概要
本章对“公平锁”的获取锁机制进行介绍(本文的公平锁指的是互斥锁的公平锁),内容包括:
基本概念
参考代码
获取公平锁(基于JDK1.7.0_40)
一.
tryAcquire()
二. addWaiter()
三. acquireQueued()
四.
selfInterrupt()
“公平锁”的释放锁的机制在后面一章再做介绍,锁的使用示例请参考“Java多线程系列--“JUC锁”02之
互斥锁ReentrantLock”。
转载请注明出处:http://www.cnblogs.com/skywang12345/p/3496147.html
基本概念
本章,我们会讲解“线程获取公平锁”的原理;在讲解之前,需要了解几个基本概念。后面的内容,都是基于这些概念的;这些概念可能比较枯燥,但从这些概念中,能窥见“java锁”的一些架构,这对我们了解锁是有帮助的。
1. AQS --
指AbstractQueuedSynchronizer类。
AQS是java中管理“锁”的抽象类,锁的许多公共方法都是在这个类中实现。AQS是独占锁(例如,ReentrantLock)和共享锁(例如,Semaphore)的公共父类。
2. AQS锁的类别
-- 分为“独占锁”和“共享锁”两种。
(01) 独占锁 --
锁在一个时间点只能被一个线程锁占有。根据锁的获取机制,它又划分为“公平锁”和“非公平锁”。公平锁,是按照通过CLH等待线程按照先来先得的规则,公平的获取锁;而非公平锁,则当线程要获取锁时,它会无视CLH等待队列而直接获取锁。独占锁的典型实例子是ReentrantLock,此外,ReentrantReadWriteLock.WriteLock也是独占锁。
(02) 共享锁 --
能被多个线程同时拥有,能被共享的锁。JUC包中的ReentrantReadWriteLock.ReadLock,CyclicBarrier, CountDownLatch和Semaphore都是共享锁。这些锁的用途和原理,在以后的章节再详细介绍。
3. CLH队列 -- Craig, Landin, and
Hagersten lock queue
CLH队列是AQS中“等待锁”的线程队列。在多线程中,为了保护竞争资源不被多个线程同时操作而起来错误,我们常常需要通过锁来保护这些资源。在独占锁中,竞争资源在一个时间点只能被一个线程锁访问;而其它线程则需要等待。CLH就是管理这些“等待锁”的线程的队列。
CLH是一个非阻塞的 FIFO
队列。也就是说往里面插入或移除一个节点的时候,在并发条件下不会阻塞,而是通过自旋锁和 CAS 保证节点插入和移除的原子性。
4. CAS函数
-- Compare And Swap
CAS函数,是比较并交换函数,它是原子操作函数;即,通过CAS操作的数据都是以原子方式进行的。例如,compareAndSetHead(),
compareAndSetTail(),
compareAndSetNext()等函数。它们共同的特点是,这些函数所执行的动作是以原子的方式进行的。
本章是围绕“公平锁”如何获取锁而层次展开。“公平锁”涉及到的知识点比较多,但总的来说,不是特别难;如果读者能读懂AQS和ReentrantLock.java这两个类的大致意思,理解锁的原理和机制也就不成问题了。本章只是作者本人对锁的一点点理解,希望这部分知识能帮助您了解“公平锁”的获取过程,认识“锁”的框架。
参考代码
下面给出Java1.7.0_40版本中,ReentrantLock和AQS的源码,仅供参考!
ReentranLock.java
1 /* 2 * ORACLE PROPRIETARY/CONFIDENTIAL. Use is subject to license terms. 3 * 4 * 5 * 6 * 7 * 8 * 9 * 10 * 11 * 12 * 13 * 14 * 15 * 16 * 17 * 18 * 19 * 20 * 21 * 22 * 23 */ 24 25 /* 26 * 27 * 28 * 29 * 30 * 31 * Written by Doug Lea with assistance from members of JCP JSR-166 32 * Expert Group and released to the public domain, as explained at 33 * http://creativecommons.org/publicdomain/zero/1.0/ 34 */ 35 36 package java.util.concurrent.locks; 37 import java.util.*; 38 import java.util.concurrent.*; 39 import java.util.concurrent.atomic.*; 40 41 /** 42 * A reentrant mutual exclusion {@link Lock} with the same basic 43 * behavior and semantics as the implicit monitor lock accessed using 44 * {@code synchronized} methods and statements, but with extended 45 * capabilities. 46 * 47 * <p>A {@code ReentrantLock} is <em>owned</em> by the thread last 48 * successfully locking, but not yet unlocking it. A thread invoking 49 * {@code lock} will return, successfully acquiring the lock, when 50 * the lock is not owned by another thread. The method will return 51 * immediately if the current thread already owns the lock. This can 52 * be checked using methods {@link #isHeldByCurrentThread}, and {@link 53 * #getHoldCount}. 54 * 55 * <p>The constructor for this class accepts an optional 56 * <em>fairness</em> parameter. When set {@code true}, under 57 * contention, locks favor granting access to the longest-waiting 58 * thread. Otherwise this lock does not guarantee any particular 59 * access order. Programs using fair locks accessed by many threads 60 * may display lower overall throughput (i.e., are slower; often much 61 * slower) than those using the default setting, but have smaller 62 * variances in times to obtain locks and guarantee lack of 63 * starvation. Note however, that fairness of locks does not guarantee 64 * fairness of thread scheduling. Thus, one of many threads using a 65 * fair lock may obtain it multiple times in succession while other 66 * active threads are not progressing and not currently holding the 67 * lock. 68 * Also note that the untimed {@link #tryLock() tryLock} method does not 69 * honor the fairness setting. It will succeed if the lock 70 * is available even if other threads are waiting. 71 * 72 * <p>It is recommended practice to <em>always</em> immediately 73 * follow a call to {@code lock} with a {@code try} block, most 74 * typically in a before/after construction such as: 75 * 76 * <pre> 77 * class X { 78 * private final ReentrantLock lock = new ReentrantLock(); 79 * // ... 80 * 81 * public void m() { 82 * lock.lock(); // block until condition holds 83 * try { 84 * // ... method body 85 * } finally { 86 * lock.unlock() 87 * } 88 * } 89 * } 90 * </pre> 91 * 92 * <p>In addition to implementing the {@link Lock} interface, this 93 * class defines methods {@code isLocked} and 94 * {@code getLockQueueLength}, as well as some associated 95 * {@code protected} access methods that may be useful for 96 * instrumentation and monitoring. 97 * 98 * <p>Serialization of this class behaves in the same way as built-in 99 * locks: a deserialized lock is in the unlocked state, regardless of 100 * its state when serialized. 101 * 102 * <p>This lock supports a maximum of 2147483647 recursive locks by 103 * the same thread. Attempts to exceed this limit result in 104 * {@link Error} throws from locking methods. 105 * 106 * @since 1.5 107 * @author Doug Lea 108 */ 109 public class ReentrantLock implements Lock, java.io.Serializable { 110 private static final long serialVersionUID = 7373984872572414699L; 111 /** Synchronizer providing all implementation mechanics */ 112 private final Sync sync; 113 114 /** 115 * Base of synchronization control for this lock. Subclassed 116 * into fair and nonfair versions below. Uses AQS state to 117 * represent the number of holds on the lock. 118 */ 119 abstract static class Sync extends AbstractQueuedSynchronizer { 120 private static final long serialVersionUID = -5179523762034025860L; 121 122 /** 123 * Performs {@link Lock#lock}. The main reason for subclassing 124 * is to allow fast path for nonfair version. 125 */ 126 abstract void lock(); 127 128 /** 129 * Performs non-fair tryLock. tryAcquire is 130 * implemented in subclasses, but both need nonfair 131 * try for trylock method. 132 */ 133 final boolean nonfairTryAcquire(int acquires) { 134 final Thread current = Thread.currentThread(); 135 int c = getState(); 136 if (c == 0) { 137 if (compareAndSetState(0, acquires)) { 138 setExclusiveOwnerThread(current); 139 return true; 140 } 141 } 142 else if (current == getExclusiveOwnerThread()) { 143 int nextc = c + acquires; 144 if (nextc < 0) // overflow 145 throw new Error("Maximum lock count exceeded"); 146 setState(nextc); 147 return true; 148 } 149 return false; 150 } 151 152 protected final boolean tryRelease(int releases) { 153 int c = getState() - releases; 154 if (Thread.currentThread() != getExclusiveOwnerThread()) 155 throw new IllegalMonitorStateException(); 156 boolean free = false; 157 if (c == 0) { 158 free = true; 159 setExclusiveOwnerThread(null); 160 } 161 setState(c); 162 return free; 163 } 164 165 protected final boolean isHeldExclusively() { 166 // While we must in general read state before owner, 167 // we don‘t need to do so to check if current thread is owner 168 return getExclusiveOwnerThread() == Thread.currentThread(); 169 } 170 171 final ConditionObject newCondition() { 172 return new ConditionObject(); 173 } 174 175 // Methods relayed from outer class 176 177 final Thread getOwner() { 178 return getState() == 0 ? null : getExclusiveOwnerThread(); 179 } 180 181 final int getHoldCount() { 182 return isHeldExclusively() ? getState() : 0; 183 } 184 185 final boolean isLocked() { 186 return getState() != 0; 187 } 188 189 /** 190 * Reconstitutes this lock instance from a stream. 191 * @param s the stream 192 */ 193 private void readObject(java.io.ObjectInputStream s) 194 throws java.io.IOException, ClassNotFoundException { 195 s.defaultReadObject(); 196 setState(0); // reset to unlocked state 197 } 198 } 199 200 /** 201 * Sync object for non-fair locks 202 */ 203 static final class NonfairSync extends Sync { 204 private static final long serialVersionUID = 7316153563782823691L; 205 206 /** 207 * Performs lock. Try immediate barge, backing up to normal 208 * acquire on failure. 209 */ 210 final void lock() { 211 if (compareAndSetState(0, 1)) 212 setExclusiveOwnerThread(Thread.currentThread()); 213 else 214 acquire(1); 215 } 216 217 protected final boolean tryAcquire(int acquires) { 218 return nonfairTryAcquire(acquires); 219 } 220 } 221 222 /** 223 * Sync object for fair locks 224 */ 225 static final class FairSync extends Sync { 226 private static final long serialVersionUID = -3000897897090466540L; 227 228 final void lock() { 229 acquire(1); 230 } 231 232 /** 233 * Fair version of tryAcquire. Don‘t grant access unless 234 * recursive call or no waiters or is first. 235 */ 236 protected final boolean tryAcquire(int acquires) { 237 final Thread current = Thread.currentThread(); 238 int c = getState(); 239 if (c == 0) { 240 if (!hasQueuedPredecessors() && 241 compareAndSetState(0, acquires)) { 242 setExclusiveOwnerThread(current); 243 return true; 244 } 245 } 246 else if (current == getExclusiveOwnerThread()) { 247 int nextc = c + acquires; 248 if (nextc < 0) 249 throw new Error("Maximum lock count exceeded"); 250 setState(nextc); 251 return true; 252 } 253 return false; 254 } 255 } 256 257 /** 258 * Creates an instance of {@code ReentrantLock}. 259 * This is equivalent to using {@code ReentrantLock(false)}. 260 */ 261 public ReentrantLock() { 262 sync = new NonfairSync(); 263 } 264 265 /** 266 * Creates an instance of {@code ReentrantLock} with the 267 * given fairness policy. 268 * 269 * @param fair {@code true} if this lock should use a fair ordering policy 270 */ 271 public ReentrantLock(boolean fair) { 272 sync = fair ? new FairSync() : new NonfairSync(); 273 } 274 275 /** 276 * Acquires the lock. 277 * 278 * <p>Acquires the lock if it is not held by another thread and returns 279 * immediately, setting the lock hold count to one. 280 * 281 * <p>If the current thread already holds the lock then the hold 282 * count is incremented by one and the method returns immediately. 283 * 284 * <p>If the lock is held by another thread then the 285 * current thread becomes disabled for thread scheduling 286 * purposes and lies dormant until the lock has been acquired, 287 * at which time the lock hold count is set to one. 288 */ 289 public void lock() { 290 sync.lock(); 291 } 292 293 /** 294 * Acquires the lock unless the current thread is 295 * {@linkplain Thread#interrupt interrupted}. 296 * 297 * <p>Acquires the lock if it is not held by another thread and returns 298 * immediately, setting the lock hold count to one. 299 * 300 * <p>If the current thread already holds this lock then the hold count 301 * is incremented by one and the method returns immediately. 302 * 303 * <p>If the lock is held by another thread then the 304 * current thread becomes disabled for thread scheduling 305 * purposes and lies dormant until one of two things happens: 306 * 307 * <ul> 308 * 309 * <li>The lock is acquired by the current thread; or 310 * 311 * <li>Some other thread {@linkplain Thread#interrupt interrupts} the 312 * current thread. 313 * 314 * </ul> 315 * 316 * <p>If the lock is acquired by the current thread then the lock hold 317 * count is set to one. 318 * 319 * <p>If the current thread: 320 * 321 * <ul> 322 * 323 * <li>has its interrupted status set on entry to this method; or 324 * 325 * <li>is {@linkplain Thread#interrupt interrupted} while acquiring 326 * the lock, 327 * 328 * </ul> 329 * 330 * then {@link InterruptedException} is thrown and the current thread‘s 331 * interrupted status is cleared. 332 * 333 * <p>In this implementation, as this method is an explicit 334 * interruption point, preference is given to responding to the 335 * interrupt over normal or reentrant acquisition of the lock. 336 * 337 * @throws InterruptedException if the current thread is interrupted 338 */ 339 public void lockInterruptibly() throws InterruptedException { 340 sync.acquireInterruptibly(1); 341 } 342 343 /** 344 * Acquires the lock only if it is not held by another thread at the time 345 * of invocation. 346 * 347 * <p>Acquires the lock if it is not held by another thread and 348 * returns immediately with the value {@code true}, setting the 349 * lock hold count to one. Even when this lock has been set to use a 350 * fair ordering policy, a call to {@code tryLock()} <em>will</em> 351 * immediately acquire the lock if it is available, whether or not 352 * other threads are currently waiting for the lock. 353 * This "barging" behavior can be useful in certain 354 * circumstances, even though it breaks fairness. If you want to honor 355 * the fairness setting for this lock, then use 356 * {@link #tryLock(long, TimeUnit) tryLock(0, TimeUnit.SECONDS) } 357 * which is almost equivalent (it also detects interruption). 358 * 359 * <p> If the current thread already holds this lock then the hold 360 * count is incremented by one and the method returns {@code true}. 361 * 362 * <p>If the lock is held by another thread then this method will return 363 * immediately with the value {@code false}. 364 * 365 * @return {@code true} if the lock was free and was acquired by the 366 * current thread, or the lock was already held by the current 367 * thread; and {@code false} otherwise 368 */ 369 public boolean tryLock() { 370 return sync.nonfairTryAcquire(1); 371 } 372 373 /** 374 * Acquires the lock if it is not held by another thread within the given 375 * waiting time and the current thread has not been 376 * {@linkplain Thread#interrupt interrupted}. 377 * 378 * <p>Acquires the lock if it is not held by another thread and returns 379 * immediately with the value {@code true}, setting the lock hold count 380 * to one. If this lock has been set to use a fair ordering policy then 381 * an available lock <em>will not</em> be acquired if any other threads 382 * are waiting for the lock. This is in contrast to the {@link #tryLock()} 383 * method. If you want a timed {@code tryLock} that does permit barging on 384 * a fair lock then combine the timed and un-timed forms together: 385 * 386 * <pre>if (lock.tryLock() || lock.tryLock(timeout, unit) ) { ... } 387 * </pre> 388 * 389 * <p>If the current thread 390 * already holds this lock then the hold count is incremented by one and 391 * the method returns {@code true}. 392 * 393 * <p>If the lock is held by another thread then the 394 * current thread becomes disabled for thread scheduling 395 * purposes and lies dormant until one of three things happens: 396 * 397 * <ul> 398 * 399 * <li>The lock is acquired by the current thread; or 400 * 401 * <li>Some other thread {@linkplain Thread#interrupt interrupts} 402 * the current thread; or 403 * 404 * <li>The specified waiting time elapses 405 * 406 * </ul> 407 * 408 * <p>If the lock is acquired then the value {@code true} is returned and 409 * the lock hold count is set to one. 410 * 411 * <p>If the current thread: 412 * 413 * <ul> 414 * 415 * <li>has its interrupted status set on entry to this method; or 416 * 417 * <li>is {@linkplain Thread#interrupt interrupted} while 418 * acquiring the lock, 419 * 420 * </ul> 421 * then {@link InterruptedException} is thrown and the current thread‘s 422 * interrupted status is cleared. 423 * 424 * <p>If the specified waiting time elapses then the value {@code false} 425 * is returned. If the time is less than or equal to zero, the method 426 * will not wait at all. 427 * 428 * <p>In this implementation, as this method is an explicit 429 * interruption point, preference is given to responding to the 430 * interrupt over normal or reentrant acquisition of the lock, and 431 * over reporting the elapse of the waiting time. 432 * 433 * @param timeout the time to wait for the lock 434 * @param unit the time unit of the timeout argument 435 * @return {@code true} if the lock was free and was acquired by the 436 * current thread, or the lock was already held by the current 437 * thread; and {@code false} if the waiting time elapsed before 438 * the lock could be acquired 439 * @throws InterruptedException if the current thread is interrupted 440 * @throws NullPointerException if the time unit is null 441 * 442 */ 443 public boolean tryLock(long timeout, TimeUnit unit) 444 throws InterruptedException { 445 return sync.tryAcquireNanos(1, unit.toNanos(timeout)); 446 } 447 448 /** 449 * Attempts to release this lock. 450 * 451 * <p>If the current thread is the holder of this lock then the hold 452 * count is decremented. If the hold count is now zero then the lock 453 * is released. If the current thread is not the holder of this 454 * lock then {@link IllegalMonitorStateException} is thrown. 