]> bbs.cooldavid.org Git - net-next-2.6.git/blame - kernel/hrtimer.c
[PATCH] hrtimers: posix-timer: cleanup common_timer_get()
[net-next-2.6.git] / kernel / hrtimer.c
CommitLineData
c0a31329
TG
1/*
2 * linux/kernel/hrtimer.c
3 *
4 * Copyright(C) 2005, Thomas Gleixner <tglx@linutronix.de>
5 * Copyright(C) 2005, Red Hat, Inc., Ingo Molnar
6 *
7 * High-resolution kernel timers
8 *
9 * In contrast to the low-resolution timeout API implemented in
10 * kernel/timer.c, hrtimers provide finer resolution and accuracy
11 * depending on system configuration and capabilities.
12 *
13 * These timers are currently used for:
14 * - itimers
15 * - POSIX timers
16 * - nanosleep
17 * - precise in-kernel timing
18 *
19 * Started by: Thomas Gleixner and Ingo Molnar
20 *
21 * Credits:
22 * based on kernel/timer.c
23 *
66188fae
TG
24 * Help, testing, suggestions, bugfixes, improvements were
25 * provided by:
26 *
27 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
28 * et. al.
29 *
c0a31329
TG
30 * For licencing details see kernel-base/COPYING
31 */
32
33#include <linux/cpu.h>
34#include <linux/module.h>
35#include <linux/percpu.h>
36#include <linux/hrtimer.h>
37#include <linux/notifier.h>
38#include <linux/syscalls.h>
39#include <linux/interrupt.h>
40
41#include <asm/uaccess.h>
42
43/**
44 * ktime_get - get the monotonic time in ktime_t format
45 *
46 * returns the time in ktime_t format
47 */
48static ktime_t ktime_get(void)
49{
50 struct timespec now;
51
52 ktime_get_ts(&now);
53
54 return timespec_to_ktime(now);
55}
56
57/**
58 * ktime_get_real - get the real (wall-) time in ktime_t format
59 *
60 * returns the time in ktime_t format
61 */
62static ktime_t ktime_get_real(void)
63{
64 struct timespec now;
65
66 getnstimeofday(&now);
67
68 return timespec_to_ktime(now);
69}
70
71EXPORT_SYMBOL_GPL(ktime_get_real);
72
73/*
74 * The timer bases:
7978672c
GA
75 *
76 * Note: If we want to add new timer bases, we have to skip the two
77 * clock ids captured by the cpu-timers. We do this by holding empty
78 * entries rather than doing math adjustment of the clock ids.
79 * This ensures that we capture erroneous accesses to these clock ids
80 * rather than moving them into the range of valid clock id's.
c0a31329
TG
81 */
82
83#define MAX_HRTIMER_BASES 2
84
85static DEFINE_PER_CPU(struct hrtimer_base, hrtimer_bases[MAX_HRTIMER_BASES]) =
86{
87 {
88 .index = CLOCK_REALTIME,
89 .get_time = &ktime_get_real,
90 .resolution = KTIME_REALTIME_RES,
91 },
92 {
93 .index = CLOCK_MONOTONIC,
94 .get_time = &ktime_get,
95 .resolution = KTIME_MONOTONIC_RES,
96 },
97};
98
99/**
100 * ktime_get_ts - get the monotonic clock in timespec format
101 *
102 * @ts: pointer to timespec variable
103 *
104 * The function calculates the monotonic clock from the realtime
105 * clock and the wall_to_monotonic offset and stores the result
106 * in normalized timespec format in the variable pointed to by ts.
107 */
108void ktime_get_ts(struct timespec *ts)
109{
110 struct timespec tomono;
111 unsigned long seq;
112
113 do {
114 seq = read_seqbegin(&xtime_lock);
115 getnstimeofday(ts);
116 tomono = wall_to_monotonic;
117
118 } while (read_seqretry(&xtime_lock, seq));
119
120 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
121 ts->tv_nsec + tomono.tv_nsec);
122}
69778e32 123EXPORT_SYMBOL_GPL(ktime_get_ts);
c0a31329 124
92127c7a
TG
125/*
126 * Get the coarse grained time at the softirq based on xtime and
127 * wall_to_monotonic.
