projects/linux·blob v6.6 · refs

1833 lines · 49.6 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * RT-Mutexes: simple blocking mutual exclusion locks with PI support
4 *
5 * started by Ingo Molnar and Thomas Gleixner.
6 *
7 * Copyright (C) 2004-2006 Red Hat, Inc., Ingo Molnar <[email protected]>
8 * Copyright (C) 2005-2006 Timesys Corp., Thomas Gleixner <[email protected]>
9 * Copyright (C) 2005 Kihon Technologies Inc., Steven Rostedt
10 * Copyright (C) 2006 Esben Nielsen
11 * Adaptive Spinlocks:
12 * Copyright (C) 2008 Novell, Inc., Gregory Haskins, Sven Dietrich,
13 * and Peter Morreale,
14 * Adaptive Spinlocks simplification:
15 * Copyright (C) 2008 Red Hat, Inc., Steven Rostedt <[email protected]>
16 *
17 * See Documentation/locking/rt-mutex-design.rst for details.
18 */
19#include <linux/sched.h>
20#include <linux/sched/debug.h>
21#include <linux/sched/deadline.h>
22#include <linux/sched/signal.h>
23#include <linux/sched/rt.h>
24#include <linux/sched/wake_q.h>
25#include <linux/ww_mutex.h>
26
27#include <trace/events/lock.h>
28
29#include "rtmutex_common.h"
30
31#ifndef WW_RT
32# define build_ww_mutex() (false)
33# define ww_container_of(rtm) NULL
34
35static inline int __ww_mutex_add_waiter(struct rt_mutex_waiter *waiter,
36 struct rt_mutex *lock,
37 struct ww_acquire_ctx *ww_ctx)
38{
39 return 0;
40}
41
42static inline void __ww_mutex_check_waiters(struct rt_mutex *lock,
43 struct ww_acquire_ctx *ww_ctx)
44{
45}
46
47static inline void ww_mutex_lock_acquired(struct ww_mutex *lock,
48 struct ww_acquire_ctx *ww_ctx)
49{
50}
51
52static inline int __ww_mutex_check_kill(struct rt_mutex *lock,
53 struct rt_mutex_waiter *waiter,
54 struct ww_acquire_ctx *ww_ctx)
55{
56 return 0;
57}
58
59#else
60# define build_ww_mutex() (true)
61# define ww_container_of(rtm) container_of(rtm, struct ww_mutex, base)
62# include "ww_mutex.h"
63#endif
64
65/*
66 * lock->owner state tracking:
67 *
68 * lock->owner holds the task_struct pointer of the owner. Bit 0
69 * is used to keep track of the "lock has waiters" state.
70 *
71 * owner bit0
72 * NULL 0 lock is free (fast acquire possible)
73 * NULL 1 lock is free and has waiters and the top waiter
74 * is going to take the lock*
75 * taskpointer 0 lock is held (fast release possible)
76 * taskpointer 1 lock is held and has waiters**
77 *
78 * The fast atomic compare exchange based acquire and release is only
79 * possible when bit 0 of lock->owner is 0.
80 *
81 * (*) It also can be a transitional state when grabbing the lock
82 * with ->wait_lock is held. To prevent any fast path cmpxchg to the lock,
83 * we need to set the bit0 before looking at the lock, and the owner may be
84 * NULL in this small time, hence this can be a transitional state.
85 *
86 * (**) There is a small time when bit 0 is set but there are no
87 * waiters. This can happen when grabbing the lock in the slow path.
88 * To prevent a cmpxchg of the owner releasing the lock, we need to
89 * set this bit before looking at the lock.
90 */
91
92static __always_inline struct task_struct *
93rt_mutex_owner_encode(struct rt_mutex_base *lock, struct task_struct *owner)
94{
95 unsigned long val = (unsigned long)owner;
96
97 if (rt_mutex_has_waiters(lock))
98 val |= RT_MUTEX_HAS_WAITERS;
99
100 return (struct task_struct *)val;
101}
102
103static __always_inline void
104rt_mutex_set_owner(struct rt_mutex_base *lock, struct task_struct *owner)
105{
106 /*
107 * lock->wait_lock is held but explicit acquire semantics are needed
108 * for a new lock owner so WRITE_ONCE is insufficient.
109 */
110 xchg_acquire(&lock->owner, rt_mutex_owner_encode(lock, owner));
111}
112
113static __always_inline void rt_mutex_clear_owner(struct rt_mutex_base *lock)
114{
115 /* lock->wait_lock is held so the unlock provides release semantics. */
116 WRITE_ONCE(lock->owner, rt_mutex_owner_encode(lock, NULL));
117}
118
119static __always_inline void clear_rt_mutex_waiters(struct rt_mutex_base *lock)
120{
121 lock->owner = (struct task_struct *)
122 ((unsigned long)lock->owner & ~RT_MUTEX_HAS_WAITERS);
123}
124
125static __always_inline void
126fixup_rt_mutex_waiters(struct rt_mutex_base *lock, bool acquire_lock)
127{
128 unsigned long owner, *p = (unsigned long *) &lock->owner;
129
130 if (rt_mutex_has_waiters(lock))
131 return;
132
133 /*
134 * The rbtree has no waiters enqueued, now make sure that the
135 * lock->owner still has the waiters bit set, otherwise the
136 * following can happen:
137 *
138 * CPU 0 CPU 1 CPU2
139 * l->owner=T1
140 * rt_mutex_lock(l)
141 * lock(l->lock)
142 * l->owner = T1 | HAS_WAITERS;
143 * enqueue(T2)
144 * boost()
145 * unlock(l->lock)
146 * block()
147 *
148 * rt_mutex_lock(l)
149 * lock(l->lock)
150 * l->owner = T1 | HAS_WAITERS;
151 * enqueue(T3)
152 * boost()
153 * unlock(l->lock)
154 * block()
155 * signal(->T2) signal(->T3)
156 * lock(l->lock)
157 * dequeue(T2)
158 * deboost()
159 * unlock(l->lock)
160 * lock(l->lock)
161 * dequeue(T3)
162 * ==> wait list is empty
163 * deboost()
164 * unlock(l->lock)
165 * lock(l->lock)
166 * fixup_rt_mutex_waiters()
167 * if (wait_list_empty(l) {
168 * l->owner = owner
169 * owner = l->owner & ~HAS_WAITERS;
170 * ==> l->owner = T1
171 * }
172 * lock(l->lock)
173 * rt_mutex_unlock(l) fixup_rt_mutex_waiters()
174 * if (wait_list_empty(l) {
175 * owner = l->owner & ~HAS_WAITERS;
176 * cmpxchg(l->owner, T1, NULL)
177 * ===> Success (l->owner = NULL)
178 *
179 * l->owner = owner
180 * ==> l->owner = T1
181 * }
182 *
183 * With the check for the waiter bit in place T3 on CPU2 will not
184 * overwrite. All tasks fiddling with the waiters bit are
185 * serialized by l->lock, so nothing else can modify the waiters
186 * bit. If the bit is set then nothing can change l->owner either
187 * so the simple RMW is safe. The cmpxchg() will simply fail if it
188 * happens in the middle of the RMW because the waiters bit is
189 * still set.
190 */
191 owner = READ_ONCE(*p);
192 if (owner & RT_MUTEX_HAS_WAITERS) {
193 /*
194 * See rt_mutex_set_owner() and rt_mutex_clear_owner() on
195 * why xchg_acquire() is used for updating owner for
196 * locking and WRITE_ONCE() for unlocking.
197 *
198 * WRITE_ONCE() would work for the acquire case too, but
199 * in case that the lock acquisition failed it might
200 * force other lockers into the slow path unnecessarily.
201 */
202 if (acquire_lock)
203 xchg_acquire(p, owner & ~RT_MUTEX_HAS_WAITERS);
204 else
205 WRITE_ONCE(*p, owner & ~RT_MUTEX_HAS_WAITERS);
206 }
207}
208
209/*
210 * We can speed up the acquire/release, if there's no debugging state to be
211 * set up.
212 */
213#ifndef CONFIG_DEBUG_RT_MUTEXES
214static __always_inline bool rt_mutex_cmpxchg_acquire(struct rt_mutex_base *lock,
215 struct task_struct *old,
216 struct task_struct *new)
217{
218 return try_cmpxchg_acquire(&lock->owner, &old, new);
219}
220
221static __always_inline bool rt_mutex_cmpxchg_release(struct rt_mutex_base *lock,
222 struct task_struct *old,
223 struct task_struct *new)
224{
225 return try_cmpxchg_release(&lock->owner, &old, new);
226}
227
228/*
229 * Callers must hold the ->wait_lock -- which is the whole purpose as we force
230 * all future threads that attempt to [Rmw] the lock to the slowpath. As such
231 * relaxed semantics suffice.
232 */
233static __always_inline void mark_rt_mutex_waiters(struct rt_mutex_base *lock)
234{
235 unsigned long owner, *p = (unsigned long *) &lock->owner;
236
237 do {
238 owner = *p;
239 } while (cmpxchg_relaxed(p, owner,
240 owner | RT_MUTEX_HAS_WAITERS) != owner);
241
242 /*
243 * The cmpxchg loop above is relaxed to avoid back-to-back ACQUIRE
244 * operations in the event of contention. Ensure the successful
245 * cmpxchg is visible.
246 */
247 smp_mb__after_atomic();
248}
249
250/*
251 * Safe fastpath aware unlock:
252 * 1) Clear the waiters bit
253 * 2) Drop lock->wait_lock
254 * 3) Try to unlock the lock with cmpxchg
255 */
256static __always_inline bool unlock_rt_mutex_safe(struct rt_mutex_base *lock,
257 unsigned long flags)
258 __releases(lock->wait_lock)
259{
260 struct task_struct *owner = rt_mutex_owner(lock);
261
262 clear_rt_mutex_waiters(lock);
263 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
264 /*
265 * If a new waiter comes in between the unlock and the cmpxchg
266 * we have two situations:
267 *
268 * unlock(wait_lock);
269 * lock(wait_lock);
270 * cmpxchg(p, owner, 0) == owner
271 * mark_rt_mutex_waiters(lock);
272 * acquire(lock);
273 * or:
274 *
275 * unlock(wait_lock);
276 * lock(wait_lock);
277 * mark_rt_mutex_waiters(lock);
278 *
279 * cmpxchg(p, owner, 0) != owner
280 * enqueue_waiter();
281 * unlock(wait_lock);
282 * lock(wait_lock);
283 * wake waiter();
284 * unlock(wait_lock);
285 * lock(wait_lock);
286 * acquire(lock);
287 */
288 return rt_mutex_cmpxchg_release(lock, owner, NULL);
289}
290
291#else
292static __always_inline bool rt_mutex_cmpxchg_acquire(struct rt_mutex_base *lock,
293 struct task_struct *old,
294 struct task_struct *new)
295{
296 return false;
297
298}
299
300static __always_inline bool rt_mutex_cmpxchg_release(struct rt_mutex_base *lock,
301 struct task_struct *old,
302 struct task_struct *new)
303{
304 return false;
305}
306
307static __always_inline void mark_rt_mutex_waiters(struct rt_mutex_base *lock)
308{
309 lock->owner = (struct task_struct *)
310 ((unsigned long)lock->owner | RT_MUTEX_HAS_WAITERS);
311}
312
313/*
314 * Simple slow path only version: lock->owner is protected by lock->wait_lock.
315 */
316static __always_inline bool unlock_rt_mutex_safe(struct rt_mutex_base *lock,
317 unsigned long flags)
318 __releases(lock->wait_lock)
319{
320 lock->owner = NULL;
321 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
322 return true;
323}
324#endif
325
326static __always_inline int __waiter_prio(struct task_struct *task)
327{
328 int prio = task->prio;
329
330 if (!rt_prio(prio))
331 return DEFAULT_PRIO;
332
333 return prio;
334}
335
336/*
337 * Update the waiter->tree copy of the sort keys.
338 */
339static __always_inline void
340waiter_update_prio(struct rt_mutex_waiter *waiter, struct task_struct *task)
341{
342 lockdep_assert_held(&waiter->lock->wait_lock);
343 lockdep_assert(RB_EMPTY_NODE(&waiter->tree.entry));
344
345 waiter->tree.prio = __waiter_prio(task);
346 waiter->tree.deadline = task->dl.deadline;
347}
348
349/*
350 * Update the waiter->pi_tree copy of the sort keys (from the tree copy).
351 */
352static __always_inline void
353waiter_clone_prio(struct rt_mutex_waiter *waiter, struct task_struct *task)
354{
355 lockdep_assert_held(&waiter->lock->wait_lock);
356 lockdep_assert_held(&task->pi_lock);
357 lockdep_assert(RB_EMPTY_NODE(&waiter->pi_tree.entry));
358
359 waiter->pi_tree.prio = waiter->tree.prio;
360 waiter->pi_tree.deadline = waiter->tree.deadline;
361}
362
363/*
364 * Only use with rt_waiter_node_{less,equal}()
365 */
366#define task_to_waiter_node(p) \
367 &(struct rt_waiter_node){ .prio = __waiter_prio(p), .deadline = (p)->dl.deadline }
368#define task_to_waiter(p) \
369 &(struct rt_mutex_waiter){ .tree = *task_to_waiter_node(p) }
370
371static __always_inline int rt_waiter_node_less(struct rt_waiter_node *left,
372 struct rt_waiter_node *right)
373{
374 if (left->prio < right->prio)
375 return 1;
376
377 /*
378 * If both waiters have dl_prio(), we check the deadlines of the
379 * associated tasks.
380 * If left waiter has a dl_prio(), and we didn't return 1 above,
381 * then right waiter has a dl_prio() too.
382 */
383 if (dl_prio(left->prio))
384 return dl_time_before(left->deadline, right->deadline);
385
386 return 0;
387}
388
389static __always_inline int rt_waiter_node_equal(struct rt_waiter_node *left,
390 struct rt_waiter_node *right)
391{
392 if (left->prio != right->prio)
393 return 0;
394
395 /*
396 * If both waiters have dl_prio(), we check the deadlines of the
397 * associated tasks.
398 * If left waiter has a dl_prio(), and we didn't return 0 above,
399 * then right waiter has a dl_prio() too.
400 */
401 if (dl_prio(left->prio))
402 return left->deadline == right->deadline;
403
404 return 1;
405}
406
407static inline bool rt_mutex_steal(struct rt_mutex_waiter *waiter,
408 struct rt_mutex_waiter *top_waiter)
409{
410 if (rt_waiter_node_less(&waiter->tree, &top_waiter->tree))
411 return true;
412
413#ifdef RT_MUTEX_BUILD_SPINLOCKS
414 /*
415 * Note that RT tasks are excluded from same priority (lateral)
416 * steals to prevent the introduction of an unbounded latency.
417 */
418 if (rt_prio(waiter->tree.prio) || dl_prio(waiter->tree.prio))
419 return false;
420
421 return rt_waiter_node_equal(&waiter->tree, &top_waiter->tree);
422#else
423 return false;
424#endif
425}
426
427#define __node_2_waiter(node) \
428 rb_entry((node), struct rt_mutex_waiter, tree.entry)
429
430static __always_inline bool __waiter_less(struct rb_node *a, const struct rb_node *b)
431{
432 struct rt_mutex_waiter *aw = __node_2_waiter(a);
433 struct rt_mutex_waiter *bw = __node_2_waiter(b);
434
435 if (rt_waiter_node_less(&aw->tree, &bw->tree))
436 return 1;
437
438 if (!build_ww_mutex())
439 return 0;
440
441 if (rt_waiter_node_less(&bw->tree, &aw->tree))
442 return 0;
443
444 /* NOTE: relies on waiter->ww_ctx being set before insertion */
445 if (aw->ww_ctx) {
446 if (!bw->ww_ctx)
447 return 1;
448
449 return (signed long)(aw->ww_ctx->stamp -
450 bw->ww_ctx->stamp) < 0;
451 }
452
453 return 0;
454}
455
456static __always_inline void
457rt_mutex_enqueue(struct rt_mutex_base *lock, struct rt_mutex_waiter *waiter)
458{
459 lockdep_assert_held(&lock->wait_lock);
460
461 rb_add_cached(&waiter->tree.entry, &lock->waiters, __waiter_less);
462}
463
464static __always_inline void
465rt_mutex_dequeue(struct rt_mutex_base *lock, struct rt_mutex_waiter *waiter)
466{
467 lockdep_assert_held(&lock->wait_lock);
468
469 if (RB_EMPTY_NODE(&waiter->tree.entry))
470 return;
471
472 rb_erase_cached(&waiter->tree.entry, &lock->waiters);
473 RB_CLEAR_NODE(&waiter->tree.entry);
474}
475
476#define __node_2_rt_node(node) \
477 rb_entry((node), struct rt_waiter_node, entry)
478
479static __always_inline bool __pi_waiter_less(struct rb_node *a, const struct rb_node *b)
480{
481 return rt_waiter_node_less(__node_2_rt_node(a), __node_2_rt_node(b));
482}
483
484static __always_inline void
485rt_mutex_enqueue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
486{
487 lockdep_assert_held(&task->pi_lock);
488
489 rb_add_cached(&waiter->pi_tree.entry, &task->pi_waiters, __pi_waiter_less);
490}
491
492static __always_inline void
493rt_mutex_dequeue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
494{
495 lockdep_assert_held(&task->pi_lock);
496
497 if (RB_EMPTY_NODE(&waiter->pi_tree.entry))
498 return;
499
500 rb_erase_cached(&waiter->pi_tree.entry, &task->pi_waiters);
501 RB_CLEAR_NODE(&waiter->pi_tree.entry);
502}
503
504static __always_inline void rt_mutex_adjust_prio(struct rt_mutex_base *lock,
505 struct task_struct *p)
506{
507 struct task_struct *pi_task = NULL;
508
509 lockdep_assert_held(&lock->wait_lock);
510 lockdep_assert(rt_mutex_owner(lock) == p);
511 lockdep_assert_held(&p->pi_lock);
512
513 if (task_has_pi_waiters(p))
514 pi_task = task_top_pi_waiter(p)->task;
515
516 rt_mutex_setprio(p, pi_task);
517}
518
519/* RT mutex specific wake_q wrappers */
520static __always_inline void rt_mutex_wake_q_add_task(struct rt_wake_q_head *wqh,
521 struct task_struct *task,
522 unsigned int wake_state)
523{
524 if (IS_ENABLED(CONFIG_PREEMPT_RT) && wake_state == TASK_RTLOCK_WAIT) {
525 if (IS_ENABLED(CONFIG_PROVE_LOCKING))
526 WARN_ON_ONCE(wqh->rtlock_task);
527 get_task_struct(task);
528 wqh->rtlock_task = task;
529 } else {
530 wake_q_add(&wqh->head, task);
531 }
532}
533
534static __always_inline void rt_mutex_wake_q_add(struct rt_wake_q_head *wqh,
535 struct rt_mutex_waiter *w)
536{
537 rt_mutex_wake_q_add_task(wqh, w->task, w->wake_state);
538}
539
540static __always_inline void rt_mutex_wake_up_q(struct rt_wake_q_head *wqh)
541{
542 if (IS_ENABLED(CONFIG_PREEMPT_RT) && wqh->rtlock_task) {
543 wake_up_state(wqh->rtlock_task, TASK_RTLOCK_WAIT);
544 put_task_struct(wqh->rtlock_task);
545 wqh->rtlock_task = NULL;
546 }
547
548 if (!wake_q_empty(&wqh->head))
549 wake_up_q(&wqh->head);
550
551 /* Pairs with preempt_disable() in mark_wakeup_next_waiter() */
552 preempt_enable();
553}
554
555/*
556 * Deadlock detection is conditional:
557 *
558 * If CONFIG_DEBUG_RT_MUTEXES=n, deadlock detection is only conducted
559 * if the detect argument is == RT_MUTEX_FULL_CHAINWALK.
560 *
561 * If CONFIG_DEBUG_RT_MUTEXES=y, deadlock detection is always
562 * conducted independent of the detect argument.
563 *
564 * If the waiter argument is NULL this indicates the deboost path and
565 * deadlock detection is disabled independent of the detect argument
566 * and the config settings.
567 */
568static __always_inline bool
569rt_mutex_cond_detect_deadlock(struct rt_mutex_waiter *waiter,
570 enum rtmutex_chainwalk chwalk)
571{
572 if (IS_ENABLED(CONFIG_DEBUG_RT_MUTEXES))
573 return waiter != NULL;
574 return chwalk == RT_MUTEX_FULL_CHAINWALK;
575}
576
577static __always_inline struct rt_mutex_base *task_blocked_on_lock(struct task_struct *p)
578{
579 return p->pi_blocked_on ? p->pi_blocked_on->lock : NULL;
580}
581
582/*
583 * Adjust the priority chain. Also used for deadlock detection.
584 * Decreases task's usage by one - may thus free the task.
585 *
586 * @task: the task owning the mutex (owner) for which a chain walk is
587 * probably needed
588 * @chwalk: do we have to carry out deadlock detection?
589 * @orig_lock: the mutex (can be NULL if we are walking the chain to recheck
590 * things for a task that has just got its priority adjusted, and
591 * is waiting on a mutex)
592 * @next_lock: the mutex on which the owner of @orig_lock was blocked before
593 * we dropped its pi_lock. Is never dereferenced, only used for
594 * comparison to detect lock chain changes.
595 * @orig_waiter: rt_mutex_waiter struct for the task that has just donated
596 * its priority to the mutex owner (can be NULL in the case
597 * depicted above or if the top waiter is gone away and we are
598 * actually deboosting the owner)
599 * @top_task: the current top waiter
600 *
601 * Returns 0 or -EDEADLK.
602 *
603 * Chain walk basics and protection scope
604 *
605 * [R] refcount on task
606 * [Pn] task->pi_lock held
607 * [L] rtmutex->wait_lock held
608 *
609 * Normal locking order:
610 *
611 * rtmutex->wait_lock
612 * task->pi_lock
613 *
614 * Step Description Protected by
615 * function arguments:
616 * @task [R]
617 * @orig_lock if != NULL @top_task is blocked on it
618 * @next_lock Unprotected. Cannot be
619 * dereferenced. Only used for
620 * comparison.
621 * @orig_waiter if != NULL @top_task is blocked on it
622 * @top_task current, or in case of proxy
623 * locking protected by calling
624 * code
625 * again:
626 * loop_sanity_check();
627 * retry:
628 * [1] lock(task->pi_lock); [R] acquire [P1]
629 * [2] waiter = task->pi_blocked_on; [P1]
630 * [3] check_exit_conditions_1(); [P1]
631 * [4] lock = waiter->lock; [P1]
632 * [5] if (!try_lock(lock->wait_lock)) { [P1] try to acquire [L]
633 * unlock(task->pi_lock); release [P1]
634 * goto retry;
635 * }
636 * [6] check_exit_conditions_2(); [P1] + [L]
637 * [7] requeue_lock_waiter(lock, waiter); [P1] + [L]
638 * [8] unlock(task->pi_lock); release [P1]
639 * put_task_struct(task); release [R]
640 * [9] check_exit_conditions_3(); [L]
641 * [10] task = owner(lock); [L]
642 * get_task_struct(task); [L] acquire [R]
643 * lock(task->pi_lock); [L] acquire [P2]
644 * [11] requeue_pi_waiter(tsk, waiters(lock));[P2] + [L]
645 * [12] check_exit_conditions_4(); [P2] + [L]
646 * [13] unlock(task->pi_lock); release [P2]
647 * unlock(lock->wait_lock); release [L]
648 * goto again;
649 *
650 * Where P1 is the blocking task and P2 is the lock owner; going up one step
651 * the owner becomes the next blocked task etc..
652 *
653*
654 */
655static int __sched rt_mutex_adjust_prio_chain(struct task_struct *task,
656 enum rtmutex_chainwalk chwalk,
657 struct rt_mutex_base *orig_lock,
658 struct rt_mutex_base *next_lock,
659 struct rt_mutex_waiter *orig_waiter,
660 struct task_struct *top_task)
661{
662 struct rt_mutex_waiter *waiter, *top_waiter = orig_waiter;
663 struct rt_mutex_waiter *prerequeue_top_waiter;
664 int ret = 0, depth = 0;
665 struct rt_mutex_base *lock;
666 bool detect_deadlock;
667 bool requeue = true;
668
669 detect_deadlock = rt_mutex_cond_detect_deadlock(orig_waiter, chwalk);
670
671 /*
672 * The (de)boosting is a step by step approach with a lot of
673 * pitfalls. We want this to be preemptible and we want hold a
674 * maximum of two locks per step. So we have to check
675 * carefully whether things change under us.
676 */
677 again:
678 /*
679 * We limit the lock chain length for each invocation.
680 */
681 if (++depth > max_lock_depth) {
682 static int prev_max;
683
684 /*
685 * Print this only once. If the admin changes the limit,
686 * print a new message when reaching the limit again.
687 */
688 if (prev_max != max_lock_depth) {
689 prev_max = max_lock_depth;
690 printk(KERN_WARNING "Maximum lock depth %d reached "
691 "task: %s (%d)\n", max_lock_depth,
692 top_task->comm, task_pid_nr(top_task));
693 }
694 put_task_struct(task);
695
696 return -EDEADLK;
697 }
698
699 /*
700 * We are fully preemptible here and only hold the refcount on
701 * @task. So everything can have changed under us since the
702 * caller or our own code below (goto retry/again) dropped all
703 * locks.
704 */
705 retry:
706 /*
707 * [1] Task cannot go away as we did a get_task() before !
708 */
709 raw_spin_lock_irq(&task->pi_lock);
710
711 /*
712 * [2] Get the waiter on which @task is blocked on.
713 */
714 waiter = task->pi_blocked_on;
715
716 /*
717 * [3] check_exit_conditions_1() protected by task->pi_lock.
718 */
719
720 /*
721 * Check whether the end of the boosting chain has been
722 * reached or the state of the chain has changed while we
723 * dropped the locks.
724 */
725 if (!waiter)
726 goto out_unlock_pi;
727
728 /*
729 * Check the orig_waiter state. After we dropped the locks,
730 * the previous owner of the lock might have released the lock.
731 */
732 if (orig_waiter && !rt_mutex_owner(orig_lock))
733 goto out_unlock_pi;
734
735 /*
736 * We dropped all locks after taking a refcount on @task, so
737 * the task might have moved on in the lock chain or even left
738 * the chain completely and blocks now on an unrelated lock or
739 * on @orig_lock.
740 *
741 * We stored the lock on which @task was blocked in @next_lock,
742 * so we can detect the chain change.
743 */
744 if (next_lock != waiter->lock)
745 goto out_unlock_pi;
746
747 /*
748 * There could be 'spurious' loops in the lock graph due to ww_mutex,
749 * consider:
750 *
751 * P1: A, ww_A, ww_B
752 * P2: ww_B, ww_A
753 * P3: A
754 *
755 * P3 should not return -EDEADLK because it gets trapped in the cycle
756 * created by P1 and P2 (which will resolve -- and runs into
757 * max_lock_depth above). Therefore disable detect_deadlock such that
758 * the below termination condition can trigger once all relevant tasks
759 * are boosted.
760 *
761 * Even when we start with ww_mutex we can disable deadlock detection,
762 * since we would supress a ww_mutex induced deadlock at [6] anyway.
763 * Supressing it here however is not sufficient since we might still
764 * hit [6] due to adjustment driven iteration.
765 *
766 * NOTE: if someone were to create a deadlock between 2 ww_classes we'd
767 * utterly fail to report it; lockdep should.
768 */
769 if (IS_ENABLED(CONFIG_PREEMPT_RT) && waiter->ww_ctx && detect_deadlock)
770 detect_deadlock = false;
771
772 /*
773 * Drop out, when the task has no waiters. Note,
774 * top_waiter can be NULL, when we are in the deboosting
775 * mode!
776 */
777 if (top_waiter) {
778 if (!task_has_pi_waiters(task))
779 goto out_unlock_pi;
780 /*
781 * If deadlock detection is off, we stop here if we
782 * are not the top pi waiter of the task. If deadlock
783 * detection is enabled we continue, but stop the
784 * requeueing in the chain walk.
785 */
786 if (top_waiter != task_top_pi_waiter(task)) {
787 if (!detect_deadlock)
788 goto out_unlock_pi;
789 else
790 requeue = false;
791 }
792 }
793
794 /*
795 * If the waiter priority is the same as the task priority
796 * then there is no further priority adjustment necessary. If
797 * deadlock detection is off, we stop the chain walk. If its
798 * enabled we continue, but stop the requeueing in the chain
799 * walk.
800 */
801 if (rt_waiter_node_equal(&waiter->tree, task_to_waiter_node(task))) {
802 if (!detect_deadlock)
803 goto out_unlock_pi;
804 else
805 requeue = false;
806 }
807
808 /*
809 * [4] Get the next lock; per holding task->pi_lock we can't unblock
810 * and guarantee @lock's existence.
811 */
812 lock = waiter->lock;
813 /*
814 * [5] We need to trylock here as we are holding task->pi_lock,
815 * which is the reverse lock order versus the other rtmutex
816 * operations.
817 *
818 * Per the above, holding task->pi_lock guarantees lock exists, so
819 * inverting this lock order is infeasible from a life-time
820 * perspective.
821 */
822 if (!raw_spin_trylock(&lock->wait_lock)) {
823 raw_spin_unlock_irq(&task->pi_lock);
824 cpu_relax();
825 goto retry;
826 }
827
828 /*
829 * [6] check_exit_conditions_2() protected by task->pi_lock and
830 * lock->wait_lock.
831 *
832 * Deadlock detection. If the lock is the same as the original
833 * lock which caused us to walk the lock chain or if the
834 * current lock is owned by the task which initiated the chain
835 * walk, we detected a deadlock.
836 */
837 if (lock == orig_lock || rt_mutex_owner(lock) == top_task) {
838 ret = -EDEADLK;
839
840 /*
841 * When the deadlock is due to ww_mutex; also see above. Don't
842 * report the deadlock and instead let the ww_mutex wound/die
843 * logic pick which of the contending threads gets -EDEADLK.
844 *
845 * NOTE: assumes the cycle only contains a single ww_class; any
846 * other configuration and we fail to report; also, see
847 * lockdep.
848 */
849 if (IS_ENABLED(CONFIG_PREEMPT_RT) && orig_waiter && orig_waiter->ww_ctx)
850 ret = 0;
851
852 raw_spin_unlock(&lock->wait_lock);
853 goto out_unlock_pi;
854 }
855
856 /*
857 * If we just follow the lock chain for deadlock detection, no
858 * need to do all the requeue operations. To avoid a truckload
859 * of conditionals around the various places below, just do the
860 * minimum chain walk checks.
861 */
862 if (!requeue) {
863 /*
864 * No requeue[7] here. Just release @task [8]
865 */
866 raw_spin_unlock(&task->pi_lock);
867 put_task_struct(task);
868
869 /*
870 * [9] check_exit_conditions_3 protected by lock->wait_lock.
871 * If there is no owner of the lock, end of chain.
872 */
873 if (!rt_mutex_owner(lock)) {
874 raw_spin_unlock_irq(&lock->wait_lock);
875 return 0;
876 }
877
878 /* [10] Grab the next task, i.e. owner of @lock */
879 task = get_task_struct(rt_mutex_owner(lock));
880 raw_spin_lock(&task->pi_lock);
881
882 /*
883 * No requeue [11] here. We just do deadlock detection.
884 *
885 * [12] Store whether owner is blocked
886 * itself. Decision is made after dropping the locks
887 */
888 next_lock = task_blocked_on_lock(task);
889 /*
890 * Get the top waiter for the next iteration
891 */
892 top_waiter = rt_mutex_top_waiter(lock);
893
894 /* [13] Drop locks */
895 raw_spin_unlock(&task->pi_lock);
896 raw_spin_unlock_irq(&lock->wait_lock);
897
898 /* If owner is not blocked, end of chain. */
899 if (!next_lock)
900 goto out_put_task;
901 goto again;
902 }
903
904 /*
905 * Store the current top waiter before doing the requeue
906 * operation on @lock. We need it for the boost/deboost
907 * decision below.
908 */
909 prerequeue_top_waiter = rt_mutex_top_waiter(lock);
910
911 /* [7] Requeue the waiter in the lock waiter tree. */
912 rt_mutex_dequeue(lock, waiter);
913
914 /*
915 * Update the waiter prio fields now that we're dequeued.
916 *
917 * These values can have changed through either:
918 *
919 * sys_sched_set_scheduler() / sys_sched_setattr()
920 *
921 * or
922 *
923 * DL CBS enforcement advancing the effective deadline.
924 */
925 waiter_update_prio(waiter, task);
926
927 rt_mutex_enqueue(lock, waiter);
928
929 /*
930 * [8] Release the (blocking) task in preparation for
931 * taking the owner task in [10].
932 *
933 * Since we hold lock->waiter_lock, task cannot unblock, even if we
934 * release task->pi_lock.
935 */
936 raw_spin_unlock(&task->pi_lock);
937 put_task_struct(task);
938
939 /*
940 * [9] check_exit_conditions_3 protected by lock->wait_lock.
941 *
942 * We must abort the chain walk if there is no lock owner even
943 * in the dead lock detection case, as we have nothing to
944 * follow here. This is the end of the chain we are walking.
945 */
946 if (!rt_mutex_owner(lock)) {
947 /*
948 * If the requeue [7] above changed the top waiter,
949 * then we need to wake the new top waiter up to try
950 * to get the lock.
951 */
952 top_waiter = rt_mutex_top_waiter(lock);
953 if (prerequeue_top_waiter != top_waiter)
954 wake_up_state(top_waiter->task, top_waiter->wake_state);
955 raw_spin_unlock_irq(&lock->wait_lock);
956 return 0;
957 }
958
959 /*
960 * [10] Grab the next task, i.e. the owner of @lock
961 *
962 * Per holding lock->wait_lock and checking for !owner above, there
963 * must be an owner and it cannot go away.
964 */
965 task = get_task_struct(rt_mutex_owner(lock));
966 raw_spin_lock(&task->pi_lock);
967
968 /* [11] requeue the pi waiters if necessary */
969 if (waiter == rt_mutex_top_waiter(lock)) {
970 /*
971 * The waiter became the new top (highest priority)
972 * waiter on the lock. Replace the previous top waiter
973 * in the owner tasks pi waiters tree with this waiter
974 * and adjust the priority of the owner.
975 */
976 rt_mutex_dequeue_pi(task, prerequeue_top_waiter);
977 waiter_clone_prio(waiter, task);
978 rt_mutex_enqueue_pi(task, waiter);
979 rt_mutex_adjust_prio(lock, task);
980
981 } else if (prerequeue_top_waiter == waiter) {
982 /*
983 * The waiter was the top waiter on the lock, but is
984 * no longer the top priority waiter. Replace waiter in
985 * the owner tasks pi waiters tree with the new top
986 * (highest priority) waiter and adjust the priority
987 * of the owner.
988 * The new top waiter is stored in @waiter so that
989 * @waiter == @top_waiter evaluates to true below and
990 * we continue to deboost the rest of the chain.
991 */
992 rt_mutex_dequeue_pi(task, waiter);
993 waiter = rt_mutex_top_waiter(lock);
994 waiter_clone_prio(waiter, task);
995 rt_mutex_enqueue_pi(task, waiter);
996 rt_mutex_adjust_prio(lock, task);
997 } else {
998 /*
999 * Nothing changed. No need to do any priority
1000 * adjustment.
1001 */
1002 }
1003
1004 /*
1005 * [12] check_exit_conditions_4() protected by task->pi_lock
1006 * and lock->wait_lock. The actual decisions are made after we
1007 * dropped the locks.
1008 *
1009 * Check whether the task which owns the current lock is pi
1010 * blocked itself. If yes we store a pointer to the lock for
1011 * the lock chain change detection above. After we dropped
1012 * task->pi_lock next_lock cannot be dereferenced anymore.
1013 */
1014 next_lock = task_blocked_on_lock(task);
1015 /*
1016 * Store the top waiter of @lock for the end of chain walk
1017 * decision below.
1018 */
1019 top_waiter = rt_mutex_top_waiter(lock);
1020
1021 /* [13] Drop the locks */
1022 raw_spin_unlock(&task->pi_lock);
1023 raw_spin_unlock_irq(&lock->wait_lock);
1024
1025 /*
1026 * Make the actual exit decisions [12], based on the stored
1027 * values.
1028 *
1029 * We reached the end of the lock chain. Stop right here. No
1030 * point to go back just to figure that out.
1031 */
1032 if (!next_lock)
1033 goto out_put_task;
1034
1035 /*
1036 * If the current waiter is not the top waiter on the lock,
1037 * then we can stop the chain walk here if we are not in full
1038 * deadlock detection mode.
1039 */
1040 if (!detect_deadlock && waiter != top_waiter)
1041 goto out_put_task;
1042
1043 goto again;
1044
1045 out_unlock_pi:
1046 raw_spin_unlock_irq(&task->pi_lock);
1047 out_put_task:
1048 put_task_struct(task);
1049
1050 return ret;
1051}
1052
1053/*
1054 * Try to take an rt-mutex
1055 *
1056 * Must be called with lock->wait_lock held and interrupts disabled
1057 *
1058 * @lock: The lock to be acquired.
1059 * @task: The task which wants to acquire the lock
1060 * @waiter: The waiter that is queued to the lock's wait tree if the
1061 * callsite called task_blocked_on_lock(), otherwise NULL
1062 */
1063static int __sched
1064try_to_take_rt_mutex(struct rt_mutex_base *lock, struct task_struct *task,
1065 struct rt_mutex_waiter *waiter)
1066{
1067 lockdep_assert_held(&lock->wait_lock);
1068
1069 /*
1070 * Before testing whether we can acquire @lock, we set the
1071 * RT_MUTEX_HAS_WAITERS bit in @lock->owner. This forces all
1072 * other tasks which try to modify @lock into the slow path
1073 * and they serialize on @lock->wait_lock.
1074 *
1075 * The RT_MUTEX_HAS_WAITERS bit can have a transitional state
1076 * as explained at the top of this file if and only if:
1077 *
1078 * - There is a lock owner. The caller must fixup the
1079 * transient state if it does a trylock or leaves the lock
1080 * function due to a signal or timeout.
1081 *
1082 * - @task acquires the lock and there are no other
1083 * waiters. This is undone in rt_mutex_set_owner(@task) at
1084 * the end of this function.
1085 */
1086 mark_rt_mutex_waiters(lock);
1087
1088 /*
1089 * If @lock has an owner, give up.
1090 */
1091 if (rt_mutex_owner(lock))
1092 return 0;
1093
1094 /*
1095 * If @waiter != NULL, @task has already enqueued the waiter
1096 * into @lock waiter tree. If @waiter == NULL then this is a
1097 * trylock attempt.
1098 */
1099 if (waiter) {
1100 struct rt_mutex_waiter *top_waiter = rt_mutex_top_waiter(lock);
1101
1102 /*
1103 * If waiter is the highest priority waiter of @lock,
1104 * or allowed to steal it, take it over.
1105 */
1106 if (waiter == top_waiter || rt_mutex_steal(waiter, top_waiter)) {
1107 /*
1108 * We can acquire the lock. Remove the waiter from the
1109 * lock waiters tree.
1110 */
1111 rt_mutex_dequeue(lock, waiter);
1112 } else {
1113 return 0;
1114 }
1115 } else {
1116 /*
1117 * If the lock has waiters already we check whether @task is
1118 * eligible to take over the lock.
1119 *
1120 * If there are no other waiters, @task can acquire
1121 * the lock. @task->pi_blocked_on is NULL, so it does
1122 * not need to be dequeued.
1123 */
1124 if (rt_mutex_has_waiters(lock)) {
1125 /* Check whether the trylock can steal it. */
1126 if (!rt_mutex_steal(task_to_waiter(task),
1127 rt_mutex_top_waiter(lock)))
1128 return 0;
1129
1130 /*
1131 * The current top waiter stays enqueued. We
1132 * don't have to change anything in the lock
1133 * waiters order.
1134 */
1135 } else {
1136 /*
1137 * No waiters. Take the lock without the
1138 * pi_lock dance.@task->pi_blocked_on is NULL
1139 * and we have no waiters to enqueue in @task
1140 * pi waiters tree.
1141 */
1142 goto takeit;
1143 }
1144 }
1145
1146 /*
1147 * Clear @task->pi_blocked_on. Requires protection by
1148 * @task->pi_lock. Redundant operation for the @waiter == NULL
1149 * case, but conditionals are more expensive than a redundant
1150 * store.
1151 */
1152 raw_spin_lock(&task->pi_lock);
1153 task->pi_blocked_on = NULL;
1154 /*
1155 * Finish the lock acquisition. @task is the new owner. If
1156 * other waiters exist we have to insert the highest priority
1157 * waiter into @task->pi_waiters tree.
1158 */
1159 if (rt_mutex_has_waiters(lock))
1160 rt_mutex_enqueue_pi(task, rt_mutex_top_waiter(lock));
1161 raw_spin_unlock(&task->pi_lock);
1162
1163takeit:
1164 /*
1165 * This either preserves the RT_MUTEX_HAS_WAITERS bit if there
1166 * are still waiters or clears it.
1167 */
1168 rt_mutex_set_owner(lock, task);
1169
1170 return 1;
1171}
1172
1173/*
1174 * Task blocks on lock.
1175 *
1176 * Prepare waiter and propagate pi chain
1177 *
1178 * This must be called with lock->wait_lock held and interrupts disabled
1179 */
1180static int __sched task_blocks_on_rt_mutex(struct rt_mutex_base *lock,
1181 struct rt_mutex_waiter *waiter,
1182 struct task_struct *task,
1183 struct ww_acquire_ctx *ww_ctx,
1184 enum rtmutex_chainwalk chwalk)
1185{
1186 struct task_struct *owner = rt_mutex_owner(lock);
1187 struct rt_mutex_waiter *top_waiter = waiter;
1188 struct rt_mutex_base *next_lock;
1189 int chain_walk = 0, res;
1190
1191 lockdep_assert_held(&lock->wait_lock);
1192
1193 /*
1194 * Early deadlock detection. We really don't want the task to
1195 * enqueue on itself just to untangle the mess later. It's not
1196 * only an optimization. We drop the locks, so another waiter
1197 * can come in before the chain walk detects the deadlock. So
1198 * the other will detect the deadlock and return -EDEADLOCK,
1199 * which is wrong, as the other waiter is not in a deadlock
1200 * situation.
1201 *
1202 * Except for ww_mutex, in that case the chain walk must already deal
1203 * with spurious cycles, see the comments at [3] and [6].
1204 */
1205 if (owner == task && !(build_ww_mutex() && ww_ctx))
1206 return -EDEADLK;
1207
1208 raw_spin_lock(&task->pi_lock);
1209 waiter->task = task;
1210 waiter->lock = lock;
1211 waiter_update_prio(waiter, task);
1212 waiter_clone_prio(waiter, task);
1213
1214 /* Get the top priority waiter on the lock */
1215 if (rt_mutex_has_waiters(lock))
1216 top_waiter = rt_mutex_top_waiter(lock);
1217 rt_mutex_enqueue(lock, waiter);
1218
1219 task->pi_blocked_on = waiter;
1220
1221 raw_spin_unlock(&task->pi_lock);
1222
1223 if (build_ww_mutex() && ww_ctx) {
1224 struct rt_mutex *rtm;
1225
1226 /* Check whether the waiter should back out immediately */
1227 rtm = container_of(lock, struct rt_mutex, rtmutex);
1228 res = __ww_mutex_add_waiter(waiter, rtm, ww_ctx);
1229 if (res) {
1230 raw_spin_lock(&task->pi_lock);
1231 rt_mutex_dequeue(lock, waiter);
1232 task->pi_blocked_on = NULL;
1233 raw_spin_unlock(&task->pi_lock);
1234 return res;
1235 }
1236 }
1237
1238 if (!owner)
1239 return 0;
1240
1241 raw_spin_lock(&owner->pi_lock);
1242 if (waiter == rt_mutex_top_waiter(lock)) {
1243 rt_mutex_dequeue_pi(owner, top_waiter);
1244 rt_mutex_enqueue_pi(owner, waiter);
1245
1246 rt_mutex_adjust_prio(lock, owner);
1247 if (owner->pi_blocked_on)
1248 chain_walk = 1;
1249 } else if (rt_mutex_cond_detect_deadlock(waiter, chwalk)) {
1250 chain_walk = 1;
1251 }
1252
1253 /* Store the lock on which owner is blocked or NULL */
1254 next_lock = task_blocked_on_lock(owner);
1255
1256 raw_spin_unlock(&owner->pi_lock);
1257 /*
1258 * Even if full deadlock detection is on, if the owner is not
1259 * blocked itself, we can avoid finding this out in the chain
1260 * walk.
1261 */
1262 if (!chain_walk || !next_lock)
1263 return 0;
1264
1265 /*
1266 * The owner can't disappear while holding a lock,
1267 * so the owner struct is protected by wait_lock.
1268 * Gets dropped in rt_mutex_adjust_prio_chain()!
1269 */
1270 get_task_struct(owner);
1271
1272 raw_spin_unlock_irq(&lock->wait_lock);
1273
1274 res = rt_mutex_adjust_prio_chain(owner, chwalk, lock,
1275 next_lock, waiter, task);
1276
1277 raw_spin_lock_irq(&lock->wait_lock);
1278
1279 return res;
1280}
1281
1282/*
1283 * Remove the top waiter from the current tasks pi waiter tree and
1284 * queue it up.
1285 *
1286 * Called with lock->wait_lock held and interrupts disabled.
1287 */
1288static void __sched mark_wakeup_next_waiter(struct rt_wake_q_head *wqh,
1289 struct rt_mutex_base *lock)
1290{
1291 struct rt_mutex_waiter *waiter;
1292
1293 lockdep_assert_held(&lock->wait_lock);
1294
1295 raw_spin_lock(&current->pi_lock);
1296
1297 waiter = rt_mutex_top_waiter(lock);
1298
1299 /*
1300 * Remove it from current->pi_waiters and deboost.
1301 *
1302 * We must in fact deboost here in order to ensure we call
1303 * rt_mutex_setprio() to update p->pi_top_task before the
1304 * task unblocks.
1305 */
1306 rt_mutex_dequeue_pi(current, waiter);
1307 rt_mutex_adjust_prio(lock, current);
1308
1309 /*
1310 * As we are waking up the top waiter, and the waiter stays
1311 * queued on the lock until it gets the lock, this lock
1312 * obviously has waiters. Just set the bit here and this has
1313 * the added benefit of forcing all new tasks into the
1314 * slow path making sure no task of lower priority than
1315 * the top waiter can steal this lock.
1316 */
1317 lock->owner = (void *) RT_MUTEX_HAS_WAITERS;
1318
1319 /*
1320 * We deboosted before waking the top waiter task such that we don't
1321 * run two tasks with the 'same' priority (and ensure the
1322 * p->pi_top_task pointer points to a blocked task). This however can
1323 * lead to priority inversion if we would get preempted after the
1324 * deboost but before waking our donor task, hence the preempt_disable()
1325 * before unlock.
1326 *
1327 * Pairs with preempt_enable() in rt_mutex_wake_up_q();
1328 */
1329 preempt_disable();
1330 rt_mutex_wake_q_add(wqh, waiter);
1331 raw_spin_unlock(&current->pi_lock);
1332}
1333
1334static int __sched __rt_mutex_slowtrylock(struct rt_mutex_base *lock)
1335{
1336 int ret = try_to_take_rt_mutex(lock, current, NULL);
1337
1338 /*
1339 * try_to_take_rt_mutex() sets the lock waiters bit
1340 * unconditionally. Clean this up.
1341 */
1342 fixup_rt_mutex_waiters(lock, true);
1343
1344 return ret;
1345}
1346
1347/*
1348 * Slow path try-lock function:
1349 */
1350static int __sched rt_mutex_slowtrylock(struct rt_mutex_base *lock)
1351{
1352 unsigned long flags;
1353 int ret;
1354
1355 /*
1356 * If the lock already has an owner we fail to get the lock.
1357 * This can be done without taking the @lock->wait_lock as
1358 * it is only being read, and this is a trylock anyway.
1359 */
1360 if (rt_mutex_owner(lock))
1361 return 0;
1362
1363 /*
1364 * The mutex has currently no owner. Lock the wait lock and try to
1365 * acquire the lock. We use irqsave here to support early boot calls.
1366 */
1367 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1368
1369 ret = __rt_mutex_slowtrylock(lock);
1370
1371 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1372
1373 return ret;
1374}
1375
1376static __always_inline int __rt_mutex_trylock(struct rt_mutex_base *lock)
1377{
1378 if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
1379 return 1;
1380
1381 return rt_mutex_slowtrylock(lock);
1382}
1383
1384/*
1385 * Slow path to release a rt-mutex.
1386 */
1387static void __sched rt_mutex_slowunlock(struct rt_mutex_base *lock)
1388{
1389 DEFINE_RT_WAKE_Q(wqh);
1390 unsigned long flags;
1391
1392 /* irqsave required to support early boot calls */
1393 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1394
1395 debug_rt_mutex_unlock(lock);
1396
1397 /*
1398 * We must be careful here if the fast path is enabled. If we
1399 * have no waiters queued we cannot set owner to NULL here
1400 * because of:
1401 *
1402 * foo->lock->owner = NULL;
1403 * rtmutex_lock(foo->lock); <- fast path
1404 * free = atomic_dec_and_test(foo->refcnt);
1405 * rtmutex_unlock(foo->lock); <- fast path
1406 * if (free)
1407 * kfree(foo);
1408 * raw_spin_unlock(foo->lock->wait_lock);
1409 *
1410 * So for the fastpath enabled kernel:
1411 *
1412 * Nothing can set the waiters bit as long as we hold
1413 * lock->wait_lock. So we do the following sequence:
1414 *
1415 * owner = rt_mutex_owner(lock);
1416 * clear_rt_mutex_waiters(lock);
1417 * raw_spin_unlock(&lock->wait_lock);
1418 * if (cmpxchg(&lock->owner, owner, 0) == owner)
1419 * return;
1420 * goto retry;
1421 *
1422 * The fastpath disabled variant is simple as all access to
1423 * lock->owner is serialized by lock->wait_lock:
1424 *
1425 * lock->owner = NULL;
1426 * raw_spin_unlock(&lock->wait_lock);
1427 */
1428 while (!rt_mutex_has_waiters(lock)) {
1429 /* Drops lock->wait_lock ! */
1430 if (unlock_rt_mutex_safe(lock, flags) == true)
1431 return;
1432 /* Relock the rtmutex and try again */
1433 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1434 }
1435
1436 /*
1437 * The wakeup next waiter path does not suffer from the above
1438 * race. See the comments there.
1439 *
1440 * Queue the next waiter for wakeup once we release the wait_lock.
1441 */
1442 mark_wakeup_next_waiter(&wqh, lock);
1443 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1444
1445 rt_mutex_wake_up_q(&wqh);
1446}
1447
1448static __always_inline void __rt_mutex_unlock(struct rt_mutex_base *lock)
1449{
1450 if (likely(rt_mutex_cmpxchg_release(lock, current, NULL)))
1451 return;
1452
1453 rt_mutex_slowunlock(lock);
1454}
1455
1456#ifdef CONFIG_SMP
1457static bool rtmutex_spin_on_owner(struct rt_mutex_base *lock,
1458 struct rt_mutex_waiter *waiter,
1459 struct task_struct *owner)
1460{
1461 bool res = true;
1462
1463 rcu_read_lock();
1464 for (;;) {
1465 /* If owner changed, trylock again. */
1466 if (owner != rt_mutex_owner(lock))
1467 break;
1468 /*
1469 * Ensure that @owner is dereferenced after checking that
1470 * the lock owner still matches @owner. If that fails,
1471 * @owner might point to freed memory. If it still matches,
1472 * the rcu_read_lock() ensures the memory stays valid.
1473 */
1474 barrier();
1475 /*
1476 * Stop spinning when:
1477 * - the lock owner has been scheduled out
1478 * - current is not longer the top waiter
1479 * - current is requested to reschedule (redundant
1480 * for CONFIG_PREEMPT_RCU=y)
1481 * - the VCPU on which owner runs is preempted
1482 */
1483 if (!owner_on_cpu(owner) || need_resched() ||
1484 !rt_mutex_waiter_is_top_waiter(lock, waiter)) {
1485 res = false;
1486 break;
1487 }
1488 cpu_relax();
1489 }
1490 rcu_read_unlock();
1491 return res;
1492}
1493#else
1494static bool rtmutex_spin_on_owner(struct rt_mutex_base *lock,
1495 struct rt_mutex_waiter *waiter,
1496 struct task_struct *owner)
1497{
1498 return false;
1499}
1500#endif
1501
1502#ifdef RT_MUTEX_BUILD_MUTEX
1503/*
1504 * Functions required for:
1505 * - rtmutex, futex on all kernels
1506 * - mutex and rwsem substitutions on RT kernels
1507 */
1508
1509/*
1510 * Remove a waiter from a lock and give up
1511 *
1512 * Must be called with lock->wait_lock held and interrupts disabled. It must
1513 * have just failed to try_to_take_rt_mutex().
1514 */
1515static void __sched remove_waiter(struct rt_mutex_base *lock,
1516 struct rt_mutex_waiter *waiter)
1517{
1518 bool is_top_waiter = (waiter == rt_mutex_top_waiter(lock));
1519 struct task_struct *owner = rt_mutex_owner(lock);
1520 struct rt_mutex_base *next_lock;
1521
1522 lockdep_assert_held(&lock->wait_lock);
1523
1524 raw_spin_lock(&current->pi_lock);
1525 rt_mutex_dequeue(lock, waiter);
1526 current->pi_blocked_on = NULL;
1527 raw_spin_unlock(&current->pi_lock);
1528
1529 /*
1530 * Only update priority if the waiter was the highest priority
1531 * waiter of the lock and there is an owner to update.
1532 */
1533 if (!owner || !is_top_waiter)
1534 return;
1535
1536 raw_spin_lock(&owner->pi_lock);
1537
1538 rt_mutex_dequeue_pi(owner, waiter);
1539
1540 if (rt_mutex_has_waiters(lock))
1541 rt_mutex_enqueue_pi(owner, rt_mutex_top_waiter(lock));
1542
1543 rt_mutex_adjust_prio(lock, owner);
1544
1545 /* Store the lock on which owner is blocked or NULL */
1546 next_lock = task_blocked_on_lock(owner);
1547
1548 raw_spin_unlock(&owner->pi_lock);
1549
1550 /*
1551 * Don't walk the chain, if the owner task is not blocked
1552 * itself.
1553 */
1554 if (!next_lock)
1555 return;
1556
1557 /* gets dropped in rt_mutex_adjust_prio_chain()! */
1558 get_task_struct(owner);
1559
1560 raw_spin_unlock_irq(&lock->wait_lock);
1561
1562 rt_mutex_adjust_prio_chain(owner, RT_MUTEX_MIN_CHAINWALK, lock,
1563 next_lock, NULL, current);
1564
1565 raw_spin_lock_irq(&lock->wait_lock);
1566}
1567
1568/**
1569 * rt_mutex_slowlock_block() - Perform the wait-wake-try-to-take loop
1570 * @lock: the rt_mutex to take
1571 * @ww_ctx: WW mutex context pointer
1572 * @state: the state the task should block in (TASK_INTERRUPTIBLE
1573 * or TASK_UNINTERRUPTIBLE)
1574 * @timeout: the pre-initialized and started timer, or NULL for none
1575 * @waiter: the pre-initialized rt_mutex_waiter
1576 *
1577 * Must be called with lock->wait_lock held and interrupts disabled
1578 */
1579static int __sched rt_mutex_slowlock_block(struct rt_mutex_base *lock,
1580 struct ww_acquire_ctx *ww_ctx,
1581 unsigned int state,
1582 struct hrtimer_sleeper *timeout,
1583 struct rt_mutex_waiter *waiter)
1584{
1585 struct rt_mutex *rtm = container_of(lock, struct rt_mutex, rtmutex);
1586 struct task_struct *owner;
1587 int ret = 0;
1588
1589 for (;;) {
1590 /* Try to acquire the lock: */
1591 if (try_to_take_rt_mutex(lock, current, waiter))
1592 break;
1593
1594 if (timeout && !timeout->task) {
1595 ret = -ETIMEDOUT;
1596 break;
1597 }
1598 if (signal_pending_state(state, current)) {
1599 ret = -EINTR;
1600 break;
1601 }
1602
1603 if (build_ww_mutex() && ww_ctx) {
1604 ret = __ww_mutex_check_kill(rtm, waiter, ww_ctx);
1605 if (ret)
1606 break;
1607 }
1608
1609 if (waiter == rt_mutex_top_waiter(lock))
1610 owner = rt_mutex_owner(lock);
1611 else
1612 owner = NULL;
1613 raw_spin_unlock_irq(&lock->wait_lock);
1614
1615 if (!owner || !rtmutex_spin_on_owner(lock, waiter, owner))
1616 schedule();
1617
1618 raw_spin_lock_irq(&lock->wait_lock);
1619 set_current_state(state);
1620 }
1621
1622 __set_current_state(TASK_RUNNING);
1623 return ret;
1624}
1625
1626static void __sched rt_mutex_handle_deadlock(int res, int detect_deadlock,
1627 struct rt_mutex_waiter *w)
1628{
1629 /*
1630 * If the result is not -EDEADLOCK or the caller requested
1631 * deadlock detection, nothing to do here.
