aboutsummaryrefslogtreecommitdiffstats
path: root/kernel
diff options
context:
space:
mode:
authorIngo Molnar <mingo@elte.hu>2006-06-27 05:54:53 -0400
committerLinus Torvalds <torvalds@g5.osdl.org>2006-06-27 20:32:47 -0400
commit23f78d4a03c53cbd75d87a795378ea540aa08c86 (patch)
tree27dfe06337990911380fe8c5949ae9acd8e9568a /kernel
parentb29739f902ee76a05493fb7d2303490fc75364f4 (diff)
[PATCH] pi-futex: rt mutex core
Core functions for the rt-mutex subsystem. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'kernel')
-rw-r--r--kernel/Makefile1
-rw-r--r--kernel/fork.c16
-rw-r--r--kernel/rtmutex.c904
-rw-r--r--kernel/rtmutex.h29
-rw-r--r--kernel/rtmutex_common.h93
-rw-r--r--kernel/sysctl.c15
6 files changed, 1058 insertions, 0 deletions
diff --git a/kernel/Makefile b/kernel/Makefile
index 752bd7d383af..21df9a338ff0 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -16,6 +16,7 @@ obj-$(CONFIG_FUTEX) += futex.o
16ifeq ($(CONFIG_COMPAT),y) 16ifeq ($(CONFIG_COMPAT),y)
17obj-$(CONFIG_FUTEX) += futex_compat.o 17obj-$(CONFIG_FUTEX) += futex_compat.o
18endif 18endif
19obj-$(CONFIG_RT_MUTEXES) += rtmutex.o
19obj-$(CONFIG_GENERIC_ISA_DMA) += dma.o 20obj-$(CONFIG_GENERIC_ISA_DMA) += dma.o
20obj-$(CONFIG_SMP) += cpu.o spinlock.o 21obj-$(CONFIG_SMP) += cpu.o spinlock.o
21obj-$(CONFIG_DEBUG_SPINLOCK) += spinlock.o 22obj-$(CONFIG_DEBUG_SPINLOCK) += spinlock.o
diff --git a/kernel/fork.c b/kernel/fork.c
index 9b4e54ef0225..b664a081fffa 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -104,6 +104,7 @@ static kmem_cache_t *mm_cachep;
104void free_task(struct task_struct *tsk) 104void free_task(struct task_struct *tsk)
105{ 105{
106 free_thread_info(tsk->thread_info); 106 free_thread_info(tsk->thread_info);
107 rt_mutex_debug_task_free(tsk);
107 free_task_struct(tsk); 108 free_task_struct(tsk);
108} 109}
109EXPORT_SYMBOL(free_task); 110EXPORT_SYMBOL(free_task);
@@ -913,6 +914,19 @@ asmlinkage long sys_set_tid_address(int __user *tidptr)
913 return current->pid; 914 return current->pid;
914} 915}
915 916
917static inline void rt_mutex_init_task(struct task_struct *p)
918{
919#ifdef CONFIG_RT_MUTEXES
920 spin_lock_init(&p->pi_lock);
921 plist_head_init(&p->pi_waiters, &p->pi_lock);
922 p->pi_blocked_on = NULL;
923# ifdef CONFIG_DEBUG_RT_MUTEXES
924 spin_lock_init(&p->held_list_lock);
925 INIT_LIST_HEAD(&p->held_list_head);
926# endif
927#endif
928}
929
916/* 930/*
917 * This creates a new process as a copy of the old one, 931 * This creates a new process as a copy of the old one,
918 * but does not actually start it yet. 932 * but does not actually start it yet.
@@ -1034,6 +1048,8 @@ static task_t *copy_process(unsigned long clone_flags,
1034 mpol_fix_fork_child_flag(p); 1048 mpol_fix_fork_child_flag(p);
1035#endif 1049#endif
1036 1050
1051 rt_mutex_init_task(p);
1052
1037#ifdef CONFIG_DEBUG_MUTEXES 1053#ifdef CONFIG_DEBUG_MUTEXES
1038 p->blocked_on = NULL; /* not blocked yet */ 1054 p->blocked_on = NULL; /* not blocked yet */
1039#endif 1055#endif
diff --git a/kernel/rtmutex.c b/kernel/rtmutex.c
new file mode 100644
index 000000000000..937a474fae94
--- /dev/null
+++ b/kernel/rtmutex.c
@@ -0,0 +1,904 @@
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#include <linux/spinlock.h>
12#include <linux/module.h>
13#include <linux/sched.h>
14#include <linux/timer.h>
15
16#include "rtmutex_common.h"
17
18#ifdef CONFIG_DEBUG_RT_MUTEXES
19# include "rtmutex-debug.h"
20#else
21# include "rtmutex.h"
22#endif
23
24/*
25 * lock->owner state tracking:
26 *
27 * lock->owner holds the task_struct pointer of the owner. Bit 0 and 1
28 * are used to keep track of the "owner is pending" and "lock has
29 * waiters" state.