455 * 456 * @throws IllegalMonitorStateException if the current thread does not 457 * hold this lock 458 */ 459 public void unlock() { 460 sync.release(1); 461 } 462 463 /** 464 * Returns a {@link Condition} instance for use with this 465 * {@link Lock} instance. 466 * 467 * <p>The returned {@link Condition} instance supports the same 468 * usages as do the {@link Object} monitor methods ({@link 469 * Object#wait() wait}, {@link Object#notify notify}, and {@link 470 * Object#notifyAll notifyAll}) when used with the built-in 471 * monitor lock. 472 * 473 * <ul> 474 * 475 * <li>If this lock is not held when any of the {@link Condition} 476 * {@linkplain Condition#await() waiting} or {@linkplain 477 * Condition#signal signalling} methods are called, then an {@link 478 * IllegalMonitorStateException} is thrown. 479 * 480 * <li>When the condition {@linkplain Condition#await() waiting} 481 * methods are called the lock is released and, before they 482 * return, the lock is reacquired and the lock hold count restored 483 * to what it was when the method was called. 484 * 485 * <li>If a thread is {@linkplain Thread#interrupt interrupted} 486 * while waiting then the wait will terminate, an {@link 487 * InterruptedException} will be thrown, and the thread‘s 488 * interrupted status will be cleared. 489 * 490 * <li> Waiting threads are signalled in FIFO order. 491 * 492 * <li>The ordering of lock reacquisition for threads returning 493 * from waiting methods is the same as for threads initially 494 * acquiring the lock, which is in the default case not specified, 495 * but for <em>fair</em> locks favors those threads that have been 496 * waiting the longest. 497 * 498 * </ul> 499 * 500 * @return the Condition object 501 */ 502 public Condition newCondition() { 503 return sync.newCondition(); 504 } 505 506 /** 507 * Queries the number of holds on this lock by the current thread. 508 * 509 * <p>A thread has a hold on a lock for each lock action that is not 510 * matched by an unlock action. 511 * 512 * <p>The hold count information is typically only used for testing and 513 * debugging purposes. For example, if a certain section of code should 514 * not be entered with the lock already held then we can assert that 515 * fact: 516 * 517 * <pre> 518 * class X { 519 * ReentrantLock lock = new ReentrantLock(); 520 * // ... 521 * public void m() { 522 * assert lock.getHoldCount() == 0; 523 * lock.lock(); 524 * try { 525 * // ... method body 526 * } finally { 527 * lock.unlock(); 528 * } 529 * } 530 * } 531 * </pre> 532 * 533 * @return the number of holds on this lock by the current thread, 534 * or zero if this lock is not held by the current thread 535 */ 536 public int getHoldCount() { 537 return sync.getHoldCount(); 538 } 539 540 /** 541 * Queries if this lock is held by the current thread. 542 * 543 * <p>Analogous to the {@link Thread#holdsLock} method for built-in 544 * monitor locks, this method is typically used for debugging and 545 * testing. For example, a method that should only be called while 546 * a lock is held can assert that this is the case: 547 * 548 * <pre> 549 * class X { 550 * ReentrantLock lock = new ReentrantLock(); 551 * // ... 552 * 553 * public void m() { 554 * assert lock.isHeldByCurrentThread(); 555 * // ... method body 556 * } 557 * } 558 * </pre> 559 * 560 * <p>It can also be used to ensure that a reentrant lock is used 561 * in a non-reentrant manner, for example: 562 * 563 * <pre> 564 * class X { 565 * ReentrantLock lock = new ReentrantLock(); 566 * // ... 567 * 568 * public void m() { 569 * assert !lock.isHeldByCurrentThread(); 570 * lock.lock(); 571 * try { 572 * // ... method body 573 * } finally { 574 * lock.unlock(); 575 * } 576 * } 577 * } 578 * </pre> 579 * 580 * @return {@code true} if current thread holds this lock and 581 * {@code false} otherwise 582 */ 583 public boolean isHeldByCurrentThread() { 584 return sync.isHeldExclusively(); 585 } 586 587 /** 588 * Queries if this lock is held by any thread. This method is 589 * designed for use in monitoring of the system state, 590 * not for synchronization control. 591 * 592 * @return {@code true} if any thread holds this lock and 593 * {@code false} otherwise 594 */ 595 public boolean isLocked() { 596 return sync.isLocked(); 597 } 598 599 /** 600 * Returns {@code true} if this lock has fairness set true. 601 * 602 * @return {@code true} if this lock has fairness set true 603 */ 604 public final boolean isFair() { 605 return sync instanceof FairSync; 606 } 607 608 /** 609 * Returns the thread that currently owns this lock, or 610 * {@code null} if not owned. When this method is called by a 611 * thread that is not the owner, the return value reflects a 612 * best-effort approximation of current lock status. For example, 613 * the owner may be momentarily {@code null} even if there are 614 * threads trying to acquire the lock but have not yet done so. 615 * This method is designed to facilitate construction of 616 * subclasses that provide more extensive lock monitoring 617 * facilities. 618 * 619 * @return the owner, or {@code null} if not owned 620 */ 621 protected Thread getOwner() { 622 return sync.getOwner(); 623 } 624 625 /** 626 * Queries whether any threads are waiting to acquire this lock. Note that 627 * because cancellations may occur at any time, a {@code true} 628 * return does not guarantee that any other thread will ever 629 * acquire this lock. This method is designed primarily for use in 630 * monitoring of the system state. 631 * 632 * @return {@code true} if there may be other threads waiting to 633 * acquire the lock 634 */ 635 public final boolean hasQueuedThreads() { 636 return sync.hasQueuedThreads(); 637 } 638 639 640 /** 641 * Queries whether the given thread is waiting to acquire this 642 * lock. Note that because cancellations may occur at any time, a 643 * {@code true} return does not guarantee that this thread 644 * will ever acquire this lock. This method is designed primarily for use 645 * in monitoring of the system state. 646 * 647 * @param thread the thread 648 * @return {@code true} if the given thread is queued waiting for this lock 649 * @throws NullPointerException if the thread is null 650 */ 651 public final boolean hasQueuedThread(Thread thread) { 652 return sync.isQueued(thread); 653 } 654 655 656 /** 657 * Returns an estimate of the number of threads waiting to 658 * acquire this lock. The value is only an estimate because the number of 659 * threads may change dynamically while this method traverses 660 * internal data structures. This method is designed for use in 661 * monitoring of the system state, not for synchronization 662 * control. 663 * 664 * @return the estimated number of threads waiting for this lock 665 */ 666 public final int getQueueLength() { 667 return sync.getQueueLength(); 668 } 669 670 /** 671 * Returns a collection containing threads that may be waiting to 672 * acquire this lock. Because the actual set of threads may change 673 * dynamically while constructing this result, the returned 674 * collection is only a best-effort estimate. The elements of the 675 * returned collection are in no particular order. This method is 676 * designed to facilitate construction of subclasses that provide 677 * more extensive monitoring facilities. 678 * 679 * @return the collection of threads 680 */ 681 protected Collection<Thread> getQueuedThreads() { 682 return sync.getQueuedThreads(); 683 } 684 685 /** 686 * Queries whether any threads are waiting on the given condition 687 * associated with this lock. Note that because timeouts and 688 * interrupts may occur at any time, a {@code true} return does 689 * not guarantee that a future {@code signal} will awaken any 690 * threads. This method is designed primarily for use in 691 * monitoring of the system state. 692 * 693 * @param condition the condition 694 * @return {@code true} if there are any waiting threads 695 * @throws IllegalMonitorStateException if this lock is not held 696 * @throws IllegalArgumentException if the given condition is 697 * not associated with this lock 698 * @throws NullPointerException if the condition is null 699 */ 700 public boolean hasWaiters(Condition condition) { 701 if (condition == null) 702 throw new NullPointerException(); 703 if (!(condition instanceof AbstractQueuedSynchronizer.ConditionObject)) 704 throw new IllegalArgumentException("not owner"); 705 return sync.hasWaiters((AbstractQueuedSynchronizer.ConditionObject)condition); 706 } 707 708 /** 709 * Returns an estimate of the number of threads waiting on the 710 * given condition associated with this lock. Note that because 711 * timeouts and interrupts may occur at any time, the estimate 712 * serves only as an upper bound on the actual number of waiters. 713 * This method is designed for use in monitoring of the system 714 * state, not for synchronization control. 715 * 716 * @param condition the condition 717 * @return the estimated number of waiting threads 718 * @throws IllegalMonitorStateException if this lock is not held 719 * @throws IllegalArgumentException if the given condition is 720 * not associated with this lock 721 * @throws NullPointerException if the condition is null 722 */ 723 public int getWaitQueueLength(Condition condition) { 724 if (condition == null) 725 throw new NullPointerException(); 726 if (!(condition instanceof AbstractQueuedSynchronizer.ConditionObject)) 727 throw new IllegalArgumentException("not owner"); 728 return sync.getWaitQueueLength((AbstractQueuedSynchronizer.ConditionObject)condition); 729 } 730 731 /** 732 * Returns a collection containing those threads that may be 733 * waiting on the given condition associated with this lock. 734 * Because the actual set of threads may change dynamically while 735 * constructing this result, the returned collection is only a 736 * best-effort estimate. The elements of the returned collection 737 * are in no particular order. This method is designed to 738 * facilitate construction of subclasses that provide more 739 * extensive condition monitoring facilities. 740 * 741 * @param condition the condition 742 * @return the collection of threads 743 * @throws IllegalMonitorStateException if this lock is not held 744 * @throws IllegalArgumentException if the given condition is 745 * not associated with this lock 746 * @throws NullPointerException if the condition is null 747 */ 748 protected Collection<Thread> getWaitingThreads(Condition condition) { 749 if (condition == null) 750 throw new NullPointerException(); 751 if (!(condition instanceof AbstractQueuedSynchronizer.ConditionObject)) 752 throw new IllegalArgumentException("not owner"); 753 return sync.getWaitingThreads((AbstractQueuedSynchronizer.ConditionObject)condition); 754 } 755 756 /** 757 * Returns a string identifying this lock, as well as its lock state. 758 * The state, in brackets, includes either the String {@code "Unlocked"} 759 * or the String {@code "Locked by"} followed by the 760 * {@linkplain Thread#getName name} of the owning thread. 761 * 762 * @return a string identifying this lock, as well as its lock state 763 */ 764 public String toString() { 765 Thread o = sync.getOwner(); 766 return super.toString() + ((o == null) ? 767 "[Unlocked]" : 768 "[Locked by thread " + o.getName() + "]"); 769 } 770 }
AQS(AbstractQueuedSynchronizer.java)
1 /* 2 * ORACLE PROPRIETARY/CONFIDENTIAL. Use is subject to license terms. 3 * 4 * 5 * 6 * 7 * 8 * 9 * 10 * 11 * 12 * 13 * 14 * 15 * 16 * 17 * 18 * 19 * 20 * 21 * 22 * 23 */ 24 25 /* 26 * 27 * 28 * 29 * 30 * 31 * Written by Doug Lea with assistance from members of JCP JSR-166 32 * Expert Group and released to the public domain, as explained at 33 * http://creativecommons.org/publicdomain/zero/1.0/ 34 */ 35 36 package java.util.concurrent.locks; 37 import java.util.*; 38 import java.util.concurrent.*; 39 import java.util.concurrent.atomic.*; 40 import sun.misc.Unsafe; 41 42 /** 43 * Provides a framework for implementing blocking locks and related 44 * synchronizers (semaphores, events, etc) that rely on 45 * first-in-first-out (FIFO) wait queues. This class is designed to 46 * be a useful basis for most kinds of synchronizers that rely on a 47 * single atomic <tt>int</tt> value to represent state. Subclasses 48 * must define the protected methods that change this state, and which 49 * define what that state means in terms of this object being acquired 50 * or released. Given these, the other methods in this class carry 51 * out all queuing and blocking mechanics. Subclasses can maintain 52 * other state fields, but only the atomically updated <tt>int</tt> 53 * value manipulated using methods {@link #getState}, {@link 54 * #setState} and {@link #compareAndSetState} is tracked with respect 55 * to synchronization. 56 * 57 * <p>Subclasses should be defined as non-public internal helper 58 * classes that are used to implement the synchronization properties 59 * of their enclosing class. Class 60 * <tt>AbstractQueuedSynchronizer</tt> does not implement any 61 * synchronization interface. Instead it defines methods such as 62 * {@link #acquireInterruptibly} that can be invoked as 63 * appropriate by concrete locks and related synchronizers to 64 * implement their public methods. 65 * 66 * <p>This class supports either or both a default <em>exclusive</em> 67 * mode and a <em>shared</em> mode. When acquired in exclusive mode, 68 * attempted acquires by other threads cannot succeed. Shared mode 69 * acquires by multiple threads may (but need not) succeed. This class 70 * does not "understand" these differences except in the 71 * mechanical sense that when a shared mode acquire succeeds, the next 72 * waiting thread (if one exists) must also determine whether it can 73 * acquire as well. Threads waiting in the different modes share the 74 * same FIFO queue. Usually, implementation subclasses support only 75 * one of these modes, but both can come into play for example in a 76 * {@link ReadWriteLock}. Subclasses that support only exclusive or 77 * only shared modes need not define the methods supporting the unused mode. 78 * 79 * <p>This class defines a nested {@link ConditionObject} class that 80 * can be used as a {@link Condition} implementation by subclasses 81 * supporting exclusive mode for which method {@link 82 * #isHeldExclusively} reports whether synchronization is exclusively 83 * held with respect to the current thread, method {@link #release} 84 * invoked with the current {@link #getState} value fully releases 85 * this object, and {@link #acquire}, given this saved state value, 86 * eventually restores this object to its previous acquired state. No 87 * <tt>AbstractQueuedSynchronizer</tt> method otherwise creates such a 88 * condition, so if this constraint cannot be met, do not use it. The 89 * behavior of {@link ConditionObject} depends of course on the 90 * semantics of its synchronizer implementation. 91 * 92 * <p>This class provides inspection, instrumentation, and monitoring 93 * methods for the internal queue, as well as similar methods for 94 * condition objects. These can be exported as desired into classes 95 * using an <tt>AbstractQueuedSynchronizer</tt> for their 96 * synchronization mechanics. 97 * 98 * <p>Serialization of this class stores only the underlying atomic 99 * integer maintaining state, so deserialized objects have empty 100 * thread queues. Typical subclasses requiring serializability will 101 * define a <tt>readObject</tt> method that restores this to a known 102 * initial state upon deserialization. 