128 */
129static void hrtimer_get_softirq_time(struct hrtimer_base *base)
130{
131 ktime_t xtim, tomono;
132 unsigned long seq;
133
134 do {
135 seq = read_seqbegin(&xtime_lock);
136 xtim = timespec_to_ktime(xtime);
137 tomono = timespec_to_ktime(wall_to_monotonic);
138
139 } while (read_seqretry(&xtime_lock, seq));
140
141 base[CLOCK_REALTIME].softirq_time = xtim;
142 base[CLOCK_MONOTONIC].softirq_time = ktime_add(xtim, tomono);
143}
144
c0a31329
TG
145/*
146 * Functions and macros which are different for UP/SMP systems are kept in a
147 * single place
148 */
149#ifdef CONFIG_SMP
150
151#define set_curr_timer(b, t) do { (b)->curr_timer = (t); } while (0)
152
153/*
154 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
155 * means that all timers which are tied to this base via timer->base are
156 * locked, and the base itself is locked too.
157 *
158 * So __run_timers/migrate_timers can safely modify all timers which could
159 * be found on the lists/queues.
160 *
161 * When the timer's base is locked, and the timer removed from list, it is
162 * possible to set timer->base = NULL and drop the lock: the timer remains
163 * locked.
164 */
165static struct hrtimer_base *lock_hrtimer_base(const struct hrtimer *timer,
166 unsigned long *flags)
167{
168 struct hrtimer_base *base;
169
170 for (;;) {
171 base = timer->base;
172 if (likely(base != NULL)) {
173 spin_lock_irqsave(&base->lock, *flags);
174 if (likely(base == timer->base))
175 return base;
176 /* The timer has migrated to another CPU: */
177 spin_unlock_irqrestore(&base->lock, *flags);
178 }
179 cpu_relax();
180 }
181}
182
183/*
184 * Switch the timer base to the current CPU when possible.
185 */
186static inline struct hrtimer_base *
187switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_base *base)
188{
189 struct hrtimer_base *new_base;
190
191 new_base = &__get_cpu_var(hrtimer_bases[base->index]);
192
193 if (base != new_base) {
194 /*
195 * We are trying to schedule the timer on the local CPU.
196 * However we can't change timer's base while it is running,
197 * so we keep it on the same CPU. No hassle vs. reprogramming
198 * the event source in the high resolution case. The softirq
199 * code will take care of this when the timer function has
200 * completed. There is no conflict as we hold the lock until
201 * the timer is enqueued.
202 */
203 if (unlikely(base->curr_timer == timer))
204 return base;
205
206 /* See the comment in lock_timer_base() */
207 timer->base = NULL;
208 spin_unlock(&base->lock);
209 spin_lock(&new_base->lock);
210 timer->base = new_base;
211 }
212 return new_base;
213}
214
215#else /* CONFIG_SMP */
216
217#define set_curr_timer(b, t) do { } while (0)
218
219static inline struct hrtimer_base *
220lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
221{
222 struct hrtimer_base *base = timer->base;
223
224 spin_lock_irqsave(&base->lock, *flags);
225
226 return base;
227}
228
229#define switch_hrtimer_base(t, b) (b)
230
231#endif /* !CONFIG_SMP */
232
233/*
234 * Functions for the union type storage format of ktime_t which are
235 * too large for inlining:
236 */
237#if BITS_PER_LONG < 64
238# ifndef CONFIG_KTIME_SCALAR
239/**
240 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
241 *
242 * @kt: addend
243 * @nsec: the scalar nsec value to add
244 *
245 * Returns the sum of kt and nsec in ktime_t format
246 */
247ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
248{
249 ktime_t tmp;
250
251 if (likely(nsec < NSEC_PER_SEC)) {
252 tmp.tv64 = nsec;
253 } else {
254 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
