1 | // Copyright 2016 Amanieu d'Antras |
2 | // |
3 | // Licensed under the Apache License, Version 2.0, <LICENSE-APACHE or |
4 | // http://apache.org/licenses/LICENSE-2.0> or the MIT license <LICENSE-MIT or |
5 | // http://opensource.org/licenses/MIT>, at your option. This file may not be |
6 | // copied, modified, or distributed except according to those terms. |
7 | |
8 | use crate::mutex::MutexGuard; |
9 | use crate::raw_mutex::{RawMutex, TOKEN_HANDOFF, TOKEN_NORMAL}; |
10 | use crate::{deadlock, util}; |
11 | use core::{ |
12 | fmt, ptr, |
13 | sync::atomic::{AtomicPtr, Ordering}, |
14 | }; |
15 | use lock_api::RawMutex as RawMutex_; |
16 | use parking_lot_core::{self, ParkResult, RequeueOp, UnparkResult, DEFAULT_PARK_TOKEN}; |
17 | use std::ops::DerefMut; |
18 | use std::time::{Duration, Instant}; |
19 | |
20 | /// A type indicating whether a timed wait on a condition variable returned |
21 | /// due to a time out or not. |
22 | #[derive (Debug, PartialEq, Eq, Copy, Clone)] |
23 | pub struct WaitTimeoutResult(bool); |
24 | |
25 | impl WaitTimeoutResult { |
26 | /// Returns whether the wait was known to have timed out. |
27 | #[inline ] |
28 | pub fn timed_out(self) -> bool { |
29 | self.0 |
30 | } |
31 | } |
32 | |
33 | /// A Condition Variable |
34 | /// |
35 | /// Condition variables represent the ability to block a thread such that it |
36 | /// consumes no CPU time while waiting for an event to occur. Condition |
37 | /// variables are typically associated with a boolean predicate (a condition) |
38 | /// and a mutex. The predicate is always verified inside of the mutex before |
39 | /// determining that thread must block. |
40 | /// |
41 | /// Note that this module places one additional restriction over the system |
42 | /// condition variables: each condvar can be used with only one mutex at a |
43 | /// time. Any attempt to use multiple mutexes on the same condition variable |
44 | /// simultaneously will result in a runtime panic. However it is possible to |
45 | /// switch to a different mutex if there are no threads currently waiting on |
46 | /// the condition variable. |
47 | /// |
48 | /// # Differences from the standard library `Condvar` |
49 | /// |
50 | /// - No spurious wakeups: A wait will only return a non-timeout result if it |
51 | /// was woken up by `notify_one` or `notify_all`. |
52 | /// - `Condvar::notify_all` will only wake up a single thread, the rest are |
53 | /// requeued to wait for the `Mutex` to be unlocked by the thread that was |
54 | /// woken up. |
55 | /// - Only requires 1 word of space, whereas the standard library boxes the |
56 | /// `Condvar` due to platform limitations. |
57 | /// - Can be statically constructed. |
58 | /// - Does not require any drop glue when dropped. |
59 | /// - Inline fast path for the uncontended case. |
60 | /// |
61 | /// # Examples |
62 | /// |
63 | /// ``` |
64 | /// use parking_lot::{Mutex, Condvar}; |
65 | /// use std::sync::Arc; |
66 | /// use std::thread; |
67 | /// |
68 | /// let pair = Arc::new((Mutex::new(false), Condvar::new())); |
69 | /// let pair2 = pair.clone(); |
70 | /// |
71 | /// // Inside of our lock, spawn a new thread, and then wait for it to start |
72 | /// thread::spawn(move|| { |
73 | /// let &(ref lock, ref cvar) = &*pair2; |
74 | /// let mut started = lock.lock(); |
75 | /// *started = true; |
76 | /// cvar.notify_one(); |
77 | /// }); |
78 | /// |
79 | /// // wait for the thread to start up |
80 | /// let &(ref lock, ref cvar) = &*pair; |
81 | /// let mut started = lock.lock(); |
82 | /// if !*started { |
83 | /// cvar.wait(&mut started); |
84 | /// } |
85 | /// // Note that we used an if instead of a while loop above. This is only |
86 | /// // possible because parking_lot's Condvar will never spuriously wake up. |
87 | /// // This means that wait() will only return after notify_one or notify_all is |
88 | /// // called. |
89 | /// ``` |
90 | pub struct Condvar { |
91 | state: AtomicPtr<RawMutex>, |
92 | } |
93 | |
94 | impl Condvar { |
95 | /// Creates a new condition variable which is ready to be waited on and |
96 | /// notified. |
97 | #[inline ] |
98 | pub const fn new() -> Condvar { |
99 | Condvar { |
100 | state: AtomicPtr::new(ptr::null_mut()), |
101 | } |
102 | } |
103 | |
104 | /// Wakes up one blocked thread on this condvar. |
105 | /// |
106 | /// Returns whether a thread was woken up. |
107 | /// |
108 | /// If there is a blocked thread on this condition variable, then it will |
109 | /// be woken up from its call to `wait` or `wait_timeout`. Calls to |
110 | /// `notify_one` are not buffered in any way. |
111 | /// |
112 | /// To wake up all threads, see `notify_all()`. |
113 | /// |
114 | /// # Examples |
115 | /// |
116 | /// ``` |
117 | /// use parking_lot::Condvar; |
118 | /// |
119 | /// let condvar = Condvar::new(); |
120 | /// |
121 | /// // do something with condvar, share it with other threads |
122 | /// |
123 | /// if !condvar.notify_one() { |
124 | /// println!("Nobody was listening for this." ); |
125 | /// } |
126 | /// ``` |
127 | #[inline ] |
128 | pub fn notify_one(&self) -> bool { |
129 | // Nothing to do if there are no waiting threads |
130 | let state = self.state.load(Ordering::Relaxed); |
131 | if state.is_null() { |
132 | return false; |
133 | } |
134 | |
135 | self.notify_one_slow(state) |
136 | } |
137 | |
138 | #[cold ] |