103 * 104 * <h3>Usage</h3> 105 * 106 * <p>To use this class as the basis of a synchronizer, redefine the 107 * following methods, as applicable, by inspecting and/or modifying 108 * the synchronization state using {@link #getState}, {@link 109 * #setState} and/or {@link #compareAndSetState}: 110 * 111 * <ul> 112 * <li> {@link #tryAcquire} 113 * <li> {@link #tryRelease} 114 * <li> {@link #tryAcquireShared} 115 * <li> {@link #tryReleaseShared} 116 * <li> {@link #isHeldExclusively} 117 *</ul> 118 * 119 * Each of these methods by default throws {@link 120 * UnsupportedOperationException}. Implementations of these methods 121 * must be internally thread-safe, and should in general be short and 122 * not block. Defining these methods is the <em>only</em> supported 123 * means of using this class. All other methods are declared 124 * <tt>final</tt> because they cannot be independently varied. 125 * 126 * <p>You may also find the inherited methods from {@link 127 * AbstractOwnableSynchronizer} useful to keep track of the thread 128 * owning an exclusive synchronizer. You are encouraged to use them 129 * -- this enables monitoring and diagnostic tools to assist users in 130 * determining which threads hold locks. 131 * 132 * <p>Even though this class is based on an internal FIFO queue, it 133 * does not automatically enforce FIFO acquisition policies. The core 134 * of exclusive synchronization takes the form: 135 * 136 * <pre> 137 * Acquire: 138 * while (!tryAcquire(arg)) { 139 * <em>enqueue thread if it is not already queued</em>; 140 * <em>possibly block current thread</em>; 141 * } 142 * 143 * Release: 144 * if (tryRelease(arg)) 145 * <em>unblock the first queued thread</em>; 146 * </pre> 147 * 148 * (Shared mode is similar but may involve cascading signals.) 149 * 150 * <p><a name="barging">Because checks in acquire are invoked before 151 * enqueuing, a newly acquiring thread may <em>barge</em> ahead of 152 * others that are blocked and queued. However, you can, if desired, 153 * define <tt>tryAcquire</tt> and/or <tt>tryAcquireShared</tt> to 154 * disable barging by internally invoking one or more of the inspection 155 * methods, thereby providing a <em>fair</em> FIFO acquisition order. 156 * In particular, most fair synchronizers can define <tt>tryAcquire</tt> 157 * to return <tt>false</tt> if {@link #hasQueuedPredecessors} (a method 158 * specifically designed to be used by fair synchronizers) returns 159 * <tt>true</tt>. Other variations are possible. 160 * 161 * <p>Throughput and scalability are generally highest for the 162 * default barging (also known as <em>greedy</em>, 163 * <em>renouncement</em>, and <em>convoy-avoidance</em>) strategy. 164 * While this is not guaranteed to be fair or starvation-free, earlier 165 * queued threads are allowed to recontend before later queued 166 * threads, and each recontention has an unbiased chance to succeed 167 * against incoming threads. Also, while acquires do not 168 * "spin" in the usual sense, they may perform multiple 169 * invocations of <tt>tryAcquire</tt> interspersed with other 170 * computations before blocking. This gives most of the benefits of 171 * spins when exclusive synchronization is only briefly held, without 172 * most of the liabilities when it isn‘t. If so desired, you can 173 * augment this by preceding calls to acquire methods with 174 * "fast-path" checks, possibly prechecking {@link #hasContended} 175 * and/or {@link #hasQueuedThreads} to only do so if the synchronizer 176 * is likely not to be contended. 177 * 178 * <p>This class provides an efficient and scalable basis for 179 * synchronization in part by specializing its range of use to 180 * synchronizers that can rely on <tt>int</tt> state, acquire, and 181 * release parameters, and an internal FIFO wait queue. When this does 182 * not suffice, you can build synchronizers from a lower level using 183 * {@link java.util.concurrent.atomic atomic} classes, your own custom 184 * {@link java.util.Queue} classes, and {@link LockSupport} blocking 185 * support. 186 * 187 * <h3>Usage Examples</h3> 188 * 189 * <p>Here is a non-reentrant mutual exclusion lock class that uses 190 * the value zero to represent the unlocked state, and one to 191 * represent the locked state. While a non-reentrant lock 192 * does not strictly require recording of the current owner 193 * thread, this class does so anyway to make usage easier to monitor. 194 * It also supports conditions and exposes 195 * one of the instrumentation methods: 196 * 197 * <pre> 198 * class Mutex implements Lock, java.io.Serializable { 199 * 200 * // Our internal helper class 201 * private static class Sync extends AbstractQueuedSynchronizer { 202 * // Report whether in locked state 203 * protected boolean isHeldExclusively() { 204 * return getState() == 1; 205 * } 206 * 207 * // Acquire the lock if state is zero 208 * public boolean tryAcquire(int acquires) { 209 * assert acquires == 1; // Otherwise unused 210 * if (compareAndSetState(0, 1)) { 211 * setExclusiveOwnerThread(Thread.currentThread()); 212 * return true; 213 * } 214 * return false; 215 * } 216 * 217 * // Release the lock by setting state to zero 218 * protected boolean tryRelease(int releases) { 219 * assert releases == 1; // Otherwise unused 220 * if (getState() == 0) throw new IllegalMonitorStateException(); 221 * setExclusiveOwnerThread(null); 222 * setState(0); 223 * return true; 224 * } 225 * 226 * // Provide a Condition 227 * Condition newCondition() { return new ConditionObject(); } 228 * 229 * // Deserialize properly 230 * private void readObject(ObjectInputStream s) 231 * throws IOException, ClassNotFoundException { 232 * s.defaultReadObject(); 233 * setState(0); // reset to unlocked state 234 * } 235 * } 236 * 237 * // The sync object does all the hard work. We just forward to it. 238 * private final Sync sync = new Sync(); 239 * 240 * public void lock() { sync.acquire(1); } 241 * public boolean tryLock() { return sync.tryAcquire(1); } 242 * public void unlock() { sync.release(1); } 243 * public Condition newCondition() { return sync.newCondition(); } 244 * public boolean isLocked() { return sync.isHeldExclusively(); } 245 * public boolean hasQueuedThreads() { return sync.hasQueuedThreads(); } 246 * public void lockInterruptibly() throws InterruptedException { 247 * sync.acquireInterruptibly(1); 248 * } 249 * public boolean tryLock(long timeout, TimeUnit unit) 250 * throws InterruptedException { 251 * return sync.tryAcquireNanos(1, unit.toNanos(timeout)); 252 * } 253 * } 254 * </pre> 255 * 256 * <p>Here is a latch class that is like a {@link CountDownLatch} 257 * except that it only requires a single <tt>signal</tt> to 258 * fire. Because a latch is non-exclusive, it uses the <tt>shared</tt> 259 * acquire and release methods. 260 * 261 * <pre> 262 * class BooleanLatch { 263 * 264 * private static class Sync extends AbstractQueuedSynchronizer { 265 * boolean isSignalled() { return getState() != 0; } 266 * 267 * protected int tryAcquireShared(int ignore) { 268 * return isSignalled() ? 1 : -1; 269 * } 270 * 271 * protected boolean tryReleaseShared(int ignore) { 272 * setState(1); 273 * return true; 274 * } 275 * } 276 * 277 * private final Sync sync = new Sync(); 278 * public boolean isSignalled() { return sync.isSignalled(); } 279 * public void signal() { sync.releaseShared(1); } 280 * public void await() throws InterruptedException { 281 * sync.acquireSharedInterruptibly(1); 282 * } 283 * } 284 * </pre> 285 * 286 * @since 1.5 287 * @author Doug Lea 288 */ 289 public abstract class AbstractQueuedSynchronizer 290 extends AbstractOwnableSynchronizer 291 implements java.io.Serializable { 292 293 private static final long serialVersionUID = 7373984972572414691L; 294 295 /** 296 * Creates a new <tt>AbstractQueuedSynchronizer</tt> instance 297 * with initial synchronization state of zero. 298 */ 299 protected AbstractQueuedSynchronizer() { } 300 301 /** 302 * Wait queue node class. 303 * 304 * <p>The wait queue is a variant of a "CLH" (Craig, Landin, and 305 * Hagersten) lock queue. CLH locks are normally used for 306 * spinlocks. We instead use them for blocking synchronizers, but 307 * use the same basic tactic of holding some of the control 308 * information about a thread in the predecessor of its node. A 309 * "status" field in each node keeps track of whether a thread 310 * should block. A node is signalled when its predecessor 311 * releases. Each node of the queue otherwise serves as a 312 * specific-notification-style monitor holding a single waiting 313 * thread. The status field does NOT control whether threads are 314 * granted locks etc though. A thread may try to acquire if it is 315 * first in the queue. But being first does not guarantee success; 316 * it only gives the right to contend. So the currently released 317 * contender thread may need to rewait. 318 * 319 * <p>To enqueue into a CLH lock, you atomically splice it in as new 320 * tail. To dequeue, you just set the head field. 321 * <pre> 322 * +------+ prev +-----+ +-----+ 323 * head | | <---- | | <---- | | tail 324 * +------+ +-----+ +-----+ 325 * </pre> 326 * 327 * <p>Insertion into a CLH queue requires only a single atomic 328 * operation on "tail", so there is a simple atomic point of 329 * demarcation from unqueued to queued. Similarly, dequeing 330 * involves only updating the "head". However, it takes a bit 331 * more work for nodes to determine who their successors are, 332 * in part to deal with possible cancellation due to timeouts 333 * and interrupts. 334 * 335 * <p>The "prev" links (not used in original CLH locks), are mainly 336 * needed to handle cancellation. If a node is cancelled, its 337 * successor is (normally) relinked to a non-cancelled 338 * predecessor. For explanation of similar mechanics in the case 339 * of spin locks, see the papers by Scott and Scherer at 340 * http://www.cs.rochester.edu/u/scott/synchronization/ 341 * 342 * <p>We also use "next" links to implement blocking mechanics. 343 * The thread id for each node is kept in its own node, so a 344 * predecessor signals the next node to wake up by traversing 345 * next link to determine which thread it is. Determination of 346 * successor must avoid races with newly queued nodes to set 347 * the "next" fields of their predecessors. This is solved 348 * when necessary by checking backwards from the atomically 349 * updated "tail" when a node‘s successor appears to be null. 350 * (Or, said differently, the next-links are an optimization 351 * so that we don‘t usually need a backward scan.) 352 * 353 * <p>Cancellation introduces some conservatism to the basic 354 * algorithms. Since we must poll for cancellation of other 355 * nodes, we can miss noticing whether a cancelled node is 356 * ahead or behind us. This is dealt with by always unparking 357 * successors upon cancellation, allowing them to stabilize on 358 * a new predecessor, unless we can identify an uncancelled 359 * predecessor who will carry this responsibility. 360 * 361 * <p>CLH queues need a dummy header node to get started. But 362 * we don‘t create them on construction, because it would be wasted 363 * effort if there is never contention. Instead, the node 364 * is constructed and head and tail pointers are set upon first 365 * contention. 366 * 367 * <p>Threads waiting on Conditions use the same nodes, but 368 * use an additional link. Conditions only need to link nodes 369 * in simple (non-concurrent) linked queues because they are 370 * only accessed when exclusively held. Upon await, a node is 371 * inserted into a condition queue. Upon signal, the node is 372 * transferred to the main queue. A special value of status 373 * field is used to mark which queue a node is on. 374 * 375 * <p>Thanks go to Dave Dice, Mark Moir, Victor Luchangco, Bill 376 * Scherer and Michael Scott, along with members of JSR-166 377 * expert group, for helpful ideas, discussions, and critiques 378 * on the design of this class. 379 */ 380 static final class Node { 381 /** Marker to indicate a node is waiting in shared mode */ 382 static final Node SHARED = new Node(); 383 /** Marker to indicate a node is waiting in exclusive mode */ 384 static final Node EXCLUSIVE = null; 385 386 /** waitStatus value to indicate thread has cancelled */ 387 static final int CANCELLED = 1; 388 /** waitStatus value to indicate successor‘s thread needs unparking */ 389 static final int SIGNAL = -1; 390 /** waitStatus value to indicate thread is waiting on condition */ 391 static final int CONDITION = -2; 392 /** 393 * waitStatus value to indicate the next acquireShared should 394 * unconditionally propagate 395 */ 396 static final int PROPAGATE = -3; 397 398 /** 399 * Status field, taking on only the values: 400 * SIGNAL: The successor of this node is (or will soon be) 401 * blocked (via park), so the current node must 402 * unpark its successor when it releases or 403 * cancels. To avoid races, acquire methods must 404 * first indicate they need a signal, 405 * then retry the atomic acquire, and then, 406 * on failure, block. 407 * CANCELLED: This node is cancelled due to timeout or interrupt. 408 * Nodes never leave this state. In particular, 409 * a thread with cancelled node never again blocks. 410 * CONDITION: This node is currently on a condition queue. 411 * It will not be used as a sync queue node 412 * until transferred, at which time the status 413 * will be set to 0. (Use of this value here has 414 * nothing to do with the other uses of the 415 * field, but simplifies mechanics.) 416 * PROPAGATE: A releaseShared should be propagated to other 417 * nodes. This is set (for head node only) in 418 * doReleaseShared to ensure propagation 419 * continues, even if other operations have 420 * since intervened. 421 * 0: None of the above 422 * 423 * The values are arranged numerically to simplify use. 424 * Non-negative values mean that a node doesn‘t need to 425 * signal. So, most code doesn‘t need to check for particular 426 * values, just for sign. 427 * 428 * The field is initialized to 0 for normal sync nodes, and 429 * CONDITION for condition nodes. It is modified using CAS 430 * (or when possible, unconditional volatile writes). 431 */ 432 volatile int waitStatus; 433 434 /** 435 * Link to predecessor node that current node/thread relies on 436 * for checking waitStatus. Assigned during enqueing, and nulled 437 * out (for sake of GC) only upon dequeuing. Also, upon 438 * cancellation of a predecessor, we short-circuit while 439 * finding a non-cancelled one, which will always exist 440 * because the head node is never cancelled: A node becomes 441 * head only as a result of successful acquire. A 442 * cancelled thread never succeeds in acquiring, and a thread only 443 * cancels itself, not any other node. 444 */ 445 volatile Node prev; 446 447 /** 448 * Link to the successor node that the current node/thread 449 * unparks upon release. Assigned during enqueuing, adjusted 450 * when bypassing cancelled predecessors, and nulled out (for 451 * sake of GC) when dequeued. The enq operation does not 452 * assign next field of a predecessor until after attachment, 453 * so seeing a null next field does not necessarily mean that 454 * node is at end of queue. However, if a next field appears 455 * to be null, we can scan prev‘s from the tail to 456 * double-check. The next field of cancelled nodes is set to 457 * point to the node itself instead of null, to make life 458 * easier for isOnSyncQueue. 459 */ 460 volatile Node next; 461 462 /** 463 * The thread that enqueued this node. Initialized on 464 * construction and nulled out after use. 465 */ 466 volatile Thread thread; 467 468 /** 469 * Link to next node waiting on condition, or the special 470 * value SHARED. Because condition queues are accessed only 471 * when holding in exclusive mode, we just need a simple 472 * linked queue to hold nodes while they are waiting on 473 * conditions. They are then transferred to the queue to 474 * re-acquire. And because conditions can only be exclusive, 475 * we save a field by using special value to indicate shared 476 * mode. 477 */ 478 Node nextWaiter; 479 480 /** 481 * Returns true if node is waiting in shared mode 482 */ 483 final boolean isShared() { 484 return nextWaiter == SHARED; 485 } 486 487 /** 488 * Returns previous node, or throws NullPointerException if null. 489 * Use when predecessor cannot be null. The null check could 490 * be elided, but is present to help the VM. 491 * 492 * @return the predecessor of this node 493 */ 494 final Node predecessor() throws NullPointerException { 495 Node p = prev; 496 if (p == null) 497 throw new NullPointerException(); 498 else 499 return p; 500 } 501 502 Node() { // Used to establish initial head or SHARED marker 503 } 504 505 Node(Thread thread, Node mode) { // Used by addWaiter 506 this.nextWaiter = mode; 507 this.thread = thread; 508 } 509 510 Node(Thread thread, int waitStatus) { // Used by Condition 511 this.waitStatus = waitStatus; 512 this.thread = thread; 513 } 514 } 515 516 /** 517 * Head of the wait queue, lazily initialized. Except for 518 * initialization, it is modified only via method setHead. Note: 519 * If head exists, its waitStatus is guaranteed not to be 520 * CANCELLED. 