep();
870 >                            break;
871 >                        }
872 >                    }
873 >                }
874              }
875 <            else // s is still tail; cannot clean
875 >        }
876 >    }
877 >
878 >    /**
879 >     * Unlinks matched nodes encountered in a traversal from head.
880 >     */
881 >    private void sweep() {
882 >        for (Node p = head, s, n; p != null && (s = p.next) != null; ) {
883 >            if (p == s)                    // stale
884 >                p = head;
885 >            else if (!s.isMatched())
886 >                p = s;
887 >            else if ((n = s.next) == null) // trailing node is pinned
888                  break;
889 +            else
890 +                p.casNext(s, n);
891          }
448        return pred;
892      }
893  
894      /**
895 +     * Main implementation of remove(Object)
896 +     */
897 +    private boolean findAndRemove(Object e) {
898 +        if (e != null) {
899 +            for (Node pred = null, p = head; p != null; ) {
900 +                Object item = p.item;
901 +                if (p.isData) {
902 +                    if (item != null && item != p && e.equals(item) &&
903 +                        p.tryMatchData()) {
904 +                        unsplice(pred, p);
905 +                        return true;
906 +                    }
907 +                }
908 +                else if (item == null)
909 +                    break;
910 +                pred = p;
911 +                if ((p = p.next) == pred) { // stale
912 +                    pred = null;
913 +                    p = head;
914 +                }
915 +            }
916 +        }
917 +        return false;
918 +    }
919 +
920 +
921 +    /**
922       * Creates an initially empty {@code LinkedTransferQueue}.
923       */
924      public LinkedTransferQueue() {
455        Node<E> dummy = new Node<E>(null, false);
456        head = new PaddedAtomicReference<Node<E>>(dummy);
457        tail = new PaddedAtomicReference<Node<E>>(dummy);
458        cleanMe = new PaddedAtomicReference<Node<E>>(null);
925      }
926  
927      /**
# Line 479 | Line 945 | public class LinkedTransferQueue<E> exte
945       * @throws NullPointerException if the specified element is null
946       */
947      public void put(E e) {
948 <        offer(e);
948 >        xfer(e, true, ASYNC, 0);
949      }
950  
951      /**
# Line 492 | Line 958 | public class LinkedTransferQueue<E> exte
958       * @throws NullPointerException if the specified element is null
959       */
960      public boolean offer(E e, long timeout, TimeUnit unit) {
961 <        return offer(e);
961 >        xfer(e, true, ASYNC, 0);
962 >        return true;
963      }
964  
965      /**
# Line 504 | Line 971 | public class LinkedTransferQueue<E> exte
971       * @throws NullPointerException if the specified element is null
972       */
973      public boolean offer(E e) {
974 <        if (e == null) throw new NullPointerException();
508 <        xfer(e, NOWAIT, 0);
974 >        xfer(e, true, ASYNC, 0);
975          return true;
976      }
977  
# Line 518 | Line 984 | public class LinkedTransferQueue<E> exte
984       * @throws NullPointerException if the specified element is null
985       */
986      public boolean add(E e) {
987 <        return offer(e);
987 >        xfer(e, true, ASYNC, 0);
988 >        return true;
989      }
990  
991      /**
# Line 532 | Line 999 | public class LinkedTransferQueue<E> exte
999       * @throws NullPointerException if the specified element is null
1000       */
1001      public boolean tryTransfer(E e) {
1002 <        if (e == null) throw new NullPointerException();
536 <        return fulfill(e) != null;
1002 >        return xfer(e, true, NOW, 0) == null;
1003      }
1004  
1005      /**
# Line 548 | Line 1014 | public class LinkedTransferQueue<E> exte
1014       * @throws NullPointerException if the specified element is null
1015       */
1016      public void transfer(E e) throws InterruptedException {
