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
__ww_mutex_add_waiter(struct rt_mutex_waiter * waiter,struct rt_mutex * lock,struct ww_acquire_ctx * ww_ctx)35 static 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
__ww_mutex_check_waiters(struct rt_mutex * lock,struct ww_acquire_ctx * ww_ctx)42 static inline void __ww_mutex_check_waiters(struct rt_mutex *lock,
43 struct ww_acquire_ctx *ww_ctx)
44 {
45 }
46
ww_mutex_lock_acquired(struct ww_mutex * lock,struct ww_acquire_ctx * ww_ctx)47 static inline void ww_mutex_lock_acquired(struct ww_mutex *lock,
48 struct ww_acquire_ctx *ww_ctx)
49 {
50 }
51
__ww_mutex_check_kill(struct rt_mutex * lock,struct rt_mutex_waiter * waiter,struct ww_acquire_ctx * ww_ctx)52 static 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
92 static __always_inline void
rt_mutex_set_owner(struct rt_mutex_base * lock,struct task_struct * owner)93 rt_mutex_set_owner(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 WRITE_ONCE(lock->owner, (struct task_struct *)val);
101 }
102
clear_rt_mutex_waiters(struct rt_mutex_base * lock)103 static __always_inline void clear_rt_mutex_waiters(struct rt_mutex_base *lock)
104 {
105 lock->owner = (struct task_struct *)
106 ((unsigned long)lock->owner & ~RT_MUTEX_HAS_WAITERS);
107 }
108
fixup_rt_mutex_waiters(struct rt_mutex_base * lock)109 static __always_inline void fixup_rt_mutex_waiters(struct rt_mutex_base *lock)
110 {
111 unsigned long owner, *p = (unsigned long *) &lock->owner;
112
113 if (rt_mutex_has_waiters(lock))
114 return;
115
116 /*
117 * The rbtree has no waiters enqueued, now make sure that the
118 * lock->owner still has the waiters bit set, otherwise the
119 * following can happen:
120 *
121 * CPU 0 CPU 1 CPU2
122 * l->owner=T1
123 * rt_mutex_lock(l)
124 * lock(l->lock)
125 * l->owner = T1 | HAS_WAITERS;
126 * enqueue(T2)
127 * boost()
128 * unlock(l->lock)
129 * block()
130 *
131 * rt_mutex_lock(l)
132 * lock(l->lock)
133 * l->owner = T1 | HAS_WAITERS;
134 * enqueue(T3)
135 * boost()
136 * unlock(l->lock)
137 * block()
138 * signal(->T2) signal(->T3)
139 * lock(l->lock)
140 * dequeue(T2)
141 * deboost()
142 * unlock(l->lock)
143 * lock(l->lock)
144 * dequeue(T3)
145 * ==> wait list is empty
146 * deboost()
147 * unlock(l->lock)
148 * lock(l->lock)
149 * fixup_rt_mutex_waiters()
150 * if (wait_list_empty(l) {
151 * l->owner = owner
152 * owner = l->owner & ~HAS_WAITERS;
153 * ==> l->owner = T1
154 * }
155 * lock(l->lock)
156 * rt_mutex_unlock(l) fixup_rt_mutex_waiters()
157 * if (wait_list_empty(l) {
158 * owner = l->owner & ~HAS_WAITERS;
159 * cmpxchg(l->owner, T1, NULL)
160 * ===> Success (l->owner = NULL)
161 *
162 * l->owner = owner
163 * ==> l->owner = T1
164 * }
165 *
166 * With the check for the waiter bit in place T3 on CPU2 will not
167 * overwrite. All tasks fiddling with the waiters bit are
168 * serialized by l->lock, so nothing else can modify the waiters
169 * bit. If the bit is set then nothing can change l->owner either
170 * so the simple RMW is safe. The cmpxchg() will simply fail if it
171 * happens in the middle of the RMW because the waiters bit is
172 * still set.
173 */
174 owner = READ_ONCE(*p);
175 if (owner & RT_MUTEX_HAS_WAITERS)
176 WRITE_ONCE(*p, owner & ~RT_MUTEX_HAS_WAITERS);
177 }
178
179 /*
180 * We can speed up the acquire/release, if there's no debugging state to be
181 * set up.
182 */
183 #ifndef CONFIG_DEBUG_RT_MUTEXES
rt_mutex_cmpxchg_acquire(struct rt_mutex_base * lock,struct task_struct * old,struct task_struct * new)184 static __always_inline bool rt_mutex_cmpxchg_acquire(struct rt_mutex_base *lock,
185 struct task_struct *old,
186 struct task_struct *new)
187 {
188 return try_cmpxchg_acquire(&lock->owner, &old, new);
189 }
190
rt_mutex_cmpxchg_release(struct rt_mutex_base * lock,struct task_struct * old,struct task_struct * new)191 static __always_inline bool rt_mutex_cmpxchg_release(struct rt_mutex_base *lock,
192 struct task_struct *old,
193 struct task_struct *new)
194 {
195 return try_cmpxchg_release(&lock->owner, &old, new);
196 }
197
198 /*
199 * Callers must hold the ->wait_lock -- which is the whole purpose as we force
200 * all future threads that attempt to [Rmw] the lock to the slowpath. As such
201 * relaxed semantics suffice.
202 */
mark_rt_mutex_waiters(struct rt_mutex_base * lock)203 static __always_inline void mark_rt_mutex_waiters(struct rt_mutex_base *lock)
204 {
205 unsigned long owner, *p = (unsigned long *) &lock->owner;
206
207 do {
208 owner = *p;
209 } while (cmpxchg_relaxed(p, owner,
210 owner | RT_MUTEX_HAS_WAITERS) != owner);
211 }
212
213 /*
214 * Safe fastpath aware unlock:
215 * 1) Clear the waiters bit
216 * 2) Drop lock->wait_lock
217 * 3) Try to unlock the lock with cmpxchg
218 */
unlock_rt_mutex_safe(struct rt_mutex_base * lock,unsigned long flags)219 static __always_inline bool unlock_rt_mutex_safe(struct rt_mutex_base *lock,
220 unsigned long flags)
221 __releases(lock->wait_lock)
222 {
223 struct task_struct *owner = rt_mutex_owner(lock);
224
225 clear_rt_mutex_waiters(lock);
226 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
227 /*
228 * If a new waiter comes in between the unlock and the cmpxchg
229 * we have two situations:
230 *
231 * unlock(wait_lock);
232 * lock(wait_lock);
233 * cmpxchg(p, owner, 0) == owner
234 * mark_rt_mutex_waiters(lock);
235 * acquire(lock);
236 * or:
237 *
238 * unlock(wait_lock);
239 * lock(wait_lock);
240 * mark_rt_mutex_waiters(lock);
241 *
242 * cmpxchg(p, owner, 0) != owner
243 * enqueue_waiter();
244 * unlock(wait_lock);
245 * lock(wait_lock);
246 * wake waiter();
247 * unlock(wait_lock);
248 * lock(wait_lock);
249 * acquire(lock);
250 */
251 return rt_mutex_cmpxchg_release(lock, owner, NULL);
252 }
253
254 #else
rt_mutex_cmpxchg_acquire(struct rt_mutex_base * lock,struct task_struct * old,struct task_struct * new)255 static __always_inline bool rt_mutex_cmpxchg_acquire(struct rt_mutex_base *lock,
256 struct task_struct *old,
257 struct task_struct *new)
258 {
259 return false;
260
261 }
262
rt_mutex_cmpxchg_release(struct rt_mutex_base * lock,struct task_struct * old,struct task_struct * new)263 static __always_inline bool rt_mutex_cmpxchg_release(struct rt_mutex_base *lock,
264 struct task_struct *old,
265 struct task_struct *new)
266 {
267 return false;
268 }
269
mark_rt_mutex_waiters(struct rt_mutex_base * lock)270 static __always_inline void mark_rt_mutex_waiters(struct rt_mutex_base *lock)
271 {
272 lock->owner = (struct task_struct *)
273 ((unsigned long)lock->owner | RT_MUTEX_HAS_WAITERS);
274 }
275
276 /*
277 * Simple slow path only version: lock->owner is protected by lock->wait_lock.