521 */ 522 private transient volatile Node head; 523 524 /** 525 * Tail of the wait queue, lazily initialized. Modified only via 526 * method enq to add new wait node. 527 */ 528 private transient volatile Node tail; 529 530 /** 531 * The synchronization state. 532 */ 533 private volatile int state; 534 535 /** 536 * Returns the current value of synchronization state. 537 * This operation has memory semantics of a <tt>volatile</tt> read. 538 * @return current state value 539 */ 540 protected final int getState() { 541 return state; 542 } 543 544 /** 545 * Sets the value of synchronization state. 546 * This operation has memory semantics of a <tt>volatile</tt> write. 547 * @param newState the new state value 548 */ 549 protected final void setState(int newState) { 550 state = newState; 551 } 552 553 /** 554 * Atomically sets synchronization state to the given updated 555 * value if the current state value equals the expected value. 556 * This operation has memory semantics of a <tt>volatile</tt> read 557 * and write. 558 * 559 * @param expect the expected value 560 * @param update the new value 561 * @return true if successful. False return indicates that the actual 562 * value was not equal to the expected value. 563 */ 564 protected final boolean compareAndSetState(int expect, int update) { 565 // See below for intrinsics setup to support this 566 return unsafe.compareAndSwapInt(this, stateOffset, expect, update); 567 } 568 569 // Queuing utilities 570 571 /** 572 * The number of nanoseconds for which it is faster to spin 573 * rather than to use timed park. A rough estimate suffices 574 * to improve responsiveness with very short timeouts. 575 */ 576 static final long spinForTimeoutThreshold = 1000L; 577 578 /** 579 * Inserts node into queue, initializing if necessary. See picture above. 580 * @param node the node to insert 581 * @return node‘s predecessor 582 */ 583 private Node enq(final Node node) { 584 for (;;) { 585 Node t = tail; 586 if (t == null) { // Must initialize 587 if (compareAndSetHead(new Node())) 588 tail = head; 589 } else { 590 node.prev = t; 591 if (compareAndSetTail(t, node)) { 592 t.next = node; 593 return t; 594 } 595 } 596 } 597 } 598 599 /** 600 * Creates and enqueues node for current thread and given mode. 601 * 602 * @param mode Node.EXCLUSIVE for exclusive, Node.SHARED for shared 603 * @return the new node 604 */ 605 private Node addWaiter(Node mode) { 606 Node node = new Node(Thread.currentThread(), mode); 607 // Try the fast path of enq; backup to full enq on failure 608 Node pred = tail; 609 if (pred != null) { 610 node.prev = pred; 611 if (compareAndSetTail(pred, node)) { 612 pred.next = node; 613 return node; 614 } 615 } 616 enq(node); 617 return node; 618 } 619 620 /** 621 * Sets head of queue to be node, thus dequeuing. Called only by 622 * acquire methods. Also nulls out unused fields for sake of GC 623 * and to suppress unnecessary signals and traversals. 624 * 625 * @param node the node 626 */ 627 private void setHead(Node node) { 628 head = node; 629 node.thread = null; 630 node.prev = null; 631 } 632 633 /** 634 * Wakes up node‘s successor, if one exists. 635 * 636 * @param node the node 637 */ 638 private void unparkSuccessor(Node node) { 639 /* 640 * If status is negative (i.e., possibly needing signal) try 641 * to clear in anticipation of signalling. It is OK if this 642 * fails or if status is changed by waiting thread. 643 */ 644 int ws = node.waitStatus; 645 if (ws < 0) 646 compareAndSetWaitStatus(node, ws, 0); 647 648 /* 649 * Thread to unpark is held in successor, which is normally 650 * just the next node. But if cancelled or apparently null, 651 * traverse backwards from tail to find the actual 652 * non-cancelled successor. 653 */ 654 Node s = node.next; 655 if (s == null || s.waitStatus > 0) { 656 s = null; 657 for (Node t = tail; t != null && t != node; t = t.prev) 658 if (t.waitStatus <= 0) 659 s = t; 660 } 661 if (s != null) 662 LockSupport.unpark(s.thread); 663 } 664 665 /** 666 * Release action for shared mode -- signal successor and ensure 667 * propagation. (Note: For exclusive mode, release just amounts 668 * to calling unparkSuccessor of head if it needs signal.) 669 */ 670 private void doReleaseShared() { 671 /* 672 * Ensure that a release propagates, even if there are other 673 * in-progress acquires/releases. This proceeds in the usual 674 * way of trying to unparkSuccessor of head if it needs 675 * signal. But if it does not, status is set to PROPAGATE to 676 * ensure that upon release, propagation continues. 677 * Additionally, we must loop in case a new node is added 678 * while we are doing this. Also, unlike other uses of 679 * unparkSuccessor, we need to know if CAS to reset status 680 * fails, if so rechecking. 681 */ 682 for (;;) { 683 Node h = head; 684 if (h != null && h != tail) { 685 int ws = h.waitStatus; 686 if (ws == Node.SIGNAL) { 687 if (!compareAndSetWaitStatus(h, Node.SIGNAL, 0)) 688 continue; // loop to recheck cases 689 unparkSuccessor(h); 690 } 691 else if (ws == 0 && 692 !compareAndSetWaitStatus(h, 0, Node.PROPAGATE)) 693 continue; // loop on failed CAS 694 } 695 if (h == head) // loop if head changed 696 break; 697 } 698 } 699 700 /** 701 * Sets head of queue, and checks if successor may be waiting 702 * in shared mode, if so propagating if either propagate > 0 or 703 * PROPAGATE status was set. 704 * 705 * @param node the node 706 * @param propagate the return value from a tryAcquireShared 707 */ 708 private void setHeadAndPropagate(Node node, int propagate) { 709 Node h = head; // Record old head for check below 710 setHead(node); 711 /* 712 * Try to signal next queued node if: 713 * Propagation was indicated by caller, 714 * or was recorded (as h.waitStatus) by a previous operation 715 * (note: this uses sign-check of waitStatus because 716 * PROPAGATE status may transition to SIGNAL.) 717 * and 718 * The next node is waiting in shared mode, 719 * or we don‘t know, because it appears null 720 * 721 * The conservatism in both of these checks may cause 722 * unnecessary wake-ups, but only when there are multiple 723 * racing acquires/releases, so most need signals now or soon 724 * anyway. 725 */ 726 if (propagate > 0 || h == null || h.waitStatus < 0) { 727 Node s = node.next; 728 if (s == null || s.isShared()) 729 doReleaseShared(); 730 } 731 } 732 733 // Utilities for various versions of acquire 734 735 /** 736 * Cancels an ongoing attempt to acquire. 737 * 738 * @param node the node 739 */ 740 private void cancelAcquire(Node node) { 741 // Ignore if node doesn‘t exist 742 if (node == null) 743 return; 744 745 node.thread = null; 746 747 // Skip cancelled predecessors 748 Node pred = node.prev; 749 while (pred.waitStatus > 0) 750 node.prev = pred = pred.prev; 751 752 // predNext is the apparent node to unsplice. CASes below will 753 // fail if not, in which case, we lost race vs another cancel 754 // or signal, so no further action is necessary. 755 Node predNext = pred.next; 756 757 // Can use unconditional write instead of CAS here. 758 // After this atomic step, other Nodes can skip past us. 759 // Before, we are free of interference from other threads. 760 node.waitStatus = Node.CANCELLED; 761 762 // If we are the tail, remove ourselves. 763 if (node == tail && compareAndSetTail(node, pred)) { 764 compareAndSetNext(pred, predNext, null); 765 } else { 766 // If successor needs signal, try to set pred‘s next-link 767 // so it will get one. Otherwise wake it up to propagate. 768 int ws; 769 if (pred != head && 770 ((ws = pred.waitStatus) == Node.SIGNAL || 771 (ws <= 0 && compareAndSetWaitStatus(pred, ws, Node.SIGNAL))) && 772 pred.thread != null) { 773 Node next = node.next; 774 if (next != null && next.waitStatus <= 0) 775 compareAndSetNext(pred, predNext, next); 776 } else { 777 unparkSuccessor(node); 778 } 779 780 node.next = node; // help GC 781 } 782 } 783 784 /** 785 * Checks and updates status for a node that failed to acquire. 786 * Returns true if thread should block. This is the main signal 787 * control in all acquire loops. Requires that pred == node.prev 788 * 789 * @param pred node‘s predecessor holding status 790 * @param node the node 791 * @return {@code true} if thread should block 792 */ 793 private static boolean shouldParkAfterFailedAcquire(Node pred, Node node) { 794 int ws = pred.waitStatus; 795 if (ws == Node.SIGNAL) 796 /* 797 * This node has already set status asking a release 798 * to signal it, so it can safely park. 799 */ 800 return true; 801 if (ws > 0) { 802 /* 803 * Predecessor was cancelled. Skip over predecessors and 804 * indicate retry. 805 */ 806 do { 807 node.prev = pred = pred.prev; 808 } while (pred.waitStatus > 0); 809 pred.next = node; 810 } else { 811 /* 812 * waitStatus must be 0 or PROPAGATE. Indicate that we 813 * need a signal, but don‘t park yet. Caller will need to 814 * retry to make sure it cannot acquire before parking. 815 */ 816 compareAndSetWaitStatus(pred, ws, Node.SIGNAL); 817 } 818 return false; 819 } 820 821 /** 822 * Convenience method to interrupt current thread. 823 */ 824 private static void selfInterrupt() { 825 Thread.currentThread().interrupt(); 826 } 827 828 /** 829 * Convenience method to park and then check if interrupted 830 * 831 * @return {@code true} if interrupted 832 */ 833 private final boolean parkAndCheckInterrupt() { 834 LockSupport.park(this); 835 return Thread.interrupted(); 836 } 837 838 /* 839 * Various flavors of acquire, varying in exclusive/shared and 840 * control modes. Each is mostly the same, but annoyingly 841 * different. Only a little bit of factoring is possible due to 842 * interactions of exception mechanics (including ensuring that we 843 * cancel if tryAcquire throws exception) and other control, at 844 * least not without hurting performance too much. 845 */ 846 847 /** 848 * Acquires in exclusive uninterruptible mode for thread already in 849 * queue. Used by condition wait methods as well as acquire. 850 * 851 * @param node the node 852 * @param arg the acquire argument 853 * @return {@code true} if interrupted while waiting 854 */ 855 final boolean acquireQueued(final Node node, int arg) { 856 boolean failed = true; 857 try { 858 boolean interrupted = false; 859 for (;;) { 860 final Node p = node.predecessor(); 861 if (p == head && tryAcquire(arg)) { 862 setHead(node); 863 p.next = null; // help GC 864 failed = false; 865 return interrupted; 866 } 867 if (shouldParkAfterFailedAcquire(p, node) && 868 parkAndCheckInterrupt()) 869 interrupted = true; 870 } 871 } finally { 872 if (failed) 873 cancelAcquire(node); 874 } 875 } 876 877 /** 878 * Acquires in exclusive interruptible mode. 879 * @param arg the acquire argument 880 */ 881 private void doAcquireInterruptibly(int arg) 882 throws InterruptedException { 883 final Node node = addWaiter(Node.EXCLUSIVE); 884 boolean failed = true; 885 try { 886 for (;;) { 887 final Node p = node.predecessor(); 888 if (p == head && tryAcquire(arg)) { 889 setHead(node); 890 p.next = null; // help GC 891 failed = false; 892 return; 893 } 894 if (shouldParkAfterFailedAcquire(p, node) && 895 parkAndCheckInterrupt()) 896 throw new InterruptedException(); 897 } 898 } finally { 899 if (failed) 900 cancelAcquire(node); 901 } 902 } 903 904 /** 905 * Acquires in exclusive timed mode. 906 * 907 * @param arg the acquire argument 908 * @param nanosTimeout max wait time 909 * @return {@code true} if acquired 910 */ 911 private boolean doAcquireNanos(int arg, long nanosTimeout) 912 throws InterruptedException { 913 long lastTime = System.nanoTime(); 914 final Node node = addWaiter(Node.EXCLUSIVE); 915 boolean failed = true; 916 try { 917 for (;;) { 918 final Node p = node.predecessor(); 919 if (p == head && tryAcquire(arg)) { 920 setHead(node); 921 p.next = null; // help GC 922 failed = false; 923 return true; 924 } 925 if (nanosTimeout <= 0) 926 return false; 927 if (shouldParkAfterFailedAcquire(p, node) && 928 nanosTimeout > spinForTimeoutThreshold) 929 LockSupport.parkNanos(this, nanosTimeout); 930 long now = System.nanoTime(); 931 nanosTimeout -= now - lastTime; 932 lastTime = now; 933 if (Thread.interrupted()) 934 throw new InterruptedException(); 935 } 936 } finally { 937 if (failed) 938 cancelAcquire(node); 939 } 940 } 941 942 /** 943 * Acquires in shared uninterruptible mode. 944 * @param arg the acquire argument 945 */ 946 private void doAcquireShared(int arg) { 947 final Node node = addWaiter(Node.SHARED); 948 boolean failed = true; 949 try { 950 boolean interrupted = false; 951 for (;;) { 952 final Node p = node.predecessor(); 953 if (p == head) { 954 int r = tryAcquireShared(arg); 955 if (r >= 0) { 956 setHeadAndPropagate(node, r); 957 p.next = null; // help GC 958 if (interrupted) 959 selfInterrupt(); 960 failed = false; 961 return; 962 } 963 } 964 if (shouldParkAfterFailedAcquire(p, node) && 965 parkAndCheckInterrupt()) 966 interrupted = true; 967 } 968 } finally { 969 if (failed) 970 cancelAcquire(node); 971 } 972 } 973 974 /** 975 * Acquires in shared interruptible mode. 976 * @param arg the acquire argument 977 */ 978 private void doAcquireSharedInterruptibly(int arg) 979 throws InterruptedException { 980 final Node node = addWaiter(Node.SHARED); 981 boolean failed = true; 982 try { 983 for (;;) { 984 final Node p = node.predecessor(); 985 if (p == head) { 986 int r = tryAcquireShared(arg); 987 if (r >= 0) { 988 setHeadAndPropagate(node, r); 989 p.next = null; // help GC 990 failed = false; 991 return; 992 } 993 } 994 if (shouldParkAfterFailedAcquire(p, node) && 995 parkAndCheckInterrupt()) 996 throw new InterruptedException(); 997 } 998 } finally { 999 if (failed) 1000 cancelAcquire(node); 1001 } 1002 } 1003 1004 /** 1005 * Acquires in shared timed mode. 1006 * 1007 * @param arg the acquire argument 1008 * @param nanosTimeout max wait time 1009 * @return {@code true} if acquired 1010 */ 1011 private boolean doAcquireSharedNanos(int arg, long nanosTimeout) 1012 throws InterruptedException { 1013 1014 long lastTime = System.nanoTime(); 1015 final Node node = addWaiter(Node.SHARED); 1016 boolean failed = true; 1017 try { 1018 for (;;) { 1019 final Node p = node.predecessor(); 1020 if (p == head) { 1021 int r = tryAcquireShared(arg); 1022 if (r >= 0) { 1023 setHeadAndPropagate(node, r); 1024 p.next = null; // help GC 1025 failed = false; 1026 return true; 1027 } 1028 } 1029 if (nanosTimeout <= 0) 1030 return false; 1031 if (shouldParkAfterFailedAcquire(p, node) && 1032 nanosTimeout > spinForTimeoutThreshold) 1033 LockSupport.parkNanos(this, nanosTimeout); 1034 long now = System.nanoTime(); 1035 nanosTimeout -= now - lastTime; 1036 lastTime = now; 1037 if (Thread.interrupted()) 1038 throw new InterruptedException(); 1039 } 1040 } finally { 1041 if (failed) 1042 cancelAcquire(node); 1043 } 1044 } 1045 1046 // Main exported methods 1047 1048 /** 1049 * Attempts to acquire in exclusive mode. This method should query 1050 * if the state of the object permits it to be acquired in the 1051 * exclusive mode, and if so to acquire it. 1052 * 1053 * <p>This method is always invoked by the thread performing 1054 * acquire. If this method reports failure, the acquire method 1055 * may queue the thread, if it is not already queued, until it is 1056 * signalled by a release from some other thread. This can be used 1057 * to implement method {@link Lock#tryLock()}. 