1017 <        if (e == null) throw new NullPointerException();
1018 <        if (xfer(e, WAIT, 0) == null) {
553 <            Thread.interrupted();
1017 >        if (xfer(e, true, SYNC, 0) != null) {
1018 >            Thread.interrupted(); // failure possible only due to interrupt
1019              throw new InterruptedException();
1020          }
1021      }
# Line 571 | Line 1036 | public class LinkedTransferQueue<E> exte
1036       */
1037      public boolean tryTransfer(E e, long timeout, TimeUnit unit)
1038          throws InterruptedException {
1039 <        if (e == null) throw new NullPointerException();
575 <        if (xfer(e, TIMEOUT, unit.toNanos(timeout)) != null)
1039 >        if (xfer(e, true, TIMED, unit.toNanos(timeout)) == null)
1040              return true;
1041          if (!Thread.interrupted())
1042              return false;
# Line 580 | Line 1044 | public class LinkedTransferQueue<E> exte
1044      }
1045  
1046      public E take() throws InterruptedException {
1047 <        E e = xfer(null, WAIT, 0);
1047 >        E e = xfer(null, false, SYNC, 0);
1048          if (e != null)
1049              return e;
1050          Thread.interrupted();
# Line 588 | Line 1052 | public class LinkedTransferQueue<E> exte
1052      }
1053  
1054      public E poll(long timeout, TimeUnit unit) throws InterruptedException {
1055 <        E e = xfer(null, TIMEOUT, unit.toNanos(timeout));
1055 >        E e = xfer(null, false, TIMED, unit.toNanos(timeout));
1056          if (e != null || !Thread.interrupted())
1057              return e;
1058          throw new InterruptedException();
1059      }
1060  
1061      public E poll() {
1062 <        return fulfill(null);
1062 >        return xfer(null, false, NOW, 0);
1063      }
1064  
1065      /**
# Line 634 | Line 1098 | public class LinkedTransferQueue<E> exte
1098          return n;
1099      }
1100  
637    // Traversal-based methods
638
639    /**
640     * Returns head after performing any outstanding helping steps.
641     */
642    private Node<E> traversalHead() {
643        for (;;) {
644            Node<E> t = tail.get();
645            Node<E> h = head.get();
646            if (h != null && t != null) {
647                Node<E> last = t.next;
648                Node<E> first = h.next;
649                if (t == tail.get()) {
650                    if (last != null)
651                        tail.compareAndSet(t, last);
652                    else if (first != null) {
653                        Object x = first.get();
654                        if (x == first)
655                            advanceHead(h, first);
656                        else
657                            return h;
658                    }
659                    else
660                        return h;
661                }
662            }
663            reclean();
664        }
665    }
666
1101      /**
1102       * Returns an iterator over the elements in this queue in proper
1103       * sequence, from head to tail.
# Line 681 | Line 1115 | public class LinkedTransferQueue<E> exte
1115          return new Itr();
1116      }
1117  
684    /**
685     * Iterators. Basic strategy is to traverse list, treating
686     * non-data (i.e., request) nodes as terminating list.
687     * Once a valid data node is found, the item is cached
688     * so that the next call to next() will return it even
689     * if subsequently removed.
690     */
691    class Itr implements Iterator<E> {
692        Node<E> next;        // node to return next
693        Node<E> pnext;       // predecessor of next
694        Node<E> curr;        // last returned node, for remove()
695        Node<E> pcurr;       // predecessor of curr, for remove()
696        E nextItem;          // Cache of next item, once committed to in next
697
698        Itr() {
699            advance();
700        }
701
702        /**
703         * Moves to next valid node and returns item to return for
704         * next(), or null if no such.