30 *
31 * owner bit1 bit0
32 * NULL 0 0 lock is free (fast acquire possible)
33 * NULL 0 1 invalid state
34 * NULL 1 0 Transitional State*
35 * NULL 1 1 invalid state
36 * taskpointer 0 0 lock is held (fast release possible)
37 * taskpointer 0 1 task is pending owner
38 * taskpointer 1 0 lock is held and has waiters
39 * taskpointer 1 1 task is pending owner and lock has more waiters
40 *
41 * Pending ownership is assigned to the top (highest priority)
42 * waiter of the lock, when the lock is released. The thread is woken
43 * up and can now take the lock. Until the lock is taken (bit 0
44 * cleared) a competing higher priority thread can steal the lock
45 * which puts the woken up thread back on the waiters list.
46 *
47 * The fast atomic compare exchange based acquire and release is only
48 * possible when bit 0 and 1 of lock->owner are 0.
49 *
50 * (*) There's a small time where the owner can be NULL and the
51 * "lock has waiters" bit is set. This can happen when grabbing the lock.
52 * To prevent a cmpxchg of the owner releasing the lock, we need to set this
53 * bit before looking at the lock, hence the reason this is a transitional
54 * state.
55 */
56
57static void
58rt_mutex_set_owner(struct rt_mutex *lock, struct task_struct *owner,
59 unsigned long mask)
60{
61 unsigned long val = (unsigned long)owner | mask;
62
63 if (rt_mutex_has_waiters(lock))
64 val |= RT_MUTEX_HAS_WAITERS;
65
66 lock->owner = (struct task_struct *)val;
67}
68
69static inline void clear_rt_mutex_waiters(struct rt_mutex *lock)
70{
71 lock->owner = (struct task_struct *)
72 ((unsigned long)lock->owner & ~RT_MUTEX_HAS_WAITERS);
73}
74
75static void fixup_rt_mutex_waiters(struct rt_mutex *lock)
76{
77 if (!rt_mutex_has_waiters(lock))
78 clear_rt_mutex_waiters(lock);
79}
80
81/*
82 * We can speed up the acquire/release, if the architecture
83 * supports cmpxchg and if there's no debugging state to be set up
84 */
85#if defined(__HAVE_ARCH_CMPXCHG) && !defined(CONFIG_DEBUG_RT_MUTEXES)
86# define rt_mutex_cmpxchg(l,c,n) (cmpxchg(&l->owner, c, n) == c)
87static inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
88{
89 unsigned long owner, *p = (unsigned long *) &lock->owner;
90
91 do {
92 owner = *p;
93 } while (cmpxchg(p, owner, owner | RT_MUTEX_HAS_WAITERS) != owner);
94}
95#else
96# define rt_mutex_cmpxchg(l,c,n) (0)
97static inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
98{
99 lock->owner = (struct task_struct *)
100 ((unsigned long)lock->owner | RT_MUTEX_HAS_WAITERS);
101}
102#endif
103
104/*
105 * Calculate task priority from the waiter list priority
106 *
107 * Return task->normal_prio when the waiter list is empty or when
108 * the waiter is not allowed to do priority boosting
109 */
110int rt_mutex_getprio(struct task_struct *task)
111{
112 if (likely(!task_has_pi_waiters(task)))
113 return task->normal_prio;
114
115 return min(task_top_pi_waiter(task)->pi_list_entry.prio,
116 task->normal_prio);
117}
118
119/*
120 * Adjust the priority of a task, after its pi_waiters got modified.
121 *
122 * This can be both boosting and unboosting. task->pi_lock must be held.
123 */
124static void __rt_mutex_adjust_prio(struct task_struct *task)
125{
126 int prio = rt_mutex_getprio(task);
127
128 if (task->prio != prio)
129 rt_mutex_setprio(task, prio);
130}
131
132/*
133 * Adjust task priority (undo boosting). Called from the exit path of
134 * rt_mutex_slowunlock() and rt_mutex_slowlock().
135 *
136 * (Note: We do this outside of the protection of lock->wait_lock to
137 * allow the lock to be taken while or before we readjust the priority
138 * of task. We do not use the spin_xx_mutex() variants here as we are
139 * outside of the debug path.)
140 */
141static void rt_mutex_adjust_prio(struct task_struct *task)
142{
143 unsigned long flags;
144
145 spin_lock_irqsave(&task->pi_lock, flags);
146 __rt_mutex_adjust_prio(task);
147 spin_unlock_irqrestore(&task->pi_lock, flags);
148}
149
150/*
151 * Max number of times we'll walk the boosting chain:
152 */
153int max_lock_depth = 1024;
154
155/*
156 * Adjust the priority chain. Also used for deadlock detection.
157 * Decreases task's usage by one - may thus free the task.
158 * Returns 0 or -EDEADLK.
159 */
160static int rt_mutex_adjust_prio_chain(task_t *task,
161 int deadlock_detect,
162 struct rt_mutex *orig_lock,
163 struct rt_mutex_waiter *orig_waiter
164 __IP_DECL__)
165{
166 struct rt_mutex *lock;
167 struct rt_mutex_waiter *waiter, *top_waiter = orig_waiter;
168 int detect_deadlock, ret = 0, depth = 0;
169 unsigned long flags;
170
171 detect_deadlock = debug_rt_mutex_detect_deadlock(orig_waiter,
172 deadlock_detect);
173
174 /*
175 * The (de)boosting is a step by step approach with a lot of
176 * pitfalls. We want this to be preemptible and we want hold a
177 * maximum of two locks per step. So we have to check
178 * carefully whether things change under us.
179 */
180 again:
181 if (++depth > max_lock_depth) {
182 static int prev_max;
183
184 /*
185 * Print this only once. If the admin changes the limit,
186 * print a new message when reaching the limit again.
187 */
188 if (prev_max != max_lock_depth) {
189 prev_max = max_lock_depth;
190 printk(KERN_WARNING "Maximum lock depth %d reached "
191 "task: %s (%d)\n", max_lock_depth,
192 current->comm, current->pid);
193 }
194 put_task_struct(task);
195
196 return deadlock_detect ? -EDEADLK : 0;
197 }
198 retry:
199 /*
200 * Task can not go away as we did a get_task() before !
201 */
202 spin_lock_irqsave(&task->pi_lock, flags);
203
204 waiter = task->pi_blocked_on;
205 /*
206 * Check whether the end of the boosting chain has been
207 * reached or the state of the chain has changed while we
208 * dropped the locks.
209 */
210 if (!waiter || !waiter->task)
211 goto out_unlock_pi;
212
213 if (top_waiter && (!task_has_pi_waiters(task) ||
214 top_waiter != task_top_pi_waiter(task)))
215 goto out_unlock_pi;
216
217 /*
218 * When deadlock detection is off then we check, if further
219 * priority adjustment is necessary.
220 */
221 if (!detect_deadlock && waiter->list_entry.prio == task->prio)
222 goto out_unlock_pi;
223
224 lock = waiter->lock;
225 if (!spin_trylock(&lock->wait_lock)) {
226 spin_unlock_irqrestore(&task->pi_lock, flags);
227 cpu_relax();
228 goto retry;
229 }
230
231 /* Deadlock detection */
232 if (lock == orig_lock || rt_mutex_owner(lock) == current) {
233 debug_rt_mutex_deadlock(deadlock_detect, orig_waiter, lock);
234 spin_unlock(&lock->wait_lock);
235 ret = deadlock_detect ? -EDEADLK : 0;
236 goto out_unlock_pi;
237 }
238
239 top_waiter = rt_mutex_top_waiter(lock);
240
241 /* Requeue the waiter */
242 plist_del(&waiter->list_entry, &lock->wait_list);
243 waiter->list_entry.prio = task->prio;
244 plist_add(&waiter->list_entry, &lock->wait_list);
245
246 /* Release the task */
247 spin_unlock_irqrestore(&task->pi_lock, flags);
248 put_task_struct(task);
249
250 /* Grab the next task */
251 task = rt_mutex_owner(lock);
252 spin_lock_irqsave(&task->pi_lock, flags);
253
254 if (waiter == rt_mutex_top_waiter(lock)) {
255 /* Boost the owner */
256 plist_del(&top_waiter->pi_list_entry, &task->pi_waiters);
257 waiter->pi_list_entry.prio = waiter->list_entry.prio;
258 plist_add(&waiter->pi_list_entry, &task->pi_waiters);
259 __rt_mutex_adjust_prio(task);
260
261 } else if (top_waiter == waiter) {
262 /* Deboost the owner */
263 plist_del(&waiter->pi_list_entry, &task->pi_waiters);
264 waiter = rt_mutex_top_waiter(lock);
265 waiter->pi_list_entry.prio = waiter->list_entry.prio;
266 plist_add(&waiter->pi_list_entry, &task->pi_waiters);
267 __rt_mutex_adjust_prio(task);
268 }
269
270 get_task_struct(task);
271 spin_unlock_irqrestore(&task->pi_lock, flags);
272
273 top_waiter = rt_mutex_top_waiter(lock);
274 spin_unlock(&lock->wait_lock);
275
276 if (!detect_deadlock && waiter != top_waiter)
277 goto out_put_task;
278
279 goto again;
280
281 out_unlock_pi:
282 spin_unlock_irqrestore(&task->pi_lock, flags);
283 out_put_task:
284 put_task_struct(task);
285 return ret;
286}
287
288/*
289 * Optimization: check if we can steal the lock from the
290 * assigned pending owner [which might not have taken the
291 * lock yet]:
292 */
293static inline int try_to_steal_lock(struct rt_mutex *lock)
294{
295 struct task_struct *pendowner = rt_mutex_owner(lock);
296 struct rt_mutex_waiter *next;
297 unsigned long flags;
298
299 if (!rt_mutex_owner_pending(lock))
300 return 0;
301
302 if (pendowner == current)
303 return 1;
304
305 spin_lock_irqsave(&pendowner->pi_lock, flags);
306 if (current->prio >= pendowner->prio) {
307 spin_unlock_irqrestore(&pendowner->pi_lock, flags);
308 return 0;
309 }
310
311 /*
312 * Check if a waiter is enqueued on the pending owners
313 * pi_waiters list. Remove it and readjust pending owners
314 * priority.
315 */
316 if (likely(!rt_mutex_has_waiters(lock))) {
317 spin_unlock_irqrestore(&pendowner->pi_lock, flags);
318 return 1;
319 }
320
321 /* No chain handling, pending owner is not blocked on anything: */
322 next = rt_mutex_top_waiter(lock);
323 plist_del(&next->pi_list_entry, &pendowner->pi_waiters);
324 __rt_mutex_adjust_prio(pendowner);
325 spin_unlock_irqrestore(&pendowner->pi_lock, flags);
326
327 /*
328 * We are going to steal the lock and a waiter was
329 * enqueued on the pending owners pi_waiters queue. So
330 * we have to enqueue this waiter into
331 * current->pi_waiters list. This covers the case,
332 * where current is boosted because it holds another
333 * lock and gets unboosted because the booster is
334 * interrupted, so we would delay a waiter with higher
335 * priority as current->normal_prio.
336 *
337 * Note: in the rare case of a SCHED_OTHER task changing
338 * its priority and thus stealing the lock, next->task
339 * might be current:
340 */
341 if (likely(next->task != current)) {
342 spin_lock_irqsave(&current->pi_lock, flags);
343 plist_add(&next->pi_list_entry, &current->pi_waiters);
344 __rt_mutex_adjust_prio(current);
345 spin_unlock_irqrestore(&current->pi_lock, flags);
346 }
347 return 1;
348}
349
350/*
351 * Try to take an rt-mutex
352 *
353 * This fails
354 * - when the lock has a real owner
355 * - when a different pending owner exists and has higher priority than current
356 *
357 * Must be called with lock->wait_lock held.
358 */
359static int try_to_take_rt_mutex(struct rt_mutex *lock __IP_DECL__)
360{
361 /*
362 * We have to be careful here if the atomic speedups are
363 * enabled, such that, when
364 * - no other waiter is on the lock
365 * - the lock has been released since we did the cmpxchg
366 * the lock can be released or taken while we are doing the
367 * checks and marking the lock with RT_MUTEX_HAS_WAITERS.
368 *
369 * The atomic acquire/release aware variant of
370 * mark_rt_mutex_waiters uses a cmpxchg loop. After setting
371 * the WAITERS bit, the atomic release / acquire can not
372 * happen anymore and lock->wait_lock protects us from the
373 * non-atomic case.
374 *
375 * Note, that this might set lock->owner =
376 * RT_MUTEX_HAS_WAITERS in the case the lock is not contended
377 * any more. This is fixed up when we take the ownership.
378 * This is the transitional state explained at the top of this file.
379 */
380 mark_rt_mutex_waiters(lock);
381
382 if (rt_mutex_owner(lock) && !try_to_steal_lock(lock))
383 return 0;
384
385 /* We got the lock. */
386 debug_rt_mutex_lock(lock __IP__);
387
388 rt_mutex_set_owner(lock, current, 0);
389
390 rt_mutex_deadlock_account_lock(lock, current);
391
392 return 1;
393}
394
395/*
396 * Task blocks on lock.
397 *
398 * Prepare waiter and propagate pi chain
399 *
400 * This must be called with lock->wait_lock held.
401 */
402static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
403 struct rt_mutex_waiter *waiter,
404 int detect_deadlock
405 __IP_DECL__)
406{
407 struct rt_mutex_waiter *top_waiter = waiter;
408 task_t *owner = rt_mutex_owner(lock);
409 int boost = 0, res;
410 unsigned long flags;
411
412 spin_lock_irqsave(&current->pi_lock, flags);
413 __rt_mutex_adjust_prio(current);
414 waiter->task = current;
415 waiter->lock = lock;
416 plist_node_init(&waiter->list_entry, current->prio);
417 plist_node_init(&waiter->pi_list_entry, current->prio);
418
419 /* Get the top priority waiter on the lock */
420 if (rt_mutex_has_waiters(lock))
421 top_waiter = rt_mutex_top_waiter(lock);
422 plist_add(&waiter->list_entry, &lock->wait_list);
423
424 current->pi_blocked_on = waiter;
425
426 spin_unlock_irqrestore(&current->pi_lock, flags);
427
428 if (waiter == rt_mutex_top_waiter(lock)) {
429 spin_lock_irqsave(&owner->pi_lock, flags);
430 plist_del(&top_waiter->pi_list_entry, &owner->pi_waiters);
431 plist_add(&waiter->pi_list_entry, &owner->pi_waiters);
432
433 __rt_mutex_adjust_prio(owner);
434 if (owner->pi_blocked_on) {
435 boost = 1;
436 get_task_struct(owner);
437 }
438 spin_unlock_irqrestore(&owner->pi_lock, flags);
439 }
440 else if (debug_rt_mutex_detect_deadlock(waiter, detect_deadlock)) {
441 spin_lock_irqsave(&owner->pi_lock, flags);
442 if (owner->pi_blocked_on) {
443 boost = 1;
444 get_task_struct(owner);
445 }
446 spin_unlock_irqrestore(&owner->pi_lock, flags);
447 }
448 if (!boost)
449 return 0;
450
451 spin_unlock(&lock->wait_lock);
452
453 res = rt_mutex_adjust_prio_chain(owner, detect_deadlock, lock,
454 waiter __IP__);
455
456 spin_lock(&lock->wait_lock);
457
458 return res;
459}
460
461/*
462 * Wake up the next waiter on the lock.
463 *
464 * Remove the top waiter from the current tasks waiter list and from
465 * the lock waiter list. Set it as pending owner. Then wake it up.
466 *
467 * Called with lock->wait_lock held.
