1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * rtmutex API
4 */
5 #include <linux/spinlock.h>
6 #include <linux/export.h>
7
8 #define RT_MUTEX_BUILD_MUTEX
9 #include "rtmutex.c"
10
11 /*
12 * Max number of times we'll walk the boosting chain:
13 */
14 int max_lock_depth = 1024;
15
16 /*
17 * Debug aware fast / slowpath lock,trylock,unlock
18 *
19 * The atomic acquire/release ops are compiled away, when either the
20 * architecture does not support cmpxchg or when debugging is enabled.
21 */
22 static __always_inline int __rt_mutex_lock_common(struct rt_mutex *lock,
23 unsigned int state,
24 struct lockdep_map *nest_lock,
25 unsigned int subclass)
26 {
27 int ret;
28
29 might_sleep();
30 mutex_acquire_nest(&lock->dep_map, subclass, 0, nest_lock, _RET_IP_);
31 ret = __rt_mutex_lock(&lock->rtmutex, state);
32 if (ret)
33 mutex_release(&lock->dep_map, _RET_IP_);
34 return ret;
35 }
36
37 void rt_mutex_base_init(struct rt_mutex_base *rtb)
38 {
39 __rt_mutex_base_init(rtb);
40 }
41 EXPORT_SYMBOL(rt_mutex_base_init);
42
43 #ifdef CONFIG_DEBUG_LOCK_ALLOC
44 /**
45 * rt_mutex_lock_nested - lock a rt_mutex
46 *
47 * @lock: the rt_mutex to be locked
48 * @subclass: the lockdep subclass
49 */
50 void __sched rt_mutex_lock_nested(struct rt_mutex *lock, unsigned int subclass)
51 {
52 __rt_mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, NULL, subclass);
53 }
54 EXPORT_SYMBOL_GPL(rt_mutex_lock_nested);
55
56 void __sched _rt_mutex_lock_nest_lock(struct rt_mutex *lock, struct lockdep_map *nest_lock)
57 {
58 __rt_mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, nest_lock, 0);
59 }
60 EXPORT_SYMBOL_GPL(_rt_mutex_lock_nest_lock);
61
62 #else /* !CONFIG_DEBUG_LOCK_ALLOC */
63
64 /**
65 * rt_mutex_lock - lock a rt_mutex
66 *
67 * @lock: the rt_mutex to be locked
68 */
69 void __sched rt_mutex_lock(struct rt_mutex *lock)
70 {
71 __rt_mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, NULL, 0);
72 }
73 EXPORT_SYMBOL_GPL(rt_mutex_lock);
74 #endif
75
76 /**
77 * rt_mutex_lock_interruptible - lock a rt_mutex interruptible
78 *
79 * @lock: the rt_mutex to be locked
80 *
81 * Returns:
82 * 0 on success
83 * -EINTR when interrupted by a signal
84 */
85 int __sched rt_mutex_lock_interruptible(struct rt_mutex *lock)
86 {
87 return __rt_mutex_lock_common(lock, TASK_INTERRUPTIBLE, NULL, 0);
88 }
89 EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible);
90
91 /**
92 * rt_mutex_lock_killable - lock a rt_mutex killable
93 *
94 * @lock: the rt_mutex to be locked
95 *
96 * Returns:
97 * 0 on success
98 * -EINTR when interrupted by a signal
99 */
100 int __sched rt_mutex_lock_killable(struct rt_mutex *lock)
101 {
102 return __rt_mutex_lock_common(lock, TASK_KILLABLE, NULL, 0);
103 }
104 EXPORT_SYMBOL_GPL(rt_mutex_lock_killable);
105
106 /**
107 * rt_mutex_trylock - try to lock a rt_mutex
108 *
109 * @lock: the rt_mutex to be locked
110 *
111 * This function can only be called in thread context. It's safe to call it
112 * from atomic regions, but not from hard or soft interrupt context.
113 *
114 * Returns:
115 * 1 on success
116 * 0 on contention
117 */
118 int __sched rt_mutex_trylock(struct rt_mutex *lock)
119 {
120 int ret;
121
122 if (IS_ENABLED(CONFIG_DEBUG_RT_MUTEXES) && WARN_ON_ONCE(!in_task()))
123 return 0;
124
125 ret = __rt_mutex_trylock(&lock->rtmutex);
126 if (ret)
127 mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_);
128
129 return ret;
130 }
131 EXPORT_SYMBOL_GPL(rt_mutex_trylock);
132
133 /**
134 * rt_mutex_unlock - unlock a rt_mutex
135 *
136 * @lock: the rt_mutex to be unlocked
137 */
138 void __sched rt_mutex_unlock(struct rt_mutex *lock)
139 {
140 mutex_release(&lock->dep_map, _RET_IP_);
141 __rt_mutex_unlock(&lock->rtmutex);
142 }
143 EXPORT_SYMBOL_GPL(rt_mutex_unlock);
144
145 /*
146 * Futex variants, must not use fastpath.