255
256 tmp = ktime_set((long)nsec, rem);
257 }
258
259 return ktime_add(kt, tmp);
260}
261
262#else /* CONFIG_KTIME_SCALAR */
263
264# endif /* !CONFIG_KTIME_SCALAR */
265
266/*
267 * Divide a ktime value by a nanosecond value
268 */
269static unsigned long ktime_divns(const ktime_t kt, nsec_t div)
270{
271 u64 dclc, inc, dns;
272 int sft = 0;
273
274 dclc = dns = ktime_to_ns(kt);
275 inc = div;
276 /* Make sure the divisor is less than 2^32: */
277 while (div >> 32) {
278 sft++;
279 div >>= 1;
280 }
281 dclc >>= sft;
282 do_div(dclc, (unsigned long) div);
283
284 return (unsigned long) dclc;
285}
286
287#else /* BITS_PER_LONG < 64 */
288# define ktime_divns(kt, div) (unsigned long)((kt).tv64 / (div))
289#endif /* BITS_PER_LONG >= 64 */
290
291/*
292 * Counterpart to lock_timer_base above:
293 */
294static inline
295void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
296{
297 spin_unlock_irqrestore(&timer->base->lock, *flags);
298}
299
300/**
301 * hrtimer_forward - forward the timer expiry
302 *
303 * @timer: hrtimer to forward
44f21475 304 * @now: forward past this time
c0a31329
TG
305 * @interval: the interval to forward
306 *
307 * Forward the timer expiry so it will expire in the future.
8dca6f33 308 * Returns the number of overruns.
c0a31329
TG
309 */
310unsigned long
44f21475 311hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
c0a31329
TG
312{
313 unsigned long orun = 1;
44f21475 314 ktime_t delta;
c0a31329
TG
315
316 delta = ktime_sub(now, timer->expires);
317
318 if (delta.tv64 < 0)
319 return 0;
320
c9db4fa1
TG
321 if (interval.tv64 < timer->base->resolution.tv64)
322 interval.tv64 = timer->base->resolution.tv64;
323
c0a31329
TG
324 if (unlikely(delta.tv64 >= interval.tv64)) {
325 nsec_t incr = ktime_to_ns(interval);
326
327 orun = ktime_divns(delta, incr);
328 timer->expires = ktime_add_ns(timer->expires, incr * orun);
329 if (timer->expires.tv64 > now.tv64)
330 return orun;
331 /*
332 * This (and the ktime_add() below) is the
333 * correction for exact:
334 */
335 orun++;
336 }
337 timer->expires = ktime_add(timer->expires, interval);
338
339 return orun;
340}
341
342/*
343 * enqueue_hrtimer - internal function to (re)start a timer
344 *
345 * The timer is inserted in expiry order. Insertion into the
346 * red black tree is O(log(n)). Must hold the base lock.
347 */
348static void enqueue_hrtimer(struct hrtimer *timer, struct hrtimer_base *base)
349{
350 struct rb_node **link = &base->active.rb_node;
c0a31329
TG
351 struct rb_node *parent = NULL;
352 struct hrtimer *entry;
353
354 /*
355 * Find the right place in the rbtree:
356 */
357 while (*link) {
358 parent = *link;
359 entry = rb_entry(parent, struct hrtimer, node);
360 /*
361 * We dont care about collisions. Nodes with
362 * the same expiry time stay together.
363 */
364 if (timer->expires.tv64 < entry->expires.tv64)
365 link = &(*link)->rb_left;
288867ec 366 else
c0a31329 367 link = &(*link)->rb_right;
c0a31329
TG
368 }
369
370 /*
288867ec
TG
371 * Insert the timer to the rbtree and check whether it
372 * replaces the first pending timer
c0a31329
TG
373 */
374 rb_link_node(&timer->node, parent, link);
375 rb_insert_color(&timer->node, &base->active);
c0a31329
TG
376
377 timer->state = HRTIMER_PENDING;
c0a31329 378
288867ec
TG
379 if (!base->first || timer->expires.tv64 <
380 rb_entry(base->first, struct hrtimer, node)->expires.tv64)
381 base->first = &timer->node;
382}
c0a31329
TG
383
384/*
385 * __remove_hrtimer - internal function to remove a timer
386 *
387 * Caller must hold the base lock.