468 */
469static void wakeup_next_waiter(struct rt_mutex *lock)
470{
471 struct rt_mutex_waiter *waiter;
472 struct task_struct *pendowner;
473 unsigned long flags;
474
475 spin_lock_irqsave(&current->pi_lock, flags);
476
477 waiter = rt_mutex_top_waiter(lock);
478 plist_del(&waiter->list_entry, &lock->wait_list);
479
480 /*
481 * Remove it from current->pi_waiters. We do not adjust a
482 * possible priority boost right now. We execute wakeup in the
483 * boosted mode and go back to normal after releasing
484 * lock->wait_lock.
485 */
486 plist_del(&waiter->pi_list_entry, &current->pi_waiters);
487 pendowner = waiter->task;
488 waiter->task = NULL;
489
490 rt_mutex_set_owner(lock, pendowner, RT_MUTEX_OWNER_PENDING);
491
492 spin_unlock_irqrestore(&current->pi_lock, flags);
493
494 /*
495 * Clear the pi_blocked_on variable and enqueue a possible
496 * waiter into the pi_waiters list of the pending owner. This
497 * prevents that in case the pending owner gets unboosted a
498 * waiter with higher priority than pending-owner->normal_prio
499 * is blocked on the unboosted (pending) owner.
500 */
501 spin_lock_irqsave(&pendowner->pi_lock, flags);
502
503 WARN_ON(!pendowner->pi_blocked_on);
504 WARN_ON(pendowner->pi_blocked_on != waiter);
505 WARN_ON(pendowner->pi_blocked_on->lock != lock);
506
507 pendowner->pi_blocked_on = NULL;
508
509 if (rt_mutex_has_waiters(lock)) {
510 struct rt_mutex_waiter *next;
511
512 next = rt_mutex_top_waiter(lock);
513 plist_add(&next->pi_list_entry, &pendowner->pi_waiters);
514 }
515 spin_unlock_irqrestore(&pendowner->pi_lock, flags);
516
517 wake_up_process(pendowner);
518}
519
520/*
521 * Remove a waiter from a lock
522 *
523 * Must be called with lock->wait_lock held
524 */
525static void remove_waiter(struct rt_mutex *lock,
526 struct rt_mutex_waiter *waiter __IP_DECL__)
527{
528 int first = (waiter == rt_mutex_top_waiter(lock));
529 int boost = 0;
530 task_t *owner = rt_mutex_owner(lock);
531 unsigned long flags;
532
533 spin_lock_irqsave(&current->pi_lock, flags);
534 plist_del(&waiter->list_entry, &lock->wait_list);
535 waiter->task = NULL;
536 current->pi_blocked_on = NULL;
537 spin_unlock_irqrestore(&current->pi_lock, flags);
538
539 if (first && owner != current) {
540
541 spin_lock_irqsave(&owner->pi_lock, flags);
542
543 plist_del(&waiter->pi_list_entry, &owner->pi_waiters);
544
545 if (rt_mutex_has_waiters(lock)) {
546 struct rt_mutex_waiter *next;
547
548 next = rt_mutex_top_waiter(lock);
549 plist_add(&next->pi_list_entry, &owner->pi_waiters);
550 }
551 __rt_mutex_adjust_prio(owner);
552
553 if (owner->pi_blocked_on) {
554 boost = 1;
555 get_task_struct(owner);
556 }
557 spin_unlock_irqrestore(&owner->pi_lock, flags);
558 }
559
560 WARN_ON(!plist_node_empty(&waiter->pi_list_entry));
561
562 if (!boost)
563 return;
564
565 spin_unlock(&lock->wait_lock);
566
567 rt_mutex_adjust_prio_chain(owner, 0, lock, NULL __IP__);
568
569 spin_lock(&lock->wait_lock);
570}
571
572/*
573 * Slow path lock function:
574 */
575static int __sched
576rt_mutex_slowlock(struct rt_mutex *lock, int state,
577 struct hrtimer_sleeper *timeout,
578 int detect_deadlock __IP_DECL__)
579{
580 struct rt_mutex_waiter waiter;
581 int ret = 0;
582
583 debug_rt_mutex_init_waiter(&waiter);
584 waiter.task = NULL;
585
586 spin_lock(&lock->wait_lock);
587
588 /* Try to acquire the lock again: */
589 if (try_to_take_rt_mutex(lock __IP__)) {
590 spin_unlock(&lock->wait_lock);
591 return 0;
592 }
593
594 set_current_state(state);
595
596 /* Setup the timer, when timeout != NULL */
597 if (unlikely(timeout))
598 hrtimer_start(&timeout->timer, timeout->timer.expires,
599 HRTIMER_ABS);
600
601 for (;;) {
602 /* Try to acquire the lock: */
603 if (try_to_take_rt_mutex(lock __IP__))
604 break;
605
606 /*
607 * TASK_INTERRUPTIBLE checks for signals and
608 * timeout. Ignored otherwise.
609 */
610 if (unlikely(state == TASK_INTERRUPTIBLE)) {
611 /* Signal pending? */
612 if (signal_pending(current))
613 ret = -EINTR;
614 if (timeout && !timeout->task)
615 ret = -ETIMEDOUT;
616 if (ret)
617 break;
618 }
619
620 /*
621 * waiter.task is NULL the first time we come here and
622 * when we have been woken up by the previous owner
623 * but the lock got stolen by a higher prio task.
