]> Git Repo - J-linux.git/blob - kernel/locking/rtmutex_api.c
Merge tag 'vfs-6.13-rc7.fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs
[J-linux.git] / kernel / locking / rtmutex_api.c
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          * mark_wakeup_next_waiter() deboosts and retains preemption
179          * disabled when dropping the wait_lock, to avoid inversion prior
180          * to the wakeup.  preempt_disable() therein pairs with the
181          * preempt_enable() in 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  * @wake_q:             the wake_q to wake tasks after we release the wait_lock
279  *
280  * Starts the rt_mutex acquire; it enqueues the @waiter and does deadlock
281  * detection. It does not wait, see rt_mutex_wait_proxy_lock() for that.
282  *
283  * NOTE: does _NOT_ remove the @waiter on failure; must either call
284  * rt_mutex_wait_proxy_lock() or rt_mutex_cleanup_proxy_lock() after this.
285  *
286  * Returns:
287  *  0 - task blocked on lock
288  *  1 - acquired the lock for task, caller should wake it up
289  * <0 - error
290  *
291  * Special API call for PI-futex support.
292  */
293 int __sched __rt_mutex_start_proxy_lock(struct rt_mutex_base *lock,
294                                         struct rt_mutex_waiter *waiter,
295                                         struct task_struct *task,
296                                         struct wake_q_head *wake_q)
297 {
298         int ret;
299
300         lockdep_assert_held(&lock->wait_lock);
301
302         if (try_to_take_rt_mutex(lock, task, NULL))
303                 return 1;
304
305         /* We enforce deadlock detection for futexes */
306         ret = task_blocks_on_rt_mutex(lock, waiter, task, NULL,
307                                       RT_MUTEX_FULL_CHAINWALK, wake_q);
308
309         if (ret && !rt_mutex_owner(lock)) {
310                 /*
311                  * Reset the return value. We might have
312                  * returned with -EDEADLK and the owner
313                  * released the lock while we were walking the
314                  * pi chain.  Let the waiter sort it out.
315                  */
316                 ret = 0;
317         }
318
319         return ret;
320 }
321
322 /**
323  * rt_mutex_start_proxy_lock() - Start lock acquisition for another task
324  * @lock:               the rt_mutex to take
325  * @waiter:             the pre-initialized rt_mutex_waiter
326  * @task:               the task to prepare
327  *
328  * Starts the rt_mutex acquire; it enqueues the @waiter and does deadlock
329  * detection. It does not wait, see rt_mutex_wait_proxy_lock() for that.
330  *
331  * NOTE: unlike __rt_mutex_start_proxy_lock this _DOES_ remove the @waiter
332  * on failure.
333  *
334  * Returns:
335  *  0 - task blocked on lock
336  *  1 - acquired the lock for task, caller should wake it up
337  * <0 - error
338  *
339  * Special API call for PI-futex support.
340  */
341 int __sched rt_mutex_start_proxy_lock(struct rt_mutex_base *lock,
342                                       struct rt_mutex_waiter *waiter,
343                                       struct task_struct *task)
344 {
345         int ret;
346         DEFINE_WAKE_Q(wake_q);
347
348         raw_spin_lock_irq(&lock->wait_lock);
349         ret = __rt_mutex_start_proxy_lock(lock, waiter, task, &wake_q);
350         if (unlikely(ret))
351                 remove_waiter(lock, waiter);
352         preempt_disable();
353         raw_spin_unlock_irq(&lock->wait_lock);
354         wake_up_q(&wake_q);
355         preempt_enable();
356
357         return ret;
358 }
359
360 /**
361  * rt_mutex_wait_proxy_lock() - Wait for lock acquisition
362  * @lock:               the rt_mutex we were woken on
363  * @to:                 the timeout, null if none. hrtimer should already have
364  *                      been started.
365  * @waiter:             the pre-initialized rt_mutex_waiter
366  *
367  * Wait for the lock acquisition started on our behalf by
368  * rt_mutex_start_proxy_lock(). Upon failure, the caller must call
369  * rt_mutex_cleanup_proxy_lock().
370  *
371  * Returns:
372  *  0 - success
373  * <0 - error, one of -EINTR, -ETIMEDOUT
374  *
375  * Special API call for PI-futex support
376  */
377 int __sched rt_mutex_wait_proxy_lock(struct rt_mutex_base *lock,
378                                      struct hrtimer_sleeper *to,
379                                      struct rt_mutex_waiter *waiter)
380 {
381         int ret;
382
383         raw_spin_lock_irq(&lock->wait_lock);
384         /* sleep on the mutex */
385         set_current_state(TASK_INTERRUPTIBLE);
386         ret = rt_mutex_slowlock_block(lock, NULL, TASK_INTERRUPTIBLE, to, waiter, NULL);
387         /*
388          * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
389          * have to fix that up.