139 | fn notify_one_slow(&self, mutex: *mut RawMutex) -> bool { |
140 | // Unpark one thread and requeue the rest onto the mutex |
141 | let from = self as *const _ as usize; |
142 | let to = mutex as usize; |
143 | let validate = || { |
144 | // Make sure that our atomic state still points to the same |
145 | // mutex. If not then it means that all threads on the current |
146 | // mutex were woken up and a new waiting thread switched to a |
147 | // different mutex. In that case we can get away with doing |
148 | // nothing. |
149 | if self.state.load(Ordering::Relaxed) != mutex { |
150 | return RequeueOp::Abort; |
151 | } |
152 | |
153 | // Unpark one thread if the mutex is unlocked, otherwise just |
154 | // requeue everything to the mutex. This is safe to do here |
155 | // since unlocking the mutex when the parked bit is set requires |
156 | // locking the queue. There is the possibility of a race if the |
157 | // mutex gets locked after we check, but that doesn't matter in |
158 | // this case. |
159 | if unsafe { (*mutex).mark_parked_if_locked() } { |
160 | RequeueOp::RequeueOne |
161 | } else { |
162 | RequeueOp::UnparkOne |
163 | } |
164 | }; |
165 | let callback = |_op, result: UnparkResult| { |
166 | // Clear our state if there are no more waiting threads |
167 | if !result.have_more_threads { |
168 | self.state.store(ptr::null_mut(), Ordering::Relaxed); |
169 | } |
170 | TOKEN_NORMAL |
171 | }; |
172 | let res = unsafe { parking_lot_core::unpark_requeue(from, to, validate, callback) }; |
173 | |
174 | res.unparked_threads + res.requeued_threads != 0 |
175 | } |
176 | |
177 | /// Wakes up all blocked threads on this condvar. |
178 | /// |
179 | /// Returns the number of threads woken up. |
180 | /// |
181 | /// This method will ensure that any current waiters on the condition |
182 | /// variable are awoken. Calls to `notify_all()` are not buffered in any |
183 | /// way. |
184 | /// |
185 | /// To wake up only one thread, see `notify_one()`. |
186 | #[inline ] |
187 | pub fn notify_all(&self) -> usize { |
188 | // Nothing to do if there are no waiting threads |
189 | let state = self.state.load(Ordering::Relaxed); |
190 | if state.is_null() { |
191 | return 0; |
192 | } |
193 | |
194 | self.notify_all_slow(state) |
195 | } |
196 | |
197 | #[cold ] |
198 | fn notify_all_slow(&self, mutex: *mut RawMutex) -> usize { |
199 | // Unpark one thread and requeue the rest onto the mutex |
200 | let from = self as *const _ as usize; |
201 | let to = mutex as usize; |
202 | let validate = || { |
203 | // Make sure that our atomic state still points to the same |
204 | // mutex. If not then it means that all threads on the current |
205 | // mutex were woken up and a new waiting thread switched to a |
206 | // different mutex. In that case we can get away with doing |
207 | // nothing. |
208 | if self.state.load(Ordering::Relaxed) != mutex { |
209 | return RequeueOp::Abort; |
210 | } |
211 | |
212 | // Clear our state since we are going to unpark or requeue all |
213 | // threads. |
214 | self.state.store(ptr::null_mut(), Ordering::Relaxed); |
215 | |
216 | // Unpark one thread if the mutex is unlocked, otherwise just |
217 | // requeue everything to the mutex. This is safe to do here |
218 | // since unlocking the mutex when the parked bit is set requires |
219 | // locking the queue. There is the possibility of a race if the |
220 | // mutex gets locked after we check, but that doesn't matter in |
221 | // this case. |
222 | if unsafe { (*mutex).mark_parked_if_locked() } { |
223 | RequeueOp::RequeueAll |
224 | } else { |
225 | RequeueOp::UnparkOneRequeueRest |
226 | } |
227 | }; |
228 | let callback = |op, result: UnparkResult| { |
229 | // If we requeued threads to the mutex, mark it as having |
230 | // parked threads. The RequeueAll case is already handled above. |
231 | if op == RequeueOp::UnparkOneRequeueRest && result.requeued_threads != 0 { |
232 | unsafe { (*mutex).mark_parked() }; |
233 | } |
234 | TOKEN_NORMAL |
235 | }; |
236 | let res = unsafe { parking_lot_core::unpark_requeue(from, to, validate, callback) }; |
237 | |
238 | res.unparked_threads + res.requeued_threads |
239 | } |
240 | |
241 | /// Blocks the current thread until this condition variable receives a |
242 | /// notification. |
243 | /// |
244 | /// This function will atomically unlock the mutex specified (represented by |
245 | /// `mutex_guard`) and block the current thread. This means that any calls |
246 | /// to `notify_*()` which happen logically after the mutex is unlocked are |
247 | /// candidates to wake this thread up. When this function call returns, the |
248 | /// lock specified will have been re-acquired. |
249 | /// |
250 | /// # Panics |
251 | /// |
252 | /// This function will panic if another thread is waiting on the `Condvar` |
253 | /// with a different `Mutex` object. |
254 | #[inline ] |
255 | pub fn wait<T: ?Sized>(&self, mutex_guard: &mut MutexGuard<'_, T>) { |
256 | self.wait_until_internal(unsafe { MutexGuard::mutex(mutex_guard).raw() }, None); |
257 | } |
258 | |
259 | /// Waits on this condition variable for a notification, timing out after |
260 | /// the specified time instant. |
261 | /// |
262 | /// The semantics of this function are equivalent to `wait()` except that |
263 | /// the thread will be blocked roughly until `timeout` is reached. This |
264 | /// method should not be used for precise timing due to anomalies such as |