147 */
148 int __sched rt_mutex_futex_trylock(struct rt_mutex_base *lock)
149 {
150 return rt_mutex_slowtrylock(lock);
151 }
152
153 int __sched __rt_mutex_futex_trylock(struct rt_mutex_base *lock)
154 {
155 return __rt_mutex_slowtrylock(lock);
156 }
157
158 /**
159 * __rt_mutex_futex_unlock - Futex variant, that since futex variants
160 * do not use the fast-path, can be simple and will not need to retry.
161 *
162 * @lock: The rt_mutex to be unlocked
163 * @wqh: The wake queue head from which to get the next lock waiter
164 */
165 bool __sched __rt_mutex_futex_unlock(struct rt_mutex_base *lock,
166 struct rt_wake_q_head *wqh)
167 {
168 lockdep_assert_held(&lock->wait_lock);
169
170 debug_rt_mutex_unlock(lock);
171
172 if (!rt_mutex_has_waiters(lock)) {
173 lock->owner = NULL;
174 return false; /* done */
175 }
176
177 /*
178 * We've already deboosted, mark_wakeup_next_waiter() will
179 * retain preempt_disabled when we drop the wait_lock, to
180 * avoid inversion prior to the wakeup. preempt_disable()
181 * therein pairs with rt_mutex_postunlock().
182 */
183 mark_wakeup_next_waiter(wqh, lock);
184
185 return true; /* call postunlock() */
186 }
187
188 void __sched rt_mutex_futex_unlock(struct rt_mutex_base *lock)
189 {
190 DEFINE_RT_WAKE_Q(wqh);
191 unsigned long flags;
192 bool postunlock;
193
194 raw_spin_lock_irqsave(&lock->wait_lock, flags);
195 postunlock = __rt_mutex_futex_unlock(lock, &wqh);
196 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
197
198 if (postunlock)
199 rt_mutex_postunlock(&wqh);
200 }
201
202 /**
203 * __rt_mutex_init - initialize the rt_mutex
204 *
205 * @lock: The rt_mutex to be initialized
206 * @name: The lock name used for debugging
207 * @key: The lock class key used for debugging
208 *
209 * Initialize the rt_mutex to unlocked state.
210 *
211 * Initializing of a locked rt_mutex is not allowed
212 */
213 void __sched __rt_mutex_init(struct rt_mutex *lock, const char *name,
214 struct lock_class_key *key)
215 {
216 debug_check_no_locks_freed((void *)lock, sizeof(*lock));
217 __rt_mutex_base_init(&lock->rtmutex);
218 lockdep_init_map_wait(&lock->dep_map, name, key, 0, LD_WAIT_SLEEP);
219 }
220 EXPORT_SYMBOL_GPL(__rt_mutex_init);
221
222 /**
223 * rt_mutex_init_proxy_locked - initialize and lock a rt_mutex on behalf of a
224 * proxy owner
225 *
226 * @lock: the rt_mutex to be locked
227 * @proxy_owner:the task to set as owner
228 *
229 * No locking. Caller has to do serializing itself
230 *
231 * Special API call for PI-futex support. This initializes the rtmutex and
232 * assigns it to @proxy_owner. Concurrent operations on the rtmutex are not
233 * possible at this point because the pi_state which contains the rtmutex
234 * is not yet visible to other tasks.
235 */
236 void __sched rt_mutex_init_proxy_locked(struct rt_mutex_base *lock,
237 struct task_struct *proxy_owner)
238 {
239 static struct lock_class_key pi_futex_key;
240
241 __rt_mutex_base_init(lock);
242 /*
243 * On PREEMPT_RT the futex hashbucket spinlock becomes 'sleeping'
244 * and rtmutex based. That causes a lockdep false positive, because
245 * some of the futex functions invoke spin_unlock(&hb->lock) with
246 * the wait_lock of the rtmutex associated to the pi_futex held.
247 * spin_unlock() in turn takes wait_lock of the rtmutex on which
248 * the spinlock is based, which makes lockdep notice a lock
249 * recursion. Give the futex/rtmutex wait_lock a separate key.