390          */
391         fixup_rt_mutex_waiters(lock, true);
392         raw_spin_unlock_irq(&lock->wait_lock);
393
394         return ret;
395 }
396
397 /**
398  * rt_mutex_cleanup_proxy_lock() - Cleanup failed lock acquisition
399  * @lock:               the rt_mutex we were woken on
400  * @waiter:             the pre-initialized rt_mutex_waiter
401  *
402  * Attempt to clean up after a failed __rt_mutex_start_proxy_lock() or
403  * rt_mutex_wait_proxy_lock().
404  *
405  * Unless we acquired the lock; we're still enqueued on the wait-list and can
406  * in fact still be granted ownership until we're removed. Therefore we can
407  * find we are in fact the owner and must disregard the
408  * rt_mutex_wait_proxy_lock() failure.
409  *
410  * Returns:
411  *  true  - did the cleanup, we done.
412  *  false - we acquired the lock after rt_mutex_wait_proxy_lock() returned,
413  *          caller should disregards its return value.
414  *
415  * Special API call for PI-futex support
416  */
417 bool __sched rt_mutex_cleanup_proxy_lock(struct rt_mutex_base *lock,
418                                          struct rt_mutex_waiter *waiter)
419 {
420         bool cleanup = false;
421
422         raw_spin_lock_irq(&lock->wait_lock);
423         /*
424          * Do an unconditional try-lock, this deals with the lock stealing
425          * state where __rt_mutex_futex_unlock() -> mark_wakeup_next_waiter()
426          * sets a NULL owner.
427          *
428          * We're not interested in the return value, because the subsequent
429          * test on rt_mutex_owner() will infer that. If the trylock succeeded,
430          * we will own the lock and it will have removed the waiter. If we
431          * failed the trylock, we're still not owner and we need to remove
432          * ourselves.
433          */
434         try_to_take_rt_mutex(lock, current, waiter);
435         /*
436          * Unless we're the owner; we're still enqueued on the wait_list.
437          * So check if we became owner, if not, take us off the wait_list.
438          */
439         if (rt_mutex_owner(lock) != current) {
440                 remove_waiter(lock, waiter);
441                 cleanup = true;
442         }
443         /*
444          * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
445          * have to fix that up.
446          */
447         fixup_rt_mutex_waiters(lock, false);
448
449         raw_spin_unlock_irq(&lock->wait_lock);
450
451         return cleanup;
452 }
453
454 /*
455  * Recheck the pi chain, in case we got a priority setting
456  *
457  * Called from sched_setscheduler
458  */
459 void __sched rt_mutex_adjust_pi(struct task_struct *task)
460 {
461         struct rt_mutex_waiter *waiter;
462         struct rt_mutex_base *next_lock;
463         unsigned long flags;
464
465         raw_spin_lock_irqsave(&task->pi_lock, flags);
466
467         waiter = task->pi_blocked_on;
468         if (!waiter || rt_waiter_node_equal(&waiter->tree, task_to_waiter_node(task))) {
469                 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
470                 return;
471         }
472         next_lock = waiter->lock;
473         raw_spin_unlock_irqrestore(&task->pi_lock, flags);
474
475         /* gets dropped in rt_mutex_adjust_prio_chain()! */
476         get_task_struct(task);
477
478         rt_mutex_adjust_prio_chain(task, RT_MUTEX_MIN_CHAINWALK, NULL,
479                                    next_lock, NULL, task);
480 }
481
482 /*
483  * Performs the wakeup of the top-waiter and re-enables preemption.