265 | /// preemption or platform differences that may not cause the maximum |
266 | /// amount of time waited to be precisely `timeout`. |
267 | /// |
268 | /// Note that the best effort is made to ensure that the time waited is |
269 | /// measured with a monotonic clock, and not affected by the changes made to |
270 | /// the system time. |
271 | /// |
272 | /// The returned `WaitTimeoutResult` value indicates if the timeout is |
273 | /// known to have elapsed. |
274 | /// |
275 | /// Like `wait`, the lock specified will be re-acquired when this function |
276 | /// returns, regardless of whether the timeout elapsed or not. |
277 | /// |
278 | /// # Panics |
279 | /// |
280 | /// This function will panic if another thread is waiting on the `Condvar` |
281 | /// with a different `Mutex` object. |
282 | #[inline ] |
283 | pub fn wait_until<T: ?Sized>( |
284 | &self, |
285 | mutex_guard: &mut MutexGuard<'_, T>, |
286 | timeout: Instant, |
287 | ) -> WaitTimeoutResult { |
288 | self.wait_until_internal( |
289 | unsafe { MutexGuard::mutex(mutex_guard).raw() }, |
290 | Some(timeout), |
291 | ) |
292 | } |
293 | |
294 | // This is a non-generic function to reduce the monomorphization cost of |
295 | // using `wait_until`. |
296 | fn wait_until_internal(&self, mutex: &RawMutex, timeout: Option<Instant>) -> WaitTimeoutResult { |
297 | let result; |
298 | let mut bad_mutex = false; |
299 | let mut requeued = false; |
300 | { |
301 | let addr = self as *const _ as usize; |
302 | let lock_addr = mutex as *const _ as *mut _; |
303 | let validate = || { |
304 | // Ensure we don't use two different mutexes with the same |
305 | // Condvar at the same time. This is done while locked to |
306 | // avoid races with notify_one |
307 | let state = self.state.load(Ordering::Relaxed); |
308 | if state.is_null() { |
309 | self.state.store(lock_addr, Ordering::Relaxed); |
310 | } else if state != lock_addr { |
311 | bad_mutex = true; |
312 | return false; |
313 | } |
314 | true |
315 | }; |
316 | let before_sleep = || { |
317 | // Unlock the mutex before sleeping... |
318 | unsafe { mutex.unlock() }; |
319 | }; |
320 | let timed_out = |k, was_last_thread| { |
321 | // If we were requeued to a mutex, then we did not time out. |
322 | // We'll just park ourselves on the mutex again when we try |
323 | // to lock it later. |
324 | requeued = k != addr; |
325 | |
326 | // If we were the last thread on the queue then we need to |
327 | // clear our state. This is normally done by the |
328 | // notify_{one,all} functions when not timing out. |
329 | if !requeued && was_last_thread { |
330 | self.state.store(ptr::null_mut(), Ordering::Relaxed); |
331 | } |
332 | }; |
333 | result = unsafe { parking_lot_core::park( |
334 | addr, |
335 | validate, |
336 | before_sleep, |
337 | timed_out, |
338 | DEFAULT_PARK_TOKEN, |
339 | timeout, |
340 | ) }; |
341 | } |
342 | |
343 | // Panic if we tried to use multiple mutexes with a Condvar. Note |
344 | // that at this point the MutexGuard is still locked. It will be |
345 | // unlocked by the unwinding logic. |
346 | if bad_mutex { |
347 | panic!("attempted to use a condition variable with more than one mutex" ); |
348 | } |
349 | |
350 | // ... and re-lock it once we are done sleeping |
351 | if result == ParkResult::Unparked(TOKEN_HANDOFF) { |
352 | unsafe { deadlock::acquire_resource(mutex as *const _ as usize) }; |
353 | } else { |
354 | mutex.lock(); |
355 | } |
356 | |
357 | WaitTimeoutResult(!(result.is_unparked() || requeued)) |
358 | } |
359 | |
360 | /// Waits on this condition variable for a notification, timing out after a |
361 | /// specified duration. |
362 | /// |
363 | /// The semantics of this function are equivalent to `wait()` except that |
364 | /// the thread will be blocked for roughly no longer than `timeout`. This |
365 | /// method should not be used for precise timing due to anomalies such as |
366 | /// preemption or platform differences that may not cause the maximum |
367 | /// amount of time waited to be precisely `timeout`. |
368 | /// |
369 | /// Note that the best effort is made to ensure that the time waited is |
370 | /// measured with a monotonic clock, and not affected by the changes made to |
371 | /// the system time. |
372 | /// |
373 | /// The returned `WaitTimeoutResult` value indicates if the timeout is |
374 | /// known to have elapsed. |
375 | /// |
376 | /// Like `wait`, the lock specified will be re-acquired when this function |
377 | /// returns, regardless of whether the timeout elapsed or not. |
378 | #[inline ] |
379 | pub fn wait_for<T: ?Sized>( |
380 | &self, |
381 | mutex_guard: &mut MutexGuard<'_, T>, |
382 | timeout: Duration, |
383 | ) -> WaitTimeoutResult { |
384 | let deadline = util::to_deadline(timeout); |
385 | self.wait_until_internal(unsafe { MutexGuard::mutex(mutex_guard).raw() }, deadline) |
386 | } |
387 | |
388 | #[inline ] |
389 | fn wait_while_until_internal<T, F>( |
390 | &self, |
391 | mutex_guard: &mut MutexGuard<'_, T>, |
392 | mut condition: F, |
393 | timeout: Option<Instant>, |
394 | ) -> WaitTimeoutResult |
395 | where |
396 | T: ?Sized, |
397 | F: FnMut(&mut T) -> bool, |
398 | { |
399 | let mut result = WaitTimeoutResult(false); |
400 | |
401 | while !result.timed_out() && condition(mutex_guard.deref_mut()) { |
402 | result = |
403 | self.wait_until_internal(unsafe { MutexGuard::mutex(mutex_guard).raw() }, timeout); |