624 */
625 if (!waiter.task) {
626 ret = task_blocks_on_rt_mutex(lock, &waiter,
627 detect_deadlock __IP__);
628 /*
629 * If we got woken up by the owner then start loop
630 * all over without going into schedule to try
631 * to get the lock now:
632 */
633 if (unlikely(!waiter.task))
634 continue;
635
636 if (unlikely(ret))
637 break;
638 }
639 spin_unlock(&lock->wait_lock);
640
641 debug_rt_mutex_print_deadlock(&waiter);
642
643 schedule();
644
645 spin_lock(&lock->wait_lock);
646 set_current_state(state);
647 }
648
649 set_current_state(TASK_RUNNING);
650
651 if (unlikely(waiter.task))
652 remove_waiter(lock, &waiter __IP__);
653
654 /*
655 * try_to_take_rt_mutex() sets the waiter bit
656 * unconditionally. We might have to fix that up.
657 */
658 fixup_rt_mutex_waiters(lock);
659
660 spin_unlock(&lock->wait_lock);
661
662 /* Remove pending timer: */
663 if (unlikely(timeout))
664 hrtimer_cancel(&timeout->timer);
665
666 /*
667 * Readjust priority, when we did not get the lock. We might
668 * have been the pending owner and boosted. Since we did not
669 * take the lock, the PI boost has to go.
670 */
671 if (unlikely(ret))
672 rt_mutex_adjust_prio(current);
673
674 debug_rt_mutex_free_waiter(&waiter);
675
676 return ret;
677}
678
679/*
680 * Slow path try-lock function:
681 */
682static inline int
683rt_mutex_slowtrylock(struct rt_mutex *lock __IP_DECL__)
684{
685 int ret = 0;
686
687 spin_lock(&lock->wait_lock);
688
689 if (likely(rt_mutex_owner(lock) != current)) {
690
691 ret = try_to_take_rt_mutex(lock __IP__);
692 /*
693 * try_to_take_rt_mutex() sets the lock waiters
694 * bit unconditionally. Clean this up.
695 */
696 fixup_rt_mutex_waiters(lock);
697 }
698
699 spin_unlock(&lock->wait_lock);
700
701 return ret;
702}
703
704/*
705 * Slow path to release a rt-mutex:
706 */
707static void __sched
708rt_mutex_slowunlock(struct rt_mutex *lock)
709{
710 spin_lock(&lock->wait_lock);
711
712 debug_rt_mutex_unlock(lock);
713
714 rt_mutex_deadlock_account_unlock(current);
715
716 if (!rt_mutex_has_waiters(lock)) {
717 lock->owner = NULL;
718 spin_unlock(&lock->wait_lock);
719 return;
720 }
721
722 wakeup_next_waiter(lock);
723
724 spin_unlock(&lock->wait_lock);
725
726 /* Undo pi boosting if necessary: */
727 rt_mutex_adjust_prio(current);
728}
729
730/*
731 * debug aware fast / slowpath lock,trylock,unlock
732 *
733 * The atomic acquire/release ops are compiled away, when either the
734 * architecture does not support cmpxchg or when debugging is enabled.
735 */
736static inline int
737rt_mutex_fastlock(struct rt_mutex *lock, int state,
738 int detect_deadlock,
739 int (*slowfn)(struct rt_mutex *lock, int state,
740 struct hrtimer_sleeper *timeout,
741 int detect_deadlock __IP_DECL__))
742{
743 if (!detect_deadlock && likely(rt_mutex_cmpxchg(lock, NULL, current))) {
744 rt_mutex_deadlock_account_lock(lock, current);
745 return 0;
746 } else
747 return slowfn(lock, state, NULL, detect_deadlock __RET_IP__);
748}
749
750static inline int
751rt_mutex_timed_fastlock(struct rt_mutex *lock, int state,
752 struct hrtimer_sleeper *timeout, int detect_deadlock,
753 int (*slowfn)(struct rt_mutex *lock, int state,
754 struct hrtimer_sleeper *timeout,
755 int detect_deadlock __IP_DECL__))
756{
757 if (!detect_deadlock && likely(rt_mutex_cmpxchg(lock, NULL, current))) {
758 rt_mutex_deadlock_account_lock(lock, current);
759 return 0;
760 } else
761 return slowfn(lock, state, timeout, detect_deadlock __RET_IP__);
762}
763
764static inline int
765rt_mutex_fasttrylock(struct rt_mutex *lock,
766 int (*slowfn)(struct rt_mutex *lock __IP_DECL__))
767{
768 if (likely(rt_mutex_cmpxchg(lock, NULL, current))) {
769 rt_mutex_deadlock_account_lock(lock, current);
770 return 1;
771 }
772 return slowfn(lock __RET_IP__);
773}
774
775static inline void
776rt_mutex_fastunlock(struct rt_mutex *lock,