705         */
706        private E advance() {
707            pcurr = pnext;
708            curr = next;
709            E item = nextItem;
710
711            for (;;) {
712                pnext = (next == null) ? traversalHead() : next;
713                next = pnext.next;
714                if (next == pnext) {
715                    next = null;
716                    continue;  // restart
717                }
718                if (next == null)
719                    break;
720                Object x = next.get();
721                if (x != null && x != next) {
722                    nextItem = (E) x;
723                    break;
724                }
725            }
726            return item;
727        }
728
729        public boolean hasNext() {
730            return next != null;
731        }
732
733        public E next() {
734            if (next == null)
735                throw new NoSuchElementException();
736            return advance();
737        }
738
739        public void remove() {
740            Node<E> p = curr;
741            if (p == null)
742                throw new IllegalStateException();
743            Object x = p.get();
744            if (x != null && x != p && p.compareAndSet(x, p))
745                clean(pcurr, p);
746        }
747    }
748
1118      public E peek() {
1119 <        for (;;) {
751 <            Node<E> h = traversalHead();
752 <            Node<E> p = h.next;
753 <            if (p == null)
754 <                return null;
755 <            Object x = p.get();
756 <            if (p != x) {
757 <                if (!p.isData)
758 <                    return null;
759 <                if (x != null)
760 <                    return (E) x;
761 <            }
762 <        }
1119 >        return firstDataItem();
1120      }
1121  
1122      /**
# Line 768 | Line 1125 | public class LinkedTransferQueue<E> exte
1125       * @return {@code true} if this queue contains no elements
1126       */
1127      public boolean isEmpty() {
1128 <        for (;;) {
772 <            Node<E> h = traversalHead();
773 <            Node<E> p = h.next;
774 <            if (p == null)
775 <                return true;
776 <            Object x = p.get();
777 <            if (p != x) {
778 <                if (!p.isData)
779 <                    return true;
780 <                if (x != null)
781 <                    return false;
782 <            }
783 <        }
1128 >        return firstOfMode(true) == null;
1129      }
1130  
1131      public boolean hasWaitingConsumer() {
1132 <        for (;;) {
788 <            Node<E> h = traversalHead();
789 <            Node<E> p = h.next;
790 <            if (p == null)
791 <                return false;
792 <            Object x = p.get();
793 <            if (p != x)
794 <                return !p.isData;
795 <        }
1132 >        return firstOfMode(false) != null;
1133      }
1134  
1135      /**
# Line 808 | Line 1145 | public class LinkedTransferQueue<E> exte
1145       * @return the number of elements in this queue
1146       */
1147      public int size() {
1148 <        for (;;) {
812 <            int count = 0;
813 <            Node<E> pred = traversalHead();
814 <            for (;;) {
815 <                Node<E> q = pred.next;
816 <                if (q == pred) // restart
817 <                    break;
818 <                if (q == null || !q.isData)
819 <                    return count;
820 <                Object x = q.get();
821 <                if (x != null && x != q) {
822 <                    if (++count == Integer.MAX_VALUE) // saturated
823 <                        return count;
824 <                }
825 <                pred = q;
826 <            }
827 <        }
1148 >        return countOfMode(true);
1149      }
1150  
1151      public int getWaitingConsumerCount() {
1152 <        // converse of size -- count valid non-data nodes
832 <        for (;;) {
833 <            int count = 0;
834 <            Node<E> pred = traversalHead();
835 <            for (;;) {
836 <                Node<E> q = pred.next;
837 <                if (q == pred) // restart
838 <                    break;
839 <                if (q == null || q.isData)
840 <                    return count;
841 <                Object x = q.get();
842 <                if (x == null) {
843 <                    if (++count == Integer.MAX_VALUE) // saturated
844 <                        return count;
845 <                }
846 <                pred = q;
847 <            }
848 <        }
1152 >        return countOfMode(false);
1153      }
1154  
1155 +    /**
1156 +     * Removes a single instance of the specified element from this queue,
1157 +     * if it is present.  More formally, removes an element {@code e} such
1158 +     * that {@code o.equals(e)}, if this queue contains one or more such
1159 +     * elements.
1160 +     * Returns {@code true} if this queue contained the specified element
1161 +     * (or equivalently, if this queue changed as a result of the call).