404 | } |
405 | |
406 | result |
407 | } |
408 | /// Blocks the current thread until this condition variable receives a |
409 | /// notification. If the provided condition evaluates to `false`, then the |
410 | /// thread is no longer blocked and the operation is completed. If the |
411 | /// condition evaluates to `true`, then the thread is blocked again and |
412 | /// waits for another notification before repeating this process. |
413 | /// |
414 | /// This function will atomically unlock the mutex specified (represented by |
415 | /// `mutex_guard`) and block the current thread. This means that any calls |
416 | /// to `notify_*()` which happen logically after the mutex is unlocked are |
417 | /// candidates to wake this thread up. When this function call returns, the |
418 | /// lock specified will have been re-acquired. |
419 | /// |
420 | /// # Panics |
421 | /// |
422 | /// This function will panic if another thread is waiting on the `Condvar` |
423 | /// with a different `Mutex` object. |
424 | #[inline ] |
425 | pub fn wait_while<T, F>(&self, mutex_guard: &mut MutexGuard<'_, T>, condition: F) |
426 | where |
427 | T: ?Sized, |
428 | F: FnMut(&mut T) -> bool, |
429 | { |
430 | self.wait_while_until_internal(mutex_guard, condition, None); |
431 | } |
432 | |
433 | /// Waits on this condition variable for a notification, timing out after |
434 | /// the specified time instant. If the provided condition evaluates to |
435 | /// `false`, then the thread is no longer blocked and the operation is |
436 | /// completed. If the condition evaluates to `true`, then the thread is |
437 | /// blocked again and waits for another notification before repeating |
438 | /// this process. |
439 | /// |
440 | /// The semantics of this function are equivalent to `wait()` except that |
441 | /// the thread will be blocked roughly until `timeout` is reached. This |
442 | /// method should not be used for precise timing due to anomalies such as |
443 | /// preemption or platform differences that may not cause the maximum |
444 | /// amount of time waited to be precisely `timeout`. |
445 | /// |
446 | /// Note that the best effort is made to ensure that the time waited is |
447 | /// measured with a monotonic clock, and not affected by the changes made to |
448 | /// the system time. |
449 | /// |
450 | /// The returned `WaitTimeoutResult` value indicates if the timeout is |
451 | /// known to have elapsed. |
452 | /// |
453 | /// Like `wait`, the lock specified will be re-acquired when this function |
454 | /// returns, regardless of whether the timeout elapsed or not. |
455 | /// |
456 | /// # Panics |
457 | /// |
458 | /// This function will panic if another thread is waiting on the `Condvar` |
459 | /// with a different `Mutex` object. |
460 | #[inline ] |
461 | pub fn wait_while_until<T, F>( |
462 | &self, |
463 | mutex_guard: &mut MutexGuard<'_, T>, |
464 | condition: F, |
465 | timeout: Instant, |
466 | ) -> WaitTimeoutResult |
467 | where |
468 | T: ?Sized, |
469 | F: FnMut(&mut T) -> bool, |
470 | { |
471 | self.wait_while_until_internal(mutex_guard, condition, Some(timeout)) |
472 | } |
473 | |
474 | /// Waits on this condition variable for a notification, timing out after a |
475 | /// specified duration. If the provided condition evaluates to `false`, |
476 | /// then the thread is no longer blocked and the operation is completed. |
477 | /// If the condition evaluates to `true`, then the thread is blocked again |
478 | /// and waits for another notification before repeating this process. |
479 | /// |
480 | /// The semantics of this function are equivalent to `wait()` except that |
481 | /// the thread will be blocked for roughly no longer than `timeout`. This |
482 | /// method should not be used for precise timing due to anomalies such as |
483 | /// preemption or platform differences that may not cause the maximum |
484 | /// amount of time waited to be precisely `timeout`. |
485 | /// |
486 | /// Note that the best effort is made to ensure that the time waited is |
487 | /// measured with a monotonic clock, and not affected by the changes made to |
488 | /// the system time. |
489 | /// |
490 | /// The returned `WaitTimeoutResult` value indicates if the timeout is |
491 | /// known to have elapsed. |
492 | /// |
493 | /// Like `wait`, the lock specified will be re-acquired when this function |
494 | /// returns, regardless of whether the timeout elapsed or not. |
495 | #[inline ] |
496 | pub fn wait_while_for<T: ?Sized, F>( |
497 | &self, |
498 | mutex_guard: &mut MutexGuard<'_, T>, |
499 | condition: F, |
500 | timeout: Duration, |
501 | ) -> WaitTimeoutResult |
502 | where |
503 | F: FnMut(&mut T) -> bool, |
504 | { |
505 | let deadline = util::to_deadline(timeout); |
506 | self.wait_while_until_internal(mutex_guard, condition, deadline) |
507 | } |
508 | } |
509 | |
510 | impl Default for Condvar { |
511 | #[inline ] |
512 | fn default() -> Condvar { |
513 | Condvar::new() |
514 | } |
515 | } |
516 | |
517 | impl fmt::Debug for Condvar { |
518 | fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { |
519 | f.pad("Condvar { .. }" ) |
520 | } |
521 | } |
522 | |
523 | #[cfg (test)] |
524 | mod tests { |
525 | use crate::{Condvar, Mutex, MutexGuard}; |
526 | use std::sync::mpsc::channel; |
527 | use std::sync::Arc; |
528 | use std::thread; |
529 | use std::thread::sleep; |