484  */
485 void __sched rt_mutex_postunlock(struct rt_wake_q_head *wqh)
486 {
487         rt_mutex_wake_up_q(wqh);
488 }
489
490 #ifdef CONFIG_DEBUG_RT_MUTEXES
491 void rt_mutex_debug_task_free(struct task_struct *task)
492 {
493         DEBUG_LOCKS_WARN_ON(!RB_EMPTY_ROOT(&task->pi_waiters.rb_root));
494         DEBUG_LOCKS_WARN_ON(task->pi_blocked_on);
495 }
496 #endif
497
498 #ifdef CONFIG_PREEMPT_RT
499 /* Mutexes */
500 void __mutex_rt_init(struct mutex *mutex, const char *name,
501                      struct lock_class_key *key)
502 {
503         debug_check_no_locks_freed((void *)mutex, sizeof(*mutex));
504         lockdep_init_map_wait(&mutex->dep_map, name, key, 0, LD_WAIT_SLEEP);
505 }
506 EXPORT_SYMBOL(__mutex_rt_init);
507
508 static __always_inline int __mutex_lock_common(struct mutex *lock,
509                                                unsigned int state,
510                                                unsigned int subclass,
511                                                struct lockdep_map *nest_lock,
512                                                unsigned long ip)
513 {
514         int ret;
515
516         might_sleep();
517         mutex_acquire_nest(&lock->dep_map, subclass, 0, nest_lock, ip);
518         ret = __rt_mutex_lock(&lock->rtmutex, state);
519         if (ret)
520                 mutex_release(&lock->dep_map, ip);
521         else
522                 lock_acquired(&lock->dep_map, ip);
523         return ret;
524 }
525
526 #ifdef CONFIG_DEBUG_LOCK_ALLOC
527 void __sched mutex_lock_nested(struct mutex *lock, unsigned int subclass)
528 {
529         __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, subclass, NULL, _RET_IP_);
530 }
531 EXPORT_SYMBOL_GPL(mutex_lock_nested);
532
533 void __sched _mutex_lock_nest_lock(struct mutex *lock,
534                                    struct lockdep_map *nest_lock)
535 {
536         __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, 0, nest_lock, _RET_IP_);
537 }
538 EXPORT_SYMBOL_GPL(_mutex_lock_nest_lock);
539
540 int __sched mutex_lock_interruptible_nested(struct mutex *lock,
541                                             unsigned int subclass)
542 {
543         return __mutex_lock_common(lock, TASK_INTERRUPTIBLE, subclass, NULL, _RET_IP_);
544 }
545 EXPORT_SYMBOL_GPL(mutex_lock_interruptible_nested);
546
547 int __sched mutex_lock_killable_nested(struct mutex *lock,
548                                             unsigned int subclass)
549 {
550         return __mutex_lock_common(lock, TASK_KILLABLE, subclass, NULL, _RET_IP_);
551 }
552 EXPORT_SYMBOL_GPL(mutex_lock_killable_nested);
553
554 void __sched mutex_lock_io_nested(struct mutex *lock, unsigned int subclass)
555 {
556         int token;
557
558         might_sleep();
559
560         token = io_schedule_prepare();
561         __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, subclass, NULL, _RET_IP_);
562         io_schedule_finish(token);
563 }
564 EXPORT_SYMBOL_GPL(mutex_lock_io_nested);
565
566 #else /* CONFIG_DEBUG_LOCK_ALLOC */
567
568 void __sched mutex_lock(struct mutex *lock)
569 {
570         __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, 0, NULL, _RET_IP_);
571 }
572 EXPORT_SYMBOL(mutex_lock);
573
574 int __sched mutex_lock_interruptible(struct mutex *lock)
575 {
576         return __mutex_lock_common(lock, TASK_INTERRUPTIBLE, 0, NULL, _RET_IP_);
577 }
578 EXPORT_SYMBOL(mutex_lock_interruptible);
579
580 int __sched mutex_lock_killable(struct mutex *lock)
581 {
582         return __mutex_lock_common(lock, TASK_KILLABLE, 0, NULL, _RET_IP_);
583 }
584 EXPORT_SYMBOL(mutex_lock_killable);
585
586 void __sched mutex_lock_io(struct mutex *lock)
587 {
588         int token = io_schedule_prepare();
589
590         __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, 0, NULL, _RET_IP_);
591         io_schedule_finish(token);
592 }
593 EXPORT_SYMBOL(mutex_lock_io);
594 #endif /* !CONFIG_DEBUG_LOCK_ALLOC */
595
596 int __sched mutex_trylock(struct mutex *lock)
597 {
598         int ret;
599
600         if (IS_ENABLED(CONFIG_DEBUG_RT_MUTEXES) && WARN_ON_ONCE(!in_task()))
601                 return 0;
602
603         ret = __rt_mutex_trylock(&lock->rtmutex);
604         if (ret)
605                 mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_);
606
607         return ret;
608 }
609 EXPORT_SYMBOL(mutex_trylock);
610
611 void __sched mutex_unlock(struct mutex *lock)
612 {
613         mutex_release(&lock->dep_map, _RET_IP_);
614         __rt_mutex_unlock(&lock->rtmutex);
615 }
616 EXPORT_SYMBOL(mutex_unlock);
617
618 #endif /* CONFIG_PREEMPT_RT */
This page took 0.064268 seconds and 4 git commands to generate.