xref: /openbmc/linux/kernel/locking/rtmutex.c (revision 930beb5a)
1 /*
2  * RT-Mutexes: simple blocking mutual exclusion locks with PI support
3  *
4  * started by Ingo Molnar and Thomas Gleixner.
5  *
6  *  Copyright (C) 2004-2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
7  *  Copyright (C) 2005-2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com>
8  *  Copyright (C) 2005 Kihon Technologies Inc., Steven Rostedt
9  *  Copyright (C) 2006 Esben Nielsen
10  *
11  *  See Documentation/rt-mutex-design.txt for details.
12  */
13 #include <linux/spinlock.h>
14 #include <linux/export.h>
15 #include <linux/sched.h>
16 #include <linux/sched/rt.h>
17 #include <linux/timer.h>
18 
19 #include "rtmutex_common.h"
20 
21 /*
22  * lock->owner state tracking:
23  *
24  * lock->owner holds the task_struct pointer of the owner. Bit 0
25  * is used to keep track of the "lock has waiters" state.
26  *
27  * owner	bit0
28  * NULL		0	lock is free (fast acquire possible)
29  * NULL		1	lock is free and has waiters and the top waiter
30  *				is going to take the lock*
31  * taskpointer	0	lock is held (fast release possible)
32  * taskpointer	1	lock is held and has waiters**
33  *
34  * The fast atomic compare exchange based acquire and release is only
35  * possible when bit 0 of lock->owner is 0.
36  *
37  * (*) It also can be a transitional state when grabbing the lock
38  * with ->wait_lock is held. To prevent any fast path cmpxchg to the lock,
39  * we need to set the bit0 before looking at the lock, and the owner may be
40  * NULL in this small time, hence this can be a transitional state.
41  *
42  * (**) There is a small time when bit 0 is set but there are no
43  * waiters. This can happen when grabbing the lock in the slow path.
44  * To prevent a cmpxchg of the owner releasing the lock, we need to
45  * set this bit before looking at the lock.
46  */
47 
48 static void
49 rt_mutex_set_owner(struct rt_mutex *lock, struct task_struct *owner)
50 {
51 	unsigned long val = (unsigned long)owner;
52 
53 	if (rt_mutex_has_waiters(lock))
54 		val |= RT_MUTEX_HAS_WAITERS;
55 
56 	lock->owner = (struct task_struct *)val;
57 }
58 
59 static inline void clear_rt_mutex_waiters(struct rt_mutex *lock)
60 {
61 	lock->owner = (struct task_struct *)
62 			((unsigned long)lock->owner & ~RT_MUTEX_HAS_WAITERS);
63 }
64 
65 static void fixup_rt_mutex_waiters(struct rt_mutex *lock)
66 {
67 	if (!rt_mutex_has_waiters(lock))
68 		clear_rt_mutex_waiters(lock);
69 }
70 
71 /*
72  * We can speed up the acquire/release, if the architecture
73  * supports cmpxchg and if there's no debugging state to be set up
74  */
75 #if defined(__HAVE_ARCH_CMPXCHG) && !defined(CONFIG_DEBUG_RT_MUTEXES)
76 # define rt_mutex_cmpxchg(l,c,n)	(cmpxchg(&l->owner, c, n) == c)
77 static inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
78 {
79 	unsigned long owner, *p = (unsigned long *) &lock->owner;
80 
81 	do {
82 		owner = *p;
83 	} while (cmpxchg(p, owner, owner | RT_MUTEX_HAS_WAITERS) != owner);
84 }
85 #else
86 # define rt_mutex_cmpxchg(l,c,n)	(0)
87 static inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
88 {
89 	lock->owner = (struct task_struct *)
90 			((unsigned long)lock->owner | RT_MUTEX_HAS_WAITERS);
91 }
92 #endif
93 
94 /*
95  * Calculate task priority from the waiter list priority
96  *
97  * Return task->normal_prio when the waiter list is empty or when
98  * the waiter is not allowed to do priority boosting
99  */
100 int rt_mutex_getprio(struct task_struct *task)
101 {
102 	if (likely(!task_has_pi_waiters(task)))
103 		return task->normal_prio;
104 
105 	return min(task_top_pi_waiter(task)->pi_list_entry.prio,
106 		   task->normal_prio);
107 }
108 
109 /*
110  * Adjust the priority of a task, after its pi_waiters got modified.
111  *
112  * This can be both boosting and unboosting. task->pi_lock must be held.
113  */
114 static void __rt_mutex_adjust_prio(struct task_struct *task)
115 {
116 	int prio = rt_mutex_getprio(task);
117 
118 	if (task->prio != prio)
119 		rt_mutex_setprio(task, prio);
120 }
121 
122 /*
123  * Adjust task priority (undo boosting). Called from the exit path of
124  * rt_mutex_slowunlock() and rt_mutex_slowlock().
125  *
126  * (Note: We do this outside of the protection of lock->wait_lock to
127  * allow the lock to be taken while or before we readjust the priority
128  * of task. We do not use the spin_xx_mutex() variants here as we are
129  * outside of the debug path.)
130  */
131 static void rt_mutex_adjust_prio(struct task_struct *task)
132 {
133 	unsigned long flags;
134 
135 	raw_spin_lock_irqsave(&task->pi_lock, flags);
136 	__rt_mutex_adjust_prio(task);
137 	raw_spin_unlock_irqrestore(&task->pi_lock, flags);
138 }
139 
140 /*
141  * Max number of times we'll walk the boosting chain:
142  */
143 int max_lock_depth = 1024;
144 
145 /*
146  * Adjust the priority chain. Also used for deadlock detection.
147  * Decreases task's usage by one - may thus free the task.
148  *
149  * @task: the task owning the mutex (owner) for which a chain walk is probably
150  *	  needed
151  * @deadlock_detect: do we have to carry out deadlock detection?
152  * @orig_lock: the mutex (can be NULL if we are walking the chain to recheck
153  * 	       things for a task that has just got its priority adjusted, and
154  *	       is waiting on a mutex)
155  * @orig_waiter: rt_mutex_waiter struct for the task that has just donated
156  *		 its priority to the mutex owner (can be NULL in the case
157  *		 depicted above or if the top waiter is gone away and we are
158  *		 actually deboosting the owner)
159  * @top_task: the current top waiter
160  *
161  * Returns 0 or -EDEADLK.
162  */
163 static int rt_mutex_adjust_prio_chain(struct task_struct *task,
164 				      int deadlock_detect,
165 				      struct rt_mutex *orig_lock,
166 				      struct rt_mutex_waiter *orig_waiter,
167 				      struct task_struct *top_task)
168 {
169 	struct rt_mutex *lock;
170 	struct rt_mutex_waiter *waiter, *top_waiter = orig_waiter;
171 	int detect_deadlock, ret = 0, depth = 0;
172 	unsigned long flags;
173 
174 	detect_deadlock = debug_rt_mutex_detect_deadlock(orig_waiter,
175 							 deadlock_detect);
176 
177 	/*
178 	 * The (de)boosting is a step by step approach with a lot of
179 	 * pitfalls. We want this to be preemptible and we want hold a
180 	 * maximum of two locks per step. So we have to check
181 	 * carefully whether things change under us.
182 	 */
183  again:
184 	if (++depth > max_lock_depth) {
185 		static int prev_max;
186 
187 		/*
188 		 * Print this only once. If the admin changes the limit,
189 		 * print a new message when reaching the limit again.
190 		 */
191 		if (prev_max != max_lock_depth) {
192 			prev_max = max_lock_depth;
193 			printk(KERN_WARNING "Maximum lock depth %d reached "
194 			       "task: %s (%d)\n", max_lock_depth,
195 			       top_task->comm, task_pid_nr(top_task));
196 		}
197 		put_task_struct(task);
198 
199 		return deadlock_detect ? -EDEADLK : 0;
200 	}
201  retry:
202 	/*
203 	 * Task can not go away as we did a get_task() before !
204 	 */
205 	raw_spin_lock_irqsave(&task->pi_lock, flags);
206 
207 	waiter = task->pi_blocked_on;
208 	/*
209 	 * Check whether the end of the boosting chain has been
210 	 * reached or the state of the chain has changed while we
211 	 * dropped the locks.
212 	 */
213 	if (!waiter)
214 		goto out_unlock_pi;
215 
216 	/*
217 	 * Check the orig_waiter state. After we dropped the locks,
218 	 * the previous owner of the lock might have released the lock.
219 	 */
220 	if (orig_waiter && !rt_mutex_owner(orig_lock))
221 		goto out_unlock_pi;
222 
223 	/*
224 	 * Drop out, when the task has no waiters. Note,
225 	 * top_waiter can be NULL, when we are in the deboosting
226 	 * mode!
227 	 */
228 	if (top_waiter && (!task_has_pi_waiters(task) ||
229 			   top_waiter != task_top_pi_waiter(task)))
230 		goto out_unlock_pi;
231 
232 	/*
233 	 * When deadlock detection is off then we check, if further
234 	 * priority adjustment is necessary.
235 	 */
236 	if (!detect_deadlock && waiter->list_entry.prio == task->prio)
237 		goto out_unlock_pi;
238 
239 	lock = waiter->lock;
240 	if (!raw_spin_trylock(&lock->wait_lock)) {
241 		raw_spin_unlock_irqrestore(&task->pi_lock, flags);
242 		cpu_relax();
243 		goto retry;
244 	}
245 
246 	/* Deadlock detection */
247 	if (lock == orig_lock || rt_mutex_owner(lock) == top_task) {
248 		debug_rt_mutex_deadlock(deadlock_detect, orig_waiter, lock);
249 		raw_spin_unlock(&lock->wait_lock);
250 		ret = deadlock_detect ? -EDEADLK : 0;
251 		goto out_unlock_pi;
252 	}
253 
254 	top_waiter = rt_mutex_top_waiter(lock);
255 
256 	/* Requeue the waiter */
257 	plist_del(&waiter->list_entry, &lock->wait_list);
258 	waiter->list_entry.prio = task->prio;
259 	plist_add(&waiter->list_entry, &lock->wait_list);
260 
261 	/* Release the task */
262 	raw_spin_unlock_irqrestore(&task->pi_lock, flags);
263 	if (!rt_mutex_owner(lock)) {
264 		/*
265 		 * If the requeue above changed the top waiter, then we need
266 		 * to wake the new top waiter up to try to get the lock.