530 | use std::thread::JoinHandle; |
531 | use std::time::Duration; |
532 | use std::time::Instant; |
533 | |
534 | #[test ] |
535 | fn smoke() { |
536 | let c = Condvar::new(); |
537 | c.notify_one(); |
538 | c.notify_all(); |
539 | } |
540 | |
541 | #[test ] |
542 | fn notify_one() { |
543 | let m = Arc::new(Mutex::new(())); |
544 | let m2 = m.clone(); |
545 | let c = Arc::new(Condvar::new()); |
546 | let c2 = c.clone(); |
547 | |
548 | let mut g = m.lock(); |
549 | let _t = thread::spawn(move || { |
550 | let _g = m2.lock(); |
551 | c2.notify_one(); |
552 | }); |
553 | c.wait(&mut g); |
554 | } |
555 | |
556 | #[test ] |
557 | fn notify_all() { |
558 | const N: usize = 10; |
559 | |
560 | let data = Arc::new((Mutex::new(0), Condvar::new())); |
561 | let (tx, rx) = channel(); |
562 | for _ in 0..N { |
563 | let data = data.clone(); |
564 | let tx = tx.clone(); |
565 | thread::spawn(move || { |
566 | let &(ref lock, ref cond) = &*data; |
567 | let mut cnt = lock.lock(); |
568 | *cnt += 1; |
569 | if *cnt == N { |
570 | tx.send(()).unwrap(); |
571 | } |
572 | while *cnt != 0 { |
573 | cond.wait(&mut cnt); |
574 | } |
575 | tx.send(()).unwrap(); |
576 | }); |
577 | } |
578 | drop(tx); |
579 | |
580 | let &(ref lock, ref cond) = &*data; |
581 | rx.recv().unwrap(); |
582 | let mut cnt = lock.lock(); |
583 | *cnt = 0; |
584 | cond.notify_all(); |
585 | drop(cnt); |
586 | |
587 | for _ in 0..N { |
588 | rx.recv().unwrap(); |
589 | } |
590 | } |
591 | |
592 | #[test ] |
593 | fn notify_one_return_true() { |
594 | let m = Arc::new(Mutex::new(())); |
595 | let m2 = m.clone(); |
596 | let c = Arc::new(Condvar::new()); |
597 | let c2 = c.clone(); |
598 | |
599 | let mut g = m.lock(); |
600 | let _t = thread::spawn(move || { |
601 | let _g = m2.lock(); |
602 | assert!(c2.notify_one()); |
603 | }); |
604 | c.wait(&mut g); |
605 | } |
606 | |
607 | #[test ] |
608 | fn notify_one_return_false() { |
609 | let m = Arc::new(Mutex::new(())); |
610 | let c = Arc::new(Condvar::new()); |
611 | |
612 | let _t = thread::spawn(move || { |
613 | let _g = m.lock(); |
614 | assert!(!c.notify_one()); |
615 | }); |
616 | } |
617 | |
618 | #[test ] |
619 | fn notify_all_return() { |
620 | const N: usize = 10; |
621 | |
622 | let data = Arc::new((Mutex::new(0), Condvar::new())); |
623 | let (tx, rx) = channel(); |
624 | for _ in 0..N { |
625 | let data = data.clone(); |
626 | let tx = tx.clone(); |
627 | thread::spawn(move || { |
628 | let &(ref lock, ref cond) = &*data; |
629 | let mut cnt = lock.lock(); |
630 | *cnt += 1; |
631 | if *cnt == N { |
632 | tx.send(()).unwrap(); |
633 | } |
634 | while *cnt != 0 { |
635 | cond.wait(&mut cnt); |
636 | } |
637 | tx.send(()).unwrap(); |
638 | }); |
639 | } |
640 | drop(tx); |
641 | |
642 | let &(ref lock, ref cond) = &*data; |
643 | rx.recv().unwrap(); |
644 | let mut cnt = lock.lock(); |
645 | *cnt = 0; |
646 | assert_eq!(cond.notify_all(), N); |
647 | drop(cnt); |
648 | |
649 | for _ in 0..N { |
650 | rx.recv().unwrap(); |
651 | } |
652 | |
653 | assert_eq!(cond.notify_all(), 0); |
654 | } |
655 | |
656 | #[test ] |
657 | fn wait_for() { |
658 | let m = Arc::new(Mutex::new(())); |
659 | let m2 = m.clone(); |
660 | let c = Arc::new(Condvar::new()); |
661 | let c2 = c.clone(); |
662 | |
663 | let mut g = m.lock(); |
664 | let no_timeout = c.wait_for(&mut g, Duration::from_millis(1)); |
665 | assert!(no_timeout.timed_out()); |
666 | |
667 | let _t = thread::spawn(move || { |
668 | let _g = m2.lock(); |
669 | c2.notify_one(); |
670 | }); |
671 | let timeout_res = c.wait_for(&mut g, Duration::from_secs(u64::max_value())); |
672 | assert!(!timeout_res.timed_out()); |
673 | |
674 | drop(g); |
675 | } |
676 | |
677 | #[test ] |
678 | fn wait_until() { |
679 | let m = Arc::new(Mutex::new(())); |
680 | let m2 = m.clone(); |
681 | let c = Arc::new(Condvar::new()); |
682 | let c2 = c.clone(); |
683 | |
684 | let mut g = m.lock(); |
685 | let no_timeout = c.wait_until(&mut g, Instant::now() + Duration::from_millis(1)); |
686 | assert!(no_timeout.timed_out()); |
687 | let _t = thread::spawn(move || { |
688 | let _g = m2.lock(); |
689 | c2.notify_one(); |
690 | }); |
691 | let timeout_res = c.wait_until( |
692 | &mut g, |
693 | Instant::now() + Duration::from_millis(u32::max_value() as u64), |
694 | ); |
695 | assert!(!timeout_res.timed_out()); |
696 | drop(g); |
697 | } |
698 | |
699 | fn spawn_wait_while_notifier( |
700 | mutex: Arc<Mutex<u32>>, |
701 | cv: Arc<Condvar>, |
702 | num_iters: u32, |
703 | timeout: Option<Instant>, |
704 | ) -> JoinHandle<()> { |
705 | thread::spawn(move || { |
706 | for epoch in 1..=num_iters { |
707 | // spin to wait for main test thread to block |
708 | // before notifying it to wake back up and check |
709 | // its condition. |
710 | let mut sleep_backoff = Duration::from_millis(1); |
711 | let _mutex_guard = loop { |
712 | let mutex_guard = mutex.lock(); |
713 | |
714 | if let Some(timeout) = timeout { |
715 | if Instant::now() >= timeout { |
716 | return; |
717 | } |
718 | } |
719 | |
720 | if *mutex_guard == epoch { |
721 | break mutex_guard; |
722 | } |
723 | |
724 | drop(mutex_guard); |
725 | |
726 | // give main test thread a good chance to |
727 | // acquire the lock before this thread does. |
728 | sleep(sleep_backoff); |
729 | sleep_backoff *= 2; |
730 | }; |
731 | |
732 | cv.notify_one(); |
733 | } |
734 | }) |
735 | } |
736 | |
737 | #[test ] |