278 */
unlock_rt_mutex_safe(struct rt_mutex_base * lock,unsigned long flags)279 static __always_inline bool unlock_rt_mutex_safe(struct rt_mutex_base *lock,
280 unsigned long flags)
281 __releases(lock->wait_lock)
282 {
283 lock->owner = NULL;
284 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
285 return true;
286 }
287 #endif
288
__waiter_prio(struct task_struct * task)289 static __always_inline int __waiter_prio(struct task_struct *task)
290 {
291 int prio = task->prio;
292
293 if (!rt_prio(prio))
294 return DEFAULT_PRIO;
295
296 return prio;
297 }
298
299 static __always_inline void
waiter_update_prio(struct rt_mutex_waiter * waiter,struct task_struct * task)300 waiter_update_prio(struct rt_mutex_waiter *waiter, struct task_struct *task)
301 {
302 waiter->prio = __waiter_prio(task);
303 waiter->deadline = task->dl.deadline;
304 }
305
306 /*
307 * Only use with rt_mutex_waiter_{less,equal}()
308 */
309 #define task_to_waiter(p) \
310 &(struct rt_mutex_waiter){ .prio = __waiter_prio(p), .deadline = (p)->dl.deadline }
311
rt_mutex_waiter_less(struct rt_mutex_waiter * left,struct rt_mutex_waiter * right)312 static __always_inline int rt_mutex_waiter_less(struct rt_mutex_waiter *left,
313 struct rt_mutex_waiter *right)
314 {
315 if (left->prio < right->prio)
316 return 1;
317
318 /*
319 * If both waiters have dl_prio(), we check the deadlines of the
320 * associated tasks.
321 * If left waiter has a dl_prio(), and we didn't return 1 above,
322 * then right waiter has a dl_prio() too.
323 */
324 if (dl_prio(left->prio))
325 return dl_time_before(left->deadline, right->deadline);
326
327 return 0;
328 }
329
rt_mutex_waiter_equal(struct rt_mutex_waiter * left,struct rt_mutex_waiter * right)330 static __always_inline int rt_mutex_waiter_equal(struct rt_mutex_waiter *left,
331 struct rt_mutex_waiter *right)
332 {
333 if (left->prio != right->prio)
334 return 0;
335
336 /*
337 * If both waiters have dl_prio(), we check the deadlines of the
338 * associated tasks.
339 * If left waiter has a dl_prio(), and we didn't return 0 above,
340 * then right waiter has a dl_prio() too.
341 */
342 if (dl_prio(left->prio))
343 return left->deadline == right->deadline;
344
345 return 1;
346 }
347
rt_mutex_steal(struct rt_mutex_waiter * waiter,struct rt_mutex_waiter * top_waiter)348 static inline bool rt_mutex_steal(struct rt_mutex_waiter *waiter,
349 struct rt_mutex_waiter *top_waiter)
350 {
351 if (rt_mutex_waiter_less(waiter, top_waiter))
352 return true;
353
354 #ifdef RT_MUTEX_BUILD_SPINLOCKS
355 /*
356 * Note that RT tasks are excluded from same priority (lateral)
357 * steals to prevent the introduction of an unbounded latency.
358 */
359 if (rt_prio(waiter->prio) || dl_prio(waiter->prio))
360 return false;
361
362 return rt_mutex_waiter_equal(waiter, top_waiter);
363 #else
364 return false;
365 #endif
366 }
367
368 #define __node_2_waiter(node) \
369 rb_entry((node), struct rt_mutex_waiter, tree_entry)
370
__waiter_less(struct rb_node * a,const struct rb_node * b)371 static __always_inline bool __waiter_less(struct rb_node *a, const struct rb_node *b)
372 {
373 struct rt_mutex_waiter *aw = __node_2_waiter(a);
374 struct rt_mutex_waiter *bw = __node_2_waiter(b);
375
376 if (rt_mutex_waiter_less(aw, bw))
377 return 1;
378
379 if (!build_ww_mutex())
380 return 0;
381
382 if (rt_mutex_waiter_less(bw, aw))
383 return 0;
384
385 /* NOTE: relies on waiter->ww_ctx being set before insertion */
386 if (aw->ww_ctx) {
387 if (!bw->ww_ctx)
388 return 1;
389
390 return (signed long)(aw->ww_ctx->stamp -
391 bw->ww_ctx->stamp) < 0;
392 }
393
394 return 0;
395 }
396
397 static __always_inline void
rt_mutex_enqueue(struct rt_mutex_base * lock,struct rt_mutex_waiter * waiter)398 rt_mutex_enqueue(struct rt_mutex_base *lock, struct rt_mutex_waiter *waiter)
399 {
400 rb_add_cached(&waiter->tree_entry, &lock->waiters, __waiter_less);
401 }
402
403 static __always_inline void
rt_mutex_dequeue(struct rt_mutex_base * lock,struct rt_mutex_waiter * waiter)404 rt_mutex_dequeue(struct rt_mutex_base *lock, struct rt_mutex_waiter *waiter)
405 {
406 if (RB_EMPTY_NODE(&waiter->tree_entry))
407 return;
408
409 rb_erase_cached(&waiter->tree_entry, &lock->waiters);
410 RB_CLEAR_NODE(&waiter->tree_entry);
411 }
412
413 #define __node_2_pi_waiter(node) \
414 rb_entry((node), struct rt_mutex_waiter, pi_tree_entry)
415
416 static __always_inline bool
__pi_waiter_less(struct rb_node * a,const struct rb_node * b)417 __pi_waiter_less(struct rb_node *a, const struct rb_node *b)
418 {
419 return rt_mutex_waiter_less(__node_2_pi_waiter(a), __node_2_pi_waiter(b));
420 }
421
422 static __always_inline void
rt_mutex_enqueue_pi(struct task_struct * task,struct rt_mutex_waiter * waiter)423 rt_mutex_enqueue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
424 {
425 rb_add_cached(&waiter->pi_tree_entry, &task->pi_waiters, __pi_waiter_less);
426 }
427
428 static __always_inline void
rt_mutex_dequeue_pi(struct task_struct * task,struct rt_mutex_waiter * waiter)429 rt_mutex_dequeue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
430 {
431 if (RB_EMPTY_NODE(&waiter->pi_tree_entry))
432 return;
433
434 rb_erase_cached(&waiter->pi_tree_entry, &task->pi_waiters);
435 RB_CLEAR_NODE(&waiter->pi_tree_entry);
436 }
437
rt_mutex_adjust_prio(struct task_struct * p)438 static __always_inline void rt_mutex_adjust_prio(struct task_struct *p)
439 {
440 struct task_struct *pi_task = NULL;
441
442 lockdep_assert_held(&p->pi_lock);
443
444 if (task_has_pi_waiters(p))
445 pi_task = task_top_pi_waiter(p)->task;
446
447 rt_mutex_setprio(p, pi_task);
448 }
449
450 /* RT mutex specific wake_q wrappers */
rt_mutex_wake_q_add_task(struct rt_wake_q_head * wqh,struct task_struct * task,unsigned int wake_state)451 static __always_inline void rt_mutex_wake_q_add_task(struct rt_wake_q_head *wqh,
452 struct task_struct *task,
453 unsigned int wake_state)
454 {
455 if (IS_ENABLED(CONFIG_PREEMPT_RT) && wake_state == TASK_RTLOCK_WAIT) {
456 if (IS_ENABLED(CONFIG_PROVE_LOCKING))
457 WARN_ON_ONCE(wqh->rtlock_task);
458 get_task_struct(task);
459 wqh->rtlock_task = task;
460 } else {
461 wake_q_add(&wqh->head, task);
462 }
463 }
464
rt_mutex_wake_q_add(struct rt_wake_q_head * wqh,struct rt_mutex_waiter * w)465 static __always_inline void rt_mutex_wake_q_add(struct rt_wake_q_head *wqh,
466 struct rt_mutex_waiter *w)
467 {
468 rt_mutex_wake_q_add_task(wqh, w->task, w->wake_state);
469 }
470
rt_mutex_wake_up_q(struct rt_wake_q_head * wqh)471 static __always_inline void rt_mutex_wake_up_q(struct rt_wake_q_head *wqh)
472 {
473 if (IS_ENABLED(CONFIG_PREEMPT_RT) && wqh->rtlock_task) {
474 wake_up_state(wqh->rtlock_task, TASK_RTLOCK_WAIT);
475 put_task_struct(wqh->rtlock_task);
476 wqh->rtlock_task = NULL;
477 }
478
479 if (!wake_q_empty(&wqh->head))
480 wake_up_q(&wqh->head);
481
482 /* Pairs with preempt_disable() in mark_wakeup_next_waiter() */
483 preempt_enable();
484 }
485
486 /*
487 * Deadlock detection is conditional:
488 *
489 * If CONFIG_DEBUG_RT_MUTEXES=n, deadlock detection is only conducted
490 * if the detect argument is == RT_MUTEX_FULL_CHAINWALK.
