projects/linux·blob v2.6.23-rc7 · refs

1447 lines · 35.2 KiB

1/*
2 * linux/kernel/hrtimer.c
3 *
4 * Copyright(C) 2005-2006, Thomas Gleixner <[email protected]>
5 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
6 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
7 *
8 * High-resolution kernel timers
9 *
10 * In contrast to the low-resolution timeout API implemented in
11 * kernel/timer.c, hrtimers provide finer resolution and accuracy
12 * depending on system configuration and capabilities.
13 *
14 * These timers are currently used for:
15 * - itimers
16 * - POSIX timers
17 * - nanosleep
18 * - precise in-kernel timing
19 *
20 * Started by: Thomas Gleixner and Ingo Molnar
21 *
22 * Credits:
23 * based on kernel/timer.c
24 *
25 * Help, testing, suggestions, bugfixes, improvements were
26 * provided by:
27 *
28 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
29 * et. al.
30 *
31 * For licencing details see kernel-base/COPYING
32 */
33
34#include <linux/cpu.h>
35#include <linux/irq.h>
36#include <linux/module.h>
37#include <linux/percpu.h>
38#include <linux/hrtimer.h>
39#include <linux/notifier.h>
40#include <linux/syscalls.h>
41#include <linux/kallsyms.h>
42#include <linux/interrupt.h>
43#include <linux/tick.h>
44#include <linux/seq_file.h>
45#include <linux/err.h>
46
47#include <asm/uaccess.h>
48
49/**
50 * ktime_get - get the monotonic time in ktime_t format
51 *
52 * returns the time in ktime_t format
53 */
54ktime_t ktime_get(void)
55{
56 struct timespec now;
57
58 ktime_get_ts(&now);
59
60 return timespec_to_ktime(now);
61}
62EXPORT_SYMBOL_GPL(ktime_get);
63
64/**
65 * ktime_get_real - get the real (wall-) time in ktime_t format
66 *
67 * returns the time in ktime_t format
68 */
69ktime_t ktime_get_real(void)
70{
71 struct timespec now;
72
73 getnstimeofday(&now);
74
75 return timespec_to_ktime(now);
76}
77
78EXPORT_SYMBOL_GPL(ktime_get_real);
79
80/*
81 * The timer bases:
82 *
83 * Note: If we want to add new timer bases, we have to skip the two
84 * clock ids captured by the cpu-timers. We do this by holding empty
85 * entries rather than doing math adjustment of the clock ids.
86 * This ensures that we capture erroneous accesses to these clock ids
87 * rather than moving them into the range of valid clock id's.
88 */
89DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
90{
91
92 .clock_base =
93 {
94 {
95 .index = CLOCK_REALTIME,
96 .get_time = &ktime_get_real,
97 .resolution = KTIME_LOW_RES,
98 },
99 {
100 .index = CLOCK_MONOTONIC,
101 .get_time = &ktime_get,
102 .resolution = KTIME_LOW_RES,
103 },
104 }
105};
106
107/**
108 * ktime_get_ts - get the monotonic clock in timespec format
109 * @ts: pointer to timespec variable
110 *
111 * The function calculates the monotonic clock from the realtime
112 * clock and the wall_to_monotonic offset and stores the result
113 * in normalized timespec format in the variable pointed to by @ts.
114 */
115void ktime_get_ts(struct timespec *ts)
116{
117 struct timespec tomono;
118 unsigned long seq;
119
120 do {
121 seq = read_seqbegin(&xtime_lock);
122 getnstimeofday(ts);
123 tomono = wall_to_monotonic;
124
125 } while (read_seqretry(&xtime_lock, seq));
126
127 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
128 ts->tv_nsec + tomono.tv_nsec);
129}
130EXPORT_SYMBOL_GPL(ktime_get_ts);
131
132/*
133 * Get the coarse grained time at the softirq based on xtime and
134 * wall_to_monotonic.