1058 * 1059 * <p>The default 1060 * implementation throws {@link UnsupportedOperationException}. 1061 * 1062 * @param arg the acquire argument. This value is always the one 1063 * passed to an acquire method, or is the value saved on entry 1064 * to a condition wait. The value is otherwise uninterpreted 1065 * and can represent anything you like. 1066 * @return {@code true} if successful. Upon success, this object has 1067 * been acquired. 1068 * @throws IllegalMonitorStateException if acquiring would place this 1069 * synchronizer in an illegal state. This exception must be 1070 * thrown in a consistent fashion for synchronization to work 1071 * correctly. 1072 * @throws UnsupportedOperationException if exclusive mode is not supported 1073 */ 1074 protected boolean tryAcquire(int arg) { 1075 throw new UnsupportedOperationException(); 1076 } 1077 1078 /** 1079 * Attempts to set the state to reflect a release in exclusive 1080 * mode. 1081 * 1082 * <p>This method is always invoked by the thread performing release. 1083 * 1084 * <p>The default implementation throws 1085 * {@link UnsupportedOperationException}. 1086 * 1087 * @param arg the release argument. This value is always the one 1088 * passed to a release method, or the current state value upon 1089 * entry to a condition wait. The value is otherwise 1090 * uninterpreted and can represent anything you like. 1091 * @return {@code true} if this object is now in a fully released 1092 * state, so that any waiting threads may attempt to acquire; 1093 * and {@code false} otherwise. 1094 * @throws IllegalMonitorStateException if releasing would place this 1095 * synchronizer in an illegal state. This exception must be 1096 * thrown in a consistent fashion for synchronization to work 1097 * correctly. 1098 * @throws UnsupportedOperationException if exclusive mode is not supported 1099 */ 1100 protected boolean tryRelease(int arg) { 1101 throw new UnsupportedOperationException(); 1102 } 1103 1104 /** 1105 * Attempts to acquire in shared mode. This method should query if 1106 * the state of the object permits it to be acquired in the shared 1107 * mode, and if so to acquire it. 1108 * 1109 * <p>This method is always invoked by the thread performing 1110 * acquire. If this method reports failure, the acquire method 1111 * may queue the thread, if it is not already queued, until it is 1112 * signalled by a release from some other thread. 1113 * 1114 * <p>The default implementation throws {@link 1115 * UnsupportedOperationException}. 1116 * 1117 * @param arg the acquire argument. This value is always the one 1118 * passed to an acquire method, or is the value saved on entry 1119 * to a condition wait. The value is otherwise uninterpreted 1120 * and can represent anything you like. 1121 * @return a negative value on failure; zero if acquisition in shared 1122 * mode succeeded but no subsequent shared-mode acquire can 1123 * succeed; and a positive value if acquisition in shared 1124 * mode succeeded and subsequent shared-mode acquires might 1125 * also succeed, in which case a subsequent waiting thread 1126 * must check availability. (Support for three different 1127 * return values enables this method to be used in contexts 1128 * where acquires only sometimes act exclusively.) Upon 1129 * success, this object has been acquired. 1130 * @throws IllegalMonitorStateException if acquiring would place this 1131 * synchronizer in an illegal state. This exception must be 1132 * thrown in a consistent fashion for synchronization to work 1133 * correctly. 1134 * @throws UnsupportedOperationException if shared mode is not supported 1135 */ 1136 protected int tryAcquireShared(int arg) { 1137 throw new UnsupportedOperationException(); 1138 } 1139 1140 /** 1141 * Attempts to set the state to reflect a release in shared mode. 1142 * 1143 * <p>This method is always invoked by the thread performing release. 1144 * 1145 * <p>The default implementation throws 1146 * {@link UnsupportedOperationException}. 1147 * 1148 * @param arg the release argument. This value is always the one 1149 * passed to a release method, or the current state value upon 1150 * entry to a condition wait. The value is otherwise 1151 * uninterpreted and can represent anything you like. 1152 * @return {@code true} if this release of shared mode may permit a 1153 * waiting acquire (shared or exclusive) to succeed; and 1154 * {@code false} otherwise 1155 * @throws IllegalMonitorStateException if releasing would place this 1156 * synchronizer in an illegal state. This exception must be 1157 * thrown in a consistent fashion for synchronization to work 1158 * correctly. 1159 * @throws UnsupportedOperationException if shared mode is not supported 1160 */ 1161 protected boolean tryReleaseShared(int arg) { 1162 throw new UnsupportedOperationException(); 1163 } 1164 1165 /** 1166 * Returns {@code true} if synchronization is held exclusively with 1167 * respect to the current (calling) thread. This method is invoked 1168 * upon each call to a non-waiting {@link ConditionObject} method. 1169 * (Waiting methods instead invoke {@link #release}.) 1170 * 1171 * <p>The default implementation throws {@link 1172 * UnsupportedOperationException}. This method is invoked 1173 * internally only within {@link ConditionObject} methods, so need 1174 * not be defined if conditions are not used. 1175 * 1176 * @return {@code true} if synchronization is held exclusively; 1177 * {@code false} otherwise 1178 * @throws UnsupportedOperationException if conditions are not supported 1179 */ 1180 protected boolean isHeldExclusively() { 1181 throw new UnsupportedOperationException(); 1182 } 1183 1184 /** 1185 * Acquires in exclusive mode, ignoring interrupts. Implemented 1186 * by invoking at least once {@link #tryAcquire}, 1187 * returning on success. Otherwise the thread is queued, possibly 1188 * repeatedly blocking and unblocking, invoking {@link 1189 * #tryAcquire} until success. This method can be used 1190 * to implement method {@link Lock#lock}. 1191 * 1192 * @param arg the acquire argument. This value is conveyed to 1193 * {@link #tryAcquire} but is otherwise uninterpreted and 1194 * can represent anything you like. 1195 */ 1196 public final void acquire(int arg) { 1197 if (!tryAcquire(arg) && 1198 acquireQueued(addWaiter(Node.EXCLUSIVE), arg)) 1199 selfInterrupt(); 1200 } 1201 1202 /** 1203 * Acquires in exclusive mode, aborting if interrupted. 1204 * Implemented by first checking interrupt status, then invoking 1205 * at least once {@link #tryAcquire}, returning on 1206 * success. Otherwise the thread is queued, possibly repeatedly 1207 * blocking and unblocking, invoking {@link #tryAcquire} 1208 * until success or the thread is interrupted. This method can be 1209 * used to implement method {@link Lock#lockInterruptibly}. 1210 * 1211 * @param arg the acquire argument. This value is conveyed to 1212 * {@link #tryAcquire} but is otherwise uninterpreted and 1213 * can represent anything you like. 1214 * @throws InterruptedException if the current thread is interrupted 1215 */ 1216 public final void acquireInterruptibly(int arg) 1217 throws InterruptedException { 1218 if (Thread.interrupted()) 1219 throw new InterruptedException(); 1220 if (!tryAcquire(arg)) 1221 doAcquireInterruptibly(arg); 1222 } 1223 1224 /** 1225 * Attempts to acquire in exclusive mode, aborting if interrupted, 1226 * and failing if the given timeout elapses. Implemented by first 1227 * checking interrupt status, then invoking at least once {@link 1228 * #tryAcquire}, returning on success. Otherwise, the thread is 1229 * queued, possibly repeatedly blocking and unblocking, invoking 1230 * {@link #tryAcquire} until success or the thread is interrupted 1231 * or the timeout elapses. This method can be used to implement 1232 * method {@link Lock#tryLock(long, TimeUnit)}. 1233 * 1234 * @param arg the acquire argument. This value is conveyed to 1235 * {@link #tryAcquire} but is otherwise uninterpreted and 1236 * can represent anything you like. 1237 * @param nanosTimeout the maximum number of nanoseconds to wait 1238 * @return {@code true} if acquired; {@code false} if timed out 1239 * @throws InterruptedException if the current thread is interrupted 1240 */ 1241 public final boolean tryAcquireNanos(int arg, long nanosTimeout) 1242 throws InterruptedException { 1243 if (Thread.interrupted()) 1244 throw new InterruptedException(); 1245 return tryAcquire(arg) || 1246 doAcquireNanos(arg, nanosTimeout); 1247 } 1248 1249 /** 1250 * Releases in exclusive mode. Implemented by unblocking one or 1251 * more threads if {@link #tryRelease} returns true. 1252 * This method can be used to implement method {@link Lock#unlock}. 1253 * 1254 * @param arg the release argument. This value is conveyed to 1255 * {@link #tryRelease} but is otherwise uninterpreted and 1256 * can represent anything you like. 1257 * @return the value returned from {@link #tryRelease} 1258 */ 1259 public final boolean release(int arg) { 1260 if (tryRelease(arg)) { 1261 Node h = head; 1262 if (h != null && h.waitStatus != 0) 1263 unparkSuccessor(h); 1264 return true; 1265 } 1266 return false; 1267 } 1268 1269 /** 1270 * Acquires in shared mode, ignoring interrupts. Implemented by 1271 * first invoking at least once {@link #tryAcquireShared}, 1272 * returning on success. Otherwise the thread is queued, possibly 1273 * repeatedly blocking and unblocking, invoking {@link 1274 * #tryAcquireShared} until success. 1275 * 1276 * @param arg the acquire argument. This value is conveyed to 1277 * {@link #tryAcquireShared} but is otherwise uninterpreted 1278 * and can represent anything you like. 1279 */ 1280 public final void acquireShared(int arg) { 1281 if (tryAcquireShared(arg) < 0) 1282 doAcquireShared(arg); 1283 } 1284 1285 /** 1286 * Acquires in shared mode, aborting if interrupted. Implemented 1287 * by first checking interrupt status, then invoking at least once 1288 * {@link #tryAcquireShared}, returning on success. Otherwise the 1289 * thread is queued, possibly repeatedly blocking and unblocking, 1290 * invoking {@link #tryAcquireShared} until success or the thread 1291 * is interrupted. 1292 * @param arg the acquire argument 1293 * This value is conveyed to {@link #tryAcquireShared} but is 1294 * otherwise uninterpreted and can represent anything 1295 * you like. 1296 * @throws InterruptedException if the current thread is interrupted 1297 */ 1298 public final void acquireSharedInterruptibly(int arg) 1299 throws InterruptedException { 1300 if (Thread.interrupted()) 1301 throw new InterruptedException(); 1302 if (tryAcquireShared(arg) < 0) 1303 doAcquireSharedInterruptibly(arg); 1304 } 1305 1306 /** 1307 * Attempts to acquire in shared mode, aborting if interrupted, and 1308 * failing if the given timeout elapses. Implemented by first 1309 * checking interrupt status, then invoking at least once {@link 1310 * #tryAcquireShared}, returning on success. Otherwise, the 1311 * thread is queued, possibly repeatedly blocking and unblocking, 1312 * invoking {@link #tryAcquireShared} until success or the thread 1313 * is interrupted or the timeout elapses. 1314 * 1315 * @param arg the acquire argument. This value is conveyed to 1316 * {@link #tryAcquireShared} but is otherwise uninterpreted 1317 * and can represent anything you like. 1318 * @param nanosTimeout the maximum number of nanoseconds to wait 1319 * @return {@code true} if acquired; {@code false} if timed out 1320 * @throws InterruptedException if the current thread is interrupted 1321 */ 1322 public final boolean tryAcquireSharedNanos(int arg, long nanosTimeout) 1323 throws InterruptedException { 1324 if (Thread.interrupted()) 1325 throw new InterruptedException(); 1326 return tryAcquireShared(arg) >= 0 || 1327 doAcquireSharedNanos(arg, nanosTimeout); 1328 } 1329 1330 /** 1331 * Releases in shared mode. Implemented by unblocking one or more 1332 * threads if {@link #tryReleaseShared} returns true. 1333 * 1334 * @param arg the release argument. This value is conveyed to 1335 * {@link #tryReleaseShared} but is otherwise uninterpreted 1336 * and can represent anything you like. 1337 * @return the value returned from {@link #tryReleaseShared} 1338 */ 1339 public final boolean releaseShared(int arg) { 1340 if (tryReleaseShared(arg)) { 1341 doReleaseShared(); 1342 return true; 1343 } 1344 return false; 1345 } 1346 1347 // Queue inspection methods 1348 1349 /** 1350 * Queries whether any threads are waiting to acquire. Note that 1351 * because cancellations due to interrupts and timeouts may occur 1352 * at any time, a {@code true} return does not guarantee that any 1353 * other thread will ever acquire. 1354 * 1355 * <p>In this implementation, this operation returns in 1356 * constant time. 1357 * 1358 * @return {@code true} if there may be other threads waiting to acquire 1359 */ 1360 public final boolean hasQueuedThreads() { 1361 return head != tail; 1362 } 1363 1364 /** 1365 * Queries whether any threads have ever contended to acquire this 1366 * synchronizer; that is if an acquire method has ever blocked. 1367 * 1368 * <p>In this implementation, this operation returns in 1369 * constant time. 1370 * 1371 * @return {@code true} if there has ever been contention 1372 */ 1373 public final boolean hasContended() { 1374 return head != null; 1375 } 1376 1377 /** 1378 * Returns the first (longest-waiting) thread in the queue, or 1379 * {@code null} if no threads are currently queued. 1380 * 1381 * <p>In this implementation, this operation normally returns in 1382 * constant time, but may iterate upon contention if other threads are 1383 * concurrently modifying the queue. 1384 * 1385 * @return the first (longest-waiting) thread in the queue, or 1386 * {@code null} if no threads are currently queued 1387 */ 1388 public final Thread getFirstQueuedThread() { 1389 // handle only fast path, else relay 1390 return (head == tail) ? null : fullGetFirstQueuedThread(); 1391 } 1392 1393 /** 1394 * Version of getFirstQueuedThread called when fastpath fails 1395 */ 1396 private Thread fullGetFirstQueuedThread() { 1397 /* 1398 * The first node is normally head.next. Try to get its 1399 * thread field, ensuring consistent reads: If thread 1400 * field is nulled out or s.prev is no longer head, then 1401 * some other thread(s) concurrently performed setHead in 1402 * between some of our reads. We try this twice before 1403 * resorting to traversal. 1404 */ 1405 Node h, s; 1406 Thread st; 1407 if (((h = head) != null && (s = h.next) != null && 1408 s.prev == head && (st = s.thread) != null) || 1409 ((h = head) != null && (s = h.next) != null && 1410 s.prev == head && (st = s.thread) != null)) 1411 return st; 1412 1413 /* 1414 * Head‘s next field might not have been set yet, or may have 1415 * been unset after setHead. So we must check to see if tail 1416 * is actually first node. If not, we continue on, safely 1417 * traversing from tail back to head to find first, 1418 * guaranteeing termination. 1419 */ 1420 1421 Node t = tail; 1422 Thread firstThread = null; 1423 while (t != null && t != head) { 1424 Thread tt = t.thread; 1425 if (tt != null) 1426 firstThread = tt; 1427 t = t.prev; 1428 } 1429 return firstThread; 1430 } 1431 1432 /** 1433 * Returns true if the given thread is currently queued. 1434 * 1435 * <p>This implementation traverses the queue to determine 1436 * presence of the given thread. 1437 * 1438 * @param thread the thread 1439 * @return {@code true} if the given thread is on the queue 1440 * @throws NullPointerException if the thread is null 1441 */ 1442 public final boolean isQueued(Thread thread) { 1443 if (thread == null) 1444 throw new NullPointerException(); 1445 for (Node p = tail; p != null; p = p.prev) 1446 if (p.thread == thread) 1447 return true; 1448 return false; 1449 } 1450 1451 /** 1452 * Returns {@code true} if the apparent first queued thread, if one 1453 * exists, is waiting in exclusive mode. If this method returns 1454 * {@code true}, and the current thread is attempting to acquire in 1455 * shared mode (that is, this method is invoked from {@link 1456 * #tryAcquireShared}) then it is guaranteed that the current thread 1457 * is not the first queued thread. Used only as a heuristic in 1458 * ReentrantReadWriteLock. 