491 *
492 * If CONFIG_DEBUG_RT_MUTEXES=y, deadlock detection is always
493 * conducted independent of the detect argument.
494 *
495 * If the waiter argument is NULL this indicates the deboost path and
496 * deadlock detection is disabled independent of the detect argument
497 * and the config settings.
498 */
499 static __always_inline bool
rt_mutex_cond_detect_deadlock(struct rt_mutex_waiter * waiter,enum rtmutex_chainwalk chwalk)500 rt_mutex_cond_detect_deadlock(struct rt_mutex_waiter *waiter,
501 enum rtmutex_chainwalk chwalk)
502 {
503 if (IS_ENABLED(CONFIG_DEBUG_RT_MUTEXES))
504 return waiter != NULL;
505 return chwalk == RT_MUTEX_FULL_CHAINWALK;
506 }
507
task_blocked_on_lock(struct task_struct * p)508 static __always_inline struct rt_mutex_base *task_blocked_on_lock(struct task_struct *p)
509 {
510 return p->pi_blocked_on ? p->pi_blocked_on->lock : NULL;
511 }
512
513 /*
514 * Adjust the priority chain. Also used for deadlock detection.
515 * Decreases task's usage by one - may thus free the task.
516 *
517 * @task: the task owning the mutex (owner) for which a chain walk is
518 * probably needed
519 * @chwalk: do we have to carry out deadlock detection?
520 * @orig_lock: the mutex (can be NULL if we are walking the chain to recheck
521 * things for a task that has just got its priority adjusted, and
522 * is waiting on a mutex)
523 * @next_lock: the mutex on which the owner of @orig_lock was blocked before
524 * we dropped its pi_lock. Is never dereferenced, only used for
525 * comparison to detect lock chain changes.
526 * @orig_waiter: rt_mutex_waiter struct for the task that has just donated
527 * its priority to the mutex owner (can be NULL in the case
528 * depicted above or if the top waiter is gone away and we are
529 * actually deboosting the owner)
530 * @top_task: the current top waiter
531 *
532 * Returns 0 or -EDEADLK.
533 *
534 * Chain walk basics and protection scope
535 *
536 * [R] refcount on task
537 * [P] task->pi_lock held
538 * [L] rtmutex->wait_lock held
539 *
540 * Step Description Protected by
541 * function arguments:
542 * @task [R]
543 * @orig_lock if != NULL @top_task is blocked on it
544 * @next_lock Unprotected. Cannot be
545 * dereferenced. Only used for
546 * comparison.
547 * @orig_waiter if != NULL @top_task is blocked on it
548 * @top_task current, or in case of proxy
549 * locking protected by calling
550 * code
551 * again:
552 * loop_sanity_check();
553 * retry:
554 * [1] lock(task->pi_lock); [R] acquire [P]
555 * [2] waiter = task->pi_blocked_on; [P]
556 * [3] check_exit_conditions_1(); [P]
557 * [4] lock = waiter->lock; [P]
558 * [5] if (!try_lock(lock->wait_lock)) { [P] try to acquire [L]
559 * unlock(task->pi_lock); release [P]
560 * goto retry;
561 * }
562 * [6] check_exit_conditions_2(); [P] + [L]
563 * [7] requeue_lock_waiter(lock, waiter); [P] + [L]
564 * [8] unlock(task->pi_lock); release [P]
565 * put_task_struct(task); release [R]
566 * [9] check_exit_conditions_3(); [L]
567 * [10] task = owner(lock); [L]
568 * get_task_struct(task); [L] acquire [R]
569 * lock(task->pi_lock); [L] acquire [P]
570 * [11] requeue_pi_waiter(tsk, waiters(lock));[P] + [L]
571 * [12] check_exit_conditions_4(); [P] + [L]
572 * [13] unlock(task->pi_lock); release [P]
573 * unlock(lock->wait_lock); release [L]
574 * goto again;
575 */
rt_mutex_adjust_prio_chain(struct task_struct * task,enum rtmutex_chainwalk chwalk,struct rt_mutex_base * orig_lock,struct rt_mutex_base * next_lock,struct rt_mutex_waiter * orig_waiter,struct task_struct * top_task)576 static int __sched rt_mutex_adjust_prio_chain(struct task_struct *task,
577 enum rtmutex_chainwalk chwalk,
578 struct rt_mutex_base *orig_lock,
579 struct rt_mutex_base *next_lock,
580 struct rt_mutex_waiter *orig_waiter,
581 struct task_struct *top_task)
582 {
583 struct rt_mutex_waiter *waiter, *top_waiter = orig_waiter;
584 struct rt_mutex_waiter *prerequeue_top_waiter;
585 int ret = 0, depth = 0;
586 struct rt_mutex_base *lock;
587 bool detect_deadlock;
588 bool requeue = true;
589
590 detect_deadlock = rt_mutex_cond_detect_deadlock(orig_waiter, chwalk);
591
592 /*
593 * The (de)boosting is a step by step approach with a lot of
594 * pitfalls. We want this to be preemptible and we want hold a
595 * maximum of two locks per step. So we have to check
596 * carefully whether things change under us.
597 */
598 again:
599 /*
600 * We limit the lock chain length for each invocation.
601 */
602 if (++depth > max_lock_depth) {
603 static int prev_max;
604
605 /*
606 * Print this only once. If the admin changes the limit,
607 * print a new message when reaching the limit again.
608 */
609 if (prev_max != max_lock_depth) {
610 prev_max = max_lock_depth;
611 printk(KERN_WARNING "Maximum lock depth %d reached "
612 "task: %s (%d)\n", max_lock_depth,
613 top_task->comm, task_pid_nr(top_task));
614 }
615 put_task_struct(task);
616
617 return -EDEADLK;
618 }
619
620 /*
621 * We are fully preemptible here and only hold the refcount on
622 * @task. So everything can have changed under us since the
623 * caller or our own code below (goto retry/again) dropped all
624 * locks.
625 */
626 retry:
627 /*
628 * [1] Task cannot go away as we did a get_task() before !
629 */
630 raw_spin_lock_irq(&task->pi_lock);
631
632 /*
633 * [2] Get the waiter on which @task is blocked on.
634 */
635 waiter = task->pi_blocked_on;
636
637 /*
638 * [3] check_exit_conditions_1() protected by task->pi_lock.
639 */
640
641 /*
642 * Check whether the end of the boosting chain has been
643 * reached or the state of the chain has changed while we
644 * dropped the locks.
645 */
646 if (!waiter)
647 goto out_unlock_pi;
648
649 /*
650 * Check the orig_waiter state. After we dropped the locks,
651 * the previous owner of the lock might have released the lock.
652 */
653 if (orig_waiter && !rt_mutex_owner(orig_lock))
654 goto out_unlock_pi;
655
656 /*
657 * We dropped all locks after taking a refcount on @task, so
658 * the task might have moved on in the lock chain or even left
659 * the chain completely and blocks now on an unrelated lock or
660 * on @orig_lock.
661 *
662 * We stored the lock on which @task was blocked in @next_lock,
663 * so we can detect the chain change.
664 */
665 if (next_lock != waiter->lock)
666 goto out_unlock_pi;
667
668 /*
669 * There could be 'spurious' loops in the lock graph due to ww_mutex,
670 * consider:
671 *
672 * P1: A, ww_A, ww_B
673 * P2: ww_B, ww_A
674 * P3: A
675 *
676 * P3 should not return -EDEADLK because it gets trapped in the cycle
677 * created by P1 and P2 (which will resolve -- and runs into
678 * max_lock_depth above). Therefore disable detect_deadlock such that
679 * the below termination condition can trigger once all relevant tasks
680 * are boosted.
681 *
682 * Even when we start with ww_mutex we can disable deadlock detection,
683 * since we would supress a ww_mutex induced deadlock at [6] anyway.
684 * Supressing it here however is not sufficient since we might still
685 * hit [6] due to adjustment driven iteration.
686 *
687 * NOTE: if someone were to create a deadlock between 2 ww_classes we'd
688 * utterly fail to report it; lockdep should.
689 */
690 if (IS_ENABLED(CONFIG_PREEMPT_RT) && waiter->ww_ctx && detect_deadlock)
691 detect_deadlock = false;
692
693 /*
694 * Drop out, when the task has no waiters. Note,
695 * top_waiter can be NULL, when we are in the deboosting
696 * mode!
697 */
698 if (top_waiter) {
699 if (!task_has_pi_waiters(task))
700 goto out_unlock_pi;
701 /*
702 * If deadlock detection is off, we stop here if we
703 * are not the top pi waiter of the task. If deadlock
704 * detection is enabled we continue, but stop the
705 * requeueing in the chain walk.
706 */
707 if (top_waiter != task_top_pi_waiter(task)) {
708 if (!detect_deadlock)
709 goto out_unlock_pi;
710 else
711 requeue = false;
712 }
713 }
714
715 /*
716 * If the waiter priority is the same as the task priority
717 * then there is no further priority adjustment necessary. If
718 * deadlock detection is off, we stop the chain walk. If its
719 * enabled we continue, but stop the requeueing in the chain
720 * walk.
721 */
722 if (rt_mutex_waiter_equal(waiter, task_to_waiter(task))) {
723 if (!detect_deadlock)
724 goto out_unlock_pi;
725 else
726 requeue = false;
727 }
728
729 /*
730 * [4] Get the next lock
731 */
732 lock = waiter->lock;
733 /*
734 * [5] We need to trylock here as we are holding task->pi_lock,
735 * which is the reverse lock order versus the other rtmutex
736 * operations.
737 */
738 if (!raw_spin_trylock(&lock->wait_lock)) {
739 raw_spin_unlock_irq(&task->pi_lock);
740 cpu_relax();
741 goto retry;
742 }
743
744 /*
745 * [6] check_exit_conditions_2() protected by task->pi_lock and
746 * lock->wait_lock.
747 *
748 * Deadlock detection. If the lock is the same as the original
749 * lock which caused us to walk the lock chain or if the
750 * current lock is owned by the task which initiated the chain
751 * walk, we detected a deadlock.
752 */
753 if (lock == orig_lock || rt_mutex_owner(lock) == top_task) {
754 ret = -EDEADLK;
755
756 /*
757 * When the deadlock is due to ww_mutex; also see above. Don't
758 * report the deadlock and instead let the ww_mutex wound/die
759 * logic pick which of the contending threads gets -EDEADLK.
760 *
761 * NOTE: assumes the cycle only contains a single ww_class; any
762 * other configuration and we fail to report; also, see
763 * lockdep.
764 */
765 if (IS_ENABLED(CONFIG_PREEMPT_RT) && orig_waiter && orig_waiter->ww_ctx)
766 ret = 0;
767
768 raw_spin_unlock(&lock->wait_lock);
769 goto out_unlock_pi;
770 }
771
772 /*
773 * If we just follow the lock chain for deadlock detection, no
774 * need to do all the requeue operations. To avoid a truckload
775 * of conditionals around the various places below, just do the
776 * minimum chain walk checks.
777 */
778 if (!requeue) {
779 /*
780 * No requeue[7] here. Just release @task [8]
781 */
782 raw_spin_unlock(&task->pi_lock);
783 put_task_struct(task);
784
785 /*
786 * [9] check_exit_conditions_3 protected by lock->wait_lock.
787 * If there is no owner of the lock, end of chain.
788 */
789 if (!rt_mutex_owner(lock)) {
790 raw_spin_unlock_irq(&lock->wait_lock);
791 return 0;
792 }
793
794 /* [10] Grab the next task, i.e. owner of @lock */
795 task = get_task_struct(rt_mutex_owner(lock));
796 raw_spin_lock(&task->pi_lock);
797
798 /*
799 * No requeue [11] here. We just do deadlock detection.
800 *
801 * [12] Store whether owner is blocked
802 * itself. Decision is made after dropping the locks
803 */
804 next_lock = task_blocked_on_lock(task);
805 /*
806 * Get the top waiter for the next iteration
807 */
808 top_waiter = rt_mutex_top_waiter(lock);
809
810 /* [13] Drop locks */
811 raw_spin_unlock(&task->pi_lock);
812 raw_spin_unlock_irq(&lock->wait_lock);
813
814 /* If owner is not blocked, end of chain. */
815 if (!next_lock)
816 goto out_put_task;
817 goto again;
818 }
819
820 /*
821 * Store the current top waiter before doing the requeue
822 * operation on @lock. We need it for the boost/deboost
823 * decision below.
824 */
825 prerequeue_top_waiter = rt_mutex_top_waiter(lock);
826
827 /* [7] Requeue the waiter in the lock waiter tree. */
828 rt_mutex_dequeue(lock, waiter);
829
830 /*
831 * Update the waiter prio fields now that we're dequeued.
832 *
833 * These values can have changed through either:
834 *
835 * sys_sched_set_scheduler() / sys_sched_setattr()
836 *
837 * or
838 *
839 * DL CBS enforcement advancing the effective deadline.
840 *
841 * Even though pi_waiters also uses these fields, and that tree is only
842 * updated in [11], we can do this here, since we hold [L], which
843 * serializes all pi_waiters access and rb_erase() does not care about
844 * the values of the node being removed.
845 */
846 waiter_update_prio(waiter, task);
847
848 rt_mutex_enqueue(lock, waiter);
849
850 /* [8] Release the task */
851 raw_spin_unlock(&task->pi_lock);
852 put_task_struct(task);
853
854 /*
855 * [9] check_exit_conditions_3 protected by lock->wait_lock.
856 *
857 * We must abort the chain walk if there is no lock owner even
858 * in the dead lock detection case, as we have nothing to
859 * follow here. This is the end of the chain we are walking.
860 */
861 if (!rt_mutex_owner(lock)) {
862 /*
863 * If the requeue [7] above changed the top waiter,
864 * then we need to wake the new top waiter up to try
865 * to get the lock.
866 */
867 if (prerequeue_top_waiter != rt_mutex_top_waiter(lock))
868 wake_up_state(waiter->task, waiter->wake_state);
869 raw_spin_unlock_irq(&lock->wait_lock);
870 return 0;
871 }
872
873 /* [10] Grab the next task, i.e. the owner of @lock */
874 task = get_task_struct(rt_mutex_owner(lock));
875 raw_spin_lock(&task->pi_lock);
876
877 /* [11] requeue the pi waiters if necessary */
878 if (waiter == rt_mutex_top_waiter(lock)) {
879 /*
880 * The waiter became the new top (highest priority)
881 * waiter on the lock. Replace the previous top waiter
882 * in the owner tasks pi waiters tree with this waiter
883 * and adjust the priority of the owner.
884 */
885 rt_mutex_dequeue_pi(task, prerequeue_top_waiter);
886 rt_mutex_enqueue_pi(task, waiter);
887 rt_mutex_adjust_prio(task);
888
889 } else if (prerequeue_top_waiter == waiter) {
890 /*
891 * The waiter was the top waiter on the lock, but is
892 * no longer the top priority waiter. Replace waiter in
893 * the owner tasks pi waiters tree with the new top
894 * (highest priority) waiter and adjust the priority
895 * of the owner.
896 * The new top waiter is stored in @waiter so that
897 * @waiter == @top_waiter evaluates to true below and
898 * we continue to deboost the rest of the chain.
899 */
900 rt_mutex_dequeue_pi(task, waiter);
901 waiter = rt_mutex_top_waiter(lock);
902 rt_mutex_enqueue_pi(task, waiter);
903 rt_mutex_adjust_prio(task);
904 } else {
905 /*
906 * Nothing changed. No need to do any priority
907 * adjustment.
908 */
909 }
910
911 /*
912 * [12] check_exit_conditions_4() protected by task->pi_lock
913 * and lock->wait_lock. The actual decisions are made after we
914 * dropped the locks.
915 *
916 * Check whether the task which owns the current lock is pi
917 * blocked itself. If yes we store a pointer to the lock for
918 * the lock chain change detection above. After we dropped
919 * task->pi_lock next_lock cannot be dereferenced anymore.
920 */
921 next_lock = task_blocked_on_lock(task);
922 /*
923 * Store the top waiter of @lock for the end of chain walk
924 * decision below.
925 */
926 top_waiter = rt_mutex_top_waiter(lock);
927
928 /* [13] Drop the locks */
929 raw_spin_unlock(&task->pi_lock);
930 raw_spin_unlock_irq(&lock->wait_lock);
931
932 /*
933 * Make the actual exit decisions [12], based on the stored
934 * values.
935 *
936 * We reached the end of the lock chain. Stop right here. No
937 * point to go back just to figure that out.
938 */
939 if (!next_lock)
940 goto out_put_task;
941
942 /*
943 * If the current waiter is not the top waiter on the lock,
944 * then we can stop the chain walk here if we are not in full
945 * deadlock detection mode.
946 */
947 if (!detect_deadlock && waiter != top_waiter)
948 goto out_put_task;
949
950 goto again;
951
952 out_unlock_pi:
953 raw_spin_unlock_irq(&task->pi_lock);
954 out_put_task:
955 put_task_struct(task);
956
957 return ret;
958 }
959
960 /*
961 * Try to take an rt-mutex
962 *
963 * Must be called with lock->wait_lock held and interrupts disabled
964 *
965 * @lock: The lock to be acquired.
966 * @task: The task which wants to acquire the lock
967 * @waiter: The waiter that is queued to the lock's wait tree if the
968 * callsite called task_blocked_on_lock(), otherwise NULL
969 */
970 static int __sched
try_to_take_rt_mutex(struct rt_mutex_base * lock,struct task_struct * task,struct rt_mutex_waiter * waiter)971 try_to_take_rt_mutex(struct rt_mutex_base *lock, struct task_struct *task,
972 struct rt_mutex_waiter *waiter)
973 {
974 lockdep_assert_held(&lock->wait_lock);
975
976 /*
977 * Before testing whether we can acquire @lock, we set the
978 * RT_MUTEX_HAS_WAITERS bit in @lock->owner. This forces all
979 * other tasks which try to modify @lock into the slow path
980 * and they serialize on @lock->wait_lock.
981 *
982 * The RT_MUTEX_HAS_WAITERS bit can have a transitional state
983 * as explained at the top of this file if and only if:
984 *
985 * - There is a lock owner. The caller must fixup the
986 * transient state if it does a trylock or leaves the lock
987 * function due to a signal or timeout.
988 *
989 * - @task acquires the lock and there are no other
990 * waiters. This is undone in rt_mutex_set_owner(@task) at
991 * the end of this function.
992 */
993 mark_rt_mutex_waiters(lock);
994
995 /*
996 * If @lock has an owner, give up.
997 */
998 if (rt_mutex_owner(lock))
999 return 0;
1000
1001 /*
1002 * If @waiter != NULL, @task has already enqueued the waiter
1003 * into @lock waiter tree. If @waiter == NULL then this is a
1004 * trylock attempt.
1005 */
1006 if (waiter) {
1007 struct rt_mutex_waiter *top_waiter = rt_mutex_top_waiter(lock);
1008
1009 /*
1010 * If waiter is the highest priority waiter of @lock,
1011 * or allowed to steal it, take it over.
1012 */
1013 if (waiter == top_waiter || rt_mutex_steal(waiter, top_waiter)) {
1014 /*
1015 * We can acquire the lock. Remove the waiter from the
1016 * lock waiters tree.
1017 */
1018 rt_mutex_dequeue(lock, waiter);
1019 } else {
1020 return 0;
1021 }
1022 } else {
1023 /*
1024 * If the lock has waiters already we check whether @task is
1025 * eligible to take over the lock.
1026 *
1027 * If there are no other waiters, @task can acquire
1028 * the lock. @task->pi_blocked_on is NULL, so it does
1029 * not need to be dequeued.
1030 */
1031 if (rt_mutex_has_waiters(lock)) {
1032 /* Check whether the trylock can steal it. */
1033 if (!rt_mutex_steal(task_to_waiter(task),
1034 rt_mutex_top_waiter(lock)))
1035 return 0;
1036
1037 /*
1038 * The current top waiter stays enqueued. We
1039 * don't have to change anything in the lock
1040 * waiters order.
1041 */
1042 } else {
1043 /*
1044 * No waiters. Take the lock without the
1045 * pi_lock dance.@task->pi_blocked_on is NULL
1046 * and we have no waiters to enqueue in @task
1047 * pi waiters tree.
1048 */
1049 goto takeit;
1050 }
1051 }
1052
1053 /*
1054 * Clear @task->pi_blocked_on. Requires protection by
1055 * @task->pi_lock. Redundant operation for the @waiter == NULL
1056 * case, but conditionals are more expensive than a redundant
1057 * store.
1058 */
1059 raw_spin_lock(&task->pi_lock);
1060 task->pi_blocked_on = NULL;
1061 /*
1062 * Finish the lock acquisition. @task is the new owner. If
1063 * other waiters exist we have to insert the highest priority
1064 * waiter into @task->pi_waiters tree.
1065 */
1066 if (rt_mutex_has_waiters(lock))
1067 rt_mutex_enqueue_pi(task, rt_mutex_top_waiter(lock));
1068 raw_spin_unlock(&task->pi_lock);
1069
1070 takeit:
1071 /*
1072 * This either preserves the RT_MUTEX_HAS_WAITERS bit if there
1073 * are still waiters or clears it.
1074 */
1075 rt_mutex_set_owner(lock, task);
1076
1077 return 1;
1078 }
1079
1080 /*
1081 * Task blocks on lock.
1082 *
1083 * Prepare waiter and propagate pi chain
1084 *
1085 * This must be called with lock->wait_lock held and interrupts disabled
1086 */
task_blocks_on_rt_mutex(struct rt_mutex_base * lock,struct rt_mutex_waiter * waiter,struct task_struct * task,struct ww_acquire_ctx * ww_ctx,enum rtmutex_chainwalk chwalk)1087 static int __sched task_blocks_on_rt_mutex(struct rt_mutex_base *lock,
1088 struct rt_mutex_waiter *waiter,
1089 struct task_struct *task,
1090 struct ww_acquire_ctx *ww_ctx,
1091 enum rtmutex_chainwalk chwalk)
1092 {
1093 struct task_struct *owner = rt_mutex_owner(lock);
1094 struct rt_mutex_waiter *top_waiter = waiter;
1095 struct rt_mutex_base *next_lock;
1096 int chain_walk = 0, res;
1097
1098 lockdep_assert_held(&lock->wait_lock);
1099
1100 /*
1101 * Early deadlock detection. We really don't want the task to
1102 * enqueue on itself just to untangle the mess later. It's not
1103 * only an optimization. We drop the locks, so another waiter
1104 * can come in before the chain walk detects the deadlock. So
1105 * the other will detect the deadlock and return -EDEADLOCK,
1106 * which is wrong, as the other waiter is not in a deadlock
1107 * situation.
1108 *
1109 * Except for ww_mutex, in that case the chain walk must already deal
1110 * with spurious cycles, see the comments at [3] and [6].
1111 */
1112 if (owner == task && !(build_ww_mutex() && ww_ctx))
1113 return -EDEADLK;
1114
1115 raw_spin_lock(&task->pi_lock);
1116 waiter->task = task;
1117 waiter->lock = lock;
1118 waiter_update_prio(waiter, task);
1119
1120 /* Get the top priority waiter on the lock */
1121 if (rt_mutex_has_waiters(lock))
1122 top_waiter = rt_mutex_top_waiter(lock);
1123 rt_mutex_enqueue(lock, waiter);
1124
1125 task->pi_blocked_on = waiter;
1126
1127 raw_spin_unlock(&task->pi_lock);
1128
1129 if (build_ww_mutex() && ww_ctx) {
1130 struct rt_mutex *rtm;
1131
1132 /* Check whether the waiter should back out immediately */
1133 rtm = container_of(lock, struct rt_mutex, rtmutex);
1134 res = __ww_mutex_add_waiter(waiter, rtm, ww_ctx);
1135 if (res) {
1136 raw_spin_lock(&task->pi_lock);
1137 rt_mutex_dequeue(lock, waiter);
1138 task->pi_blocked_on = NULL;
1139 raw_spin_unlock(&task->pi_lock);
1140 return res;
1141 }
1142 }
1143
1144 if (!owner)
1145 return 0;
1146
1147 raw_spin_lock(&owner->pi_lock);
1148 if (waiter == rt_mutex_top_waiter(lock)) {
1149 rt_mutex_dequeue_pi(owner, top_waiter);
1150 rt_mutex_enqueue_pi(owner, waiter);
1151
1152 rt_mutex_adjust_prio(owner);
1153 if (owner->pi_blocked_on)
1154 chain_walk = 1;
1155 } else if (rt_mutex_cond_detect_deadlock(waiter, chwalk)) {
1156 chain_walk = 1;
1157 }
1158
1159 /* Store the lock on which owner is blocked or NULL */
1160 next_lock = task_blocked_on_lock(owner);
1161
1162 raw_spin_unlock(&owner->pi_lock);
1163 /*
1164 * Even if full deadlock detection is on, if the owner is not
1165 * blocked itself, we can avoid finding this out in the chain
1166 * walk.
1167 */
1168 if (!chain_walk || !next_lock)
1169 return 0;
1170
1171 /*
1172 * The owner can't disappear while holding a lock,
1173 * so the owner struct is protected by wait_lock.
1174 * Gets dropped in rt_mutex_adjust_prio_chain()!
1175 */
1176 get_task_struct(owner);
1177
1178 raw_spin_unlock_irq(&lock->wait_lock);
1179
1180 res = rt_mutex_adjust_prio_chain(owner, chwalk, lock,
1181 next_lock, waiter, task);
1182
1183 raw_spin_lock_irq(&lock->wait_lock);
1184
1185 return res;
1186 }
1187
1188 /*
1189 * Remove the top waiter from the current tasks pi waiter tree and
1190 * queue it up.
1191 *
1192 * Called with lock->wait_lock held and interrupts disabled.
1193 */
mark_wakeup_next_waiter(struct rt_wake_q_head * wqh,struct rt_mutex_base * lock)1194 static void __sched mark_wakeup_next_waiter(struct rt_wake_q_head *wqh,
1195 struct rt_mutex_base *lock)
1196 {
1197 struct rt_mutex_waiter *waiter;
1198
1199 raw_spin_lock(¤t->pi_lock);
1200
1201 waiter = rt_mutex_top_waiter(lock);
1202
1203 /*
1204 * Remove it from current->pi_waiters and deboost.
1205 *
1206 * We must in fact deboost here in order to ensure we call
1207 * rt_mutex_setprio() to update p->pi_top_task before the
1208 * task unblocks.
1209 */
1210 rt_mutex_dequeue_pi(current, waiter);
1211 rt_mutex_adjust_prio(current);
1212
1213 /*
1214 * As we are waking up the top waiter, and the waiter stays
1215 * queued on the lock until it gets the lock, this lock
1216 * obviously has waiters. Just set the bit here and this has
1217 * the added benefit of forcing all new tasks into the
1218 * slow path making sure no task of lower priority than
1219 * the top waiter can steal this lock.
1220 */
1221 lock->owner = (void *) RT_MUTEX_HAS_WAITERS;
1222
1223 /*
1224 * We deboosted before waking the top waiter task such that we don't
1225 * run two tasks with the 'same' priority (and ensure the
1226 * p->pi_top_task pointer points to a blocked task). This however can
1227 * lead to priority inversion if we would get preempted after the
1228 * deboost but before waking our donor task, hence the preempt_disable()
1229 * before unlock.
1230 *
1231 * Pairs with preempt_enable() in rt_mutex_wake_up_q();
1232 */
1233 preempt_disable();
1234 rt_mutex_wake_q_add(wqh, waiter);
1235 raw_spin_unlock(¤t->pi_lock);
1236 }
1237
__rt_mutex_slowtrylock(struct rt_mutex_base * lock)1238 static int __sched __rt_mutex_slowtrylock(struct rt_mutex_base *lock)
1239 {
1240 int ret = try_to_take_rt_mutex(lock, current, NULL);
1241
1242 /*
1243 * try_to_take_rt_mutex() sets the lock waiters bit
1244 * unconditionally. Clean this up.
1245 */
1246 fixup_rt_mutex_waiters(lock);
1247
1248 return ret;
1249 }
1250
1251 /*
1252 * Slow path try-lock function:
1253 */
rt_mutex_slowtrylock(struct rt_mutex_base * lock)1254 static int __sched rt_mutex_slowtrylock(struct rt_mutex_base *lock)
1255 {
1256 unsigned long flags;
1257 int ret;
1258
1259 /*
1260 * If the lock already has an owner we fail to get the lock.
1261 * This can be done without taking the @lock->wait_lock as
1262 * it is only being read, and this is a trylock anyway.
1263 */
1264 if (rt_mutex_owner(lock))
1265 return 0;
1266
1267 /*
1268 * The mutex has currently no owner. Lock the wait lock and try to
1269 * acquire the lock. We use irqsave here to support early boot calls.
1270 */
1271 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1272
1273 ret = __rt_mutex_slowtrylock(lock);
1274
1275 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1276
1277 return ret;
1278 }
1279
__rt_mutex_trylock(struct rt_mutex_base * lock)1280 static __always_inline int __rt_mutex_trylock(struct rt_mutex_base *lock)
1281 {
1282 if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
1283 return 1;
1284
1285 return rt_mutex_slowtrylock(lock);
1286 }
1287
1288 /*
1289 * Slow path to release a rt-mutex.
1290 */
rt_mutex_slowunlock(struct rt_mutex_base * lock)1291 static void __sched rt_mutex_slowunlock(struct rt_mutex_base *lock)
1292 {
1293 DEFINE_RT_WAKE_Q(wqh);
1294 unsigned long flags;
1295
1296 /* irqsave required to support early boot calls */
1297 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1298
1299 debug_rt_mutex_unlock(lock);
1300
1301 /*
1302 * We must be careful here if the fast path is enabled. If we
1303 * have no waiters queued we cannot set owner to NULL here
1304 * because of:
1305 *
1306 * foo->lock->owner = NULL;
1307 * rtmutex_lock(foo->lock); <- fast path
1308 * free = atomic_dec_and_test(foo->refcnt);
1309 * rtmutex_unlock(foo->lock); <- fast path
1310 * if (free)
1311 * kfree(foo);
1312 * raw_spin_unlock(foo->lock->wait_lock);
1313 *
1314 * So for the fastpath enabled kernel:
1315 *
1316 * Nothing can set the waiters bit as long as we hold
1317 * lock->wait_lock. So we do the following sequence:
1318 *
1319 * owner = rt_mutex_owner(lock);
1320 * clear_rt_mutex_waiters(lock);
1321 * raw_spin_unlock(&lock->wait_lock);
1322 * if (cmpxchg(&lock->owner, owner, 0) == owner)
1323 * return;
1324 * goto retry;
1325 *
1326 * The fastpath disabled variant is simple as all access to
1327 * lock->owner is serialized by lock->wait_lock:
1328 *
1329 * lock->owner = NULL;
1330 * raw_spin_unlock(&lock->wait_lock);
1331 */
1332 while (!rt_mutex_has_waiters(lock)) {
1333 /* Drops lock->wait_lock ! */
1334 if (unlock_rt_mutex_safe(lock, flags) == true)
1335 return;
1336 /* Relock the rtmutex and try again */
1337 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1338 }
1339
1340 /*
1341 * The wakeup next waiter path does not suffer from the above
1342 * race. See the comments there.
1343 *
1344 * Queue the next waiter for wakeup once we release the wait_lock.
1345 */
1346 mark_wakeup_next_waiter(&wqh, lock);
1347 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1348
1349 rt_mutex_wake_up_q(&wqh);
1350 }
1351
__rt_mutex_unlock(struct rt_mutex_base * lock)1352 static __always_inline void __rt_mutex_unlock(struct rt_mutex_base *lock)
1353 {
1354 if (likely(rt_mutex_cmpxchg_release(lock, current, NULL)))
1355 return;
1356
1357 rt_mutex_slowunlock(lock);
1358 }
1359
1360 #ifdef CONFIG_SMP
rtmutex_spin_on_owner(struct rt_mutex_base * lock,struct rt_mutex_waiter * waiter,struct task_struct * owner)1361 static bool rtmutex_spin_on_owner(struct rt_mutex_base *lock,
1362 struct rt_mutex_waiter *waiter,
1363 struct task_struct *owner)
1364 {
1365 bool res = true;
1366
1367 rcu_read_lock();
1368 for (;;) {
1369 /* If owner changed, trylock again. */
1370 if (owner != rt_mutex_owner(lock))
1371 break;
1372 /*
1373 * Ensure that @owner is dereferenced after checking that
1374 * the lock owner still matches @owner. If that fails,
1375 * @owner might point to freed memory. If it still matches,
1376 * the rcu_read_lock() ensures the memory stays valid.
1377 */
1378 barrier();
1379 /*
1380 * Stop spinning when:
1381 * - the lock owner has been scheduled out
1382 * - current is not longer the top waiter
1383 * - current is requested to reschedule (redundant
1384 * for CONFIG_PREEMPT_RCU=y)
1385 * - the VCPU on which owner runs is preempted
1386 */
1387 if (!owner_on_cpu(owner) || need_resched() ||
1388 !rt_mutex_waiter_is_top_waiter(lock, waiter)) {
1389 res = false;
1390 break;
1391 }
1392 cpu_relax();
1393 }
1394 rcu_read_unlock();
1395 return res;
1396 }
1397 #else
rtmutex_spin_on_owner(struct rt_mutex_base * lock,struct rt_mutex_waiter * waiter,struct task_struct * owner)1398 static bool rtmutex_spin_on_owner(struct rt_mutex_base *lock,
1399 struct rt_mutex_waiter *waiter,
1400 struct task_struct *owner)
1401 {
1402 return false;
1403 }
1404 #endif
1405
1406 #ifdef RT_MUTEX_BUILD_MUTEX
1407 /*
1408 * Functions required for:
1409 * - rtmutex, futex on all kernels
1410 * - mutex and rwsem substitutions on RT kernels
1411 */
1412
1413 /*
1414 * Remove a waiter from a lock and give up
1415 *
1416 * Must be called with lock->wait_lock held and interrupts disabled. It must
1417 * have just failed to try_to_take_rt_mutex().
1418 */
remove_waiter(struct rt_mutex_base * lock,struct rt_mutex_waiter * waiter)1419 static void __sched remove_waiter(struct rt_mutex_base *lock,
1420 struct rt_mutex_waiter *waiter)
1421 {
1422 bool is_top_waiter = (waiter == rt_mutex_top_waiter(lock));
1423 struct task_struct *owner = rt_mutex_owner(lock);
1424 struct rt_mutex_base *next_lock;
1425
1426 lockdep_assert_held(&lock->wait_lock);
1427
1428 raw_spin_lock(¤t->pi_lock);
1429 rt_mutex_dequeue(lock, waiter);
1430 current->pi_blocked_on = NULL;
1431 raw_spin_unlock(¤t->pi_lock);
1432
1433 /*
1434 * Only update priority if the waiter was the highest priority
1435 * waiter of the lock and there is an owner to update.
1436 */
1437 if (!owner || !is_top_waiter)
1438 return;
1439
1440 raw_spin_lock(&owner->pi_lock);
1441
1442 rt_mutex_dequeue_pi(owner, waiter);
1443
1444 if (rt_mutex_has_waiters(lock))
1445 rt_mutex_enqueue_pi(owner, rt_mutex_top_waiter(lock));
1446
1447 rt_mutex_adjust_prio(owner);
1448
1449 /* Store the lock on which owner is blocked or NULL */
1450 next_lock = task_blocked_on_lock(owner);
1451
1452 raw_spin_unlock(&owner->pi_lock);
1453
1454 /*
1455 * Don't walk the chain, if the owner task is not blocked
1456 * itself.
1457 */
1458 if (!next_lock)
1459 return;
1460
1461 /* gets dropped in rt_mutex_adjust_prio_chain()! */
1462 get_task_struct(owner);
1463
1464 raw_spin_unlock_irq(&lock->wait_lock);
1465
1466 rt_mutex_adjust_prio_chain(owner, RT_MUTEX_MIN_CHAINWALK, lock,
1467 next_lock, NULL, current);
1468
1469 raw_spin_lock_irq(&lock->wait_lock);
1470 }
1471
1472 /**
1473 * rt_mutex_slowlock_block() - Perform the wait-wake-try-to-take loop
1474 * @lock: the rt_mutex to take
1475 * @ww_ctx: WW mutex context pointer
1476 * @state: the state the task should block in (TASK_INTERRUPTIBLE
1477 * or TASK_UNINTERRUPTIBLE)
1478 * @timeout: the pre-initialized and started timer, or NULL for none
1479 * @waiter: the pre-initialized rt_mutex_waiter
1480 *
1481 * Must be called with lock->wait_lock held and interrupts disabled
1482 */
rt_mutex_slowlock_block(struct rt_mutex_base * lock,struct ww_acquire_ctx * ww_ctx,unsigned int state,struct hrtimer_sleeper * timeout,struct rt_mutex_waiter * waiter)1483 static int __sched rt_mutex_slowlock_block(struct rt_mutex_base *lock,
1484 struct ww_acquire_ctx *ww_ctx,
1485 unsigned int state,
1486 struct hrtimer_sleeper *timeout,
1487 struct rt_mutex_waiter *waiter)
1488 {
1489 struct rt_mutex *rtm = container_of(lock, struct rt_mutex, rtmutex);
1490 struct task_struct *owner;
1491 int ret = 0;
1492
1493 for (;;) {
1494 /* Try to acquire the lock: */
1495 if (try_to_take_rt_mutex(lock, current, waiter))
1496 break;
1497
1498 if (timeout && !timeout->task) {
1499 ret = -ETIMEDOUT;
1500 break;
1501 }
1502 if (signal_pending_state(state, current)) {
1503 ret = -EINTR;
1504 break;
1505 }
1506
1507 if (build_ww_mutex() && ww_ctx) {
1508 ret = __ww_mutex_check_kill(rtm, waiter, ww_ctx);
1509 if (ret)
1510 break;
1511 }
1512
1513 if (waiter == rt_mutex_top_waiter(lock))
1514 owner = rt_mutex_owner(lock);
1515 else
1516 owner = NULL;
1517 raw_spin_unlock_irq(&lock->wait_lock);
1518
1519 if (!owner || !rtmutex_spin_on_owner(lock, waiter, owner))
1520 schedule();
1521
1522 raw_spin_lock_irq(&lock->wait_lock);
1523 set_current_state(state);
1524 }
1525
1526 __set_current_state(TASK_RUNNING);
1527 return ret;
1528 }
1529
rt_mutex_handle_deadlock(int res,int detect_deadlock,struct rt_mutex_waiter * w)1530 static void __sched rt_mutex_handle_deadlock(int res, int detect_deadlock,
1531 struct rt_mutex_waiter *w)
1532 {
1533 /*
1534 * If the result is not -EDEADLOCK or the caller requested
1535 * deadlock detection, nothing to do here.
1536 */
1537 if (res != -EDEADLOCK || detect_deadlock)
1538 return;
1539
1540 if (build_ww_mutex() && w->ww_ctx)
1541 return;
1542
1543 /*
1544 * Yell loudly and stop the task right here.
1545 */
1546 WARN(1, "rtmutex deadlock detected\n");
1547 while (1) {
1548 set_current_state(TASK_INTERRUPTIBLE);
1549 schedule();
1550 }
1551 }
1552
1553 /**
1554 * __rt_mutex_slowlock - Locking slowpath invoked with lock::wait_lock held
1555 * @lock: The rtmutex to block lock
1556 * @ww_ctx: WW mutex context pointer
1557 * @state: The task state for sleeping
1558 * @chwalk: Indicator whether full or partial chainwalk is requested
1559 * @waiter: Initializer waiter for blocking
1560 */
__rt_mutex_slowlock(struct rt_mutex_base * lock,struct ww_acquire_ctx * ww_ctx,unsigned int state,enum rtmutex_chainwalk chwalk,struct rt_mutex_waiter * waiter)1561 static int __sched __rt_mutex_slowlock(struct rt_mutex_base *lock,
1562 struct ww_acquire_ctx *ww_ctx,
1563 unsigned int state,
1564 enum rtmutex_chainwalk chwalk,
1565 struct rt_mutex_waiter *waiter)
1566 {
1567 struct rt_mutex *rtm = container_of(lock, struct rt_mutex, rtmutex);
1568 struct ww_mutex *ww = ww_container_of(rtm);
1569 int ret;
1570
1571 lockdep_assert_held(&lock->wait_lock);
1572
1573 /* Try to acquire the lock again: */
1574 if (try_to_take_rt_mutex(lock, current, NULL)) {
1575 if (build_ww_mutex() && ww_ctx) {
1576 __ww_mutex_check_waiters(rtm, ww_ctx);
1577 ww_mutex_lock_acquired(ww, ww_ctx);
1578 }
1579 return 0;
1580 }
1581
1582 set_current_state(state);
1583
1584 trace_contention_begin(lock, LCB_F_RT);
1585
1586 ret = task_blocks_on_rt_mutex(lock, waiter, current, ww_ctx, chwalk);
1587 if (likely(!ret))
1588 ret = rt_mutex_slowlock_block(lock, ww_ctx, state, NULL, waiter);
1589
1590 if (likely(!ret)) {
1591 /* acquired the lock */
1592 if (build_ww_mutex() && ww_ctx) {
1593 if (!ww_ctx->is_wait_die)
1594 __ww_mutex_check_waiters(rtm, ww_ctx);
1595 ww_mutex_lock_acquired(ww, ww_ctx);
1596 }
1597 } else {
1598 __set_current_state(TASK_RUNNING);
1599 remove_waiter(lock, waiter);
1600 rt_mutex_handle_deadlock(ret, chwalk, waiter);
1601 }
1602
1603 /*
1604 * try_to_take_rt_mutex() sets the waiter bit
1605 * unconditionally. We might have to fix that up.
1606 */
1607 fixup_rt_mutex_waiters(lock);
1608
1609 trace_contention_end(lock, ret);
1610
1611 return ret;
1612 }
1613
__rt_mutex_slowlock_locked(struct rt_mutex_base * lock,struct ww_acquire_ctx * ww_ctx,unsigned int state)1614 static inline int __rt_mutex_slowlock_locked(struct rt_mutex_base *lock,
1615 struct ww_acquire_ctx *ww_ctx,
1616 unsigned int state)
1617 {
1618 struct rt_mutex_waiter waiter;
1619 int ret;
1620
1621 rt_mutex_init_waiter(&waiter);
1622 waiter.ww_ctx = ww_ctx;
1623
1624 ret = __rt_mutex_slowlock(lock, ww_ctx, state, RT_MUTEX_MIN_CHAINWALK,
1625 &waiter);
1626
1627 debug_rt_mutex_free_waiter(&waiter);
1628 return ret;
1629 }
1630
1631 /*
1632 * rt_mutex_slowlock - Locking slowpath invoked when fast path fails
1633 * @lock: The rtmutex to block lock
1634 * @ww_ctx: WW mutex context pointer
1635 * @state: The task state for sleeping
1636 */
rt_mutex_slowlock(struct rt_mutex_base * lock,struct ww_acquire_ctx * ww_ctx,unsigned int state)1637 static int __sched rt_mutex_slowlock(struct rt_mutex_base *lock,
1638 struct ww_acquire_ctx *ww_ctx,
1639 unsigned int state)
1640 {
1641 unsigned long flags;
1642 int ret;
1643
1644 /*
1645 * Technically we could use raw_spin_[un]lock_irq() here, but this can
1646 * be called in early boot if the cmpxchg() fast path is disabled
1647 * (debug, no architecture support). In this case we will acquire the
1648 * rtmutex with lock->wait_lock held. But we cannot unconditionally
1649 * enable interrupts in that early boot case. So we need to use the
1650 * irqsave/restore variants.
1651 */
1652 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1653 ret = __rt_mutex_slowlock_locked(lock, ww_ctx, state);
1654 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1655
1656 return ret;
1657 }
1658
__rt_mutex_lock(struct rt_mutex_base * lock,unsigned int state)1659 static __always_inline int __rt_mutex_lock(struct rt_mutex_base *lock,
1660 unsigned int state)
1661 {
1662 if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
1663 return 0;
1664
1665 return rt_mutex_slowlock(lock, NULL, state);
1666 }
1667 #endif /* RT_MUTEX_BUILD_MUTEX */
1668
1669 #ifdef RT_MUTEX_BUILD_SPINLOCKS
1670 /*
1671 * Functions required for spin/rw_lock substitution on RT kernels
1672 */
1673
1674 /**
1675 * rtlock_slowlock_locked - Slow path lock acquisition for RT locks
1676 * @lock: The underlying RT mutex
1677 */
rtlock_slowlock_locked(struct rt_mutex_base * lock)1678 static void __sched rtlock_slowlock_locked(struct rt_mutex_base *lock)
1679 {
1680 struct rt_mutex_waiter waiter;
1681 struct task_struct *owner;
1682
1683 lockdep_assert_held(&lock->wait_lock);
1684
1685 if (try_to_take_rt_mutex(lock, current, NULL))
1686 return;
1687
1688 rt_mutex_init_rtlock_waiter(&waiter);
1689
1690 /* Save current state and set state to TASK_RTLOCK_WAIT */
1691 current_save_and_set_rtlock_wait_state();
1692
1693 trace_contention_begin(lock, LCB_F_RT);
1694
1695 task_blocks_on_rt_mutex(lock, &waiter, current, NULL, RT_MUTEX_MIN_CHAINWALK);
1696
1697 for (;;) {
1698 /* Try to acquire the lock again */
1699 if (try_to_take_rt_mutex(lock, current, &waiter))
1700 break;
1701
1702 if (&waiter == rt_mutex_top_waiter(lock))
1703 owner = rt_mutex_owner(lock);
1704 else
1705 owner = NULL;
1706 raw_spin_unlock_irq(&lock->wait_lock);
1707
1708 if (!owner || !rtmutex_spin_on_owner(lock, &waiter, owner))
1709 schedule_rtlock();
1710
1711 raw_spin_lock_irq(&lock->wait_lock);
1712 set_current_state(TASK_RTLOCK_WAIT);
1713 }
1714
1715 /* Restore the task state */
1716 current_restore_rtlock_saved_state();
1717
1718 /*
1719 * try_to_take_rt_mutex() sets the waiter bit unconditionally.
1720 * We might have to fix that up:
1721 */
1722 fixup_rt_mutex_waiters(lock);
1723 debug_rt_mutex_free_waiter(&waiter);
1724
1725 trace_contention_end(lock, 0);
1726 }
1727
rtlock_slowlock(struct rt_mutex_base * lock)1728 static __always_inline void __sched rtlock_slowlock(struct rt_mutex_base *lock)
1729 {
1730 unsigned long flags;
1731
1732 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1733 rtlock_slowlock_locked(lock);
1734 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1735 }
1736
1737 #endif /* RT_MUTEX_BUILD_SPINLOCKS */
1738