135 */
136static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
137{
138 ktime_t xtim, tomono;
139 struct timespec xts, tom;
140 unsigned long seq;
141
142 do {
143 seq = read_seqbegin(&xtime_lock);
144 xts = current_kernel_time();
145 tom = wall_to_monotonic;
146 } while (read_seqretry(&xtime_lock, seq));
147
148 xtim = timespec_to_ktime(xts);
149 tomono = timespec_to_ktime(tom);
150 base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
151 base->clock_base[CLOCK_MONOTONIC].softirq_time =
152 ktime_add(xtim, tomono);
153}
154
155/*
156 * Helper function to check, whether the timer is running the callback
157 * function
158 */
159static inline int hrtimer_callback_running(struct hrtimer *timer)
160{
161 return timer->state & HRTIMER_STATE_CALLBACK;
162}
163
164/*
165 * Functions and macros which are different for UP/SMP systems are kept in a
166 * single place
167 */
168#ifdef CONFIG_SMP
169
170/*
171 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
172 * means that all timers which are tied to this base via timer->base are
173 * locked, and the base itself is locked too.
174 *
175 * So __run_timers/migrate_timers can safely modify all timers which could
176 * be found on the lists/queues.
177 *
178 * When the timer's base is locked, and the timer removed from list, it is
179 * possible to set timer->base = NULL and drop the lock: the timer remains
180 * locked.
181 */
182static
183struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
184 unsigned long *flags)
185{
186 struct hrtimer_clock_base *base;
187
188 for (;;) {
189 base = timer->base;
190 if (likely(base != NULL)) {
191 spin_lock_irqsave(&base->cpu_base->lock, *flags);
192 if (likely(base == timer->base))
193 return base;
194 /* The timer has migrated to another CPU: */
195 spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
196 }
197 cpu_relax();
198 }
199}
200
201/*
202 * Switch the timer base to the current CPU when possible.
203 */
204static inline struct hrtimer_clock_base *
205switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base)
206{
207 struct hrtimer_clock_base *new_base;
208 struct hrtimer_cpu_base *new_cpu_base;
209
210 new_cpu_base = &__get_cpu_var(hrtimer_bases);
211 new_base = &new_cpu_base->clock_base[base->index];
212
213 if (base != new_base) {
214 /*
215 * We are trying to schedule the timer on the local CPU.
216 * However we can't change timer's base while it is running,
217 * so we keep it on the same CPU. No hassle vs. reprogramming
218 * the event source in the high resolution case. The softirq
219 * code will take care of this when the timer function has
220 * completed. There is no conflict as we hold the lock until
221 * the timer is enqueued.
222 */
223 if (unlikely(hrtimer_callback_running(timer)))
224 return base;
225
226 /* See the comment in lock_timer_base() */
227 timer->base = NULL;
228 spin_unlock(&base->cpu_base->lock);
229 spin_lock(&new_base->cpu_base->lock);
230 timer->base = new_base;
231 }
232 return new_base;
233}
234
235#else /* CONFIG_SMP */
236
237static inline struct hrtimer_clock_base *
238lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
239{
240 struct hrtimer_clock_base *base = timer->base;
241
242 spin_lock_irqsave(&base->cpu_base->lock, *flags);
243
244 return base;
245}
246
247# define switch_hrtimer_base(t, b) (b)
248
249#endif /* !CONFIG_SMP */
250
251/*
252 * Functions for the union type storage format of ktime_t which are
253 * too large for inlining:
254 */
255#if BITS_PER_LONG < 64
256# ifndef CONFIG_KTIME_SCALAR
257/**
258 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
259 * @kt: addend
260 * @nsec: the scalar nsec value to add
261 *
262 * Returns the sum of kt and nsec in ktime_t format
263 */
264ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
265{
266 ktime_t tmp;
267
268 if (likely(nsec < NSEC_PER_SEC)) {
269 tmp.tv64 = nsec;
270 } else {
271 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
272
273 tmp = ktime_set((long)nsec, rem);
274 }
275
276 return ktime_add(kt, tmp);
277}
278
279EXPORT_SYMBOL_GPL(ktime_add_ns);
280# endif /* !CONFIG_KTIME_SCALAR */
281
282/*
283 * Divide a ktime value by a nanosecond value
284 */
285unsigned long ktime_divns(const ktime_t kt, s64 div)
286{
287 u64 dclc, inc, dns;
288 int sft = 0;
289
290 dclc = dns = ktime_to_ns(kt);
291 inc = div;
292 /* Make sure the divisor is less than 2^32: */
293 while (div >> 32) {
294 sft++;
295 div >>= 1;
296 }
297 dclc >>= sft;
298 do_div(dclc, (unsigned long) div);
299
300 return (unsigned long) dclc;
301}
302#endif /* BITS_PER_LONG >= 64 */
303
304/* High resolution timer related functions */
305#ifdef CONFIG_HIGH_RES_TIMERS
306
307/*
308 * High resolution timer enabled ?