1459 */ 1460 final boolean apparentlyFirstQueuedIsExclusive() { 1461 Node h, s; 1462 return (h = head) != null && 1463 (s = h.next) != null && 1464 !s.isShared() && 1465 s.thread != null; 1466 } 1467 1468 /** 1469 * Queries whether any threads have been waiting to acquire longer 1470 * than the current thread. 1471 * 1472 * <p>An invocation of this method is equivalent to (but may be 1473 * more efficient than): 1474 * <pre> {@code 1475 * getFirstQueuedThread() != Thread.currentThread() && 1476 * hasQueuedThreads()}</pre> 1477 * 1478 * <p>Note that because cancellations due to interrupts and 1479 * timeouts may occur at any time, a {@code true} return does not 1480 * guarantee that some other thread will acquire before the current 1481 * thread. Likewise, it is possible for another thread to win a 1482 * race to enqueue after this method has returned {@code false}, 1483 * due to the queue being empty. 1484 * 1485 * <p>This method is designed to be used by a fair synchronizer to 1486 * avoid <a href="AbstractQueuedSynchronizer#barging">barging</a>. 1487 * Such a synchronizer‘s {@link #tryAcquire} method should return 1488 * {@code false}, and its {@link #tryAcquireShared} method should 1489 * return a negative value, if this method returns {@code true} 1490 * (unless this is a reentrant acquire). For example, the {@code 1491 * tryAcquire} method for a fair, reentrant, exclusive mode 1492 * synchronizer might look like this: 1493 * 1494 * <pre> {@code 1495 * protected boolean tryAcquire(int arg) { 1496 * if (isHeldExclusively()) { 1497 * // A reentrant acquire; increment hold count 1498 * return true; 1499 * } else if (hasQueuedPredecessors()) { 1500 * return false; 1501 * } else { 1502 * // try to acquire normally 1503 * } 1504 * }}</pre> 1505 * 1506 * @return {@code true} if there is a queued thread preceding the 1507 * current thread, and {@code false} if the current thread 1508 * is at the head of the queue or the queue is empty 1509 * @since 1.7 1510 */ 1511 public final boolean hasQueuedPredecessors() { 1512 // The correctness of this depends on head being initialized 1513 // before tail and on head.next being accurate if the current 1514 // thread is first in queue. 1515 Node t = tail; // Read fields in reverse initialization order 1516 Node h = head; 1517 Node s; 1518 return h != t && 1519 ((s = h.next) == null || s.thread != Thread.currentThread()); 1520 } 1521 1522 1523 // Instrumentation and monitoring methods 1524 1525 /** 1526 * Returns an estimate of the number of threads waiting to 1527 * acquire. The value is only an estimate because the number of 1528 * threads may change dynamically while this method traverses 1529 * internal data structures. This method is designed for use in 1530 * monitoring system state, not for synchronization 1531 * control. 1532 * 1533 * @return the estimated number of threads waiting to acquire 1534 */ 1535 public final int getQueueLength() { 1536 int n = 0; 1537 for (Node p = tail; p != null; p = p.prev) { 1538 if (p.thread != null) 1539 ++n; 1540 } 1541 return n; 1542 } 1543 1544 /** 1545 * Returns a collection containing threads that may be waiting to 1546 * acquire. Because the actual set of threads may change 1547 * dynamically while constructing this result, the returned 1548 * collection is only a best-effort estimate. The elements of the 1549 * returned collection are in no particular order. This method is 1550 * designed to facilitate construction of subclasses that provide 1551 * more extensive monitoring facilities. 1552 * 1553 * @return the collection of threads 1554 */ 1555 public final Collection<Thread> getQueuedThreads() { 1556 ArrayList<Thread> list = new ArrayList<Thread>(); 1557 for (Node p = tail; p != null; p = p.prev) { 1558 Thread t = p.thread; 1559 if (t != null) 1560 list.add(t); 1561 } 1562 return list; 1563 } 1564 1565 /** 1566 * Returns a collection containing threads that may be waiting to 1567 * acquire in exclusive mode. This has the same properties 1568 * as {@link #getQueuedThreads} except that it only returns 1569 * those threads waiting due to an exclusive acquire. 1570 * 1571 * @return the collection of threads 1572 */ 1573 public final Collection<Thread> getExclusiveQueuedThreads() { 1574 ArrayList<Thread> list = new ArrayList<Thread>(); 1575 for (Node p = tail; p != null; p = p.prev) { 1576 if (!p.isShared()) { 1577 Thread t = p.thread; 1578 if (t != null) 1579 list.add(t); 1580 } 1581 } 1582 return list; 1583 } 1584 1585 /** 1586 * Returns a collection containing threads that may be waiting to 1587 * acquire in shared mode. This has the same properties 1588 * as {@link #getQueuedThreads} except that it only returns 1589 * those threads waiting due to a shared acquire. 1590 * 1591 * @return the collection of threads 1592 */ 1593 public final Collection<Thread> getSharedQueuedThreads() { 1594 ArrayList<Thread> list = new ArrayList<Thread>(); 1595 for (Node p = tail; p != null; p = p.prev) { 1596 if (p.isShared()) { 1597 Thread t = p.thread; 1598 if (t != null) 1599 list.add(t); 1600 } 1601 } 1602 return list; 1603 } 1604 1605 /** 1606 * Returns a string identifying this synchronizer, as well as its state. 1607 * The state, in brackets, includes the String {@code "State ="} 1608 * followed by the current value of {@link #getState}, and either 1609 * {@code "nonempty"} or {@code "empty"} depending on whether the 1610 * queue is empty. 1611 * 1612 * @return a string identifying this synchronizer, as well as its state 1613 */ 1614 public String toString() { 1615 int s = getState(); 1616 String q = hasQueuedThreads() ? "non" : ""; 1617 return super.toString() + 1618 "[State = " + s + ", " + q + "empty queue]"; 1619 } 1620 1621 1622 // Internal support methods for Conditions 1623 1624 /** 1625 * Returns true if a node, always one that was initially placed on 1626 * a condition queue, is now waiting to reacquire on sync queue. 1627 * @param node the node 1628 * @return true if is reacquiring 1629 */ 1630 final boolean isOnSyncQueue(Node node) { 1631 if (node.waitStatus == Node.CONDITION || node.prev == null) 1632 return false; 1633 if (node.next != null) // If has successor, it must be on queue 1634 return true; 1635 /* 1636 * node.prev can be non-null, but not yet on queue because 1637 * the CAS to place it on queue can fail. So we have to 1638 * traverse from tail to make sure it actually made it. It 1639 * will always be near the tail in calls to this method, and 1640 * unless the CAS failed (which is unlikely), it will be 1641 * there, so we hardly ever traverse much. 1642 */ 1643 return findNodeFromTail(node); 1644 } 1645 1646 /** 1647 * Returns true if node is on sync queue by searching backwards from tail. 1648 * Called only when needed by isOnSyncQueue. 1649 * @return true if present 1650 */ 1651 private boolean findNodeFromTail(Node node) { 1652 Node t = tail; 1653 for (;;) { 1654 if (t == node) 1655 return true; 1656 if (t == null) 1657 return false; 1658 t = t.prev; 1659 } 1660 } 1661 1662 /** 1663 * Transfers a node from a condition queue onto sync queue. 1664 * Returns true if successful. 1665 * @param node the node 1666 * @return true if successfully transferred (else the node was 1667 * cancelled before signal). 1668 */ 1669 final boolean transferForSignal(Node node) { 1670 /* 1671 * If cannot change waitStatus, the node has been cancelled. 1672 */ 1673 if (!compareAndSetWaitStatus(node, Node.CONDITION, 0)) 1674 return false; 1675 1676 /* 1677 * Splice onto queue and try to set waitStatus of predecessor to 1678 * indicate that thread is (probably) waiting. If cancelled or 1679 * attempt to set waitStatus fails, wake up to resync (in which 1680 * case the waitStatus can be transiently and harmlessly wrong). 1681 */ 1682 Node p = enq(node); 1683 int ws = p.waitStatus; 1684 if (ws > 0 || !compareAndSetWaitStatus(p, ws, Node.SIGNAL)) 1685 LockSupport.unpark(node.thread); 1686 return true; 1687 } 1688 1689 /** 1690 * Transfers node, if necessary, to sync queue after a cancelled 1691 * wait. Returns true if thread was cancelled before being 1692 * signalled. 1693 * @param current the waiting thread 1694 * @param node its node 1695 * @return true if cancelled before the node was signalled 1696 */ 1697 final boolean transferAfterCancelledWait(Node node) { 1698 if (compareAndSetWaitStatus(node, Node.CONDITION, 0)) { 1699 enq(node); 1700 return true; 1701 } 1702 /* 1703 * If we lost out to a signal(), then we can‘t proceed 1704 * until it finishes its enq(). Cancelling during an 1705 * incomplete transfer is both rare and transient, so just 1706 * spin. 1707 */ 1708 while (!isOnSyncQueue(node)) 1709 Thread.yield(); 1710 return false; 1711 } 1712 1713 /** 1714 * Invokes release with current state value; returns saved state. 1715 * Cancels node and throws exception on failure. 1716 * @param node the condition node for this wait 1717 * @return previous sync state 1718 */ 1719 final int fullyRelease(Node node) { 1720 boolean failed = true; 1721 try { 1722 int savedState = getState(); 1723 if (release(savedState)) { 1724 failed = false; 1725 return savedState; 1726 } else { 1727 throw new IllegalMonitorStateException(); 1728 } 1729 } finally { 1730 if (failed) 1731 node.waitStatus = Node.CANCELLED; 1732 } 1733 } 1734 1735 // Instrumentation methods for conditions 1736 1737 /** 1738 * Queries whether the given ConditionObject 1739 * uses this synchronizer as its lock. 1740 * 1741 * @param condition the condition 1742 * @return <tt>true</tt> if owned 1743 * @throws NullPointerException if the condition is null 1744 */ 1745 public final boolean owns(ConditionObject condition) { 1746 if (condition == null) 1747 throw new NullPointerException(); 1748 return condition.isOwnedBy(this); 1749 } 1750 1751 /** 1752 * Queries whether any threads are waiting on the given condition 1753 * associated with this synchronizer. Note that because timeouts 1754 * and interrupts may occur at any time, a <tt>true</tt> return 1755 * does not guarantee that a future <tt>signal</tt> will awaken 1756 * any threads. This method is designed primarily for use in 1757 * monitoring of the system state. 1758 * 1759 * @param condition the condition 1760 * @return <tt>true</tt> if there are any waiting threads 1761 * @throws IllegalMonitorStateException if exclusive synchronization 1762 * is not held 1763 * @throws IllegalArgumentException if the given condition is 1764 * not associated with this synchronizer 1765 * @throws NullPointerException if the condition is null 1766 */ 1767 public final boolean hasWaiters(ConditionObject condition) { 1768 if (!owns(condition)) 1769 throw new IllegalArgumentException("Not owner"); 1770 return condition.hasWaiters(); 1771 } 1772 1773 /** 1774 * Returns an estimate of the number of threads waiting on the 1775 * given condition associated with this synchronizer. Note that 1776 * because timeouts and interrupts may occur at any time, the 1777 * estimate serves only as an upper bound on the actual number of 1778 * waiters. This method is designed for use in monitoring of the 1779 * system state, not for synchronization control. 1780 * 1781 * @param condition the condition 1782 * @return the estimated number of waiting threads 1783 * @throws IllegalMonitorStateException if exclusive synchronization 1784 * is not held 1785 * @throws IllegalArgumentException if the given condition is 1786 * not associated with this synchronizer 1787 * @throws NullPointerException if the condition is null 1788 */ 1789 public final int getWaitQueueLength(ConditionObject condition) { 1790 if (!owns(condition)) 1791 throw new IllegalArgumentException("Not owner"); 1792 return condition.getWaitQueueLength(); 1793 } 1794 1795 /** 1796 * Returns a collection containing those threads that may be 1797 * waiting on the given condition associated with this 1798 * synchronizer. Because the actual set of threads may change 1799 * dynamically while constructing this result, the returned 1800 * collection is only a best-effort estimate. The elements of the 1801 * returned collection are in no particular order. 1802 * 1803 * @param condition the condition 1804 * @return the collection of threads 1805 * @throws IllegalMonitorStateException if exclusive synchronization 1806 * is not held 1807 * @throws IllegalArgumentException if the given condition is 1808 * not associated with this synchronizer 1809 * @throws NullPointerException if the condition is null 1810 */ 1811 public final Collection<Thread> getWaitingThreads(ConditionObject condition) { 1812 if (!owns(condition)) 1813 throw new IllegalArgumentException("Not owner"); 1814 return condition.getWaitingThreads(); 1815 } 1816 1817 /** 1818 * Condition implementation for a {@link 1819 * AbstractQueuedSynchronizer} serving as the basis of a {@link 1820 * Lock} implementation. 1821 * 1822 * <p>Method documentation for this class describes mechanics, 1823 * not behavioral specifications from the point of view of Lock 1824 * and Condition users. Exported versions of this class will in 1825 * general need to be accompanied by documentation describing 1826 * condition semantics that rely on those of the associated 1827 * <tt>AbstractQueuedSynchronizer</tt>. 1828 * 1829 * <p>This class is Serializable, but all fields are transient, 1830 * so deserialized conditions have no waiters. 1831 */ 1832 public class ConditionObject implements Condition, java.io.Serializable { 1833 private static final long serialVersionUID = 1173984872572414699L; 1834 /** First node of condition queue. */ 1835 private transient Node firstWaiter; 1836 /** Last node of condition queue. */ 1837 private transient Node lastWaiter; 1838 1839 /** 1840 * Creates a new <tt>ConditionObject</tt> instance. 1841 */ 1842 public ConditionObject() { } 1843 1844 // Internal methods 1845 1846 /** 1847 * Adds a new waiter to wait queue. 1848 * @return its new wait node 1849 */ 1850 private Node addConditionWaiter() { 1851 Node t = lastWaiter; 1852 // If lastWaiter is cancelled, clean out. 1853 if (t != null && t.waitStatus != Node.CONDITION) { 1854 unlinkCancelledWaiters(); 1855 t = lastWaiter; 1856 } 1857 Node node = new Node(Thread.currentThread(), Node.CONDITION); 1858 if (t == null) 1859 firstWaiter = node; 1860 else 1861 t.nextWaiter = node; 1862 lastWaiter = node; 1863 return node; 1864 } 1865 1866 /** 1867 * Removes and transfers nodes until hit non-cancelled one or 1868 * null. Split out from signal in part to encourage compilers 1869 * to inline the case of no waiters. 1870 * @param first (non-null) the first node on condition queue 1871 */ 1872 private void doSignal(Node first) { 1873 do { 1874 if ( (firstWaiter = first.nextWaiter) == null) 1875 lastWaiter = null; 1876 first.nextWaiter = null; 1877 } while (!transferForSignal(first) && 1878 (first = firstWaiter) != null); 1879 } 1880 1881 /** 1882 * Removes and transfers all nodes. 1883 * @param first (non-null) the first node on condition queue 1884 */ 1885 private void doSignalAll(Node first) { 1886 lastWaiter = firstWaiter = null; 1887 do { 1888 Node next = first.nextWaiter; 1889 first.nextWaiter = null; 1890 transferForSignal(first); 1891 first = next; 1892 } while (first != null); 1893 } 1894 1895 /** 1896 * Unlinks cancelled waiter nodes from condition queue. 1897 * Called only while holding lock. This is called when 1898 * cancellation occurred during condition wait, and upon 1899 * insertion of a new waiter when lastWaiter is seen to have 1900 * been cancelled. This method is needed to avoid garbage 1901 * retention in the absence of signals. So even though it may 1902 * require a full traversal, it comes into play only when 1903 * timeouts or cancellations occur in the absence of 1904 * signals. It traverses all nodes rather than stopping at a 1905 * particular target to unlink all pointers to garbage nodes 1906 * without requiring many re-traversals during cancellation 1907 * storms. 