738 | fn wait_while_until_internal_does_not_wait_if_initially_false() { |
739 | let mutex = Arc::new(Mutex::new(0)); |
740 | let cv = Arc::new(Condvar::new()); |
741 | |
742 | let condition = |counter: &mut u32| { |
743 | *counter += 1; |
744 | false |
745 | }; |
746 | |
747 | let mut mutex_guard = mutex.lock(); |
748 | let timeout_result = cv |
749 | .wait_while_until_internal(&mut mutex_guard, condition, None); |
750 | |
751 | assert!(!timeout_result.timed_out()); |
752 | assert!(*mutex_guard == 1); |
753 | } |
754 | |
755 | #[test ] |
756 | fn wait_while_until_internal_times_out_before_false() { |
757 | let mutex = Arc::new(Mutex::new(0)); |
758 | let cv = Arc::new(Condvar::new()); |
759 | |
760 | let num_iters = 3; |
761 | let condition = |counter: &mut u32| { |
762 | *counter += 1; |
763 | true |
764 | }; |
765 | |
766 | let mut mutex_guard = mutex.lock(); |
767 | let timeout = Some(Instant::now() + Duration::from_millis(500)); |
768 | let handle = spawn_wait_while_notifier(mutex.clone(), cv.clone(), num_iters, timeout); |
769 | |
770 | let timeout_result = |
771 | cv.wait_while_until_internal(&mut mutex_guard, condition, timeout); |
772 | |
773 | assert!(timeout_result.timed_out()); |
774 | assert!(*mutex_guard == num_iters + 1); |
775 | |
776 | // prevent deadlock with notifier |
777 | drop(mutex_guard); |
778 | handle.join().unwrap(); |
779 | } |
780 | |
781 | #[test ] |
782 | fn wait_while_until_internal() { |
783 | let mutex = Arc::new(Mutex::new(0)); |
784 | let cv = Arc::new(Condvar::new()); |
785 | |
786 | let num_iters = 4; |
787 | |
788 | let condition = |counter: &mut u32| { |
789 | *counter += 1; |
790 | *counter <= num_iters |
791 | }; |
792 | |
793 | let mut mutex_guard = mutex.lock(); |
794 | let handle = spawn_wait_while_notifier(mutex.clone(), cv.clone(), num_iters, None); |
795 | |
796 | let timeout_result = |
797 | cv.wait_while_until_internal(&mut mutex_guard, condition, None); |
798 | |
799 | assert!(!timeout_result.timed_out()); |
800 | assert!(*mutex_guard == num_iters + 1); |
801 | |
802 | let timeout_result = cv.wait_while_until_internal(&mut mutex_guard, condition, None); |
803 | handle.join().unwrap(); |
804 | |
805 | assert!(!timeout_result.timed_out()); |
806 | assert!(*mutex_guard == num_iters + 2); |
807 | } |
808 | |
809 | #[test ] |
810 | #[should_panic ] |
811 | fn two_mutexes() { |
812 | let m = Arc::new(Mutex::new(())); |
813 | let m2 = m.clone(); |
814 | let m3 = Arc::new(Mutex::new(())); |
815 | let c = Arc::new(Condvar::new()); |
816 | let c2 = c.clone(); |
817 | |
818 | // Make sure we don't leave the child thread dangling |
819 | struct PanicGuard<'a>(&'a Condvar); |
820 | impl<'a> Drop for PanicGuard<'a> { |
821 | fn drop(&mut self) { |
822 | self.0.notify_one(); |
823 | } |
824 | } |
825 | |
826 | let (tx, rx) = channel(); |
827 | let g = m.lock(); |
828 | let _t = thread::spawn(move || { |
829 | let mut g = m2.lock(); |
830 | tx.send(()).unwrap(); |
831 | c2.wait(&mut g); |
832 | }); |
833 | drop(g); |
834 | rx.recv().unwrap(); |
835 | let _g = m.lock(); |
836 | let _guard = PanicGuard(&*c); |
837 | c.wait(&mut m3.lock()); |
838 | } |
839 | |
840 | #[test ] |
841 | fn two_mutexes_disjoint() { |
842 | let m = Arc::new(Mutex::new(())); |
843 | let m2 = m.clone(); |
844 | let m3 = Arc::new(Mutex::new(())); |
845 | let c = Arc::new(Condvar::new()); |
846 | let c2 = c.clone(); |
847 | |
848 | let mut g = m.lock(); |
849 | let _t = thread::spawn(move || { |
850 | let _g = m2.lock(); |
851 | c2.notify_one(); |
852 | }); |
853 | c.wait(&mut g); |
854 | drop(g); |
855 | |
856 | let _ = c.wait_for(&mut m3.lock(), Duration::from_millis(1)); |
857 | } |
858 | |
859 | #[test ] |
860 | fn test_debug_condvar() { |
861 | let c = Condvar::new(); |
862 | assert_eq!(format!(" {:?}" , c), "Condvar { .. }" ); |
863 | } |
864 | |
865 | #[test ] |
866 | fn test_condvar_requeue() { |
867 | let m = Arc::new(Mutex::new(())); |
868 | let m2 = m.clone(); |
869 | let c = Arc::new(Condvar::new()); |
870 | let c2 = c.clone(); |
871 | let t = thread::spawn(move || { |
872 | let mut g = m2.lock(); |
873 | c2.wait(&mut g); |
874 | }); |
875 | |
876 | let mut g = m.lock(); |
877 | while !c.notify_one() { |
878 | // Wait for the thread to get into wait() |
879 | MutexGuard::bump(&mut g); |
880 | // Yield, so the other thread gets a chance to do something. |
881 | // (At least Miri needs this, because it doesn't preempt threads.) |
882 | thread::yield_now(); |
883 | } |
884 | // The thread should have been requeued to the mutex, which we wake up now. |
885 | drop(g); |
886 | t.join().unwrap(); |
887 | } |
888 | |
889 | #[test ] |
890 | fn test_issue_129() { |
891 | let locks = Arc::new((Mutex::new(()), Condvar::new())); |
892 | |
893 | let (tx, rx) = channel(); |
894 | for _ in 0..4 { |
895 | let locks = locks.clone(); |
896 | let tx = tx.clone(); |
897 | thread::spawn(move || { |
898 | let mut guard = locks.0.lock(); |
899 | locks.1.wait(&mut guard); |
900 | locks.1.wait_for(&mut guard, Duration::from_millis(1)); |
901 | locks.1.notify_one(); |
902 | tx.send(()).unwrap(); |
903 | }); |
904 | } |
905 | |
906 | thread::sleep(Duration::from_millis(100)); |
907 | locks.1.notify_one(); |
908 | |
909 | for _ in 0..4 { |
910 | assert_eq!(rx.recv_timeout(Duration::from_millis(500)), Ok(())); |
911 | } |
912 | } |
913 | } |
914 | |
915 | /// This module contains an integration test that is heavily inspired from WebKit's own integration |