309 */
310static int hrtimer_hres_enabled __read_mostly = 1;
311
312/*
313 * Enable / Disable high resolution mode
314 */
315static int __init setup_hrtimer_hres(char *str)
316{
317 if (!strcmp(str, "off"))
318 hrtimer_hres_enabled = 0;
319 else if (!strcmp(str, "on"))
320 hrtimer_hres_enabled = 1;
321 else
322 return 0;
323 return 1;
324}
325
326__setup("highres=", setup_hrtimer_hres);
327
328/*
329 * hrtimer_high_res_enabled - query, if the highres mode is enabled
330 */
331static inline int hrtimer_is_hres_enabled(void)
332{
333 return hrtimer_hres_enabled;
334}
335
336/*
337 * Is the high resolution mode active ?
338 */
339static inline int hrtimer_hres_active(void)
340{
341 return __get_cpu_var(hrtimer_bases).hres_active;
342}
343
344/*
345 * Reprogram the event source with checking both queues for the
346 * next event
347 * Called with interrupts disabled and base->lock held
348 */
349static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base)
350{
351 int i;
352 struct hrtimer_clock_base *base = cpu_base->clock_base;
353 ktime_t expires;
354
355 cpu_base->expires_next.tv64 = KTIME_MAX;
356
357 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
358 struct hrtimer *timer;
359
360 if (!base->first)
361 continue;
362 timer = rb_entry(base->first, struct hrtimer, node);
363 expires = ktime_sub(timer->expires, base->offset);
364 if (expires.tv64 < cpu_base->expires_next.tv64)
365 cpu_base->expires_next = expires;
366 }
367
368 if (cpu_base->expires_next.tv64 != KTIME_MAX)
369 tick_program_event(cpu_base->expires_next, 1);
370}
371
372/*
373 * Shared reprogramming for clock_realtime and clock_monotonic
374 *
375 * When a timer is enqueued and expires earlier than the already enqueued
376 * timers, we have to check, whether it expires earlier than the timer for
377 * which the clock event device was armed.
378 *
379 * Called with interrupts disabled and base->cpu_base.lock held
380 */
381static int hrtimer_reprogram(struct hrtimer *timer,
382 struct hrtimer_clock_base *base)
383{
384 ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next;
385 ktime_t expires = ktime_sub(timer->expires, base->offset);
386 int res;
387
388 /*
389 * When the callback is running, we do not reprogram the clock event
390 * device. The timer callback is either running on a different CPU or
391 * the callback is executed in the hrtimer_interupt context. The
392 * reprogramming is handled either by the softirq, which called the
393 * callback or at the end of the hrtimer_interrupt.
394 */
395 if (hrtimer_callback_running(timer))
396 return 0;
397
398 if (expires.tv64 >= expires_next->tv64)
399 return 0;
400
401 /*
402 * Clockevents returns -ETIME, when the event was in the past.