250 */
251 lockdep_set_class(&lock->wait_lock, &pi_futex_key);
252 rt_mutex_set_owner(lock, proxy_owner);
253 }
254
255 /**
256 * rt_mutex_proxy_unlock - release a lock on behalf of owner
257 *
258 * @lock: the rt_mutex to be locked
259 *
260 * No locking. Caller has to do serializing itself
261 *
262 * Special API call for PI-futex support. This just cleans up the rtmutex
263 * (debugging) state. Concurrent operations on this rt_mutex are not
264 * possible because it belongs to the pi_state which is about to be freed
265 * and it is not longer visible to other tasks.
266 */
267 void __sched rt_mutex_proxy_unlock(struct rt_mutex_base *lock)
268 {
269 debug_rt_mutex_proxy_unlock(lock);
270 rt_mutex_clear_owner(lock);
271 }
272
273 /**
274 * __rt_mutex_start_proxy_lock() - Start lock acquisition for another task
275 * @lock: the rt_mutex to take
276 * @waiter: the pre-initialized rt_mutex_waiter
277 * @task: the task to prepare
278 *
279 * Starts the rt_mutex acquire; it enqueues the @waiter and does deadlock
280 * detection. It does not wait, see rt_mutex_wait_proxy_lock() for that.
281 *
282 * NOTE: does _NOT_ remove the @waiter on failure; must either call
283 * rt_mutex_wait_proxy_lock() or rt_mutex_cleanup_proxy_lock() after this.
284 *
285 * Returns:
286 * 0 - task blocked on lock
287 * 1 - acquired the lock for task, caller should wake it up
288 * <0 - error
289 *
290 * Special API call for PI-futex support.
291 */
292 int __sched __rt_mutex_start_proxy_lock(struct rt_mutex_base *lock,
293 struct rt_mutex_waiter *waiter,
294 struct task_struct *task)
295 {
296 int ret;
297
298 lockdep_assert_held(&lock->wait_lock);
299
300 if (try_to_take_rt_mutex(lock, task, NULL))
301 return 1;
302
303 /* We enforce deadlock detection for futexes */
304 ret = task_blocks_on_rt_mutex(lock, waiter, task, NULL,
305 RT_MUTEX_FULL_CHAINWALK);
306
307 if (ret && !rt_mutex_owner(lock)) {
308 /*
309 * Reset the return value. We might have
310 * returned with -EDEADLK and the owner
311 * released the lock while we were walking the
312 * pi chain. Let the waiter sort it out.
313 */
314 ret = 0;
315 }
316
317 return ret;
318 }
319
320 /**
321 * rt_mutex_start_proxy_lock() - Start lock acquisition for another task
322 * @lock: the rt_mutex to take
323 * @waiter: the pre-initialized rt_mutex_waiter
324 * @task: the task to prepare
325 *
326 * Starts the rt_mutex acquire; it enqueues the @waiter and does deadlock
327 * detection. It does not wait, see rt_mutex_wait_proxy_lock() for that.
328 *
329 * NOTE: unlike __rt_mutex_start_proxy_lock this _DOES_ remove the @waiter
330 * on failure.
331 *
332 * Returns:
333 * 0 - task blocked on lock
334 * 1 - acquired the lock for task, caller should wake it up
335 * <0 - error
336 *
337 * Special API call for PI-futex support.
338 */
339 int __sched rt_mutex_start_proxy_lock(struct rt_mutex_base *lock,
340 struct rt_mutex_waiter *waiter,
341 struct task_struct *task)
342 {
343 int ret;
344
345 raw_spin_lock_irq(&lock->wait_lock);
346 ret = __rt_mutex_start_proxy_lock(lock, waiter, task);
347 if (unlikely(ret))
348 remove_waiter(lock, waiter);
349 raw_spin_unlock_irq(&lock->wait_lock);
350
351 return ret;
352 }
353
354 /**
355 * rt_mutex_wait_proxy_lock() - Wait for lock acquisition
356 * @lock: the rt_mutex we were woken on
357 * @to: the timeout, null if none. hrtimer should already have
358 * been started.
359 * @waiter: the pre-initialized rt_mutex_waiter
360 *
361 * Wait for the lock acquisition started on our behalf by
362 * rt_mutex_start_proxy_lock(). Upon failure, the caller must call
363 * rt_mutex_cleanup_proxy_lock().