916 | /// tests for it's own Condvar. |
917 | #[cfg (test)] |
918 | mod webkit_queue_test { |
919 | use crate::{Condvar, Mutex, MutexGuard}; |
920 | use std::{collections::VecDeque, sync::Arc, thread, time::Duration}; |
921 | |
922 | #[derive (Clone, Copy)] |
923 | enum Timeout { |
924 | Bounded(Duration), |
925 | Forever, |
926 | } |
927 | |
928 | #[derive (Clone, Copy)] |
929 | enum NotifyStyle { |
930 | One, |
931 | All, |
932 | } |
933 | |
934 | struct Queue { |
935 | items: VecDeque<usize>, |
936 | should_continue: bool, |
937 | } |
938 | |
939 | impl Queue { |
940 | fn new() -> Self { |
941 | Self { |
942 | items: VecDeque::new(), |
943 | should_continue: true, |
944 | } |
945 | } |
946 | } |
947 | |
948 | fn wait<T: ?Sized>( |
949 | condition: &Condvar, |
950 | lock: &mut MutexGuard<'_, T>, |
951 | predicate: impl Fn(&mut MutexGuard<'_, T>) -> bool, |
952 | timeout: &Timeout, |
953 | ) { |
954 | while !predicate(lock) { |
955 | match timeout { |
956 | Timeout::Forever => condition.wait(lock), |
957 | Timeout::Bounded(bound) => { |
958 | condition.wait_for(lock, *bound); |
959 | } |
960 | } |
961 | } |
962 | } |
963 | |
964 | fn notify(style: NotifyStyle, condition: &Condvar, should_notify: bool) { |
965 | match style { |
966 | NotifyStyle::One => { |
967 | condition.notify_one(); |
968 | } |
969 | NotifyStyle::All => { |
970 | if should_notify { |
971 | condition.notify_all(); |
972 | } |
973 | } |
974 | } |
975 | } |
976 | |
977 | fn run_queue_test( |
978 | num_producers: usize, |
979 | num_consumers: usize, |
980 | max_queue_size: usize, |
981 | messages_per_producer: usize, |
982 | notify_style: NotifyStyle, |
983 | timeout: Timeout, |
984 | delay: Duration, |
985 | ) { |
986 | let input_queue = Arc::new(Mutex::new(Queue::new())); |
987 | let empty_condition = Arc::new(Condvar::new()); |
988 | let full_condition = Arc::new(Condvar::new()); |
989 | |
990 | let output_vec = Arc::new(Mutex::new(vec![])); |
991 | |
992 | let consumers = (0..num_consumers) |
993 | .map(|_| { |
994 | consumer_thread( |
995 | input_queue.clone(), |
996 | empty_condition.clone(), |
997 | full_condition.clone(), |
998 | timeout, |
999 | notify_style, |
1000 | output_vec.clone(), |
1001 | max_queue_size, |
1002 | ) |
1003 | }) |
1004 | .collect::<Vec<_>>(); |
1005 | let producers = (0..num_producers) |
1006 | .map(|_| { |
1007 | producer_thread( |
1008 | messages_per_producer, |
1009 | input_queue.clone(), |
1010 | empty_condition.clone(), |
1011 | full_condition.clone(), |
1012 | timeout, |
1013 | notify_style, |
1014 | max_queue_size, |
1015 | ) |
1016 | }) |
1017 | .collect::<Vec<_>>(); |
1018 | |
1019 | thread::sleep(delay); |
1020 | |
1021 | for producer in producers.into_iter() { |
1022 | producer.join().expect("Producer thread panicked" ); |
1023 | } |
1024 | |
1025 | { |
1026 | let mut input_queue = input_queue.lock(); |
1027 | input_queue.should_continue = false; |
1028 | } |
1029 | empty_condition.notify_all(); |
1030 | |
1031 | for consumer in consumers.into_iter() { |
1032 | consumer.join().expect("Consumer thread panicked" ); |
1033 | } |
1034 | |
1035 | let mut output_vec = output_vec.lock(); |
1036 | assert_eq!(output_vec.len(), num_producers * messages_per_producer); |
1037 | output_vec.sort(); |
1038 | for msg_idx in 0..messages_per_producer { |
1039 | for producer_idx in 0..num_producers { |
1040 | assert_eq!(msg_idx, output_vec[msg_idx * num_producers + producer_idx]); |
1041 | } |
1042 | } |
1043 | } |
1044 | |
1045 | fn consumer_thread( |
1046 | input_queue: Arc<Mutex<Queue>>, |
1047 | empty_condition: Arc<Condvar>, |
1048 | full_condition: Arc<Condvar>, |
1049 | timeout: Timeout, |
1050 | notify_style: NotifyStyle, |
1051 | output_queue: Arc<Mutex<Vec<usize>>>, |
1052 | max_queue_size: usize, |
1053 | ) -> thread::JoinHandle<()> { |
1054 | thread::spawn(move || loop { |
1055 | let (should_notify, result) = { |
1056 | let mut queue = input_queue.lock(); |
1057 | wait( |
1058 | &*empty_condition, |
1059 | &mut queue, |
1060 | |state| -> bool { !state.items.is_empty() || !state.should_continue }, |
1061 | &timeout, |
1062 | ); |
1063 | if queue.items.is_empty() && !queue.should_continue { |
1064 | return; |
1065 | } |
1066 | let should_notify = queue.items.len() == max_queue_size; |
1067 | let result = queue.items.pop_front(); |
1068 | std::mem::drop(queue); |
1069 | (should_notify, result) |
1070 | }; |
1071 | notify(notify_style, &*full_condition, should_notify); |
1072 | |
1073 | if let Some(result) = result { |
1074 | output_queue.lock().push(result); |
1075 | } |
1076 | }) |
1077 | } |
1078 | |
1079 | fn producer_thread( |
1080 | num_messages: usize, |
1081 | queue: Arc<Mutex<Queue>>, |
1082 | empty_condition: Arc<Condvar>, |
1083 | full_condition: Arc<Condvar>, |
1084 | timeout: Timeout, |
1085 | notify_style: NotifyStyle, |
1086 | max_queue_size: usize, |
1087 | ) -> thread::JoinHandle<()> { |
1088 | thread::spawn(move || { |
1089 | for message in 0..num_messages { |
1090 | let should_notify = { |
1091 | let mut queue = queue.lock(); |
1092 | wait( |
1093 | &*full_condition, |
1094 | &mut queue, |
1095 | |state| state.items.len() < max_queue_size, |
1096 | &timeout, |
1097 | ); |
1098 | let should_notify = queue.items.is_empty(); |
1099 | queue.items.push_back(message); |
1100 | std::mem::drop(queue); |
1101 | should_notify |
1102 | }; |