777 void (*slowfn)(struct rt_mutex *lock))
778{
779 if (likely(rt_mutex_cmpxchg(lock, current, NULL)))
780 rt_mutex_deadlock_account_unlock(current);
781 else
782 slowfn(lock);
783}
784
785/**
786 * rt_mutex_lock - lock a rt_mutex
787 *
788 * @lock: the rt_mutex to be locked
789 */
790void __sched rt_mutex_lock(struct rt_mutex *lock)
791{
792 might_sleep();
793
794 rt_mutex_fastlock(lock, TASK_UNINTERRUPTIBLE, 0, rt_mutex_slowlock);
795}
796EXPORT_SYMBOL_GPL(rt_mutex_lock);
797
798/**
799 * rt_mutex_lock_interruptible - lock a rt_mutex interruptible
800 *
801 * @lock: the rt_mutex to be locked
802 * @detect_deadlock: deadlock detection on/off
803 *
804 * Returns:
805 * 0 on success
806 * -EINTR when interrupted by a signal
807 * -EDEADLK when the lock would deadlock (when deadlock detection is on)
808 */
809int __sched rt_mutex_lock_interruptible(struct rt_mutex *lock,
810 int detect_deadlock)
811{
812 might_sleep();
813
814 return rt_mutex_fastlock(lock, TASK_INTERRUPTIBLE,
815 detect_deadlock, rt_mutex_slowlock);
816}
817EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible);
818
819/**
820 * rt_mutex_lock_interruptible_ktime - lock a rt_mutex interruptible
821 * the timeout structure is provided
822 * by the caller
823 *
824 * @lock: the rt_mutex to be locked
825 * @timeout: timeout structure or NULL (no timeout)
826 * @detect_deadlock: deadlock detection on/off
827 *
828 * Returns:
829 * 0 on success
830 * -EINTR when interrupted by a signal
831 * -ETIMEOUT when the timeout expired
832 * -EDEADLK when the lock would deadlock (when deadlock detection is on)
833 */
834int
835rt_mutex_timed_lock(struct rt_mutex *lock, struct hrtimer_sleeper *timeout,
836 int detect_deadlock)
837{
838 might_sleep();
839
840 return rt_mutex_timed_fastlock(lock, TASK_INTERRUPTIBLE, timeout,
841 detect_deadlock, rt_mutex_slowlock);
842}
843EXPORT_SYMBOL_GPL(rt_mutex_timed_lock);
844
845/**
846 * rt_mutex_trylock - try to lock a rt_mutex
847 *
848 * @lock: the rt_mutex to be locked
849 *
850 * Returns 1 on success and 0 on contention
851 */
852int __sched rt_mutex_trylock(struct rt_mutex *lock)
853{
854 return rt_mutex_fasttrylock(lock, rt_mutex_slowtrylock);
855}
856EXPORT_SYMBOL_GPL(rt_mutex_trylock);
857
858/**
859 * rt_mutex_unlock - unlock a rt_mutex
860 *
861 * @lock: the rt_mutex to be unlocked
862 */
863void __sched rt_mutex_unlock(struct rt_mutex *lock)
864{
865 rt_mutex_fastunlock(lock, rt_mutex_slowunlock);
866}
867EXPORT_SYMBOL_GPL(rt_mutex_unlock);
868
869/***
870 * rt_mutex_destroy - mark a mutex unusable
871 * @lock: the mutex to be destroyed
872 *
873 * This function marks the mutex uninitialized, and any subsequent
874 * use of the mutex is forbidden. The mutex must not be locked when
875 * this function is called.
876 */
877void rt_mutex_destroy(struct rt_mutex *lock)
878{
879 WARN_ON(rt_mutex_is_locked(lock));
880#ifdef CONFIG_DEBUG_RT_MUTEXES
881 lock->magic = NULL;
882#endif
883}
884
885EXPORT_SYMBOL_GPL(rt_mutex_destroy);
886
887/**
888 * __rt_mutex_init - initialize the rt lock
889 *
890 * @lock: the rt lock to be initialized
891 *
892 * Initialize the rt lock to unlocked state.
893 *
894 * Initializing of a locked rt lock is not allowed
895 */
896void __rt_mutex_init(struct rt_mutex *lock, const char *name)
897{
898 lock->owner = NULL;
899 spin_lock_init(&lock->wait_lock);
900 plist_head_init(&lock->wait_list, &lock->wait_lock);
901
902 debug_rt_mutex_init(lock, name);
903}
904EXPORT_SYMBOL_GPL(__rt_mutex_init);
diff --git a/kernel/rtmutex.h b/kernel/rtmutex.h
new file mode 100644
index 000000000000..1e0fca13ff72
--- /dev/null
+++ b/kernel/rtmutex.h
@@ -0,0 +1,29 @@
1/*
2 * RT-Mutexes: 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) 2006, Timesys Corp., Thomas Gleixner <tglx@timesys.com>
8 *
9 * This file contains macros used solely by rtmutex.c.
10 * Non-debug version.