364 *
365 * Returns:
366 * 0 - success
367 * <0 - error, one of -EINTR, -ETIMEDOUT
368 *
369 * Special API call for PI-futex support
370 */
371 int __sched rt_mutex_wait_proxy_lock(struct rt_mutex_base *lock,
372 struct hrtimer_sleeper *to,
373 struct rt_mutex_waiter *waiter)
374 {
375 int ret;
376
377 raw_spin_lock_irq(&lock->wait_lock);
378 /* sleep on the mutex */
379 set_current_state(TASK_INTERRUPTIBLE);
380 ret = rt_mutex_slowlock_block(lock, NULL, TASK_INTERRUPTIBLE, to, waiter);
381 /*
382 * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
383 * have to fix that up.
384 */
385 fixup_rt_mutex_waiters(lock, true);
386 raw_spin_unlock_irq(&lock->wait_lock);
387
388 return ret;
389 }
390
391 /**
392 * rt_mutex_cleanup_proxy_lock() - Cleanup failed lock acquisition
393 * @lock: the rt_mutex we were woken on
394 * @waiter: the pre-initialized rt_mutex_waiter
395 *
396 * Attempt to clean up after a failed __rt_mutex_start_proxy_lock() or
397 * rt_mutex_wait_proxy_lock().
398 *
399 * Unless we acquired the lock; we're still enqueued on the wait-list and can
400 * in fact still be granted ownership until we're removed. Therefore we can
401 * find we are in fact the owner and must disregard the
402 * rt_mutex_wait_proxy_lock() failure.
403 *
404 * Returns:
405 * true - did the cleanup, we done.
406 * false - we acquired the lock after rt_mutex_wait_proxy_lock() returned,
407 * caller should disregards its return value.
408 *
409 * Special API call for PI-futex support
410 */
411 bool __sched rt_mutex_cleanup_proxy_lock(struct rt_mutex_base *lock,
412 struct rt_mutex_waiter *waiter)
413 {
414 bool cleanup = false;
415
416 raw_spin_lock_irq(&lock->wait_lock);
417 /*
418 * Do an unconditional try-lock, this deals with the lock stealing
419 * state where __rt_mutex_futex_unlock() -> mark_wakeup_next_waiter()
420 * sets a NULL owner.
421 *
422 * We're not interested in the return value, because the subsequent
423 * test on rt_mutex_owner() will infer that. If the trylock succeeded,
424 * we will own the lock and it will have removed the waiter. If we
425 * failed the trylock, we're still not owner and we need to remove
426 * ourselves.
427 */
428 try_to_take_rt_mutex(lock, current, waiter);
429 /*
430 * Unless we're the owner; we're still enqueued on the wait_list.
431 * So check if we became owner, if not, take us off the wait_list.
432 */
433 if (rt_mutex_owner(lock) != current) {
434 remove_waiter(lock, waiter);
435 cleanup = true;
436 }
437 /*
438 * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
439 * have to fix that up.
440 */
441 fixup_rt_mutex_waiters(lock, false);
442
443 raw_spin_unlock_irq(&lock->wait_lock);
444
445 return cleanup;
446 }
447
448 /*
449 * Recheck the pi chain, in case we got a priority setting
450 *
451 * Called from sched_setscheduler
452 */
453 void __sched rt_mutex_adjust_pi(struct task_struct *task)
454 {
455 struct rt_mutex_waiter *waiter;
456 struct rt_mutex_base *next_lock;
457 unsigned long flags;
458
459 raw_spin_lock_irqsave(&task->pi_lock, flags);
460
461 waiter = task->pi_blocked_on;
462 if (!waiter || rt_waiter_node_equal(&waiter->tree, task_to_waiter_node(task))) {
463 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
464 return;
465 }
466 next_lock = waiter->lock;
467 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
468
469 /* gets dropped in rt_mutex_adjust_prio_chain()! */
470 get_task_struct(task);
471
472 rt_mutex_adjust_prio_chain(task, RT_MUTEX_MIN_CHAINWALK, NULL,
473 next_lock, NULL, task);
474 }
475
476 /*
477 * Performs the wakeup of the top-waiter and re-enables preemption.