1908 */ 1909 private void unlinkCancelledWaiters() { 1910 Node t = firstWaiter; 1911 Node trail = null; 1912 while (t != null) { 1913 Node next = t.nextWaiter; 1914 if (t.waitStatus != Node.CONDITION) { 1915 t.nextWaiter = null; 1916 if (trail == null) 1917 firstWaiter = next; 1918 else 1919 trail.nextWaiter = next; 1920 if (next == null) 1921 lastWaiter = trail; 1922 } 1923 else 1924 trail = t; 1925 t = next; 1926 } 1927 } 1928 1929 // public methods 1930 1931 /** 1932 * Moves the longest-waiting thread, if one exists, from the 1933 * wait queue for this condition to the wait queue for the 1934 * owning lock. 1935 * 1936 * @throws IllegalMonitorStateException if {@link #isHeldExclusively} 1937 * returns {@code false} 1938 */ 1939 public final void signal() { 1940 if (!isHeldExclusively()) 1941 throw new IllegalMonitorStateException(); 1942 Node first = firstWaiter; 1943 if (first != null) 1944 doSignal(first); 1945 } 1946 1947 /** 1948 * Moves all threads from the wait queue for this condition to 1949 * the wait queue for the owning lock. 1950 * 1951 * @throws IllegalMonitorStateException if {@link #isHeldExclusively} 1952 * returns {@code false} 1953 */ 1954 public final void signalAll() { 1955 if (!isHeldExclusively()) 1956 throw new IllegalMonitorStateException(); 1957 Node first = firstWaiter; 1958 if (first != null) 1959 doSignalAll(first); 1960 } 1961 1962 /** 1963 * Implements uninterruptible condition wait. 1964 * <ol> 1965 * <li> Save lock state returned by {@link #getState}. 1966 * <li> Invoke {@link #release} with 1967 * saved state as argument, throwing 1968 * IllegalMonitorStateException if it fails. 1969 * <li> Block until signalled. 1970 * <li> Reacquire by invoking specialized version of 1971 * {@link #acquire} with saved state as argument. 1972 * </ol> 1973 */ 1974 public final void awaitUninterruptibly() { 1975 Node node = addConditionWaiter(); 1976 int savedState = fullyRelease(node); 1977 boolean interrupted = false; 1978 while (!isOnSyncQueue(node)) { 1979 LockSupport.park(this); 1980 if (Thread.interrupted()) 1981 interrupted = true; 1982 } 1983 if (acquireQueued(node, savedState) || interrupted) 1984 selfInterrupt(); 1985 } 1986 1987 /* 1988 * For interruptible waits, we need to track whether to throw 1989 * InterruptedException, if interrupted while blocked on 1990 * condition, versus reinterrupt current thread, if 1991 * interrupted while blocked waiting to re-acquire. 1992 */ 1993 1994 /** Mode meaning to reinterrupt on exit from wait */ 1995 private static final int REINTERRUPT = 1; 1996 /** Mode meaning to throw InterruptedException on exit from wait */ 1997 private static final int THROW_IE = -1; 1998 1999 /** 2000 * Checks for interrupt, returning THROW_IE if interrupted 2001 * before signalled, REINTERRUPT if after signalled, or 2002 * 0 if not interrupted. 2003 */ 2004 private int checkInterruptWhileWaiting(Node node) { 2005 return Thread.interrupted() ? 2006 (transferAfterCancelledWait(node) ? THROW_IE : REINTERRUPT) : 2007 0; 2008 } 2009 2010 /** 2011 * Throws InterruptedException, reinterrupts current thread, or 2012 * does nothing, depending on mode. 2013 */ 2014 private void reportInterruptAfterWait(int interruptMode) 2015 throws InterruptedException { 2016 if (interruptMode == THROW_IE) 2017 throw new InterruptedException(); 2018 else if (interruptMode == REINTERRUPT) 2019 selfInterrupt(); 2020 } 2021 2022 /** 2023 * Implements interruptible condition wait. 2024 * <ol> 2025 * <li> If current thread is interrupted, throw InterruptedException. 2026 * <li> Save lock state returned by {@link #getState}. 2027 * <li> Invoke {@link #release} with 2028 * saved state as argument, throwing 2029 * IllegalMonitorStateException if it fails. 2030 * <li> Block until signalled or interrupted. 2031 * <li> Reacquire by invoking specialized version of 2032 * {@link #acquire} with saved state as argument. 2033 * <li> If interrupted while blocked in step 4, throw InterruptedException. 2034 * </ol> 2035 */ 2036 public final void await() throws InterruptedException { 2037 if (Thread.interrupted()) 2038 throw new InterruptedException(); 2039 Node node = addConditionWaiter(); 2040 int savedState = fullyRelease(node); 2041 int interruptMode = 0; 2042 while (!isOnSyncQueue(node)) { 2043 LockSupport.park(this); 2044 if ((interruptMode = checkInterruptWhileWaiting(node)) != 0) 2045 break; 2046 } 2047 if (acquireQueued(node, savedState) && interruptMode != THROW_IE) 2048 interruptMode = REINTERRUPT; 2049 if (node.nextWaiter != null) // clean up if cancelled 2050 unlinkCancelledWaiters(); 2051 if (interruptMode != 0) 2052 reportInterruptAfterWait(interruptMode); 2053 } 2054 2055 /** 2056 * Implements timed condition wait. 2057 * <ol> 2058 * <li> If current thread is interrupted, throw InterruptedException. 2059 * <li> Save lock state returned by {@link #getState}. 2060 * <li> Invoke {@link #release} with 2061 * saved state as argument, throwing 2062 * IllegalMonitorStateException if it fails. 2063 * <li> Block until signalled, interrupted, or timed out. 2064 * <li> Reacquire by invoking specialized version of 2065 * {@link #acquire} with saved state as argument. 2066 * <li> If interrupted while blocked in step 4, throw InterruptedException. 2067 * </ol> 2068 */ 2069 public final long awaitNanos(long nanosTimeout) 2070 throws InterruptedException { 2071 if (Thread.interrupted()) 2072 throw new InterruptedException(); 2073 Node node = addConditionWaiter(); 2074 int savedState = fullyRelease(node); 2075 long lastTime = System.nanoTime(); 2076 int interruptMode = 0; 2077 while (!isOnSyncQueue(node)) { 2078 if (nanosTimeout <= 0L) { 2079 transferAfterCancelledWait(node); 2080 break; 2081 } 2082 LockSupport.parkNanos(this, nanosTimeout); 2083 if ((interruptMode = checkInterruptWhileWaiting(node)) != 0) 2084 break; 2085 2086 long now = System.nanoTime(); 2087 nanosTimeout -= now - lastTime; 2088 lastTime = now; 2089 } 2090 if (acquireQueued(node, savedState) && interruptMode != THROW_IE) 2091 interruptMode = REINTERRUPT; 2092 if (node.nextWaiter != null) 2093 unlinkCancelledWaiters(); 2094 if (interruptMode != 0) 2095 reportInterruptAfterWait(interruptMode); 2096 return nanosTimeout - (System.nanoTime() - lastTime); 2097 } 2098 2099 /** 2100 * Implements absolute timed condition wait. 2101 * <ol> 2102 * <li> If current thread is interrupted, throw InterruptedException. 2103 * <li> Save lock state returned by {@link #getState}. 2104 * <li> Invoke {@link #release} with 2105 * saved state as argument, throwing 2106 * IllegalMonitorStateException if it fails. 2107 * <li> Block until signalled, interrupted, or timed out. 2108 * <li> Reacquire by invoking specialized version of 2109 * {@link #acquire} with saved state as argument. 2110 * <li> If interrupted while blocked in step 4, throw InterruptedException. 2111 * <li> If timed out while blocked in step 4, return false, else true. 2112 * </ol> 2113 */ 2114 public final boolean awaitUntil(Date deadline) 2115 throws InterruptedException { 2116 if (deadline == null) 2117 throw new NullPointerException(); 2118 long abstime = deadline.getTime(); 2119 if (Thread.interrupted()) 2120 throw new InterruptedException(); 2121 Node node = addConditionWaiter(); 2122 int savedState = fullyRelease(node); 2123 boolean timedout = false; 2124 int interruptMode = 0; 2125 while (!isOnSyncQueue(node)) { 2126 if (System.currentTimeMillis() > abstime) { 2127 timedout = transferAfterCancelledWait(node); 2128 break; 2129 } 2130 LockSupport.parkUntil(this, abstime); 2131 if ((interruptMode = checkInterruptWhileWaiting(node)) != 0) 2132 break; 2133 } 2134 if (acquireQueued(node, savedState) && interruptMode != THROW_IE) 2135 interruptMode = REINTERRUPT; 2136 if (node.nextWaiter != null) 2137 unlinkCancelledWaiters(); 2138 if (interruptMode != 0) 2139 reportInterruptAfterWait(interruptMode); 2140 return !timedout; 2141 } 2142 2143 /** 2144 * Implements timed condition wait. 2145 * <ol> 2146 * <li> If current thread is interrupted, throw InterruptedException. 2147 * <li> Save lock state returned by {@link #getState}. 2148 * <li> Invoke {@link #release} with 2149 * saved state as argument, throwing 2150 * IllegalMonitorStateException if it fails. 2151 * <li> Block until signalled, interrupted, or timed out. 2152 * <li> Reacquire by invoking specialized version of 2153 * {@link #acquire} with saved state as argument. 2154 * <li> If interrupted while blocked in step 4, throw InterruptedException. 2155 * <li> If timed out while blocked in step 4, return false, else true. 2156 * </ol> 2157 */ 2158 public final boolean await(long time, TimeUnit unit) 2159 throws InterruptedException { 2160 if (unit == null) 2161 throw new NullPointerException(); 2162 long nanosTimeout = unit.toNanos(time); 2163 if (Thread.interrupted()) 2164 throw new InterruptedException(); 2165 Node node = addConditionWaiter(); 2166 int savedState = fullyRelease(node); 2167 long lastTime = System.nanoTime(); 2168 boolean timedout = false; 2169 int interruptMode = 0; 2170 while (!isOnSyncQueue(node)) { 2171 if (nanosTimeout <= 0L) { 2172 timedout = transferAfterCancelledWait(node); 2173 break; 2174 } 2175 if (nanosTimeout >= spinForTimeoutThreshold) 2176 LockSupport.parkNanos(this, nanosTimeout); 2177 if ((interruptMode = checkInterruptWhileWaiting(node)) != 0) 2178 break; 2179 long now = System.nanoTime(); 2180 nanosTimeout -= now - lastTime; 2181 lastTime = now; 2182 } 2183 if (acquireQueued(node, savedState) && interruptMode != THROW_IE) 2184 interruptMode = REINTERRUPT; 2185 if (node.nextWaiter != null) 2186 unlinkCancelledWaiters(); 2187 if (interruptMode != 0) 2188 reportInterruptAfterWait(interruptMode); 2189 return !timedout; 2190 } 2191 2192 // support for instrumentation 2193 2194 /** 2195 * Returns true if this condition was created by the given 2196 * synchronization object. 2197 * 2198 * @return {@code true} if owned 2199 */ 2200 final boolean isOwnedBy(AbstractQueuedSynchronizer sync) { 2201 return sync == AbstractQueuedSynchronizer.this; 2202 } 2203 2204 /** 2205 * Queries whether any threads are waiting on this condition. 2206 * Implements {@link AbstractQueuedSynchronizer#hasWaiters}. 2207 * 2208 * @return {@code true} if there are any waiting threads 2209 * @throws IllegalMonitorStateException if {@link #isHeldExclusively} 2210 * returns {@code false} 2211 */ 2212 protected final boolean hasWaiters() { 2213 if (!isHeldExclusively()) 2214 throw new IllegalMonitorStateException(); 2215 for (Node w = firstWaiter; w != null; w = w.nextWaiter) { 2216 if (w.waitStatus == Node.CONDITION) 2217 return true; 2218 } 2219 return false; 2220 } 2221 2222 /** 2223 * Returns an estimate of the number of threads waiting on 2224 * this condition. 2225 * Implements {@link AbstractQueuedSynchronizer#getWaitQueueLength}. 2226 * 2227 * @return the estimated number of waiting threads 2228 * @throws IllegalMonitorStateException if {@link #isHeldExclusively} 2229 * returns {@code false} 2230 */ 2231 protected final int getWaitQueueLength() { 2232 if (!isHeldExclusively()) 2233 throw new IllegalMonitorStateException(); 2234 int n = 0; 2235 for (Node w = firstWaiter; w != null; w = w.nextWaiter) { 2236 if (w.waitStatus == Node.CONDITION) 2237 ++n; 2238 } 2239 return n; 2240 } 2241 2242 /** 2243 * Returns a collection containing those threads that may be 2244 * waiting on this Condition. 2245 * Implements {@link AbstractQueuedSynchronizer#getWaitingThreads}. 2246 * 2247 * @return the collection of threads 2248 * @throws IllegalMonitorStateException if {@link #isHeldExclusively} 2249 * returns {@code false} 2250 */ 2251 protected final Collection<Thread> getWaitingThreads() { 2252 if (!isHeldExclusively()) 2253 throw new IllegalMonitorStateException(); 2254 ArrayList<Thread> list = new ArrayList<Thread>(); 2255 for (Node w = firstWaiter; w != null; w = w.nextWaiter) { 2256 if (w.waitStatus == Node.CONDITION) { 2257 Thread t = w.thread; 2258 if (t != null) 2259 list.add(t); 2260 } 2261 } 2262 return list; 2263 } 2264 } 2265 2266 /** 2267 * Setup to support compareAndSet. We need to natively implement 2268 * this here: For the sake of permitting future enhancements, we 2269 * cannot explicitly subclass AtomicInteger, which would be 2270 * efficient and useful otherwise. So, as the lesser of evils, we 2271 * natively implement using hotspot intrinsics API. And while we 2272 * are at it, we do the same for other CASable fields (which could 2273 * otherwise be done with atomic field updaters). 2274 */ 2275 private static final Unsafe unsafe = Unsafe.getUnsafe(); 2276 private static final long stateOffset; 2277 private static final long headOffset; 2278 private static final long tailOffset; 2279 private static final long waitStatusOffset; 2280 private static final long nextOffset; 2281 2282 static { 2283 try { 2284 stateOffset = unsafe.objectFieldOffset 2285 (AbstractQueuedSynchronizer.class.getDeclaredField("state")); 2286 headOffset = unsafe.objectFieldOffset 2287 (AbstractQueuedSynchronizer.class.getDeclaredField("head")); 2288 tailOffset = unsafe.objectFieldOffset 2289 (AbstractQueuedSynchronizer.class.getDeclaredField("tail")); 2290 waitStatusOffset = unsafe.objectFieldOffset 2291 (Node.class.getDeclaredField("waitStatus")); 2292 nextOffset = unsafe.objectFieldOffset 2293 (Node.class.getDeclaredField("next")); 2294 2295 } catch (Exception ex) { throw new Error(ex); } 2296 } 2297 2298 /** 2299 * CAS head field. Used only by enq. 2300 */ 2301 private final boolean compareAndSetHead(Node update) { 2302 return unsafe.compareAndSwapObject(this, headOffset, null, update); 2303 } 2304 2305 /** 2306 * CAS tail field. Used only by enq. 2307 */ 2308 private final boolean compareAndSetTail(Node expect, Node update) { 2309 return unsafe.compareAndSwapObject(this, tailOffset, expect, update); 2310 } 2311 2312 /** 2313 * CAS waitStatus field of a node. 2314 */ 2315 private static final boolean compareAndSetWaitStatus(Node node, 2316 int expect, 2317 int update) { 2318 return unsafe.compareAndSwapInt(node, waitStatusOffset, 2319 expect, update); 2320 } 2321 2322 /** 2323 * CAS next field of a node. 2324 */ 2325 private static final boolean compareAndSetNext(Node node, 2326 Node expect, 2327 Node update) { 2328 return unsafe.compareAndSwapObject(node, nextOffset, expect, update); 2329 } 2330 }
获取公平锁(基于JDK1.7.0_40)
通过前面“Java多线程系列--“JUC锁”02之 互斥锁ReentrantLock”的“示例1”,我们知道,获取锁是通过lock()函数。下面,我们以lock()对获取公平锁的过程进行展开。
1. lock()
lock()在ReentrantLock.java的FairSync类中实现,它的源码如下:
final void lock() { acquire(1); }
说明:“当前线程”实际上是通过acquire(1)获取锁的。
这里说明一下“1”的含义,它是设置“锁的状态”的参数。对于“独占锁”而言,锁处于可获取状态时,它的状态值是0;锁被线程初次获取到了,它的状态值就变成了1。
由于ReentrantLock(公平锁/非公平锁)是可重入锁,所以“独占锁”可以被单个线程多此获取,每获取1次就将锁的状态+1。也就是说,初次获取锁时,通过acquire(1)将锁的状态值设为1;再次获取锁时,将锁的状态值设为2;依次类推...这就是为什么获取锁时,传入的参数是1的原因了。
可重入就是指锁可以被单个线程多次获取。
2. acquire()
acquire()在AQS中实现的,它的源码如下:
public final void acquire(int arg) { if (!tryAcquire(arg) && acquireQueued(addWaiter(Node.EXCLUSIVE), arg)) selfInterrupt(); }
(01)
“当前线程”首先通过tryAcquire()尝试获取锁。获取成功的话,直接返回;尝试失败的话,进入到等待队列排序等待(前面还有可能有需要线程在等待该锁)。
(02)
“当前线程”尝试失败的情况下,先通过addWaiter(Node.EXCLUSIVE)来将“当前线程”加入到"CLH队列(非阻塞的FIFO队列)"末尾。CLH队列就是线程等待队列。
(03)
再执行完addWaiter(Node.EXCLUSIVE)之后,会调用acquireQueued()来获取锁。由于此时ReentrantLock是公平锁,它会根据公平性原则来获取锁。
(04)
“当前线程”在执行acquireQueued()时,会进入到CLH队列中休眠等待,直到获取锁了才返回!如果“当前线程”在休眠等待过程中被中断过,acquireQueued会返回true,此时"当前线程"会调用selfInterrupt()来自己给自己产生一个中断。至于为什么要自己给自己产生一个中断,后面再介绍。
上面是对acquire()的概括性说明。下面,我们将该函数分为4部分来逐步解析。
一.
tryAcquire()
二.
addWaiter()
三.
acquireQueued()
四.
selfInterrupt()
一. tryAcquire()
1. tryAcquire()
公平锁的tryAcquire()在ReentrantLock.java的FairSync类中实现,源码如下:
protected final boolean tryAcquire(int acquires) { // 获取“当前线程” final Thread current = Thread.currentThread(); // 获取“独占锁”的状态 int c = getState(); // c=0意味着“锁没有被任何线程锁拥有”, if (c == 0) { // 若“锁没有被任何线程锁拥有”, // 则判断“当前线程”是不是CLH队列中的第一个线程线程, // 若是的话,则获取该锁,设置锁的状态,并切设置锁的拥有者为“当前线程”。 if (!hasQueuedPredecessors() && compareAndSetState(0, acquires)) { setExclusiveOwnerThread(current); return true; } } else if (current == getExclusiveOwnerThread()) { // 如果“独占锁”的拥有者已经为“当前线程”, // 则将更新锁的状态。 int nextc = c + acquires; if (nextc < 0) throw new Error("Maximum lock count exceeded"); setState(nextc); return true; } return false; }
说明:根据代码,我们可以分析出,tryAcquire()的作用就是尝试去获取锁。注意,这里只是尝试!