11 */
12
13#define __IP_DECL__
14#define __IP__
15#define __RET_IP__
16#define rt_mutex_deadlock_check(l) (0)
17#define rt_mutex_deadlock_account_lock(m, t) do { } while (0)
18#define rt_mutex_deadlock_account_unlock(l) do { } while (0)
19#define debug_rt_mutex_init_waiter(w) do { } while (0)
20#define debug_rt_mutex_free_waiter(w) do { } while (0)
21#define debug_rt_mutex_lock(l) do { } while (0)
22#define debug_rt_mutex_proxy_lock(l,p) do { } while (0)
23#define debug_rt_mutex_proxy_unlock(l) do { } while (0)
24#define debug_rt_mutex_unlock(l) do { } while (0)
25#define debug_rt_mutex_init(m, n) do { } while (0)
26#define debug_rt_mutex_deadlock(d, a ,l) do { } while (0)
27#define debug_rt_mutex_print_deadlock(w) do { } while (0)
28#define debug_rt_mutex_detect_deadlock(w,d) (d)
29#define debug_rt_mutex_reset_waiter(w) do { } while (0)
diff --git a/kernel/rtmutex_common.h b/kernel/rtmutex_common.h
new file mode 100644
index 000000000000..50eed60eb085
--- /dev/null
+++ b/kernel/rtmutex_common.h
@@ -0,0 +1,93 @@
1/*
2 * RT Mutexes: 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) 2006, Timesys Corp., Thomas Gleixner <tglx@timesys.com>
8 *
9 * This file contains the private data structure and API definitions.
10 */
11
12#ifndef __KERNEL_RTMUTEX_COMMON_H
13#define __KERNEL_RTMUTEX_COMMON_H
14
15#include <linux/rtmutex.h>
16
17/*
18 * This is the control structure for tasks blocked on a rt_mutex,
19 * which is allocated on the kernel stack on of the blocked task.
20 *
21 * @list_entry: pi node to enqueue into the mutex waiters list
22 * @pi_list_entry: pi node to enqueue into the mutex owner waiters list
23 * @task: task reference to the blocked task
24 */
25struct rt_mutex_waiter {
26 struct plist_node list_entry;
27 struct plist_node pi_list_entry;
28 struct task_struct *task;
29 struct rt_mutex *lock;
30#ifdef CONFIG_DEBUG_RT_MUTEXES
31 unsigned long ip;
32 pid_t deadlock_task_pid;
33 struct rt_mutex *deadlock_lock;
34#endif
35};
36
37/*
38 * Various helpers to access the waiters-plist:
39 */
40static inline int rt_mutex_has_waiters(struct rt_mutex *lock)
41{
42 return !plist_head_empty(&lock->wait_list);
43}
44
45static inline struct rt_mutex_waiter *
46rt_mutex_top_waiter(struct rt_mutex *lock)
47{
48 struct rt_mutex_waiter *w;
49
50 w = plist_first_entry(&lock->wait_list, struct rt_mutex_waiter,
51 list_entry);
52 BUG_ON(w->lock != lock);
53
54 return w;
55}
56
57static inline int task_has_pi_waiters(struct task_struct *p)
58{
59 return !plist_head_empty(&p->pi_waiters);
60}
61
62static inline struct rt_mutex_waiter *
63task_top_pi_waiter(struct task_struct *p)
64{
65 return plist_first_entry(&p->pi_waiters, struct rt_mutex_waiter,
66 pi_list_entry);
67}
68
69/*
70 * lock->owner state tracking:
71 */
72#define RT_MUTEX_OWNER_PENDING 1UL
73#define RT_MUTEX_HAS_WAITERS 2UL
74#define RT_MUTEX_OWNER_MASKALL 3UL
75
76static inline struct task_struct *rt_mutex_owner(struct rt_mutex *lock)
77{
78 return (struct task_struct *)
79 ((unsigned long)lock->owner & ~RT_MUTEX_OWNER_MASKALL);
80}
81
82static inline struct task_struct *rt_mutex_real_owner(struct rt_mutex *lock)
83{
84 return (struct task_struct *)
85 ((unsigned long)lock->owner & ~RT_MUTEX_HAS_WAITERS);
86}
87
88static inline unsigned long rt_mutex_owner_pending(struct rt_mutex *lock)
89{
90 return (unsigned long)lock->owner & RT_MUTEX_OWNER_PENDING;
91}
92
93#endif
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index f54afed8426f..93a2c5398648 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -133,6 +133,10 @@ extern int acct_parm[];
133extern int no_unaligned_warning; 133extern int no_unaligned_warning;
134#endif 134#endif
135 135
136#ifdef CONFIG_RT_MUTEXES
137extern int max_lock_depth;
138#endif
139
136static int parse_table(int __user *, int, void __user *, size_t __user *, void __user *, size_t, 140static int parse_table(int __user *, int, void __user *, size_t __user *, void __user *, size_t,
137 ctl_table *, void **); 141 ctl_table *, void **);
138static int proc_doutsstring(ctl_table *table, int write, struct file *filp, 142static int proc_doutsstring(ctl_table *table, int write, struct file *filp,
@@ -688,6 +692,17 @@ static ctl_table kern_table[] = {
688 .proc_handler = &proc_dointvec, 692 .proc_handler = &proc_dointvec,
689 }, 693 },
690#endif 694#endif
695#ifdef CONFIG_RT_MUTEXES
696 {
697 .ctl_name = KERN_MAX_LOCK_DEPTH,
698 .procname = "max_lock_depth",
699 .data = &max_lock_depth,
700 .maxlen = sizeof(int),
701 .mode = 0644,
702 .proc_handler = &proc_dointvec,
703 },
704#endif
705
691 { .ctl_name = 0 } 706 { .ctl_name = 0 }
692}; 707};
693 708