478 */
479 void __sched rt_mutex_postunlock(struct rt_wake_q_head *wqh)
480 {
481 rt_mutex_wake_up_q(wqh);
482 }
483
484 #ifdef CONFIG_DEBUG_RT_MUTEXES
485 void rt_mutex_debug_task_free(struct task_struct *task)
486 {
487 DEBUG_LOCKS_WARN_ON(!RB_EMPTY_ROOT(&task->pi_waiters.rb_root));
488 DEBUG_LOCKS_WARN_ON(task->pi_blocked_on);
489 }
490 #endif
491
492 #ifdef CONFIG_PREEMPT_RT
493 /* Mutexes */
494 void __mutex_rt_init(struct mutex *mutex, const char *name,
495 struct lock_class_key *key)
496 {
497 debug_check_no_locks_freed((void *)mutex, sizeof(*mutex));
498 lockdep_init_map_wait(&mutex->dep_map, name, key, 0, LD_WAIT_SLEEP);
499 }
500 EXPORT_SYMBOL(__mutex_rt_init);
501
502 static __always_inline int __mutex_lock_common(struct mutex *lock,
503 unsigned int state,
504 unsigned int subclass,
505 struct lockdep_map *nest_lock,
506 unsigned long ip)
507 {
508 int ret;
509
510 might_sleep();
511 mutex_acquire_nest(&lock->dep_map, subclass, 0, nest_lock, ip);
512 ret = __rt_mutex_lock(&lock->rtmutex, state);
513 if (ret)
514 mutex_release(&lock->dep_map, ip);
515 else
516 lock_acquired(&lock->dep_map, ip);
517 return ret;
518 }
519
520 #ifdef CONFIG_DEBUG_LOCK_ALLOC
521 void __sched mutex_lock_nested(struct mutex *lock, unsigned int subclass)
522 {
523 __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, subclass, NULL, _RET_IP_);
524 }
525 EXPORT_SYMBOL_GPL(mutex_lock_nested);
526
527 void __sched _mutex_lock_nest_lock(struct mutex *lock,
528 struct lockdep_map *nest_lock)
529 {
530 __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, 0, nest_lock, _RET_IP_);
531 }
532 EXPORT_SYMBOL_GPL(_mutex_lock_nest_lock);
533
534 int __sched mutex_lock_interruptible_nested(struct mutex *lock,
535 unsigned int subclass)
536 {
537 return __mutex_lock_common(lock, TASK_INTERRUPTIBLE, subclass, NULL, _RET_IP_);
538 }
539 EXPORT_SYMBOL_GPL(mutex_lock_interruptible_nested);
540
541 int __sched mutex_lock_killable_nested(struct mutex *lock,
542 unsigned int subclass)
543 {
544 return __mutex_lock_common(lock, TASK_KILLABLE, subclass, NULL, _RET_IP_);
545 }
546 EXPORT_SYMBOL_GPL(mutex_lock_killable_nested);
547
548 void __sched mutex_lock_io_nested(struct mutex *lock, unsigned int subclass)
549 {
550 int token;
551
552 might_sleep();
553
554 token = io_schedule_prepare();
555 __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, subclass, NULL, _RET_IP_);
556 io_schedule_finish(token);
557 }
558 EXPORT_SYMBOL_GPL(mutex_lock_io_nested);
559
560 #else /* CONFIG_DEBUG_LOCK_ALLOC */
561
562 void __sched mutex_lock(struct mutex *lock)
563 {
564 __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, 0, NULL, _RET_IP_);
565 }
566 EXPORT_SYMBOL(mutex_lock);
567
568 int __sched mutex_lock_interruptible(struct mutex *lock)
569 {
570 return __mutex_lock_common(lock, TASK_INTERRUPTIBLE, 0, NULL, _RET_IP_);
571 }
572 EXPORT_SYMBOL(mutex_lock_interruptible);
573
574 int __sched mutex_lock_killable(struct mutex *lock)
575 {
576 return __mutex_lock_common(lock, TASK_KILLABLE, 0, NULL, _RET_IP_);
577 }
578 EXPORT_SYMBOL(mutex_lock_killable);
579
580 void __sched mutex_lock_io(struct mutex *lock)
581 {
582 int token = io_schedule_prepare();
583
584 __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, 0, NULL, _RET_IP_);
585 io_schedule_finish(token);
586 }
587 EXPORT_SYMBOL(mutex_lock_io);
588 #endif /* !CONFIG_DEBUG_LOCK_ALLOC */
589
590 int __sched mutex_trylock(struct mutex *lock)
591 {
592 int ret;
593
594 if (IS_ENABLED(CONFIG_DEBUG_RT_MUTEXES) && WARN_ON_ONCE(!in_task()))
595 return 0;
596
597 ret = __rt_mutex_trylock(&lock->rtmutex);
598 if (ret)
599 mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_);
600
601 return ret;
602 }
603 EXPORT_SYMBOL(mutex_trylock);
604
605 void __sched mutex_unlock(struct mutex *lock)
606 {
607 mutex_release(&lock->dep_map, _RET_IP_);
608 __rt_mutex_unlock(&lock->rtmutex);
609 }
610 EXPORT_SYMBOL(mutex_unlock);
611
612 #endif /* CONFIG_PREEMPT_RT */
613
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.