1162 +     *
1163 +     * @param o element to be removed from this queue, if present
1164 +     * @return {@code true} if this queue changed as a result of the call
1165 +     */
1166      public boolean remove(Object o) {
1167 <        if (o == null)
853 <            return false;
854 <        for (;;) {
855 <            Node<E> pred = traversalHead();
856 <            for (;;) {
857 <                Node<E> q = pred.next;
858 <                if (q == pred) // restart
859 <                    break;
860 <                if (q == null || !q.isData)
861 <                    return false;
862 <                Object x = q.get();
863 <                if (x != null && x != q && o.equals(x) &&
864 <                    q.compareAndSet(x, q)) {
865 <                    clean(pred, q);
866 <                    return true;
867 <                }
868 <                pred = q;
869 <            }
870 <        }
1167 >        return findAndRemove(o);
1168      }
1169  
1170      /**
# Line 882 | Line 1179 | public class LinkedTransferQueue<E> exte
1179      }
1180  
1181      /**
1182 <     * Save the state to a stream (that is, serialize it).
1182 >     * Saves the state to a stream (that is, serializes it).
1183       *
1184       * @serialData All of the elements (each an {@code E}) in
1185       * the proper order, followed by a null
# Line 898 | Line 1195 | public class LinkedTransferQueue<E> exte
1195      }
1196  
1197      /**
1198 <     * Reconstitute the Queue instance from a stream (that is,
1199 <     * deserialize it).
1198 >     * Reconstitutes the Queue instance from a stream (that is,
1199 >     * deserializes it).
1200       *
1201       * @param s the stream
1202       */
1203      private void readObject(java.io.ObjectInputStream s)
1204          throws java.io.IOException, ClassNotFoundException {
1205          s.defaultReadObject();
909        resetHeadAndTail();
1206          for (;;) {
1207              @SuppressWarnings("unchecked") E item = (E) s.readObject();
1208              if (item == null)
# Line 916 | Line 1212 | public class LinkedTransferQueue<E> exte
1212          }
1213      }
1214  
919    // Support for resetting head/tail while deserializing
920    private void resetHeadAndTail() {
921        Node<E> dummy = new Node<E>(null, false);
922        UNSAFE.putObjectVolatile(this, headOffset,
923                                 new PaddedAtomicReference<Node<E>>(dummy));
924        UNSAFE.putObjectVolatile(this, tailOffset,
925                                 new PaddedAtomicReference<Node<E>>(dummy));
926        UNSAFE.putObjectVolatile(this, cleanMeOffset,
927                                 new PaddedAtomicReference<Node<E>>(null));
928    }
929
1215      // Unsafe mechanics
1216  
1217      private static final sun.misc.Unsafe UNSAFE = getUnsafe();
# Line 934 | Line 1219 | public class LinkedTransferQueue<E> exte
1219          objectFieldOffset(UNSAFE, "head", LinkedTransferQueue.class);
1220      private static final long tailOffset =
1221          objectFieldOffset(UNSAFE, "tail", LinkedTransferQueue.class);
1222 <    private static final long cleanMeOffset =
1223 <        objectFieldOffset(UNSAFE, "cleanMe", LinkedTransferQueue.class);
939 <
1222 >    private static final long sweepVotesOffset =
1223 >        objectFieldOffset(UNSAFE, "sweepVotes", LinkedTransferQueue.class);
1224  
1225      static long objectFieldOffset(sun.misc.Unsafe UNSAFE,
1226                                    String field, Class<?> klazz) {
# Line 957 | Line 1241 | public class LinkedTransferQueue<E> exte
1241       *
1242       * @return a sun.misc.Unsafe
1243       */
1244 <    private static sun.misc.Unsafe getUnsafe() {
1244 >    static sun.misc.Unsafe getUnsafe() {
1245          try {
1246              return sun.misc.Unsafe.getUnsafe();
1247          } catch (SecurityException se) {
# Line 977 | Line 1261 | public class LinkedTransferQueue<E> exte
1261              }
1262          }
1263      }
1264 +
1265   }

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines