Loading...
Note: File does not exist in v3.1.
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 <mingo@redhat.com>
8 * Copyright (C) 2005-2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com>
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 <srostedt@redhat.com>
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
336static __always_inline void
337waiter_update_prio(struct rt_mutex_waiter *waiter, struct task_struct *task)
338{
339 waiter->prio = __waiter_prio(task);
340 waiter->deadline = task->dl.deadline;
341}
342
343/*
344 * Only use with rt_mutex_waiter_{less,equal}()
345 */
346#define task_to_waiter(p) \
347 &(struct rt_mutex_waiter){ .prio = __waiter_prio(p), .deadline = (p)->dl.deadline }
348
349static __always_inline int rt_mutex_waiter_less(struct rt_mutex_waiter *left,
350 struct rt_mutex_waiter *right)
351{
352 if (left->prio < right->prio)
353 return 1;
354
355 /*
356 * If both waiters have dl_prio(), we check the deadlines of the
357 * associated tasks.
358 * If left waiter has a dl_prio(), and we didn't return 1 above,
359 * then right waiter has a dl_prio() too.
360 */
361 if (dl_prio(left->prio))
362 return dl_time_before(left->deadline, right->deadline);
363
364 return 0;
365}
366
367static __always_inline int rt_mutex_waiter_equal(struct rt_mutex_waiter *left,
368 struct rt_mutex_waiter *right)
369{
370 if (left->prio != right->prio)
371 return 0;
372
373 /*
374 * If both waiters have dl_prio(), we check the deadlines of the
375 * associated tasks.
376 * If left waiter has a dl_prio(), and we didn't return 0 above,
377 * then right waiter has a dl_prio() too.
378 */
379 if (dl_prio(left->prio))
380 return left->deadline == right->deadline;
381
382 return 1;
383}
384
385static inline bool rt_mutex_steal(struct rt_mutex_waiter *waiter,
386 struct rt_mutex_waiter *top_waiter)
387{
388 if (rt_mutex_waiter_less(waiter, top_waiter))
389 return true;
390
391#ifdef RT_MUTEX_BUILD_SPINLOCKS
392 /*
393 * Note that RT tasks are excluded from same priority (lateral)
394 * steals to prevent the introduction of an unbounded latency.
395 */
396 if (rt_prio(waiter->prio) || dl_prio(waiter->prio))
397 return false;
398
399 return rt_mutex_waiter_equal(waiter, top_waiter);
400#else
401 return false;
402#endif
403}
404
405#define __node_2_waiter(node) \
406 rb_entry((node), struct rt_mutex_waiter, tree_entry)
407
408static __always_inline bool __waiter_less(struct rb_node *a, const struct rb_node *b)
409{
410 struct rt_mutex_waiter *aw = __node_2_waiter(a);
411 struct rt_mutex_waiter *bw = __node_2_waiter(b);
412
413 if (rt_mutex_waiter_less(aw, bw))
414 return 1;
415
416 if (!build_ww_mutex())
417 return 0;
418
419 if (rt_mutex_waiter_less(bw, aw))
420 return 0;
421
422 /* NOTE: relies on waiter->ww_ctx being set before insertion */
423 if (aw->ww_ctx) {
424 if (!bw->ww_ctx)
425 return 1;
426
427 return (signed long)(aw->ww_ctx->stamp -
428 bw->ww_ctx->stamp) < 0;
429 }
430
431 return 0;
432}
433
434static __always_inline void
435rt_mutex_enqueue(struct rt_mutex_base *lock, struct rt_mutex_waiter *waiter)
436{
437 rb_add_cached(&waiter->tree_entry, &lock->waiters, __waiter_less);
438}
439
440static __always_inline void
441rt_mutex_dequeue(struct rt_mutex_base *lock, struct rt_mutex_waiter *waiter)
442{
443 if (RB_EMPTY_NODE(&waiter->tree_entry))
444 return;
445
446 rb_erase_cached(&waiter->tree_entry, &lock->waiters);
447 RB_CLEAR_NODE(&waiter->tree_entry);
448}
449
450#define __node_2_pi_waiter(node) \
451 rb_entry((node), struct rt_mutex_waiter, pi_tree_entry)
452
453static __always_inline bool
454__pi_waiter_less(struct rb_node *a, const struct rb_node *b)
455{
456 return rt_mutex_waiter_less(__node_2_pi_waiter(a), __node_2_pi_waiter(b));
457}
458
459static __always_inline void
460rt_mutex_enqueue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
461{
462 rb_add_cached(&waiter->pi_tree_entry, &task->pi_waiters, __pi_waiter_less);
463}
464
465static __always_inline void
466rt_mutex_dequeue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
467{
468 if (RB_EMPTY_NODE(&waiter->pi_tree_entry))
469 return;
470
471 rb_erase_cached(&waiter->pi_tree_entry, &task->pi_waiters);
472 RB_CLEAR_NODE(&waiter->pi_tree_entry);
473}
474
475static __always_inline void rt_mutex_adjust_prio(struct task_struct *p)
476{
477 struct task_struct *pi_task = NULL;
478
479 lockdep_assert_held(&p->pi_lock);
480
481 if (task_has_pi_waiters(p))
482 pi_task = task_top_pi_waiter(p)->task;
483
484 rt_mutex_setprio(p, pi_task);
485}
486
487/* RT mutex specific wake_q wrappers */
488static __always_inline void rt_mutex_wake_q_add_task(struct rt_wake_q_head *wqh,
489 struct task_struct *task,
490 unsigned int wake_state)
491{
492 if (IS_ENABLED(CONFIG_PREEMPT_RT) && wake_state == TASK_RTLOCK_WAIT) {
493 if (IS_ENABLED(CONFIG_PROVE_LOCKING))
494 WARN_ON_ONCE(wqh->rtlock_task);
495 get_task_struct(task);
496 wqh->rtlock_task = task;
497 } else {
498 wake_q_add(&wqh->head, task);
499 }
500}
501
502static __always_inline void rt_mutex_wake_q_add(struct rt_wake_q_head *wqh,
503 struct rt_mutex_waiter *w)
504{
505 rt_mutex_wake_q_add_task(wqh, w->task, w->wake_state);
506}
507
508static __always_inline void rt_mutex_wake_up_q(struct rt_wake_q_head *wqh)
509{
510 if (IS_ENABLED(CONFIG_PREEMPT_RT) && wqh->rtlock_task) {
511 wake_up_state(wqh->rtlock_task, TASK_RTLOCK_WAIT);
512 put_task_struct(wqh->rtlock_task);
513 wqh->rtlock_task = NULL;
514 }
515
516 if (!wake_q_empty(&wqh->head))
517 wake_up_q(&wqh->head);
518
519 /* Pairs with preempt_disable() in mark_wakeup_next_waiter() */
520 preempt_enable();
521}
522
523/*
524 * Deadlock detection is conditional:
525 *
526 * If CONFIG_DEBUG_RT_MUTEXES=n, deadlock detection is only conducted
527 * if the detect argument is == RT_MUTEX_FULL_CHAINWALK.
528 *
529 * If CONFIG_DEBUG_RT_MUTEXES=y, deadlock detection is always
530 * conducted independent of the detect argument.
531 *
532 * If the waiter argument is NULL this indicates the deboost path and
533 * deadlock detection is disabled independent of the detect argument
534 * and the config settings.
535 */
536static __always_inline bool
537rt_mutex_cond_detect_deadlock(struct rt_mutex_waiter *waiter,
538 enum rtmutex_chainwalk chwalk)
539{
540 if (IS_ENABLED(CONFIG_DEBUG_RT_MUTEXES))
541 return waiter != NULL;
542 return chwalk == RT_MUTEX_FULL_CHAINWALK;
543}
544
545static __always_inline struct rt_mutex_base *task_blocked_on_lock(struct task_struct *p)
546{
547 return p->pi_blocked_on ? p->pi_blocked_on->lock : NULL;
548}
549
550/*
551 * Adjust the priority chain. Also used for deadlock detection.
552 * Decreases task's usage by one - may thus free the task.
553 *
554 * @task: the task owning the mutex (owner) for which a chain walk is
555 * probably needed
556 * @chwalk: do we have to carry out deadlock detection?
557 * @orig_lock: the mutex (can be NULL if we are walking the chain to recheck
558 * things for a task that has just got its priority adjusted, and
559 * is waiting on a mutex)
560 * @next_lock: the mutex on which the owner of @orig_lock was blocked before
561 * we dropped its pi_lock. Is never dereferenced, only used for
562 * comparison to detect lock chain changes.
563 * @orig_waiter: rt_mutex_waiter struct for the task that has just donated
564 * its priority to the mutex owner (can be NULL in the case
565 * depicted above or if the top waiter is gone away and we are
566 * actually deboosting the owner)
567 * @top_task: the current top waiter
568 *
569 * Returns 0 or -EDEADLK.
570 *
571 * Chain walk basics and protection scope
572 *
573 * [R] refcount on task
574 * [P] task->pi_lock held
575 * [L] rtmutex->wait_lock held
576 *
577 * Step Description Protected by
578 * function arguments:
579 * @task [R]
580 * @orig_lock if != NULL @top_task is blocked on it
581 * @next_lock Unprotected. Cannot be
582 * dereferenced. Only used for
583 * comparison.
584 * @orig_waiter if != NULL @top_task is blocked on it
585 * @top_task current, or in case of proxy
586 * locking protected by calling
587 * code
588 * again:
589 * loop_sanity_check();
590 * retry:
591 * [1] lock(task->pi_lock); [R] acquire [P]
592 * [2] waiter = task->pi_blocked_on; [P]
593 * [3] check_exit_conditions_1(); [P]
594 * [4] lock = waiter->lock; [P]
595 * [5] if (!try_lock(lock->wait_lock)) { [P] try to acquire [L]
596 * unlock(task->pi_lock); release [P]
597 * goto retry;
598 * }
599 * [6] check_exit_conditions_2(); [P] + [L]
600 * [7] requeue_lock_waiter(lock, waiter); [P] + [L]
601 * [8] unlock(task->pi_lock); release [P]
602 * put_task_struct(task); release [R]
603 * [9] check_exit_conditions_3(); [L]
604 * [10] task = owner(lock); [L]
605 * get_task_struct(task); [L] acquire [R]
606 * lock(task->pi_lock); [L] acquire [P]
607 * [11] requeue_pi_waiter(tsk, waiters(lock));[P] + [L]
608 * [12] check_exit_conditions_4(); [P] + [L]
609 * [13] unlock(task->pi_lock); release [P]
610 * unlock(lock->wait_lock); release [L]
611 * goto again;
612 */
613static int __sched rt_mutex_adjust_prio_chain(struct task_struct *task,
614 enum rtmutex_chainwalk chwalk,
615 struct rt_mutex_base *orig_lock,
616 struct rt_mutex_base *next_lock,
617 struct rt_mutex_waiter *orig_waiter,
618 struct task_struct *top_task)
619{
620 struct rt_mutex_waiter *waiter, *top_waiter = orig_waiter;
621 struct rt_mutex_waiter *prerequeue_top_waiter;
622 int ret = 0, depth = 0;
623 struct rt_mutex_base *lock;
624 bool detect_deadlock;
625 bool requeue = true;
626
627 detect_deadlock = rt_mutex_cond_detect_deadlock(orig_waiter, chwalk);
628
629 /*
630 * The (de)boosting is a step by step approach with a lot of
631 * pitfalls. We want this to be preemptible and we want hold a
632 * maximum of two locks per step. So we have to check
633 * carefully whether things change under us.
634 */
635 again:
636 /*
637 * We limit the lock chain length for each invocation.
638 */
639 if (++depth > max_lock_depth) {
640 static int prev_max;
641
642 /*
643 * Print this only once. If the admin changes the limit,
644 * print a new message when reaching the limit again.
645 */
646 if (prev_max != max_lock_depth) {
647 prev_max = max_lock_depth;
648 printk(KERN_WARNING "Maximum lock depth %d reached "
649 "task: %s (%d)\n", max_lock_depth,
650 top_task->comm, task_pid_nr(top_task));
651 }
652 put_task_struct(task);
653
654 return -EDEADLK;
655 }
656
657 /*
658 * We are fully preemptible here and only hold the refcount on
659 * @task. So everything can have changed under us since the
660 * caller or our own code below (goto retry/again) dropped all
661 * locks.
662 */
663 retry:
664 /*
665 * [1] Task cannot go away as we did a get_task() before !
666 */
667 raw_spin_lock_irq(&task->pi_lock);
668
669 /*
670 * [2] Get the waiter on which @task is blocked on.
671 */
672 waiter = task->pi_blocked_on;
673
674 /*
675 * [3] check_exit_conditions_1() protected by task->pi_lock.
676 */
677
678 /*
679 * Check whether the end of the boosting chain has been
680 * reached or the state of the chain has changed while we
681 * dropped the locks.
682 */
683 if (!waiter)
684 goto out_unlock_pi;
685
686 /*
687 * Check the orig_waiter state. After we dropped the locks,
688 * the previous owner of the lock might have released the lock.
689 */
690 if (orig_waiter && !rt_mutex_owner(orig_lock))
691 goto out_unlock_pi;
692
693 /*
694 * We dropped all locks after taking a refcount on @task, so
695 * the task might have moved on in the lock chain or even left
696 * the chain completely and blocks now on an unrelated lock or
697 * on @orig_lock.
698 *
699 * We stored the lock on which @task was blocked in @next_lock,
700 * so we can detect the chain change.
701 */
702 if (next_lock != waiter->lock)
703 goto out_unlock_pi;
704
705 /*
706 * There could be 'spurious' loops in the lock graph due to ww_mutex,
707 * consider:
708 *
709 * P1: A, ww_A, ww_B
710 * P2: ww_B, ww_A
711 * P3: A
712 *
713 * P3 should not return -EDEADLK because it gets trapped in the cycle
714 * created by P1 and P2 (which will resolve -- and runs into
715 * max_lock_depth above). Therefore disable detect_deadlock such that
716 * the below termination condition can trigger once all relevant tasks
717 * are boosted.
718 *
719 * Even when we start with ww_mutex we can disable deadlock detection,
720 * since we would supress a ww_mutex induced deadlock at [6] anyway.
721 * Supressing it here however is not sufficient since we might still
722 * hit [6] due to adjustment driven iteration.
723 *
724 * NOTE: if someone were to create a deadlock between 2 ww_classes we'd
725 * utterly fail to report it; lockdep should.
726 */
727 if (IS_ENABLED(CONFIG_PREEMPT_RT) && waiter->ww_ctx && detect_deadlock)
728 detect_deadlock = false;
729
730 /*
731 * Drop out, when the task has no waiters. Note,
732 * top_waiter can be NULL, when we are in the deboosting
733 * mode!
734 */
735 if (top_waiter) {
736 if (!task_has_pi_waiters(task))
737 goto out_unlock_pi;
738 /*
739 * If deadlock detection is off, we stop here if we
740 * are not the top pi waiter of the task. If deadlock
741 * detection is enabled we continue, but stop the
742 * requeueing in the chain walk.
743 */
744 if (top_waiter != task_top_pi_waiter(task)) {
745 if (!detect_deadlock)
746 goto out_unlock_pi;
747 else
748 requeue = false;
749 }
750 }
751
752 /*
753 * If the waiter priority is the same as the task priority
754 * then there is no further priority adjustment necessary. If
755 * deadlock detection is off, we stop the chain walk. If its
756 * enabled we continue, but stop the requeueing in the chain
757 * walk.
758 */
759 if (rt_mutex_waiter_equal(waiter, task_to_waiter(task))) {
760 if (!detect_deadlock)
761 goto out_unlock_pi;
762 else
763 requeue = false;
764 }
765
766 /*
767 * [4] Get the next lock
768 */
769 lock = waiter->lock;
770 /*
771 * [5] We need to trylock here as we are holding task->pi_lock,
772 * which is the reverse lock order versus the other rtmutex
773 * operations.
774 */
775 if (!raw_spin_trylock(&lock->wait_lock)) {
776 raw_spin_unlock_irq(&task->pi_lock);
777 cpu_relax();
778 goto retry;
779 }
780
781 /*
782 * [6] check_exit_conditions_2() protected by task->pi_lock and
783 * lock->wait_lock.
784 *
785 * Deadlock detection. If the lock is the same as the original
786 * lock which caused us to walk the lock chain or if the
787 * current lock is owned by the task which initiated the chain
788 * walk, we detected a deadlock.
789 */
790 if (lock == orig_lock || rt_mutex_owner(lock) == top_task) {
791 ret = -EDEADLK;
792
793 /*
794 * When the deadlock is due to ww_mutex; also see above. Don't
795 * report the deadlock and instead let the ww_mutex wound/die
796 * logic pick which of the contending threads gets -EDEADLK.
797 *
798 * NOTE: assumes the cycle only contains a single ww_class; any
799 * other configuration and we fail to report; also, see
800 * lockdep.
801 */
802 if (IS_ENABLED(CONFIG_PREEMPT_RT) && orig_waiter && orig_waiter->ww_ctx)
803 ret = 0;
804
805 raw_spin_unlock(&lock->wait_lock);
806 goto out_unlock_pi;
807 }
808
809 /*
810 * If we just follow the lock chain for deadlock detection, no
811 * need to do all the requeue operations. To avoid a truckload
812 * of conditionals around the various places below, just do the
813 * minimum chain walk checks.
814 */
815 if (!requeue) {
816 /*
817 * No requeue[7] here. Just release @task [8]
818 */
819 raw_spin_unlock(&task->pi_lock);
820 put_task_struct(task);
821
822 /*
823 * [9] check_exit_conditions_3 protected by lock->wait_lock.
824 * If there is no owner of the lock, end of chain.
825 */
826 if (!rt_mutex_owner(lock)) {
827 raw_spin_unlock_irq(&lock->wait_lock);
828 return 0;
829 }
830
831 /* [10] Grab the next task, i.e. owner of @lock */
832 task = get_task_struct(rt_mutex_owner(lock));
833 raw_spin_lock(&task->pi_lock);
834
835 /*
836 * No requeue [11] here. We just do deadlock detection.
837 *
838 * [12] Store whether owner is blocked
839 * itself. Decision is made after dropping the locks
840 */
841 next_lock = task_blocked_on_lock(task);
842 /*
843 * Get the top waiter for the next iteration
844 */
845 top_waiter = rt_mutex_top_waiter(lock);
846
847 /* [13] Drop locks */
848 raw_spin_unlock(&task->pi_lock);
849 raw_spin_unlock_irq(&lock->wait_lock);
850
851 /* If owner is not blocked, end of chain. */
852 if (!next_lock)
853 goto out_put_task;
854 goto again;
855 }
856
857 /*
858 * Store the current top waiter before doing the requeue
859 * operation on @lock. We need it for the boost/deboost
860 * decision below.
861 */
862 prerequeue_top_waiter = rt_mutex_top_waiter(lock);
863
864 /* [7] Requeue the waiter in the lock waiter tree. */
865 rt_mutex_dequeue(lock, waiter);
866
867 /*
868 * Update the waiter prio fields now that we're dequeued.
869 *
870 * These values can have changed through either:
871 *
872 * sys_sched_set_scheduler() / sys_sched_setattr()
873 *
874 * or
875 *
876 * DL CBS enforcement advancing the effective deadline.
877 *
878 * Even though pi_waiters also uses these fields, and that tree is only
879 * updated in [11], we can do this here, since we hold [L], which
880 * serializes all pi_waiters access and rb_erase() does not care about
881 * the values of the node being removed.
882 */
883 waiter_update_prio(waiter, task);
884
885 rt_mutex_enqueue(lock, waiter);
886
887 /* [8] Release the task */
888 raw_spin_unlock(&task->pi_lock);
889 put_task_struct(task);
890
891 /*
892 * [9] check_exit_conditions_3 protected by lock->wait_lock.
893 *
894 * We must abort the chain walk if there is no lock owner even
895 * in the dead lock detection case, as we have nothing to
896 * follow here. This is the end of the chain we are walking.
897 */
898 if (!rt_mutex_owner(lock)) {
899 /*
900 * If the requeue [7] above changed the top waiter,
901 * then we need to wake the new top waiter up to try
902 * to get the lock.
903 */
904 top_waiter = rt_mutex_top_waiter(lock);
905 if (prerequeue_top_waiter != top_waiter)
906 wake_up_state(top_waiter->task, top_waiter->wake_state);
907 raw_spin_unlock_irq(&lock->wait_lock);
908 return 0;
909 }
910
911 /* [10] Grab the next task, i.e. the owner of @lock */
912 task = get_task_struct(rt_mutex_owner(lock));
913 raw_spin_lock(&task->pi_lock);
914
915 /* [11] requeue the pi waiters if necessary */
916 if (waiter == rt_mutex_top_waiter(lock)) {
917 /*
918 * The waiter became the new top (highest priority)
919 * waiter on the lock. Replace the previous top waiter
920 * in the owner tasks pi waiters tree with this waiter
921 * and adjust the priority of the owner.
922 */
923 rt_mutex_dequeue_pi(task, prerequeue_top_waiter);
924 rt_mutex_enqueue_pi(task, waiter);
925 rt_mutex_adjust_prio(task);
926
927 } else if (prerequeue_top_waiter == waiter) {
928 /*
929 * The waiter was the top waiter on the lock, but is
930 * no longer the top priority waiter. Replace waiter in
931 * the owner tasks pi waiters tree with the new top
932 * (highest priority) waiter and adjust the priority
933 * of the owner.
934 * The new top waiter is stored in @waiter so that
935 * @waiter == @top_waiter evaluates to true below and
936 * we continue to deboost the rest of the chain.
937 */
938 rt_mutex_dequeue_pi(task, waiter);
939 waiter = rt_mutex_top_waiter(lock);
940 rt_mutex_enqueue_pi(task, waiter);
941 rt_mutex_adjust_prio(task);
942 } else {
943 /*
944 * Nothing changed. No need to do any priority
945 * adjustment.
946 */
947 }
948
949 /*
950 * [12] check_exit_conditions_4() protected by task->pi_lock
951 * and lock->wait_lock. The actual decisions are made after we
952 * dropped the locks.
953 *
954 * Check whether the task which owns the current lock is pi
955 * blocked itself. If yes we store a pointer to the lock for
956 * the lock chain change detection above. After we dropped
957 * task->pi_lock next_lock cannot be dereferenced anymore.
958 */
959 next_lock = task_blocked_on_lock(task);
960 /*
961 * Store the top waiter of @lock for the end of chain walk
962 * decision below.
963 */
964 top_waiter = rt_mutex_top_waiter(lock);
965
966 /* [13] Drop the locks */
967 raw_spin_unlock(&task->pi_lock);
968 raw_spin_unlock_irq(&lock->wait_lock);
969
970 /*
971 * Make the actual exit decisions [12], based on the stored
972 * values.
973 *
974 * We reached the end of the lock chain. Stop right here. No
975 * point to go back just to figure that out.
976 */
977 if (!next_lock)
978 goto out_put_task;
979
980 /*
981 * If the current waiter is not the top waiter on the lock,
982 * then we can stop the chain walk here if we are not in full
983 * deadlock detection mode.
984 */
985 if (!detect_deadlock && waiter != top_waiter)
986 goto out_put_task;
987
988 goto again;
989
990 out_unlock_pi:
991 raw_spin_unlock_irq(&task->pi_lock);
992 out_put_task:
993 put_task_struct(task);
994
995 return ret;
996}
997
998/*
999 * Try to take an rt-mutex
1000 *
1001 * Must be called with lock->wait_lock held and interrupts disabled
1002 *
1003 * @lock: The lock to be acquired.
1004 * @task: The task which wants to acquire the lock
1005 * @waiter: The waiter that is queued to the lock's wait tree if the
1006 * callsite called task_blocked_on_lock(), otherwise NULL
1007 */
1008static int __sched
1009try_to_take_rt_mutex(struct rt_mutex_base *lock, struct task_struct *task,
1010 struct rt_mutex_waiter *waiter)
1011{
1012 lockdep_assert_held(&lock->wait_lock);
1013
1014 /*
1015 * Before testing whether we can acquire @lock, we set the
1016 * RT_MUTEX_HAS_WAITERS bit in @lock->owner. This forces all
1017 * other tasks which try to modify @lock into the slow path
1018 * and they serialize on @lock->wait_lock.
1019 *
1020 * The RT_MUTEX_HAS_WAITERS bit can have a transitional state
1021 * as explained at the top of this file if and only if:
1022 *
1023 * - There is a lock owner. The caller must fixup the
1024 * transient state if it does a trylock or leaves the lock
1025 * function due to a signal or timeout.
1026 *
1027 * - @task acquires the lock and there are no other
1028 * waiters. This is undone in rt_mutex_set_owner(@task) at
1029 * the end of this function.
1030 */
1031 mark_rt_mutex_waiters(lock);
1032
1033 /*
1034 * If @lock has an owner, give up.
1035 */
1036 if (rt_mutex_owner(lock))
1037 return 0;
1038
1039 /*
1040 * If @waiter != NULL, @task has already enqueued the waiter
1041 * into @lock waiter tree. If @waiter == NULL then this is a
1042 * trylock attempt.
1043 */
1044 if (waiter) {
1045 struct rt_mutex_waiter *top_waiter = rt_mutex_top_waiter(lock);
1046
1047 /*
1048 * If waiter is the highest priority waiter of @lock,
1049 * or allowed to steal it, take it over.
1050 */
1051 if (waiter == top_waiter || rt_mutex_steal(waiter, top_waiter)) {
1052 /*
1053 * We can acquire the lock. Remove the waiter from the
1054 * lock waiters tree.
1055 */
1056 rt_mutex_dequeue(lock, waiter);
1057 } else {
1058 return 0;
1059 }
1060 } else {
1061 /*
1062 * If the lock has waiters already we check whether @task is
1063 * eligible to take over the lock.
1064 *
1065 * If there are no other waiters, @task can acquire
1066 * the lock. @task->pi_blocked_on is NULL, so it does
1067 * not need to be dequeued.
1068 */
1069 if (rt_mutex_has_waiters(lock)) {
1070 /* Check whether the trylock can steal it. */
1071 if (!rt_mutex_steal(task_to_waiter(task),
1072 rt_mutex_top_waiter(lock)))
1073 return 0;
1074
1075 /*
1076 * The current top waiter stays enqueued. We
1077 * don't have to change anything in the lock
1078 * waiters order.
1079 */
1080 } else {
1081 /*
1082 * No waiters. Take the lock without the
1083 * pi_lock dance.@task->pi_blocked_on is NULL
1084 * and we have no waiters to enqueue in @task
1085 * pi waiters tree.
1086 */
1087 goto takeit;
1088 }
1089 }
1090
1091 /*
1092 * Clear @task->pi_blocked_on. Requires protection by
1093 * @task->pi_lock. Redundant operation for the @waiter == NULL
1094 * case, but conditionals are more expensive than a redundant
1095 * store.
1096 */
1097 raw_spin_lock(&task->pi_lock);
1098 task->pi_blocked_on = NULL;
1099 /*
1100 * Finish the lock acquisition. @task is the new owner. If
1101 * other waiters exist we have to insert the highest priority
1102 * waiter into @task->pi_waiters tree.
1103 */
1104 if (rt_mutex_has_waiters(lock))
1105 rt_mutex_enqueue_pi(task, rt_mutex_top_waiter(lock));
1106 raw_spin_unlock(&task->pi_lock);
1107
1108takeit:
1109 /*
1110 * This either preserves the RT_MUTEX_HAS_WAITERS bit if there
1111 * are still waiters or clears it.
1112 */
1113 rt_mutex_set_owner(lock, task);
1114
1115 return 1;
1116}
1117
1118/*
1119 * Task blocks on lock.
1120 *
1121 * Prepare waiter and propagate pi chain
1122 *
1123 * This must be called with lock->wait_lock held and interrupts disabled
1124 */
1125static int __sched task_blocks_on_rt_mutex(struct rt_mutex_base *lock,
1126 struct rt_mutex_waiter *waiter,
1127 struct task_struct *task,
1128 struct ww_acquire_ctx *ww_ctx,
1129 enum rtmutex_chainwalk chwalk)
1130{
1131 struct task_struct *owner = rt_mutex_owner(lock);
1132 struct rt_mutex_waiter *top_waiter = waiter;
1133 struct rt_mutex_base *next_lock;
1134 int chain_walk = 0, res;
1135
1136 lockdep_assert_held(&lock->wait_lock);
1137
1138 /*
1139 * Early deadlock detection. We really don't want the task to
1140 * enqueue on itself just to untangle the mess later. It's not
1141 * only an optimization. We drop the locks, so another waiter
1142 * can come in before the chain walk detects the deadlock. So
1143 * the other will detect the deadlock and return -EDEADLOCK,
1144 * which is wrong, as the other waiter is not in a deadlock
1145 * situation.
1146 *
1147 * Except for ww_mutex, in that case the chain walk must already deal
1148 * with spurious cycles, see the comments at [3] and [6].
1149 */
1150 if (owner == task && !(build_ww_mutex() && ww_ctx))
1151 return -EDEADLK;
1152
1153 raw_spin_lock(&task->pi_lock);
1154 waiter->task = task;
1155 waiter->lock = lock;
1156 waiter_update_prio(waiter, task);
1157
1158 /* Get the top priority waiter on the lock */
1159 if (rt_mutex_has_waiters(lock))
1160 top_waiter = rt_mutex_top_waiter(lock);
1161 rt_mutex_enqueue(lock, waiter);
1162
1163 task->pi_blocked_on = waiter;
1164
1165 raw_spin_unlock(&task->pi_lock);
1166
1167 if (build_ww_mutex() && ww_ctx) {
1168 struct rt_mutex *rtm;
1169
1170 /* Check whether the waiter should back out immediately */
1171 rtm = container_of(lock, struct rt_mutex, rtmutex);
1172 res = __ww_mutex_add_waiter(waiter, rtm, ww_ctx);
1173 if (res) {
1174 raw_spin_lock(&task->pi_lock);
1175 rt_mutex_dequeue(lock, waiter);
1176 task->pi_blocked_on = NULL;
1177 raw_spin_unlock(&task->pi_lock);
1178 return res;
1179 }
1180 }
1181
1182 if (!owner)
1183 return 0;
1184
1185 raw_spin_lock(&owner->pi_lock);
1186 if (waiter == rt_mutex_top_waiter(lock)) {
1187 rt_mutex_dequeue_pi(owner, top_waiter);
1188 rt_mutex_enqueue_pi(owner, waiter);
1189
1190 rt_mutex_adjust_prio(owner);
1191 if (owner->pi_blocked_on)
1192 chain_walk = 1;
1193 } else if (rt_mutex_cond_detect_deadlock(waiter, chwalk)) {
1194 chain_walk = 1;
1195 }
1196
1197 /* Store the lock on which owner is blocked or NULL */
1198 next_lock = task_blocked_on_lock(owner);
1199
1200 raw_spin_unlock(&owner->pi_lock);
1201 /*
1202 * Even if full deadlock detection is on, if the owner is not
1203 * blocked itself, we can avoid finding this out in the chain
1204 * walk.
1205 */
1206 if (!chain_walk || !next_lock)
1207 return 0;
1208
1209 /*
1210 * The owner can't disappear while holding a lock,
1211 * so the owner struct is protected by wait_lock.
1212 * Gets dropped in rt_mutex_adjust_prio_chain()!
1213 */
1214 get_task_struct(owner);
1215
1216 raw_spin_unlock_irq(&lock->wait_lock);
1217
1218 res = rt_mutex_adjust_prio_chain(owner, chwalk, lock,
1219 next_lock, waiter, task);
1220
1221 raw_spin_lock_irq(&lock->wait_lock);
1222
1223 return res;
1224}
1225
1226/*
1227 * Remove the top waiter from the current tasks pi waiter tree and
1228 * queue it up.
1229 *
1230 * Called with lock->wait_lock held and interrupts disabled.
1231 */
1232static void __sched mark_wakeup_next_waiter(struct rt_wake_q_head *wqh,
1233 struct rt_mutex_base *lock)
1234{
1235 struct rt_mutex_waiter *waiter;
1236
1237 raw_spin_lock(¤t->pi_lock);
1238
1239 waiter = rt_mutex_top_waiter(lock);
1240
1241 /*
1242 * Remove it from current->pi_waiters and deboost.
1243 *
1244 * We must in fact deboost here in order to ensure we call
1245 * rt_mutex_setprio() to update p->pi_top_task before the
1246 * task unblocks.
1247 */
1248 rt_mutex_dequeue_pi(current, waiter);
1249 rt_mutex_adjust_prio(current);
1250
1251 /*
1252 * As we are waking up the top waiter, and the waiter stays
1253 * queued on the lock until it gets the lock, this lock
1254 * obviously has waiters. Just set the bit here and this has
1255 * the added benefit of forcing all new tasks into the
1256 * slow path making sure no task of lower priority than
1257 * the top waiter can steal this lock.
1258 */
1259 lock->owner = (void *) RT_MUTEX_HAS_WAITERS;
1260
1261 /*
1262 * We deboosted before waking the top waiter task such that we don't
1263 * run two tasks with the 'same' priority (and ensure the
1264 * p->pi_top_task pointer points to a blocked task). This however can
1265 * lead to priority inversion if we would get preempted after the
1266 * deboost but before waking our donor task, hence the preempt_disable()
1267 * before unlock.
1268 *
1269 * Pairs with preempt_enable() in rt_mutex_wake_up_q();
1270 */
1271 preempt_disable();
1272 rt_mutex_wake_q_add(wqh, waiter);
1273 raw_spin_unlock(¤t->pi_lock);
1274}
1275
1276static int __sched __rt_mutex_slowtrylock(struct rt_mutex_base *lock)
1277{
1278 int ret = try_to_take_rt_mutex(lock, current, NULL);
1279
1280 /*
1281 * try_to_take_rt_mutex() sets the lock waiters bit
1282 * unconditionally. Clean this up.
1283 */
1284 fixup_rt_mutex_waiters(lock, true);
1285
1286 return ret;
1287}
1288
1289/*
1290 * Slow path try-lock function:
1291 */
1292static int __sched rt_mutex_slowtrylock(struct rt_mutex_base *lock)
1293{
1294 unsigned long flags;
1295 int ret;
1296
1297 /*
1298 * If the lock already has an owner we fail to get the lock.
1299 * This can be done without taking the @lock->wait_lock as
1300 * it is only being read, and this is a trylock anyway.
1301 */
1302 if (rt_mutex_owner(lock))
1303 return 0;
1304
1305 /*
1306 * The mutex has currently no owner. Lock the wait lock and try to
1307 * acquire the lock. We use irqsave here to support early boot calls.
1308 */
1309 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1310
1311 ret = __rt_mutex_slowtrylock(lock);
1312
1313 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1314
1315 return ret;
1316}
1317
1318static __always_inline int __rt_mutex_trylock(struct rt_mutex_base *lock)
1319{
1320 if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
1321 return 1;
1322
1323 return rt_mutex_slowtrylock(lock);
1324}
1325
1326/*
1327 * Slow path to release a rt-mutex.
1328 */
1329static void __sched rt_mutex_slowunlock(struct rt_mutex_base *lock)
1330{
1331 DEFINE_RT_WAKE_Q(wqh);
1332 unsigned long flags;
1333
1334 /* irqsave required to support early boot calls */
1335 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1336
1337 debug_rt_mutex_unlock(lock);
1338
1339 /*
1340 * We must be careful here if the fast path is enabled. If we
1341 * have no waiters queued we cannot set owner to NULL here
1342 * because of:
1343 *
1344 * foo->lock->owner = NULL;
1345 * rtmutex_lock(foo->lock); <- fast path
1346 * free = atomic_dec_and_test(foo->refcnt);
1347 * rtmutex_unlock(foo->lock); <- fast path
1348 * if (free)
1349 * kfree(foo);
1350 * raw_spin_unlock(foo->lock->wait_lock);
1351 *
1352 * So for the fastpath enabled kernel:
1353 *
1354 * Nothing can set the waiters bit as long as we hold
1355 * lock->wait_lock. So we do the following sequence:
1356 *
1357 * owner = rt_mutex_owner(lock);
1358 * clear_rt_mutex_waiters(lock);
1359 * raw_spin_unlock(&lock->wait_lock);
1360 * if (cmpxchg(&lock->owner, owner, 0) == owner)
1361 * return;
1362 * goto retry;
1363 *
1364 * The fastpath disabled variant is simple as all access to
1365 * lock->owner is serialized by lock->wait_lock:
1366 *
1367 * lock->owner = NULL;
1368 * raw_spin_unlock(&lock->wait_lock);
1369 */
1370 while (!rt_mutex_has_waiters(lock)) {
1371 /* Drops lock->wait_lock ! */
1372 if (unlock_rt_mutex_safe(lock, flags) == true)
1373 return;
1374 /* Relock the rtmutex and try again */
1375 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1376 }
1377
1378 /*
1379 * The wakeup next waiter path does not suffer from the above
1380 * race. See the comments there.
1381 *
1382 * Queue the next waiter for wakeup once we release the wait_lock.
1383 */
1384 mark_wakeup_next_waiter(&wqh, lock);
1385 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1386
1387 rt_mutex_wake_up_q(&wqh);
1388}
1389
1390static __always_inline void __rt_mutex_unlock(struct rt_mutex_base *lock)
1391{
1392 if (likely(rt_mutex_cmpxchg_release(lock, current, NULL)))
1393 return;
1394
1395 rt_mutex_slowunlock(lock);
1396}
1397
1398#ifdef CONFIG_SMP
1399static bool rtmutex_spin_on_owner(struct rt_mutex_base *lock,
1400 struct rt_mutex_waiter *waiter,
1401 struct task_struct *owner)
1402{
1403 bool res = true;
1404
1405 rcu_read_lock();
1406 for (;;) {
1407 /* If owner changed, trylock again. */
1408 if (owner != rt_mutex_owner(lock))
1409 break;
1410 /*
1411 * Ensure that @owner is dereferenced after checking that
1412 * the lock owner still matches @owner. If that fails,
1413 * @owner might point to freed memory. If it still matches,
1414 * the rcu_read_lock() ensures the memory stays valid.