403 */
404 res = tick_program_event(expires, 0);
405 if (!IS_ERR_VALUE(res))
406 *expires_next = expires;
407 return res;
408}
409
410
411/*
412 * Retrigger next event is called after clock was set
413 *
414 * Called with interrupts disabled via on_each_cpu()
415 */
416static void retrigger_next_event(void *arg)
417{
418 struct hrtimer_cpu_base *base;
419 struct timespec realtime_offset;
420 unsigned long seq;
421
422 if (!hrtimer_hres_active())
423 return;
424
425 do {
426 seq = read_seqbegin(&xtime_lock);
427 set_normalized_timespec(&realtime_offset,
428 -wall_to_monotonic.tv_sec,
429 -wall_to_monotonic.tv_nsec);
430 } while (read_seqretry(&xtime_lock, seq));
431
432 base = &__get_cpu_var(hrtimer_bases);
433
434 /* Adjust CLOCK_REALTIME offset */
435 spin_lock(&base->lock);
436 base->clock_base[CLOCK_REALTIME].offset =
437 timespec_to_ktime(realtime_offset);
438
439 hrtimer_force_reprogram(base);
440 spin_unlock(&base->lock);
441}
442
443/*
444 * Clock realtime was set
445 *
446 * Change the offset of the realtime clock vs. the monotonic
447 * clock.
448 *
449 * We might have to reprogram the high resolution timer interrupt. On
450 * SMP we call the architecture specific code to retrigger _all_ high
451 * resolution timer interrupts. On UP we just disable interrupts and
452 * call the high resolution interrupt code.
453 */
454void clock_was_set(void)
455{
456 /* Retrigger the CPU local events everywhere */
457 on_each_cpu(retrigger_next_event, NULL, 0, 1);
458}
459
460/*
461 * During resume we might have to reprogram the high resolution timer
462 * interrupt (on the local CPU):
463 */
464void hres_timers_resume(void)
465{
466 WARN_ON_ONCE(num_online_cpus() > 1);
467
468 /* Retrigger the CPU local events: */
469 retrigger_next_event(NULL);
470}
471
472/*
473 * Check, whether the timer is on the callback pending list
474 */
475static inline int hrtimer_cb_pending(const struct hrtimer *timer)
476{
477 return timer->state & HRTIMER_STATE_PENDING;
478}
479
480/*
481 * Remove a timer from the callback pending list
482 */
483static inline void hrtimer_remove_cb_pending(struct hrtimer *timer)
484{
485 list_del_init(&timer->cb_entry);
486}
487
488/*
489 * Initialize the high resolution related parts of cpu_base
490 */
491static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
492{
493 base->expires_next.tv64 = KTIME_MAX;
494 base->hres_active = 0;
495 INIT_LIST_HEAD(&base->cb_pending);
496}
497
498/*
499 * Initialize the high resolution related parts of a hrtimer
500 */
501static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
502{
503 INIT_LIST_HEAD(&timer->cb_entry);
504}
505
506/*
507 * When High resolution timers are active, try to reprogram. Note, that in case
508 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
509 * check happens. The timer gets enqueued into the rbtree. The reprogramming
510 * and expiry check is done in the hrtimer_interrupt or in the softirq.
511 */
512static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
513 struct hrtimer_clock_base *base)
514{
515 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
516
517 /* Timer is expired, act upon the callback mode */
518 switch(timer->cb_mode) {
519 case HRTIMER_CB_IRQSAFE_NO_RESTART:
520 /*
521 * We can call the callback from here. No restart
522 * happens, so no danger of recursion
523 */
524 BUG_ON(timer->function(timer) != HRTIMER_NORESTART);
525 return 1;
526 case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ:
527 /*
528 * This is solely for the sched tick emulation with
529 * dynamic tick support to ensure that we do not
530 * restart the tick right on the edge and end up with
531 * the tick timer in the softirq ! The calling site
532 * takes care of this.
533 */
534 return 1;
535 case HRTIMER_CB_IRQSAFE:
536 case HRTIMER_CB_SOFTIRQ:
537 /*
538 * Move everything else into the softirq pending list !
539 */
540 list_add_tail(&timer->cb_entry,
541 &base->cpu_base->cb_pending);
542 timer->state = HRTIMER_STATE_PENDING;
543 raise_softirq(HRTIMER_SOFTIRQ);
544 return 1;
545 default:
546 BUG();
547 }
548 }
549 return 0;
550}
551
552/*
553 * Switch to high resolution mode
554 */
555static int hrtimer_switch_to_hres(void)
556{
557 int cpu = smp_processor_id();
558 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
559 unsigned long flags;
560
561 if (base->hres_active)
562 return 1;
563
564 local_irq_save(flags);
565
566 if (tick_init_highres()) {
567 local_irq_restore(flags);
568 printk(KERN_WARNING "Could not switch to high resolution "
569 "mode on CPU %d\n", cpu);
570 return 0;
571 }
572 base->hres_active = 1;
573 base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
574 base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
575
576 tick_setup_sched_timer();
577
578 /* "Retrigger" the interrupt to get things going */
579 retrigger_next_event(NULL);
580 local_irq_restore(flags);
581 printk(KERN_INFO "Switched to high resolution mode on CPU %d\n",
582 smp_processor_id());
583 return 1;
584}
585
586#else
587
588static inline int hrtimer_hres_active(void) { return 0; }
589static inline int hrtimer_is_hres_enabled(void) { return 0; }
590static inline int hrtimer_switch_to_hres(void) { return 0; }
591static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { }
592static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
593 struct hrtimer_clock_base *base)
594{
595 return 0;
596}
597static inline int hrtimer_cb_pending(struct hrtimer *timer) { return 0; }
598static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) { }
599static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
600static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
601
602#endif /* CONFIG_HIGH_RES_TIMERS */
603
604#ifdef CONFIG_TIMER_STATS
605void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr)
606{
607 if (timer->start_site)
608 return;
609
610 timer->start_site = addr;
611 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
612 timer->start_pid = current->pid;
613}
614#endif
615
616/*
617 * Counterpart to lock_timer_base above:
618 */
619static inline
620void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
621{
622 spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
623}
624
625/**
626 * hrtimer_forward - forward the timer expiry
627 * @timer: hrtimer to forward
628 * @now: forward past this time
629 * @interval: the interval to forward
630 *
631 * Forward the timer expiry so it will expire in the future.
632 * Returns the number of overruns.
633 */
634unsigned long
635hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
636{
637 unsigned long orun = 1;
638 ktime_t delta;
639
640 delta = ktime_sub(now, timer->expires);
641
642 if (delta.tv64 < 0)
643 return 0;
644
645 if (interval.tv64 < timer->base->resolution.tv64)
646 interval.tv64 = timer->base->resolution.tv64;
647
648 if (unlikely(delta.tv64 >= interval.tv64)) {
649 s64 incr = ktime_to_ns(interval);
650
651 orun = ktime_divns(delta, incr);
652 timer->expires = ktime_add_ns(timer->expires, incr * orun);
653 if (timer->expires.tv64 > now.tv64)
654 return orun;
655 /*
656 * This (and the ktime_add() below) is the
657 * correction for exact:
658 */
659 orun++;
660 }
661 timer->expires = ktime_add(timer->expires, interval);
662 /*
663 * Make sure, that the result did not wrap with a very large
664 * interval.
665 */
666 if (timer->expires.tv64 < 0)
667 timer->expires = ktime_set(KTIME_SEC_MAX, 0);
668
669 return orun;
670}
671EXPORT_SYMBOL_GPL(hrtimer_forward);
672
673/*
674 * enqueue_hrtimer - internal function to (re)start a timer
675 *
676 * The timer is inserted in expiry order. Insertion into the
677 * red black tree is O(log(n)). Must hold the base lock.
678 */
679static void enqueue_hrtimer(struct hrtimer *timer,
680 struct hrtimer_clock_base *base, int reprogram)
681{
682 struct rb_node **link = &base->active.rb_node;
683 struct rb_node *parent = NULL;
684 struct hrtimer *entry;
685 int leftmost = 1;
686
687 /*
688 * Find the right place in the rbtree:
689 */
690 while (*link) {
691 parent = *link;
692 entry = rb_entry(parent, struct hrtimer, node);
693 /*
694 * We dont care about collisions. Nodes with
695 * the same expiry time stay together.