尝试成功的话,返回true;尝试失败的话,返回false,后续再通过其它办法来获取该锁。后面我们会说明,在尝试失败的情况下,是如何一步步获取锁的。
2. hasQueuedPredecessors()
hasQueuedPredecessors()在AQS中实现,源码如下:
public final boolean hasQueuedPredecessors() { Node t = tail; Node h = head; Node s; return h != t && ((s = h.next) == null || s.thread != Thread.currentThread()); }
说明: 通过代码,能分析出,hasQueuedPredecessors() 是通过判断"当前线程"是不是在CLH队列的队首,来返回AQS中是不是有比“当前线程”等待更久的线程。下面对head、tail和Node进行说明。
3. Node的源码
Node就是CLH队列的节点。Node在AQS中实现,它的数据结构如下:
private transient volatile Node head; // CLH队列的队首 private transient volatile Node tail; // CLH队列的队尾 // CLH队列的节点 static final class Node { static final Node SHARED = new Node(); static final Node EXCLUSIVE = null; // 线程已被取消,对应的waitStatus的值 static final int CANCELLED = 1; // “当前线程的后继线程需要被unpark(唤醒)”,对应的waitStatus的值。 // 一般发生情况是:当前线程的后继线程处于阻塞状态,而当前线程被release或cancel掉,因此需要唤醒当前线程的后继线程。 static final int SIGNAL = -1; // 线程(处在Condition休眠状态)在等待Condition唤醒,对应的waitStatus的值 static final int CONDITION = -2; // (共享锁)其它线程获取到“共享锁”,对应的waitStatus的值 static final int PROPAGATE = -3; // waitStatus为“CANCELLED, SIGNAL, CONDITION, PROPAGATE”时分别表示不同状态, // 若waitStatus=0,则意味着当前线程不属于上面的任何一种状态。 volatile int waitStatus; // 前一节点 volatile Node prev; // 后一节点 volatile Node next; // 节点所对应的线程 volatile Thread thread; // nextWaiter是“区别当前CLH队列是 ‘独占锁’队列 还是 ‘共享锁’队列 的标记” // 若nextWaiter=SHARED,则CLH队列是“独占锁”队列; // 若nextWaiter=EXCLUSIVE,(即nextWaiter=null),则CLH队列是“共享锁”队列。 Node nextWaiter; // “共享锁”则返回true,“独占锁”则返回false。 final boolean isShared() { return nextWaiter == SHARED; } // 返回前一节点 final Node predecessor() throws NullPointerException { Node p = prev; if (p == null) throw new NullPointerException(); else return p; } Node() { // Used to establish initial head or SHARED marker } // 构造函数。thread是节点所对应的线程,mode是用来表示thread的锁是“独占锁”还是“共享锁”。 Node(Thread thread, Node mode) { // Used by addWaiter this.nextWaiter = mode; this.thread = thread; } // 构造函数。thread是节点所对应的线程,waitStatus是线程的等待状态。 Node(Thread thread, int waitStatus) { // Used by Condition this.waitStatus = waitStatus; this.thread = thread; } }
说明:
Node是CLH队列的节点,代表“等待锁的线程队列”。
(01) 每个Node都会一个线程对应。
(02)
每个Node会通过prev和next分别指向上一个节点和下一个节点,这分别代表上一个等待线程和下一个等待线程。
(03)
Node通过waitStatus保存线程的等待状态。
(04)
Node通过nextWaiter来区分线程是“独占锁”线程还是“共享锁”线程。如果是“独占锁”线程,则nextWaiter的值为EXCLUSIVE;如果是“共享锁”线程,则nextWaiter的值是SHARED。
4. compareAndSetState()
compareAndSetState()在AQS中实现。它的源码如下:
protected final boolean compareAndSetState(int expect, int update) { return unsafe.compareAndSwapInt(this, stateOffset, expect, update); }
说明: compareAndSwapInt() 是sun.misc.Unsafe类中的一个本地方法。对此,我们需要了解的是 compareAndSetState(expect, update) 是以原子的方式操作当前线程;若当前线程的状态为expect,则设置它的状态为update。
5. setExclusiveOwnerThread()
setExclusiveOwnerThread()在AbstractOwnableSynchronizer.java中实现,它的源码如下:
// exclusiveOwnerThread是当前拥有“独占锁”的线程 private transient Thread exclusiveOwnerThread; protected final void setExclusiveOwnerThread(Thread t) { exclusiveOwnerThread = t; }
说明:setExclusiveOwnerThread()的作用就是,设置线程t为当前拥有“独占锁”的线程。
6. getState(), setState()
getState()和setState()都在AQS中实现,源码如下:
// 锁的状态 private volatile int state; // 设置锁的状态 protected final void setState(int newState) { state = newState; } // 获取锁的状态 protected final int getState() { return state; }
说明:state表示锁的状态,对于“独占锁”而已,state=0表示锁是可获取状态(即,锁没有被任何线程锁持有)。由于java中的独占锁是可重入的,state的值可以>1。
小结:tryAcquire()的作用就是让“当前线程”尝试获取锁。获取成功返回true,失败则返回false。
二. addWaiter(Node.EXCLUSIVE)
addWaiter(Node.EXCLUSIVE)的作用是,创建“当前线程”的Node节点,且Node中记录“当前线程”对应的锁是“独占锁”类型,并且将该节点添加到CLH队列的末尾。
1.addWaiter()
addWaiter()在AQS中实现,源码如下:
private Node addWaiter(Node mode) { // 新建一个Node节点,节点对应的线程是“当前线程”,“当前线程”的锁的模型是mode。 Node node = new Node(Thread.currentThread(), mode); Node pred = tail; // 若CLH队列不为空,则将“当前线程”添加到CLH队列末尾 if (pred != null) { node.prev = pred; if (compareAndSetTail(pred, node)) { pred.next = node; return node; } } // 若CLH队列为空,则调用enq()新建CLH队列,然后再将“当前线程”添加到CLH队列中。 enq(node); return node; }
说明:对于“公平锁”而言,addWaiter(Node.EXCLUSIVE)会首先创建一个Node节点,节点的类型是“独占锁”(Node.EXCLUSIVE)类型。然后,再将该节点添加到CLH队列的末尾。
2. compareAndSetTail()
compareAndSetTail()在AQS中实现,源码如下:
private final boolean compareAndSetTail(Node expect, Node update) { return unsafe.compareAndSwapObject(this, tailOffset, expect, update); }
说明:compareAndSetTail也属于CAS函数,也是通过“本地方法”实现的。compareAndSetTail(expect, update)会以原子的方式进行操作,它的作用是判断CLH队列的队尾是不是为expect,是的话,就将队尾设为update。
3. enq()
enq()在AQS中实现,源码如下:
private Node enq(final Node node) { for (;;) { Node t = tail; if (t == null) { // Must initialize if (compareAndSetHead(new Node())) tail = head; } else { node.prev = t; if (compareAndSetTail(t, node)) { t.next = node; return t; } } } }
说明: enq()的作用很简单。如果CLH队列为空,则新建一个CLH表头;然后将node添加到CLH末尾。否则,直接将node添加到CLH末尾。
小结:addWaiter()的作用,就是将当前线程添加到CLH队列中。这就意味着将当前线程添加到等待获取“锁”的等待线程队列中了。
三. acquireQueued()
前面,我们已经将当前线程添加到CLH队列中了。而acquireQueued()的作用就是逐步的去执行CLH队列的线程,如果当前线程获取到了锁,则返回;否则,当前线程进行休眠,直到唤醒并重新获取锁了才返回。下面,我们看看acquireQueued()的具体流程。
1. acquireQueued()
acquireQueued()在AQS中实现,源码如下:
final boolean acquireQueued(final Node node, int arg) { boolean failed = true; try { // interrupted表示在CLH队列的调度中, // “当前线程”在休眠时,有没有被中断过。 boolean interrupted = false; for (;;) { // 获取上一个节点。 // node是“当前线程”对应的节点,这里就意味着“获取上一个等待锁的线程”。 final Node p = node.predecessor(); if (p == head && tryAcquire(arg)) { setHead(node); p.next = null; // help GC failed = false; return interrupted; } if (shouldParkAfterFailedAcquire(p, node) && parkAndCheckInterrupt()) interrupted = true; } } finally { if (failed) cancelAcquire(node); } }
说明:acquireQueued()的目的是从队列中获取锁。
2. shouldParkAfterFailedAcquire()
shouldParkAfterFailedAcquire()在AQS中实现,源码如下:
// 返回“当前线程是否应该阻塞” private static boolean shouldParkAfterFailedAcquire(Node pred, Node node) { // 前继节点的状态 int ws = pred.waitStatus; // 如果前继节点是SIGNAL状态,则意味这当前线程需要被unpark唤醒。此时,返回true。 if (ws == Node.SIGNAL) return true; // 如果前继节点是“取消”状态,则设置 “当前节点”的 “当前前继节点” 为 “‘原前继节点’的前继节点”。 if (ws > 0) { do { node.prev = pred = pred.prev; } while (pred.waitStatus > 0); pred.next = node; } else { // 如果前继节点为“0”或者“共享锁”状态,则设置前继节点为SIGNAL状态。 compareAndSetWaitStatus(pred, ws, Node.SIGNAL); } return false; }
说明:
(01)
关于waitStatus请参考下表(中扩号内为waitStatus的值),更多关于waitStatus的内容,可以参考前面的Node类的介绍。
CANCELLED[1] -- 当前线程已被取消
SIGNAL[-1] -- “当前线程的后继线程需要被unpark(唤醒)”。一般发生情况是:当前线程的后继线程处于阻塞状态,而当前线程被release或cancel掉,因此需要唤醒当前线程的后继线程。
CONDITION[-2] -- 当前线程(处在Condition休眠状态)在等待Condition唤醒
PROPAGATE[-3] -- (共享锁)其它线程获取到“共享锁”
[0] -- 当前线程不属于上面的任何一种状态。
(02) shouldParkAfterFailedAcquire()通过以下规则,判断“当前线程”是否需要被阻塞。
规则1:如果前继节点状态为SIGNAL,表明当前节点需要被unpark(唤醒),此时则返回true。
规则2:如果前继节点状态为CANCELLED(ws>0),说明前继节点已经被取消,则通过先前回溯找到一个有效(非CANCELLED状态)的节点,并返回false。
规则3:如果前继节点状态为非SIGNAL、非CANCELLED,则设置前继的状态为SIGNAL,并返回false。
如果“规则1”发生,即“前继节点是SIGNAL”状态,则意味着“当前线程”需要被阻塞。接下来会调用parkAndCheckInterrupt()阻塞当前线程,直到当前先被唤醒才从parkAndCheckInterrupt()中返回。
3. parkAndCheckInterrupt())
parkAndCheckInterrupt()在AQS中实现,源码如下:
private final boolean parkAndCheckInterrupt() { // 通过LockSupport的park()阻塞“当前线程”。 LockSupport.park(this); // 返回线程的中断状态。 return Thread.interrupted(); }
说明:parkAndCheckInterrupt()的作用是阻塞当前线程,并且返回“线程被唤醒之后”的中断状态。
它会先通过LockSupport.park()阻塞“当前线程”,然后通过Thread.interrupted()返回线程的中断状态。
这里介绍一下线程被阻塞之后如何唤醒。一般有2种情况:
第1种情况:unpark()唤醒。“前继节点对应的线程”使用完锁之后,通过unpark()方式唤醒当前线程。
第2种情况:中断唤醒。其它线程通过interrupt()中断当前线程。
补充:LockSupport()中的park(),unpark()的作用 和
Object中的wait(),notify()作用类似,是阻塞/唤醒。
它们的用法不同,park(),unpark()是轻量级的,而wait(),notify()是必须先通过Synchronized获取同步锁。
关于LockSupport,我们会在之后的章节再专门进行介绍!
4. 再次tryAcquire()
了解了shouldParkAfterFailedAcquire()和parkAndCheckInterrupt()函数之后。我们接着分析acquireQueued()的for循环部分。
final Node p = node.predecessor(); if (p == head && tryAcquire(arg)) { setHead(node); p.next = null; // help GC failed = false; return interrupted; }
说明:
(01)
通过node.predecessor()获取前继节点。predecessor()就是返回node的前继节点,若对此有疑惑可以查看下面关于Node类的介绍。
(02) p == head &&
tryAcquire(arg)
首先,判断“前继节点”是不是CHL表头。如果是的话,则通过tryAcquire()尝试获取锁。
其实,这样做的目的是为了“让当前线程获取锁”,但是为什么需要先判断p==head呢?理解这个对理解“公平锁”的机制很重要,因为这么做的原因就是为了保证公平性!
(a)
前面,我们在shouldParkAfterFailedAcquire()我们判断“当前线程”是否需要阻塞;
(b)
接着,“当前线程”阻塞的话,会调用parkAndCheckInterrupt()来阻塞线程。当线程被解除阻塞的时候,我们会返回线程的中断状态。而线程被解决阻塞,可能是由于“线程被中断”,也可能是由于“其它线程调用了该线程的unpark()函数”。
(c)
再回到p==head这里。如果当前线程是因为其它线程调用了unpark()函数而被唤醒,那么唤醒它的线程,应该是它的前继节点所对应的线程(关于这一点,后面在“释放锁”的过程中会看到)。
OK,是前继节点调用unpark()唤醒了当前线程!
此时,再来理解p==head就很简单了:当前继节点是CLH队列的头节点,并且它释放锁之后;就轮到当前节点获取锁了。然后,当前节点通过tryAcquire()获取锁;获取成功的话,通过setHead(node)设置当前节点为头节点,并返回。
总之,如果“前继节点调用unpark()唤醒了当前线程”并且“前继节点是CLH表头”,此时就是满足p==head,也就是符合公平性原则的。否则,如果当前线程是因为“线程被中断”而唤醒,那么显然就不是公平了。这就是为什么说p==head就是保证公平性!
小结:acquireQueued()的作用就是“当前线程”会根据公平性原则进行阻塞等待,直到获取锁为止;并且返回当前线程在等待过程中有没有并中断过。
四. selfInterrupt()
selfInterrupt()是AQS中实现,源码如下:
private static void selfInterrupt() { Thread.currentThread().interrupt(); }
说明:selfInterrupt()的代码很简单,就是“当前线程”自己产生一个中断。但是,为什么需要这么做呢?
这必须结合acquireQueued()进行分析。如果在acquireQueued()中,当前线程被中断过,则执行selfInterrupt();否则不会执行。
在acquireQueued()中,即使是线程在阻塞状态被中断唤醒而获取到cpu执行权利;但是,如果该线程的前面还有其它等待锁的线程,根据公平性原则,该线程依然无法获取到锁。它会再次阻塞!
该线程再次阻塞,直到该线程被它的前面等待锁的线程锁唤醒;线程才会获取锁,然后“真正执行起来”!
也就是说,在该线程“成功获取锁并真正执行起来”之前,它的中断会被忽略并且中断标记会被清除!
因为在parkAndCheckInterrupt()中,我们线程的中断状态时调用了Thread.interrupted()。该函数不同于Thread的isInterrupted()函数,isInterrupted()仅仅返回中断状态,而interrupted()在返回当前中断状态之后,还会清除中断状态。
正因为之前的中断状态被清除了,所以这里需要调用selfInterrupt()重新产生一个中断!
小结:selfInterrupt()的作用就是当前线程自己产生一个中断。
总结
再回过头看看acquire()函数,它最终的目的是获取锁!
public final void acquire(int arg) { if (!tryAcquire(arg) && acquireQueued(addWaiter(Node.EXCLUSIVE), arg)) selfInterrupt(); }
(01)
先是通过tryAcquire()尝试获取锁。获取成功的话,直接返回;尝试失败的话,再通过acquireQueued()获取锁。
(02)
尝试失败的情况下,会先通过addWaiter()来将“当前线程”加入到"CLH队列"末尾;然后调用acquireQueued(),在CLH队列中排序等待获取锁,在此过程中,线程处于休眠状态。直到获取锁了才返回。
如果在休眠等待过程中被中断过,则调用selfInterrupt()来自己产生一个中断。
更多内容