388 */
389static void __remove_hrtimer(struct hrtimer *timer, struct hrtimer_base *base)
390{
391 /*
288867ec
TG
392 * Remove the timer from the rbtree and replace the
393 * first entry pointer if necessary.
c0a31329 394 */
288867ec
TG
395 if (base->first == &timer->node)
396 base->first = rb_next(&timer->node);
c0a31329
TG
397 rb_erase(&timer->node, &base->active);
398}
399
400/*
401 * remove hrtimer, called with base lock held
402 */
403static inline int
404remove_hrtimer(struct hrtimer *timer, struct hrtimer_base *base)
405{
406 if (hrtimer_active(timer)) {
407 __remove_hrtimer(timer, base);
408 timer->state = HRTIMER_INACTIVE;
409 return 1;
410 }
411 return 0;
412}
413
414/**
415 * hrtimer_start - (re)start an relative timer on the current CPU
416 *
417 * @timer: the timer to be added
418 * @tim: expiry time
419 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
420 *
421 * Returns:
422 * 0 on success
423 * 1 when the timer was active
424 */
425int
426hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
427{
428 struct hrtimer_base *base, *new_base;
429 unsigned long flags;
430 int ret;
431
432 base = lock_hrtimer_base(timer, &flags);
433
434 /* Remove an active timer from the queue: */
435 ret = remove_hrtimer(timer, base);
436
437 /* Switch the timer base, if necessary: */
438 new_base = switch_hrtimer_base(timer, base);
439
06027bdd 440 if (mode == HRTIMER_REL) {
c0a31329 441 tim = ktime_add(tim, new_base->get_time());
06027bdd
IM
442 /*
443 * CONFIG_TIME_LOW_RES is a temporary way for architectures
444 * to signal that they simply return xtime in
445 * do_gettimeoffset(). In this case we want to round up by
446 * resolution when starting a relative timer, to avoid short
447 * timeouts. This will go away with the GTOD framework.
448 */
449#ifdef CONFIG_TIME_LOW_RES
450 tim = ktime_add(tim, base->resolution);
451#endif
452 }
c0a31329
TG
453 timer->expires = tim;
454
455 enqueue_hrtimer(timer, new_base);
456
457 unlock_hrtimer_base(timer, &flags);
458
459 return ret;
460}
461
462/**
463 * hrtimer_try_to_cancel - try to deactivate a timer
464 *
465 * @timer: hrtimer to stop
466 *
467 * Returns:
468 * 0 when the timer was not active
469 * 1 when the timer was active
470 * -1 when the timer is currently excuting the callback function and
471 * can not be stopped
472 */
473int hrtimer_try_to_cancel(struct hrtimer *timer)
474{
475 struct hrtimer_base *base;
476 unsigned long flags;
477 int ret = -1;
478
479 base = lock_hrtimer_base(timer, &flags);
480
481 if (base->curr_timer != timer)
482 ret = remove_hrtimer(timer, base);
483
484 unlock_hrtimer_base(timer, &flags);
485
486 return ret;
487
488}
489
490/**
491 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
492 *
493 * @timer: the timer to be cancelled
494 *
495 * Returns:
496 * 0 when the timer was not active
497 * 1 when the timer was active
498 */
499int hrtimer_cancel(struct hrtimer *timer)
500{
501 for (;;) {
502 int ret = hrtimer_try_to_cancel(timer);
503
504 if (ret >= 0)
505 return ret;
506 }
507}
508
509/**
510 * hrtimer_get_remaining - get remaining time for the timer
511 *
512 * @timer: the timer to read
513 */
514ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
515{
516 struct hrtimer_base *base;
517 unsigned long flags;
518 ktime_t rem;
519
520 base = lock_hrtimer_base(timer, &flags);
521 rem = ktime_sub(timer->expires, timer->base->get_time());
522 unlock_hrtimer_base(timer, &flags);
523
524 return rem;
525}
526
69239749
TL
527#ifdef CONFIG_NO_IDLE_HZ
528/**
529 * hrtimer_get_next_event - get the time until next expiry event
530 *
531 * Returns the delta to the next expiry event or KTIME_MAX if no timer
532 * is pending.