1415 */
1416 barrier();
1417 /*
1418 * Stop spinning when:
1419 * - the lock owner has been scheduled out
1420 * - current is not longer the top waiter
1421 * - current is requested to reschedule (redundant
1422 * for CONFIG_PREEMPT_RCU=y)
1423 * - the VCPU on which owner runs is preempted
1424 */
1425 if (!owner_on_cpu(owner) || need_resched() ||
1426 !rt_mutex_waiter_is_top_waiter(lock, waiter)) {
1427 res = false;
1428 break;
1429 }
1430 cpu_relax();
1431 }
1432 rcu_read_unlock();
1433 return res;
1434}
1435#else
1436static bool rtmutex_spin_on_owner(struct rt_mutex_base *lock,
1437 struct rt_mutex_waiter *waiter,
1438 struct task_struct *owner)
1439{
1440 return false;
1441}
1442#endif
1443
1444#ifdef RT_MUTEX_BUILD_MUTEX
1445/*
1446 * Functions required for:
1447 * - rtmutex, futex on all kernels
1448 * - mutex and rwsem substitutions on RT kernels
1449 */
1450
1451/*
1452 * Remove a waiter from a lock and give up
1453 *
1454 * Must be called with lock->wait_lock held and interrupts disabled. It must
1455 * have just failed to try_to_take_rt_mutex().
1456 */
1457static void __sched remove_waiter(struct rt_mutex_base *lock,
1458 struct rt_mutex_waiter *waiter)
1459{
1460 bool is_top_waiter = (waiter == rt_mutex_top_waiter(lock));
1461 struct task_struct *owner = rt_mutex_owner(lock);
1462 struct rt_mutex_base *next_lock;
1463
1464 lockdep_assert_held(&lock->wait_lock);
1465
1466 raw_spin_lock(¤t->pi_lock);
1467 rt_mutex_dequeue(lock, waiter);
1468 current->pi_blocked_on = NULL;
1469 raw_spin_unlock(¤t->pi_lock);
1470
1471 /*
1472 * Only update priority if the waiter was the highest priority
1473 * waiter of the lock and there is an owner to update.
1474 */
1475 if (!owner || !is_top_waiter)
1476 return;
1477
1478 raw_spin_lock(&owner->pi_lock);
1479
1480 rt_mutex_dequeue_pi(owner, waiter);
1481
1482 if (rt_mutex_has_waiters(lock))
1483 rt_mutex_enqueue_pi(owner, rt_mutex_top_waiter(lock));
1484
1485 rt_mutex_adjust_prio(owner);
1486
1487 /* Store the lock on which owner is blocked or NULL */
1488 next_lock = task_blocked_on_lock(owner);
1489
1490 raw_spin_unlock(&owner->pi_lock);
1491
1492 /*
1493 * Don't walk the chain, if the owner task is not blocked
1494 * itself.
1495 */
1496 if (!next_lock)
1497 return;
1498
1499 /* gets dropped in rt_mutex_adjust_prio_chain()! */
1500 get_task_struct(owner);
1501
1502 raw_spin_unlock_irq(&lock->wait_lock);
1503
1504 rt_mutex_adjust_prio_chain(owner, RT_MUTEX_MIN_CHAINWALK, lock,
1505 next_lock, NULL, current);
1506
1507 raw_spin_lock_irq(&lock->wait_lock);
1508}
1509
1510/**
1511 * rt_mutex_slowlock_block() - Perform the wait-wake-try-to-take loop
1512 * @lock: the rt_mutex to take
1513 * @ww_ctx: WW mutex context pointer
1514 * @state: the state the task should block in (TASK_INTERRUPTIBLE
1515 * or TASK_UNINTERRUPTIBLE)
1516 * @timeout: the pre-initialized and started timer, or NULL for none
1517 * @waiter: the pre-initialized rt_mutex_waiter
1518 *
1519 * Must be called with lock->wait_lock held and interrupts disabled
1520 */
1521static int __sched rt_mutex_slowlock_block(struct rt_mutex_base *lock,
1522 struct ww_acquire_ctx *ww_ctx,
1523 unsigned int state,
1524 struct hrtimer_sleeper *timeout,
1525 struct rt_mutex_waiter *waiter)
1526{
1527 struct rt_mutex *rtm = container_of(lock, struct rt_mutex, rtmutex);
1528 struct task_struct *owner;
1529 int ret = 0;
1530
1531 for (;;) {
1532 /* Try to acquire the lock: */
1533 if (try_to_take_rt_mutex(lock, current, waiter))
1534 break;
1535
1536 if (timeout && !timeout->task) {
1537 ret = -ETIMEDOUT;
1538 break;
1539 }
1540 if (signal_pending_state(state, current)) {
1541 ret = -EINTR;
1542 break;
1543 }
1544
1545 if (build_ww_mutex() && ww_ctx) {
1546 ret = __ww_mutex_check_kill(rtm, waiter, ww_ctx);
1547 if (ret)
1548 break;
1549 }
1550
1551 if (waiter == rt_mutex_top_waiter(lock))
1552 owner = rt_mutex_owner(lock);
1553 else
1554 owner = NULL;
1555 raw_spin_unlock_irq(&lock->wait_lock);
1556
1557 if (!owner || !rtmutex_spin_on_owner(lock, waiter, owner))
1558 schedule();
1559
1560 raw_spin_lock_irq(&lock->wait_lock);
1561 set_current_state(state);
1562 }
1563
1564 __set_current_state(TASK_RUNNING);
1565 return ret;
1566}
1567
1568static void __sched rt_mutex_handle_deadlock(int res, int detect_deadlock,
1569 struct rt_mutex_waiter *w)
1570{
1571 /*
1572 * If the result is not -EDEADLOCK or the caller requested
1573 * deadlock detection, nothing to do here.
1574 */
1575 if (res != -EDEADLOCK || detect_deadlock)
1576 return;
1577
1578 if (build_ww_mutex() && w->ww_ctx)
1579 return;
1580
1581 /*
1582 * Yell loudly and stop the task right here.