267 		 */
268 
269 		if (top_waiter != rt_mutex_top_waiter(lock))
270 			wake_up_process(rt_mutex_top_waiter(lock)->task);
271 		raw_spin_unlock(&lock->wait_lock);
272 		goto out_put_task;
273 	}
274 	put_task_struct(task);
275 
276 	/* Grab the next task */
277 	task = rt_mutex_owner(lock);
278 	get_task_struct(task);
279 	raw_spin_lock_irqsave(&task->pi_lock, flags);
280 
281 	if (waiter == rt_mutex_top_waiter(lock)) {
282 		/* Boost the owner */
283 		plist_del(&top_waiter->pi_list_entry, &task->pi_waiters);
284 		waiter->pi_list_entry.prio = waiter->list_entry.prio;
285 		plist_add(&waiter->pi_list_entry, &task->pi_waiters);
286 		__rt_mutex_adjust_prio(task);
287 
288 	} else if (top_waiter == waiter) {
289 		/* Deboost the owner */
290 		plist_del(&waiter->pi_list_entry, &task->pi_waiters);
291 		waiter = rt_mutex_top_waiter(lock);
292 		waiter->pi_list_entry.prio = waiter->list_entry.prio;
293 		plist_add(&waiter->pi_list_entry, &task->pi_waiters);
294 		__rt_mutex_adjust_prio(task);
295 	}
296 
297 	raw_spin_unlock_irqrestore(&task->pi_lock, flags);
298 
299 	top_waiter = rt_mutex_top_waiter(lock);
300 	raw_spin_unlock(&lock->wait_lock);
301 
302 	if (!detect_deadlock && waiter != top_waiter)
303 		goto out_put_task;
304 
305 	goto again;
306 
307  out_unlock_pi:
308 	raw_spin_unlock_irqrestore(&task->pi_lock, flags);
309  out_put_task:
310 	put_task_struct(task);
311 
312 	return ret;
313 }
314 
315 /*
316  * Try to take an rt-mutex
317  *
318  * Must be called with lock->wait_lock held.
319  *
320  * @lock:   the lock to be acquired.
321  * @task:   the task which wants to acquire the lock
322  * @waiter: the waiter that is queued to the lock's wait list. (could be NULL)
323  */
324 static int try_to_take_rt_mutex(struct rt_mutex *lock, struct task_struct *task,
325 		struct rt_mutex_waiter *waiter)
326 {
327 	/*
328 	 * We have to be careful here if the atomic speedups are
329 	 * enabled, such that, when
330 	 *  - no other waiter is on the lock
331 	 *  - the lock has been released since we did the cmpxchg
332 	 * the lock can be released or taken while we are doing the
333 	 * checks and marking the lock with RT_MUTEX_HAS_WAITERS.
334 	 *
335 	 * The atomic acquire/release aware variant of
336 	 * mark_rt_mutex_waiters uses a cmpxchg loop. After setting
337 	 * the WAITERS bit, the atomic release / acquire can not
338 	 * happen anymore and lock->wait_lock protects us from the
339 	 * non-atomic case.
340 	 *
341 	 * Note, that this might set lock->owner =
342 	 * RT_MUTEX_HAS_WAITERS in the case the lock is not contended
343 	 * any more. This is fixed up when we take the ownership.
344 	 * This is the transitional state explained at the top of this file.
345 	 */
346 	mark_rt_mutex_waiters(lock);
347 
348 	if (rt_mutex_owner(lock))
349 		return 0;
350 
351 	/*
352 	 * It will get the lock because of one of these conditions:
353 	 * 1) there is no waiter
354 	 * 2) higher priority than waiters
355 	 * 3) it is top waiter
356 	 */
357 	if (rt_mutex_has_waiters(lock)) {
358 		if (task->prio >= rt_mutex_top_waiter(lock)->list_entry.prio) {
359 			if (!waiter || waiter != rt_mutex_top_waiter(lock))
360 				return 0;
361 		}
362 	}
363 
364 	if (waiter || rt_mutex_has_waiters(lock)) {
365 		unsigned long flags;
366 		struct rt_mutex_waiter *top;
367 
368 		raw_spin_lock_irqsave(&task->pi_lock, flags);
369 
370 		/* remove the queued waiter. */
371 		if (waiter) {
372 			plist_del(&waiter->list_entry, &lock->wait_list);
373 			task->pi_blocked_on = NULL;
374 		}
375 
376 		/*
377 		 * We have to enqueue the top waiter(if it exists) into
378 		 * task->pi_waiters list.
379 		 */
380 		if (rt_mutex_has_waiters(lock)) {
381 			top = rt_mutex_top_waiter(lock);
382 			top->pi_list_entry.prio = top->list_entry.prio;
383 			plist_add(&top->pi_list_entry, &task->pi_waiters);
384 		}
385 		raw_spin_unlock_irqrestore(&task->pi_lock, flags);
386 	}
387 
388 	/* We got the lock. */
389 	debug_rt_mutex_lock(lock);
390 
391 	rt_mutex_set_owner(lock, task);
392 
393 	rt_mutex_deadlock_account_lock(lock, task);
394 
395 	return 1;
396 }
397 
398 /*
399  * Task blocks on lock.
400  *
401  * Prepare waiter and propagate pi chain
402  *
403  * This must be called with lock->wait_lock held.
404  */
405 static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
406 				   struct rt_mutex_waiter *waiter,
407 				   struct task_struct *task,
408 				   int detect_deadlock)
409 {
410 	struct task_struct *owner = rt_mutex_owner(lock);
411 	struct rt_mutex_waiter *top_waiter = waiter;
412 	unsigned long flags;
413 	int chain_walk = 0, res;
414 
415 	raw_spin_lock_irqsave(&task->pi_lock, flags);
416 	__rt_mutex_adjust_prio(task);
417 	waiter->task = task;
418 	waiter->lock = lock;
419 	plist_node_init(&waiter->list_entry, task->prio);
420 	plist_node_init(&waiter->pi_list_entry, task->prio);
421 
422 	/* Get the top priority waiter on the lock */
423 	if (rt_mutex_has_waiters(lock))
424 		top_waiter = rt_mutex_top_waiter(lock);
425 	plist_add(&waiter->list_entry, &lock->wait_list);
426 
427 	task->pi_blocked_on = waiter;
428 
429 	raw_spin_unlock_irqrestore(&task->pi_lock, flags);
430 
431 	if (!owner)
432 		return 0;
433 
434 	if (waiter == rt_mutex_top_waiter(lock)) {
435 		raw_spin_lock_irqsave(&owner->pi_lock, flags);
436 		plist_del(&top_waiter->pi_list_entry, &owner->pi_waiters);
437 		plist_add(&waiter->pi_list_entry, &owner->pi_waiters);
438 
439 		__rt_mutex_adjust_prio(owner);
440 		if (owner->pi_blocked_on)
441 			chain_walk = 1;
442 		raw_spin_unlock_irqrestore(&owner->pi_lock, flags);
443 	}
444 	else if (debug_rt_mutex_detect_deadlock(waiter, detect_deadlock))
445 		chain_walk = 1;
446 
447 	if (!chain_walk)
448 		return 0;
449 
450 	/*
451 	 * The owner can't disappear while holding a lock,
452 	 * so the owner struct is protected by wait_lock.
453 	 * Gets dropped in rt_mutex_adjust_prio_chain()!
454 	 */
455 	get_task_struct(owner);
456 
457 	raw_spin_unlock(&lock->wait_lock);
458 
459 	res = rt_mutex_adjust_prio_chain(owner, detect_deadlock, lock, waiter,
460 					 task);
461 
462 	raw_spin_lock(&lock->wait_lock);
463 
464 	return res;
465 }
466 
467 /*
468  * Wake up the next waiter on the lock.
469  *
470  * Remove the top waiter from the current tasks waiter list and wake it up.
471  *
472  * Called with lock->wait_lock held.
473  */
474 static void wakeup_next_waiter(struct rt_mutex *lock)
475 {
476 	struct rt_mutex_waiter *waiter;
477 	unsigned long flags;
478 
479 	raw_spin_lock_irqsave(&current->pi_lock, flags);
480 
481 	waiter = rt_mutex_top_waiter(lock);
482 
483 	/*
484 	 * Remove it from current->pi_waiters. We do not adjust a
485 	 * possible priority boost right now. We execute wakeup in the
486 	 * boosted mode and go back to normal after releasing
487 	 * lock->wait_lock.
488 	 */
489 	plist_del(&waiter->pi_list_entry, &current->pi_waiters);
490 
491 	rt_mutex_set_owner(lock, NULL);
492 
493 	raw_spin_unlock_irqrestore(&current->pi_lock, flags);
494 
495 	wake_up_process(waiter->task);
496 }
497 
498 /*
499  * Remove a waiter from a lock and give up
500  *
501  * Must be called with lock->wait_lock held and
502  * have just failed to try_to_take_rt_mutex().
503  */
504 static void remove_waiter(struct rt_mutex *lock,
505 			  struct rt_mutex_waiter *waiter)
506 {
507 	int first = (waiter == rt_mutex_top_waiter(lock));
508 	struct task_struct *owner = rt_mutex_owner(lock);
509 	unsigned long flags;
510 	int chain_walk = 0;
511 
512 	raw_spin_lock_irqsave(&current->pi_lock, flags);
513 	plist_del(&waiter->list_entry, &lock->wait_list);
514 	current->pi_blocked_on = NULL;
515 	raw_spin_unlock_irqrestore(&current->pi_lock, flags);
516 
517 	if (!owner)
518 		return;
519 
520 	if (first) {
521 
522 		raw_spin_lock_irqsave(&owner->pi_lock, flags);
523 
524 		plist_del(&waiter->pi_list_entry, &owner->pi_waiters);
525 
526 		if (rt_mutex_has_waiters(lock)) {
527 			struct rt_mutex_waiter *next;
528 
529 			next = rt_mutex_top_waiter(lock);
530 			plist_add(&next->pi_list_entry, &owner->pi_waiters);
531 		}
532 		__rt_mutex_adjust_prio(owner);
533 
534 		if (owner->pi_blocked_on)
535 			chain_walk = 1;
536 
537 		raw_spin_unlock_irqrestore(&owner->pi_lock, flags);
538 	}
539 
540 	WARN_ON(!plist_node_empty(&waiter->pi_list_entry));
541 
542 	if (!chain_walk)
543 		return;
544 
545 	/* gets dropped in rt_mutex_adjust_prio_chain()! */
546 	get_task_struct(owner);
547 
548 	raw_spin_unlock(&lock->wait_lock);
549 
550 	rt_mutex_adjust_prio_chain(owner, 0, lock, NULL, current);
551 
552 	raw_spin_lock(&lock->wait_lock);
553 }
554 
555 /*
556  * Recheck the pi chain, in case we got a priority setting
557  *
558  * Called from sched_setscheduler
559  */
560 void rt_mutex_adjust_pi(struct task_struct *task)
561 {
562 	struct rt_mutex_waiter *waiter;
563 	unsigned long flags;
564 
565 	raw_spin_lock_irqsave(&task->pi_lock, flags);
566 
567 	waiter = task->pi_blocked_on;
568 	if (!waiter || waiter->list_entry.prio == task->prio) {
569 		raw_spin_unlock_irqrestore(&task->pi_lock, flags);
570 		return;
571 	}
572 
573 	raw_spin_unlock_irqrestore(&task->pi_lock, flags);
574 
575 	/* gets dropped in rt_mutex_adjust_prio_chain()! */
576 	get_task_struct(task);
577 	rt_mutex_adjust_prio_chain(task, 0, NULL, NULL, task);
578 }
579 
580 /**
581  * __rt_mutex_slowlock() - Perform the wait-wake-try-to-take loop
582  * @lock:		 the rt_mutex to take
583  * @state:		 the state the task should block in (TASK_INTERRUPTIBLE
584  * 			 or TASK_UNINTERRUPTIBLE)
585  * @timeout:		 the pre-initialized and started timer, or NULL for none
586  * @waiter:		 the pre-initialized rt_mutex_waiter
587  *
588  * lock->wait_lock must be held by the caller.
589  */
590 static int __sched
591 __rt_mutex_slowlock(struct rt_mutex *lock, int state,
592 		    struct hrtimer_sleeper *timeout,
593 		    struct rt_mutex_waiter *waiter)
594 {
595 	int ret = 0;
596 
597 	for (;;) {
598 		/* Try to acquire the lock: */
599 		if (try_to_take_rt_mutex(lock, current, waiter))
600 			break;
601 
602 		/*
603 		 * TASK_INTERRUPTIBLE checks for signals and
604 		 * timeout. Ignored otherwise.
605 		 */
606 		if (unlikely(state == TASK_INTERRUPTIBLE)) {
607 			/* Signal pending? */
608 			if (signal_pending(current))
609 				ret = -EINTR;
610 			if (timeout && !timeout->task)
611 				ret = -ETIMEDOUT;
612 			if (ret)
613 				break;
614 		}
615 
616 		raw_spin_unlock(&lock->wait_lock);
617 
618 		debug_rt_mutex_print_deadlock(waiter);
619 
620 		schedule_rt_mutex(lock);
621 
622 		raw_spin_lock(&lock->wait_lock);
623 		set_current_state(state);
624 	}
625 
626 	return ret;
627 }
628 
629 /*
630  * Slow path lock function:
631  */
632 static int __sched
633 rt_mutex_slowlock(struct rt_mutex *lock, int state,
634 		  struct hrtimer_sleeper *timeout,
635 		  int detect_deadlock)
636 {
637 	struct rt_mutex_waiter waiter;
638 	int ret = 0;
639 
640 	debug_rt_mutex_init_waiter(&waiter);
641 
642 	raw_spin_lock(&lock->wait_lock);
643 
644 	/* Try to acquire the lock again: */
645 	if (try_to_take_rt_mutex(lock, current, NULL)) {
646 		raw_spin_unlock(&lock->wait_lock);
647 		return 0;
648 	}
649 
650 	set_current_state(state);
651 
652 	/* Setup the timer, when timeout != NULL */
653 	if (unlikely(timeout)) {
654 		hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS);
655 		if (!hrtimer_active(&timeout->timer))
656 			timeout->task = NULL;
657 	}
658 
659 	ret = task_blocks_on_rt_mutex(lock, &waiter, current, detect_deadlock);
660 
661 	if (likely(!ret))
662 		ret = __rt_mutex_slowlock(lock, state, timeout, &waiter);
663 
664 	set_current_state(TASK_RUNNING);
665 
666 	if (unlikely(ret))
667 		remove_waiter(lock, &waiter);
668 
669 	/*
670 	 * try_to_take_rt_mutex() sets the waiter bit
671 	 * unconditionally. We might have to fix that up.
672 	 */
673 	fixup_rt_mutex_waiters(lock);
674 
675 	raw_spin_unlock(&lock->wait_lock);
676 
677 	/* Remove pending timer: */
678 	if (unlikely(timeout))
679 		hrtimer_cancel(&timeout->timer);
680 
681 	debug_rt_mutex_free_waiter(&waiter);
682 
683 	return ret;
684 }
685 
686 /*
687  * Slow path try-lock function:
688  */
689 static inline int
690 rt_mutex_slowtrylock(struct rt_mutex *lock)
691 {
692 	int ret = 0;
693 
694 	raw_spin_lock(&lock->wait_lock);
695 
696 	if (likely(rt_mutex_owner(lock) != current)) {
697 
698 		ret = try_to_take_rt_mutex(lock, current, NULL);
699 		/*
700 		 * try_to_take_rt_mutex() sets the lock waiters
701 		 * bit unconditionally. Clean this up.
702 		 */
703 		fixup_rt_mutex_waiters(lock);
704 	}
705 
706 	raw_spin_unlock(&lock->wait_lock);
707 
708 	return ret;
709 }
710 
711 /*
712  * Slow path to release a rt-mutex:
713  */
714 static void __sched
715 rt_mutex_slowunlock(struct rt_mutex *lock)
716 {
717 	raw_spin_lock(&lock->wait_lock);
718 
719 	debug_rt_mutex_unlock(lock);
720 
721 	rt_mutex_deadlock_account_unlock(current);
722 
723 	if (!rt_mutex_has_waiters(lock)) {
724 		lock->owner = NULL;
725 		raw_spin_unlock(&lock->wait_lock);
726 		return;
727 	}
728 
729 	wakeup_next_waiter(lock);
730 
731 	raw_spin_unlock(&lock->wait_lock);
732 
733 	/* Undo pi boosting if necessary: */
734 	rt_mutex_adjust_prio(current);
735 }
736 
737 /*
738  * debug aware fast / slowpath lock,trylock,unlock
739  *
740  * The atomic acquire/release ops are compiled away, when either the
741  * architecture does not support cmpxchg or when debugging is enabled.
742  */
743 static inline int
744 rt_mutex_fastlock(struct rt_mutex *lock, int state,
745 		  int detect_deadlock,
746 		  int (*slowfn)(struct rt_mutex *lock, int state,
747 				struct hrtimer_sleeper *timeout,
748 				int detect_deadlock))
749 {
750 	if (!detect_deadlock && likely(rt_mutex_cmpxchg(lock, NULL, current))) {
751 		rt_mutex_deadlock_account_lock(lock, current);
752 		return 0;
753 	} else
754 		return slowfn(lock, state, NULL, detect_deadlock);
755 }
756 
757 static inline int
758 rt_mutex_timed_fastlock(struct rt_mutex *lock, int state,
759 			struct hrtimer_sleeper *timeout, int detect_deadlock,
760 			int (*slowfn)(struct rt_mutex *lock, int state,
761 				      struct hrtimer_sleeper *timeout,
762 				      int detect_deadlock))
763 {
764 	if (!detect_deadlock && likely(rt_mutex_cmpxchg(lock, NULL, current))) {
765 		rt_mutex_deadlock_account_lock(lock, current);
766 		return 0;
767 	} else
768 		return slowfn(lock, state, timeout, detect_deadlock);
769 }
770 
771 static inline int
772 rt_mutex_fasttrylock(struct rt_mutex *lock,
773 		     int (*slowfn)(struct rt_mutex *lock))
774 {
775 	if (likely(rt_mutex_cmpxchg(lock, NULL, current))) {
776 		rt_mutex_deadlock_account_lock(lock, current);
777 		return 1;
778 	}
779 	return slowfn(lock);
780 }
781 
782 static inline void
783 rt_mutex_fastunlock(struct rt_mutex *lock,
784 		    void (*slowfn)(struct rt_mutex *lock))
785 {
786 	if (likely(rt_mutex_cmpxchg(lock, current, NULL)))
787 		rt_mutex_deadlock_account_unlock(current);
788 	else
789 		slowfn(lock);
790 }
791 
792 /**
793  * rt_mutex_lock - lock a rt_mutex
794  *
795  * @lock: the rt_mutex to be locked
796  */
797 void __sched rt_mutex_lock(struct rt_mutex *lock)
798 {
799 	might_sleep();
800 
801 	rt_mutex_fastlock(lock, TASK_UNINTERRUPTIBLE, 0, rt_mutex_slowlock);
802 }
803 EXPORT_SYMBOL_GPL(rt_mutex_lock);
804 
805 /**
806  * rt_mutex_lock_interruptible - lock a rt_mutex interruptible
807  *
808  * @lock: 		the rt_mutex to be locked
809  * @detect_deadlock:	deadlock detection on/off
810  *
811  * Returns:
812  *  0 		on success
813  * -EINTR 	when interrupted by a signal
814  * -EDEADLK	when the lock would deadlock (when deadlock detection is on)
815  */
816 int __sched rt_mutex_lock_interruptible(struct rt_mutex *lock,
817 						 int detect_deadlock)
818 {
819 	might_sleep();
820 
821 	return rt_mutex_fastlock(lock, TASK_INTERRUPTIBLE,
822 				 detect_deadlock, rt_mutex_slowlock);
823 }
824 EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible);
825 
826 /**
827  * rt_mutex_timed_lock - lock a rt_mutex interruptible
828  *			the timeout structure is provided
829  *			by the caller
830  *
831  * @lock: 		the rt_mutex to be locked
832  * @timeout:		timeout structure or NULL (no timeout)
833  * @detect_deadlock:	deadlock detection on/off
834  *
835  * Returns:
836  *  0 		on success
837  * -EINTR 	when interrupted by a signal
838  * -ETIMEDOUT	when the timeout expired
839  * -EDEADLK	when the lock would deadlock (when deadlock detection is on)
840  */
841 int
842 rt_mutex_timed_lock(struct rt_mutex *lock, struct hrtimer_sleeper *timeout,
843 		    int detect_deadlock)
844 {
845 	might_sleep();
846 
847 	return rt_mutex_timed_fastlock(lock, TASK_INTERRUPTIBLE, timeout,
848 				       detect_deadlock, rt_mutex_slowlock);
849 }
850 EXPORT_SYMBOL_GPL(rt_mutex_timed_lock);
851 
852 /**
853  * rt_mutex_trylock - try to lock a rt_mutex
854  *
855  * @lock:	the rt_mutex to be locked
856  *
857  * Returns 1 on success and 0 on contention
858  */
859 int __sched rt_mutex_trylock(struct rt_mutex *lock)
860 {
861 	return rt_mutex_fasttrylock(lock, rt_mutex_slowtrylock);
862 }
863 EXPORT_SYMBOL_GPL(rt_mutex_trylock);
864 
865 /**
866  * rt_mutex_unlock - unlock a rt_mutex
867  *
868  * @lock: the rt_mutex to be unlocked
869  */
870 void __sched rt_mutex_unlock(struct rt_mutex *lock)
871 {
872 	rt_mutex_fastunlock(lock, rt_mutex_slowunlock);
873 }
874 EXPORT_SYMBOL_GPL(rt_mutex_unlock);
875 
876 /**
877  * rt_mutex_destroy - mark a mutex unusable
878  * @lock: the mutex to be destroyed
879  *
880  * This function marks the mutex uninitialized, and any subsequent
881  * use of the mutex is forbidden. The mutex must not be locked when
882  * this function is called.
883  */
884 void rt_mutex_destroy(struct rt_mutex *lock)
885 {
886 	WARN_ON(rt_mutex_is_locked(lock));
887 #ifdef CONFIG_DEBUG_RT_MUTEXES
888 	lock->magic = NULL;
889 #endif
890 }
891 
892 EXPORT_SYMBOL_GPL(rt_mutex_destroy);
893 
894 /**
895  * __rt_mutex_init - initialize the rt lock
896  *
897  * @lock: the rt lock to be initialized
898  *
899  * Initialize the rt lock to unlocked state.
900  *
901  * Initializing of a locked rt lock is not allowed
902  */
903 void __rt_mutex_init(struct rt_mutex *lock, const char *name)
904 {
905 	lock->owner = NULL;
906 	raw_spin_lock_init(&lock->wait_lock);
907 	plist_head_init(&lock->wait_list);
908 
909 	debug_rt_mutex_init(lock, name);
910 }
911 EXPORT_SYMBOL_GPL(__rt_mutex_init);
912 
913 /**
914  * rt_mutex_init_proxy_locked - initialize and lock a rt_mutex on behalf of a
915  *				proxy owner
916  *
917  * @lock: 	the rt_mutex to be locked
918  * @proxy_owner:the task to set as owner
919  *
920  * No locking. Caller has to do serializing itself
921  * Special API call for PI-futex support
922  */
923 void rt_mutex_init_proxy_locked(struct rt_mutex *lock,
924 				struct task_struct *proxy_owner)
925 {
926 	__rt_mutex_init(lock, NULL);
927 	debug_rt_mutex_proxy_lock(lock, proxy_owner);
928 	rt_mutex_set_owner(lock, proxy_owner);
929 	rt_mutex_deadlock_account_lock(lock, proxy_owner);
930 }
931 
932 /**
933  * rt_mutex_proxy_unlock - release a lock on behalf of owner
934  *
935  * @lock: 	the rt_mutex to be locked
936  *
937  * No locking. Caller has to do serializing itself
938  * Special API call for PI-futex support
939  */
940 void rt_mutex_proxy_unlock(struct rt_mutex *lock,
941 			   struct task_struct *proxy_owner)
942 {
943 	debug_rt_mutex_proxy_unlock(lock);
944 	rt_mutex_set_owner(lock, NULL);
945 	rt_mutex_deadlock_account_unlock(proxy_owner);
946 }
947 
948 /**
949  * rt_mutex_start_proxy_lock() - Start lock acquisition for another task
950  * @lock:		the rt_mutex to take
951  * @waiter:		the pre-initialized rt_mutex_waiter
952  * @task:		the task to prepare
953  * @detect_deadlock:	perform deadlock detection (1) or not (0)
954  *
955  * Returns:
956  *  0 - task blocked on lock
957  *  1 - acquired the lock for task, caller should wake it up
958  * <0 - error
959  *
960  * Special API call for FUTEX_REQUEUE_PI support.
961  */
962 int rt_mutex_start_proxy_lock(struct rt_mutex *lock,
963 			      struct rt_mutex_waiter *waiter,
964 			      struct task_struct *task, int detect_deadlock)
965 {
966 	int ret;
967 
968 	raw_spin_lock(&lock->wait_lock);
969 
970 	if (try_to_take_rt_mutex(lock, task, NULL)) {
971 		raw_spin_unlock(&lock->wait_lock);
972 		return 1;
973 	}
974 
975 	ret = task_blocks_on_rt_mutex(lock, waiter, task, detect_deadlock);
976 
977 	if (ret && !rt_mutex_owner(lock)) {
978 		/*
979 		 * Reset the return value. We might have
980 		 * returned with -EDEADLK and the owner
981 		 * released the lock while we were walking the
982 		 * pi chain.  Let the waiter sort it out.
983 		 */
984 		ret = 0;
985 	}
986 
987 	if (unlikely(ret))
988 		remove_waiter(lock, waiter);
989 
990 	raw_spin_unlock(&lock->wait_lock);
991 
992 	debug_rt_mutex_print_deadlock(waiter);
993 
994 	return ret;
995 }
996 
997 /**
998  * rt_mutex_next_owner - return the next owner of the lock
999  *
1000  * @lock: the rt lock query
1001  *
1002  * Returns the next owner of the lock or NULL
1003  *
1004  * Caller has to serialize against other accessors to the lock
1005  * itself.
1006  *
1007  * Special API call for PI-futex support
1008  */
1009 struct task_struct *rt_mutex_next_owner(struct rt_mutex *lock)
1010 {
1011 	if (!rt_mutex_has_waiters(lock))
1012 		return NULL;
1013 
1014 	return rt_mutex_top_waiter(lock)->task;
1015 }
1016 
1017 /**
1018  * rt_mutex_finish_proxy_lock() - Complete lock acquisition
1019  * @lock:		the rt_mutex we were woken on
1020  * @to:			the timeout, null if none. hrtimer should already have
1021  * 			been started.
1022  * @waiter:		the pre-initialized rt_mutex_waiter
1023  * @detect_deadlock:	perform deadlock detection (1) or not (0)
1024  *
1025  * Complete the lock acquisition started our behalf by another thread.
1026  *
1027  * Returns:
1028  *  0 - success
1029  * <0 - error, one of -EINTR, -ETIMEDOUT, or -EDEADLK
1030  *
1031  * Special API call for PI-futex requeue support
1032  */
1033 int rt_mutex_finish_proxy_lock(struct rt_mutex *lock,
1034 			       struct hrtimer_sleeper *to,
1035 			       struct rt_mutex_waiter *waiter,
1036 			       int detect_deadlock)
1037 {
1038 	int ret;
1039 
1040 	raw_spin_lock(&lock->wait_lock);
1041 
1042 	set_current_state(TASK_INTERRUPTIBLE);
1043 
1044 	ret = __rt_mutex_slowlock(lock, TASK_INTERRUPTIBLE, to, waiter);
1045 
1046 	set_current_state(TASK_RUNNING);
1047 
1048 	if (unlikely(ret))
1049 		remove_waiter(lock, waiter);
1050 
1051 	/*
1052 	 * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
1053 	 * have to fix that up.
1054 	 */
1055 	fixup_rt_mutex_waiters(lock);
1056 
1057 	raw_spin_unlock(&lock->wait_lock);
1058 
1059 	return ret;
1060 }
1061