533 */
534ktime_t hrtimer_get_next_event(void)
535{
536 struct hrtimer_base *base = __get_cpu_var(hrtimer_bases);
537 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
538 unsigned long flags;
539 int i;
540
541 for (i = 0; i < MAX_HRTIMER_BASES; i++, base++) {
542 struct hrtimer *timer;
543
544 spin_lock_irqsave(&base->lock, flags);
545 if (!base->first) {
546 spin_unlock_irqrestore(&base->lock, flags);
547 continue;
548 }
549 timer = rb_entry(base->first, struct hrtimer, node);
550 delta.tv64 = timer->expires.tv64;
551 spin_unlock_irqrestore(&base->lock, flags);
552 delta = ktime_sub(delta, base->get_time());
553 if (delta.tv64 < mindelta.tv64)
554 mindelta.tv64 = delta.tv64;
555 }
556 if (mindelta.tv64 < 0)
557 mindelta.tv64 = 0;
558 return mindelta;
559}
560#endif
561
c0a31329 562/**
7978672c 563 * hrtimer_init - initialize a timer to the given clock
c0a31329 564 *
7978672c 565 * @timer: the timer to be initialized
c0a31329 566 * @clock_id: the clock to be used
7978672c 567 * @mode: timer mode abs/rel
c0a31329 568 */
7978672c
GA
569void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
570 enum hrtimer_mode mode)
c0a31329
TG
571{
572 struct hrtimer_base *bases;
573
7978672c
GA
574 memset(timer, 0, sizeof(struct hrtimer));
575
c0a31329 576 bases = per_cpu(hrtimer_bases, raw_smp_processor_id());
c0a31329 577
7978672c
GA
578 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_ABS)
579 clock_id = CLOCK_MONOTONIC;
580
581 timer->base = &bases[clock_id];
c0a31329
TG
582}
583
584/**
585 * hrtimer_get_res - get the timer resolution for a clock
586 *
587 * @which_clock: which clock to query
588 * @tp: pointer to timespec variable to store the resolution
589 *
590 * Store the resolution of the clock selected by which_clock in the
591 * variable pointed to by tp.
592 */
593int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
594{
595 struct hrtimer_base *bases;
596
c0a31329 597 bases = per_cpu(hrtimer_bases, raw_smp_processor_id());
e2787630 598 *tp = ktime_to_timespec(bases[which_clock].resolution);
c0a31329
TG
599
600 return 0;
601}
602
603/*
604 * Expire the per base hrtimer-queue:
605 */
606static inline void run_hrtimer_queue(struct hrtimer_base *base)
607{
288867ec 608 struct rb_node *node;
c0a31329 609
92127c7a
TG
610 if (base->get_softirq_time)
611 base->softirq_time = base->get_softirq_time();
612
c0a31329
TG
613 spin_lock_irq(&base->lock);
614
288867ec 615 while ((node = base->first)) {
c0a31329
TG
616 struct hrtimer *timer;
617 int (*fn)(void *);
618 int restart;
619 void *data;
620
288867ec 621 timer = rb_entry(node, struct hrtimer, node);
92127c7a 622 if (base->softirq_time.tv64 <= timer->expires.tv64)
c0a31329
TG
623 break;
624
625 fn = timer->function;
626 data = timer->data;
627 set_curr_timer(base, timer);
ff60a5dc 628 timer->state = HRTIMER_RUNNING;
c0a31329
TG
629 __remove_hrtimer(timer, base);
630 spin_unlock_irq(&base->lock);
631
632 /*
633 * fn == NULL is special case for the simplest timer
634 * variant - wake up process and do not restart:
635 */
636 if (!fn) {
637 wake_up_process(data);
638 restart = HRTIMER_NORESTART;
639 } else
640 restart = fn(data);
641
642 spin_lock_irq(&base->lock);
643
ff60a5dc
AM
644 /* Another CPU has added back the timer */
645 if (timer->state != HRTIMER_RUNNING)
646 continue;
647
c0a31329
TG
648 if (restart == HRTIMER_RESTART)
649 enqueue_hrtimer(timer, base);
650 else
651 timer->state = HRTIMER_EXPIRED;
652 }
653 set_curr_timer(base, NULL);
654 spin_unlock_irq(&base->lock);
655}
656
657/*
658 * Called from timer softirq every jiffy, expire hrtimers:
659 */
660void hrtimer_run_queues(void)
661{
662 struct hrtimer_base *base = __get_cpu_var(hrtimer_bases);
663 int i;
664
92127c7a
TG
665 hrtimer_get_softirq_time(base);
666
c0a31329
TG
667 for (i = 0; i < MAX_HRTIMER_BASES; i++)
668 run_hrtimer_queue(&base[i]);
669}
670
10c94ec1
TG
671/*
672 * Sleep related functions:
673 */
674
675/**
676 * schedule_hrtimer - sleep until timeout
677 *
678 * @timer: hrtimer variable initialized with the correct clock base
679 * @mode: timeout value is abs/rel
680 *
681 * Make the current task sleep until @timeout is
682 * elapsed.