1583 */
1584 WARN(1, "rtmutex deadlock detected\n");
1585 while (1) {
1586 set_current_state(TASK_INTERRUPTIBLE);
1587 schedule();
1588 }
1589}
1590
1591/**
1592 * __rt_mutex_slowlock - Locking slowpath invoked with lock::wait_lock held
1593 * @lock: The rtmutex to block lock
1594 * @ww_ctx: WW mutex context pointer
1595 * @state: The task state for sleeping
1596 * @chwalk: Indicator whether full or partial chainwalk is requested
1597 * @waiter: Initializer waiter for blocking
1598 */
1599static int __sched __rt_mutex_slowlock(struct rt_mutex_base *lock,
1600 struct ww_acquire_ctx *ww_ctx,
1601 unsigned int state,
1602 enum rtmutex_chainwalk chwalk,
1603 struct rt_mutex_waiter *waiter)
1604{
1605 struct rt_mutex *rtm = container_of(lock, struct rt_mutex, rtmutex);
1606 struct ww_mutex *ww = ww_container_of(rtm);
1607 int ret;
1608
1609 lockdep_assert_held(&lock->wait_lock);
1610
1611 /* Try to acquire the lock again: */
1612 if (try_to_take_rt_mutex(lock, current, NULL)) {
1613 if (build_ww_mutex() && ww_ctx) {
1614 __ww_mutex_check_waiters(rtm, ww_ctx);
1615 ww_mutex_lock_acquired(ww, ww_ctx);
1616 }
1617 return 0;
1618 }
1619
1620 set_current_state(state);
1621
1622 trace_contention_begin(lock, LCB_F_RT);
1623
1624 ret = task_blocks_on_rt_mutex(lock, waiter, current, ww_ctx, chwalk);
1625 if (likely(!ret))
1626 ret = rt_mutex_slowlock_block(lock, ww_ctx, state, NULL, waiter);
1627
1628 if (likely(!ret)) {
1629 /* acquired the lock */
1630 if (build_ww_mutex() && ww_ctx) {
1631 if (!ww_ctx->is_wait_die)
1632 __ww_mutex_check_waiters(rtm, ww_ctx);
1633 ww_mutex_lock_acquired(ww, ww_ctx);
1634 }
1635 } else {
1636 __set_current_state(TASK_RUNNING);
1637 remove_waiter(lock, waiter);
1638 rt_mutex_handle_deadlock(ret, chwalk, waiter);
1639 }
1640
1641 /*
1642 * try_to_take_rt_mutex() sets the waiter bit
1643 * unconditionally. We might have to fix that up.
1644 */
1645 fixup_rt_mutex_waiters(lock, true);
1646
1647 trace_contention_end(lock, ret);
1648
1649 return ret;
1650}
1651
1652static inline int __rt_mutex_slowlock_locked(struct rt_mutex_base *lock,
1653 struct ww_acquire_ctx *ww_ctx,
1654 unsigned int state)
1655{
1656 struct rt_mutex_waiter waiter;
1657 int ret;
1658
1659 rt_mutex_init_waiter(&waiter);
1660 waiter.ww_ctx = ww_ctx;
1661
1662 ret = __rt_mutex_slowlock(lock, ww_ctx, state, RT_MUTEX_MIN_CHAINWALK,
1663 &waiter);
1664
1665 debug_rt_mutex_free_waiter(&waiter);
1666 return ret;
1667}
1668
1669/*
1670 * rt_mutex_slowlock - Locking slowpath invoked when fast path fails
1671 * @lock: The rtmutex to block lock
1672 * @ww_ctx: WW mutex context pointer
1673 * @state: The task state for sleeping
1674 */
1675static int __sched rt_mutex_slowlock(struct rt_mutex_base *lock,
1676 struct ww_acquire_ctx *ww_ctx,
1677 unsigned int state)
1678{
1679 unsigned long flags;
1680 int ret;
1681
1682 /*
1683 * Technically we could use raw_spin_[un]lock_irq() here, but this can
1684 * be called in early boot if the cmpxchg() fast path is disabled
1685 * (debug, no architecture support). In this case we will acquire the
1686 * rtmutex with lock->wait_lock held. But we cannot unconditionally
1687 * enable interrupts in that early boot case. So we need to use the
1688 * irqsave/restore variants.
1689 */
1690 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1691 ret = __rt_mutex_slowlock_locked(lock, ww_ctx, state);
1692 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1693
1694 return ret;
1695}
1696
1697static __always_inline int __rt_mutex_lock(struct rt_mutex_base *lock,
1698 unsigned int state)
1699{
1700 if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
1701 return 0;
1702
1703 return rt_mutex_slowlock(lock, NULL, state);
1704}
1705#endif /* RT_MUTEX_BUILD_MUTEX */
1706
1707#ifdef RT_MUTEX_BUILD_SPINLOCKS
1708/*
1709 * Functions required for spin/rw_lock substitution on RT kernels
1710 */
1711
1712/**
1713 * rtlock_slowlock_locked - Slow path lock acquisition for RT locks
1714 * @lock: The underlying RT mutex
1715 */
1716static void __sched rtlock_slowlock_locked(struct rt_mutex_base *lock)
1717{
1718 struct rt_mutex_waiter waiter;
1719 struct task_struct *owner;
1720
1721 lockdep_assert_held(&lock->wait_lock);
1722
1723 if (try_to_take_rt_mutex(lock, current, NULL))
1724 return;
1725
1726 rt_mutex_init_rtlock_waiter(&waiter);
1727
1728 /* Save current state and set state to TASK_RTLOCK_WAIT */
1729 current_save_and_set_rtlock_wait_state();
1730
1731 trace_contention_begin(lock, LCB_F_RT);
1732
1733 task_blocks_on_rt_mutex(lock, &waiter, current, NULL, RT_MUTEX_MIN_CHAINWALK);
1734
1735 for (;;) {
1736 /* Try to acquire the lock again */
1737 if (try_to_take_rt_mutex(lock, current, &waiter))
1738 break;
1739
1740 if (&waiter == rt_mutex_top_waiter(lock))
1741 owner = rt_mutex_owner(lock);
1742 else
1743 owner = NULL;
1744 raw_spin_unlock_irq(&lock->wait_lock);
1745
1746 if (!owner || !rtmutex_spin_on_owner(lock, &waiter, owner))
1747 schedule_rtlock();
1748
1749 raw_spin_lock_irq(&lock->wait_lock);
1750 set_current_state(TASK_RTLOCK_WAIT);
1751 }
1752
1753 /* Restore the task state */
1754 current_restore_rtlock_saved_state();
1755
1756 /*
1757 * try_to_take_rt_mutex() sets the waiter bit unconditionally.
1758 * We might have to fix that up:
1759 */
1760 fixup_rt_mutex_waiters(lock, true);
1761 debug_rt_mutex_free_waiter(&waiter);
1762
1763 trace_contention_end(lock, 0);
1764}
1765
1766static __always_inline void __sched rtlock_slowlock(struct rt_mutex_base *lock)
1767{
1768 unsigned long flags;
1769
1770 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1771 rtlock_slowlock_locked(lock);
1772 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1773}
1774
1775#endif /* RT_MUTEX_BUILD_SPINLOCKS */