1103 | notify(notify_style, &*empty_condition, should_notify); |
1104 | } |
1105 | }) |
1106 | } |
1107 | |
1108 | macro_rules! run_queue_tests { |
1109 | ( $( $name:ident( |
1110 | num_producers: $num_producers:expr, |
1111 | num_consumers: $num_consumers:expr, |
1112 | max_queue_size: $max_queue_size:expr, |
1113 | messages_per_producer: $messages_per_producer:expr, |
1114 | notification_style: $notification_style:expr, |
1115 | timeout: $timeout:expr, |
1116 | delay_seconds: $delay_seconds:expr); |
1117 | )* ) => { |
1118 | $(#[test] |
1119 | fn $name() { |
1120 | let delay = Duration::from_secs($delay_seconds); |
1121 | run_queue_test( |
1122 | $num_producers, |
1123 | $num_consumers, |
1124 | $max_queue_size, |
1125 | $messages_per_producer, |
1126 | $notification_style, |
1127 | $timeout, |
1128 | delay, |
1129 | ); |
1130 | })* |
1131 | }; |
1132 | } |
1133 | |
1134 | run_queue_tests! { |
1135 | sanity_check_queue( |
1136 | num_producers: 1, |
1137 | num_consumers: 1, |
1138 | max_queue_size: 1, |
1139 | messages_per_producer: 100_000, |
1140 | notification_style: NotifyStyle::All, |
1141 | timeout: Timeout::Bounded(Duration::from_secs(1)), |
1142 | delay_seconds: 0 |
1143 | ); |
1144 | sanity_check_queue_timeout( |
1145 | num_producers: 1, |
1146 | num_consumers: 1, |
1147 | max_queue_size: 1, |
1148 | messages_per_producer: 100_000, |
1149 | notification_style: NotifyStyle::All, |
1150 | timeout: Timeout::Forever, |
1151 | delay_seconds: 0 |
1152 | ); |
1153 | new_test_without_timeout_5( |
1154 | num_producers: 1, |
1155 | num_consumers: 5, |
1156 | max_queue_size: 1, |
1157 | messages_per_producer: 100_000, |
1158 | notification_style: NotifyStyle::All, |
1159 | timeout: Timeout::Forever, |
1160 | delay_seconds: 0 |
1161 | ); |
1162 | one_producer_one_consumer_one_slot( |
1163 | num_producers: 1, |
1164 | num_consumers: 1, |
1165 | max_queue_size: 1, |
1166 | messages_per_producer: 100_000, |
1167 | notification_style: NotifyStyle::All, |
1168 | timeout: Timeout::Forever, |
1169 | delay_seconds: 0 |
1170 | ); |
1171 | one_producer_one_consumer_one_slot_timeout( |
1172 | num_producers: 1, |
1173 | num_consumers: 1, |
1174 | max_queue_size: 1, |
1175 | messages_per_producer: 100_000, |
1176 | notification_style: NotifyStyle::All, |
1177 | timeout: Timeout::Forever, |
1178 | delay_seconds: 1 |
1179 | ); |
1180 | one_producer_one_consumer_hundred_slots( |
1181 | num_producers: 1, |
1182 | num_consumers: 1, |
1183 | max_queue_size: 100, |
1184 | messages_per_producer: 1_000_000, |
1185 | notification_style: NotifyStyle::All, |
1186 | timeout: Timeout::Forever, |
1187 | delay_seconds: 0 |
1188 | ); |
1189 | ten_producers_one_consumer_one_slot( |
1190 | num_producers: 10, |
1191 | num_consumers: 1, |
1192 | max_queue_size: 1, |
1193 | messages_per_producer: 10000, |
1194 | notification_style: NotifyStyle::All, |
1195 | timeout: Timeout::Forever, |
1196 | delay_seconds: 0 |
1197 | ); |
1198 | ten_producers_one_consumer_hundred_slots_notify_all( |
1199 | num_producers: 10, |
1200 | num_consumers: 1, |
1201 | max_queue_size: 100, |
1202 | messages_per_producer: 10000, |
1203 | notification_style: NotifyStyle::All, |
1204 | timeout: Timeout::Forever, |
1205 | delay_seconds: 0 |
1206 | ); |
1207 | ten_producers_one_consumer_hundred_slots_notify_one( |
1208 | num_producers: 10, |
1209 | num_consumers: 1, |
1210 | max_queue_size: 100, |
1211 | messages_per_producer: 10000, |
1212 | notification_style: NotifyStyle::One, |
1213 | timeout: Timeout::Forever, |
1214 | delay_seconds: 0 |
1215 | ); |
1216 | one_producer_ten_consumers_one_slot( |
1217 | num_producers: 1, |
1218 | num_consumers: 10, |
1219 | max_queue_size: 1, |
1220 | messages_per_producer: 10000, |
1221 | notification_style: NotifyStyle::All, |
1222 | timeout: Timeout::Forever, |
1223 | delay_seconds: 0 |
1224 | ); |
1225 | one_producer_ten_consumers_hundred_slots_notify_all( |
1226 | num_producers: 1, |
1227 | num_consumers: 10, |
1228 | max_queue_size: 100, |
1229 | messages_per_producer: 100_000, |
1230 | notification_style: NotifyStyle::All, |
1231 | timeout: Timeout::Forever, |
1232 | delay_seconds: 0 |
1233 | ); |
1234 | one_producer_ten_consumers_hundred_slots_notify_one( |
1235 | num_producers: 1, |
1236 | num_consumers: 10, |
1237 | max_queue_size: 100, |
1238 | messages_per_producer: 100_000, |
1239 | notification_style: NotifyStyle::One, |
1240 | timeout: Timeout::Forever, |
1241 | delay_seconds: 0 |
1242 | ); |
1243 | ten_producers_ten_consumers_one_slot( |
1244 | num_producers: 10, |
1245 | num_consumers: 10, |
1246 | max_queue_size: 1, |
1247 | messages_per_producer: 50000, |
1248 | notification_style: NotifyStyle::All, |
1249 | timeout: Timeout::Forever, |
1250 | delay_seconds: 0 |
1251 | ); |
1252 | ten_producers_ten_consumers_hundred_slots_notify_all( |
1253 | num_producers: 10, |
1254 | num_consumers: 10, |
1255 | max_queue_size: 100, |
1256 | messages_per_producer: 50000, |
1257 | notification_style: NotifyStyle::All, |
1258 | timeout: Timeout::Forever, |
1259 | delay_seconds: 0 |
1260 | ); |
1261 | ten_producers_ten_consumers_hundred_slots_notify_one( |
1262 | num_producers: 10, |
1263 | num_consumers: 10, |
1264 | max_queue_size: 100, |
1265 | messages_per_producer: 50000, |
1266 | notification_style: NotifyStyle::One, |
1267 | timeout: Timeout::Forever, |
1268 | delay_seconds: 0 |
1269 | ); |
1270 | } |
1271 | } |
1272 | |