1632 */
1633 if (res != -EDEADLOCK || detect_deadlock)
1634 return;
1635
1636 if (build_ww_mutex() && w->ww_ctx)
1637 return;
1638
1639 /*
1640 * Yell loudly and stop the task right here.
1641 */
1642 WARN(1, "rtmutex deadlock detected\n");
1643 while (1) {
1644 set_current_state(TASK_INTERRUPTIBLE);
1645 schedule();
1646 }
1647}
1648
1649/**
1650 * __rt_mutex_slowlock - Locking slowpath invoked with lock::wait_lock held
1651 * @lock: The rtmutex to block lock
1652 * @ww_ctx: WW mutex context pointer
1653 * @state: The task state for sleeping
1654 * @chwalk: Indicator whether full or partial chainwalk is requested
1655 * @waiter: Initializer waiter for blocking
1656 */
1657static int __sched __rt_mutex_slowlock(struct rt_mutex_base *lock,
1658 struct ww_acquire_ctx *ww_ctx,
1659 unsigned int state,
1660 enum rtmutex_chainwalk chwalk,
1661 struct rt_mutex_waiter *waiter)
1662{
1663 struct rt_mutex *rtm = container_of(lock, struct rt_mutex, rtmutex);
1664 struct ww_mutex *ww = ww_container_of(rtm);
1665 int ret;
1666
1667 lockdep_assert_held(&lock->wait_lock);
1668
1669 /* Try to acquire the lock again: */
1670 if (try_to_take_rt_mutex(lock, current, NULL)) {
1671 if (build_ww_mutex() && ww_ctx) {
1672 __ww_mutex_check_waiters(rtm, ww_ctx);
1673 ww_mutex_lock_acquired(ww, ww_ctx);
1674 }
1675 return 0;
1676 }
1677
1678 set_current_state(state);
1679
1680 trace_contention_begin(lock, LCB_F_RT);
1681
1682 ret = task_blocks_on_rt_mutex(lock, waiter, current, ww_ctx, chwalk);
1683 if (likely(!ret))
1684 ret = rt_mutex_slowlock_block(lock, ww_ctx, state, NULL, waiter);
1685
1686 if (likely(!ret)) {
1687 /* acquired the lock */
1688 if (build_ww_mutex() && ww_ctx) {
1689 if (!ww_ctx->is_wait_die)
1690 __ww_mutex_check_waiters(rtm, ww_ctx);
1691 ww_mutex_lock_acquired(ww, ww_ctx);
1692 }
1693 } else {
1694 __set_current_state(TASK_RUNNING);
1695 remove_waiter(lock, waiter);
1696 rt_mutex_handle_deadlock(ret, chwalk, waiter);
1697 }
1698
1699 /*
1700 * try_to_take_rt_mutex() sets the waiter bit
1701 * unconditionally. We might have to fix that up.
1702 */
1703 fixup_rt_mutex_waiters(lock, true);
1704
1705 trace_contention_end(lock, ret);
1706
1707 return ret;
1708}
1709
1710static inline int __rt_mutex_slowlock_locked(struct rt_mutex_base *lock,
1711 struct ww_acquire_ctx *ww_ctx,
1712 unsigned int state)
1713{
1714 struct rt_mutex_waiter waiter;
1715 int ret;
1716
1717 rt_mutex_init_waiter(&waiter);
1718 waiter.ww_ctx = ww_ctx;
1719
1720 ret = __rt_mutex_slowlock(lock, ww_ctx, state, RT_MUTEX_MIN_CHAINWALK,
1721 &waiter);
1722
1723 debug_rt_mutex_free_waiter(&waiter);
1724 return ret;
1725}
1726
1727/*
1728 * rt_mutex_slowlock - Locking slowpath invoked when fast path fails
1729 * @lock: The rtmutex to block lock
1730 * @ww_ctx: WW mutex context pointer
1731 * @state: The task state for sleeping
1732 */
1733static int __sched rt_mutex_slowlock(struct rt_mutex_base *lock,
1734 struct ww_acquire_ctx *ww_ctx,
1735 unsigned int state)
1736{
1737 unsigned long flags;
1738 int ret;
1739
1740 /*
1741 * Technically we could use raw_spin_[un]lock_irq() here, but this can
1742 * be called in early boot if the cmpxchg() fast path is disabled
1743 * (debug, no architecture support). In this case we will acquire the
1744 * rtmutex with lock->wait_lock held. But we cannot unconditionally
1745 * enable interrupts in that early boot case. So we need to use the
1746 * irqsave/restore variants.
1747 */
1748 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1749 ret = __rt_mutex_slowlock_locked(lock, ww_ctx, state);
1750 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1751
1752 return ret;
1753}
1754
1755static __always_inline int __rt_mutex_lock(struct rt_mutex_base *lock,
1756 unsigned int state)
1757{
1758 if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
1759 return 0;
1760
1761 return rt_mutex_slowlock(lock, NULL, state);
1762}
1763#endif /* RT_MUTEX_BUILD_MUTEX */
1764
1765#ifdef RT_MUTEX_BUILD_SPINLOCKS
1766/*
1767 * Functions required for spin/rw_lock substitution on RT kernels
1768 */
1769
1770/**
1771 * rtlock_slowlock_locked - Slow path lock acquisition for RT locks
1772 * @lock: The underlying RT mutex
1773 */
1774static void __sched rtlock_slowlock_locked(struct rt_mutex_base *lock)
1775{
1776 struct rt_mutex_waiter waiter;
1777 struct task_struct *owner;
1778
1779 lockdep_assert_held(&lock->wait_lock);
1780
1781 if (try_to_take_rt_mutex(lock, current, NULL))
1782 return;
1783
1784 rt_mutex_init_rtlock_waiter(&waiter);
1785
1786 /* Save current state and set state to TASK_RTLOCK_WAIT */
1787 current_save_and_set_rtlock_wait_state();
1788
1789 trace_contention_begin(lock, LCB_F_RT);
1790
1791 task_blocks_on_rt_mutex(lock, &waiter, current, NULL, RT_MUTEX_MIN_CHAINWALK);
1792
1793 for (;;) {
1794 /* Try to acquire the lock again */
1795 if (try_to_take_rt_mutex(lock, current, &waiter))
1796 break;
1797
1798 if (&waiter == rt_mutex_top_waiter(lock))
1799 owner = rt_mutex_owner(lock);
1800 else
1801 owner = NULL;
1802 raw_spin_unlock_irq(&lock->wait_lock);
1803
1804 if (!owner || !rtmutex_spin_on_owner(lock, &waiter, owner))
1805 schedule_rtlock();
1806
1807 raw_spin_lock_irq(&lock->wait_lock);
1808 set_current_state(TASK_RTLOCK_WAIT);
1809 }
1810
1811 /* Restore the task state */
1812 current_restore_rtlock_saved_state();
1813
1814 /*
1815 * try_to_take_rt_mutex() sets the waiter bit unconditionally.
1816 * We might have to fix that up:
1817 */
1818 fixup_rt_mutex_waiters(lock, true);
1819 debug_rt_mutex_free_waiter(&waiter);
1820
1821 trace_contention_end(lock, 0);
1822}
1823
1824static __always_inline void __sched rtlock_slowlock(struct rt_mutex_base *lock)
1825{
1826 unsigned long flags;
1827
1828 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1829 rtlock_slowlock_locked(lock);
1830 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1831}
1832
1833#endif /* RT_MUTEX_BUILD_SPINLOCKS */