683 *
684 * You can set the task state as follows -
685 *
686 * %TASK_UNINTERRUPTIBLE - at least @timeout is guaranteed to
687 * pass before the routine returns. The routine will return 0
688 *
689 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
690 * delivered to the current task. In this case the remaining time
691 * will be returned
692 *
693 * The current task state is guaranteed to be TASK_RUNNING when this
694 * routine returns.
695 */
696static ktime_t __sched
697schedule_hrtimer(struct hrtimer *timer, const enum hrtimer_mode mode)
698{
699 /* fn stays NULL, meaning single-shot wakeup: */
700 timer->data = current;
701
702 hrtimer_start(timer, timer->expires, mode);
703
704 schedule();
705 hrtimer_cancel(timer);
706
707 /* Return the remaining time: */
708 if (timer->state != HRTIMER_EXPIRED)
709 return ktime_sub(timer->expires, timer->base->get_time());
710 else
711 return (ktime_t) {.tv64 = 0 };
712}
713
714static inline ktime_t __sched
715schedule_hrtimer_interruptible(struct hrtimer *timer,
716 const enum hrtimer_mode mode)
717{
718 set_current_state(TASK_INTERRUPTIBLE);
719
720 return schedule_hrtimer(timer, mode);
721}
722
7978672c 723static long __sched nanosleep_restart(struct restart_block *restart)
10c94ec1 724{
ea13dbc8
IM
725 struct timespec __user *rmtp;
726 struct timespec tu;
10c94ec1
TG
727 void *rfn_save = restart->fn;
728 struct hrtimer timer;
729 ktime_t rem;
730
731 restart->fn = do_no_restart_syscall;
732
7978672c 733 hrtimer_init(&timer, (clockid_t) restart->arg3, HRTIMER_ABS);
10c94ec1
TG
734
735 timer.expires.tv64 = ((u64)restart->arg1 << 32) | (u64) restart->arg0;
736
737 rem = schedule_hrtimer_interruptible(&timer, HRTIMER_ABS);
738
739 if (rem.tv64 <= 0)
740 return 0;
741
742 rmtp = (struct timespec __user *) restart->arg2;
743 tu = ktime_to_timespec(rem);
744 if (rmtp && copy_to_user(rmtp, &tu, sizeof(tu)))
745 return -EFAULT;
746
747 restart->fn = rfn_save;
748
749 /* The other values in restart are already filled in */
750 return -ERESTART_RESTARTBLOCK;
751}
752
10c94ec1
TG
753long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
754 const enum hrtimer_mode mode, const clockid_t clockid)
755{
756 struct restart_block *restart;
757 struct hrtimer timer;
758 struct timespec tu;
759 ktime_t rem;
760
7978672c 761 hrtimer_init(&timer, clockid, mode);
10c94ec1
TG
762
763 timer.expires = timespec_to_ktime(*rqtp);
764
765 rem = schedule_hrtimer_interruptible(&timer, mode);
766 if (rem.tv64 <= 0)
767 return 0;
768
7978672c 769 /* Absolute timers do not update the rmtp value and restart: */
10c94ec1
TG
770 if (mode == HRTIMER_ABS)
771 return -ERESTARTNOHAND;
772
773 tu = ktime_to_timespec(rem);
774
775 if (rmtp && copy_to_user(rmtp, &tu, sizeof(tu)))
776 return -EFAULT;
777
778 restart = &current_thread_info()->restart_block;
7978672c 779 restart->fn = nanosleep_restart;
10c94ec1
TG
780 restart->arg0 = timer.expires.tv64 & 0xFFFFFFFF;
781 restart->arg1 = timer.expires.tv64 >> 32;
782 restart->arg2 = (unsigned long) rmtp;
7978672c 783 restart->arg3 = (unsigned long) timer.base->index;
10c94ec1
TG
784
785 return -ERESTART_RESTARTBLOCK;
786}
787
6ba1b912
TG
788asmlinkage long
789sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
790{
791 struct timespec tu;
792
793 if (copy_from_user(&tu, rqtp, sizeof(tu)))
794 return -EFAULT;
795
796 if (!timespec_valid(&tu))
797 return -EINVAL;
798
799 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_REL, CLOCK_MONOTONIC);
800}
801
c0a31329
TG
802/*
803 * Functions related to boot-time initialization:
804 */
805static void __devinit init_hrtimers_cpu(int cpu)
806{
807 struct hrtimer_base *base = per_cpu(hrtimer_bases, cpu);
808 int i;
809
7978672c 810 for (i = 0; i < MAX_HRTIMER_BASES; i++, base++)
c0a31329 811 spin_lock_init(&base->lock);
c0a31329
TG
812}
813
814#ifdef CONFIG_HOTPLUG_CPU
815
816static void migrate_hrtimer_list(struct hrtimer_base *old_base,
817 struct hrtimer_base *new_base)
818{
819 struct hrtimer *timer;
820 struct rb_node *node;
821
822 while ((node = rb_first(&old_base->active))) {
823 timer = rb_entry(node, struct hrtimer, node);
824 __remove_hrtimer(timer, old_base);
825 timer->base = new_base;
826 enqueue_hrtimer(timer, new_base);
827 }
828}
829
830static void migrate_hrtimers(int cpu)
831{
832 struct hrtimer_base *old_base, *new_base;
833 int i;
834
835 BUG_ON(cpu_online(cpu));
836 old_base = per_cpu(hrtimer_bases, cpu);
837 new_base = get_cpu_var(hrtimer_bases);
838
839 local_irq_disable();
840
841 for (i = 0; i < MAX_HRTIMER_BASES; i++) {
842
843 spin_lock(&new_base->lock);
844 spin_lock(&old_base->lock);
845
846 BUG_ON(old_base->curr_timer);
847
848 migrate_hrtimer_list(old_base, new_base);
849
850 spin_unlock(&old_base->lock);
851 spin_unlock(&new_base->lock);
852 old_base++;
853 new_base++;
854 }
855
856 local_irq_enable();
857 put_cpu_var(hrtimer_bases);
858}
859#endif /* CONFIG_HOTPLUG_CPU */
860
861static int __devinit hrtimer_cpu_notify(struct notifier_block *self,
862 unsigned long action, void *hcpu)
863{
864 long cpu = (long)hcpu;
865
866 switch (action) {
867
868 case CPU_UP_PREPARE:
869 init_hrtimers_cpu(cpu);
870 break;
871
872#ifdef CONFIG_HOTPLUG_CPU
873 case CPU_DEAD:
874 migrate_hrtimers(cpu);
875 break;
876#endif
877
878 default:
879 break;
880 }
881
882 return NOTIFY_OK;
883}
884
885static struct notifier_block __devinitdata hrtimers_nb = {
886 .notifier_call = hrtimer_cpu_notify,
887};
888
889void __init hrtimers_init(void)
890{
891 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
892 (void *)(long)smp_processor_id());
893 register_cpu_notifier(&hrtimers_nb);
894}
895