696 */
697 if (timer->expires.tv64 < entry->expires.tv64) {
698 link = &(*link)->rb_left;
699 } else {
700 link = &(*link)->rb_right;
701 leftmost = 0;
702 }
703 }
704
705 /*
706 * Insert the timer to the rbtree and check whether it
707 * replaces the first pending timer
708 */
709 if (leftmost) {
710 /*
711 * Reprogram the clock event device. When the timer is already
712 * expired hrtimer_enqueue_reprogram has either called the
713 * callback or added it to the pending list and raised the
714 * softirq.
715 *
716 * This is a NOP for !HIGHRES
717 */
718 if (reprogram && hrtimer_enqueue_reprogram(timer, base))
719 return;
720
721 base->first = &timer->node;
722 }
723
724 rb_link_node(&timer->node, parent, link);
725 rb_insert_color(&timer->node, &base->active);
726 /*
727 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
728 * state of a possibly running callback.
729 */
730 timer->state |= HRTIMER_STATE_ENQUEUED;
731}
732
733/*
734 * __remove_hrtimer - internal function to remove a timer
735 *
736 * Caller must hold the base lock.
737 *
738 * High resolution timer mode reprograms the clock event device when the
739 * timer is the one which expires next. The caller can disable this by setting
740 * reprogram to zero. This is useful, when the context does a reprogramming
741 * anyway (e.g. timer interrupt)
742 */
743static void __remove_hrtimer(struct hrtimer *timer,
744 struct hrtimer_clock_base *base,
745 unsigned long newstate, int reprogram)
746{
747 /* High res. callback list. NOP for !HIGHRES */
748 if (hrtimer_cb_pending(timer))
749 hrtimer_remove_cb_pending(timer);
750 else {
751 /*
752 * Remove the timer from the rbtree and replace the
753 * first entry pointer if necessary.
754 */
755 if (base->first == &timer->node) {
756 base->first = rb_next(&timer->node);
757 /* Reprogram the clock event device. if enabled */
758 if (reprogram && hrtimer_hres_active())
759 hrtimer_force_reprogram(base->cpu_base);
760 }
761 rb_erase(&timer->node, &base->active);
762 }
763 timer->state = newstate;
764}
765
766/*
767 * remove hrtimer, called with base lock held
768 */
769static inline int
770remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
771{
772 if (hrtimer_is_queued(timer)) {
773 int reprogram;
774
775 /*
776 * Remove the timer and force reprogramming when high
777 * resolution mode is active and the timer is on the current
778 * CPU. If we remove a timer on another CPU, reprogramming is
779 * skipped. The interrupt event on this CPU is fired and
780 * reprogramming happens in the interrupt handler. This is a
781 * rare case and less expensive than a smp call.
782 */
783 timer_stats_hrtimer_clear_start_info(timer);
784 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
785 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
786 reprogram);
787 return 1;
788 }
789 return 0;
790}
791
792/**
793 * hrtimer_start - (re)start an relative timer on the current CPU
794 * @timer: the timer to be added
795 * @tim: expiry time
796 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
797 *
798 * Returns:
799 * 0 on success
800 * 1 when the timer was active
801 */
802int
803hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
804{
805 struct hrtimer_clock_base *base, *new_base;
806 unsigned long flags;
807 int ret;
808
809 base = lock_hrtimer_base(timer, &flags);
810
811 /* Remove an active timer from the queue: */
812 ret = remove_hrtimer(timer, base);
813
814 /* Switch the timer base, if necessary: */
815 new_base = switch_hrtimer_base(timer, base);
816
817 if (mode == HRTIMER_MODE_REL) {
818 tim = ktime_add(tim, new_base->get_time());
819 /*
820 * CONFIG_TIME_LOW_RES is a temporary way for architectures
821 * to signal that they simply return xtime in
822 * do_gettimeoffset(). In this case we want to round up by
823 * resolution when starting a relative timer, to avoid short
824 * timeouts. This will go away with the GTOD framework.
825 */
826#ifdef CONFIG_TIME_LOW_RES
827 tim = ktime_add(tim, base->resolution);
828#endif
829 }
830 timer->expires = tim;
831
832 timer_stats_hrtimer_set_start_info(timer);
833
834 /*
835 * Only allow reprogramming if the new base is on this CPU.
836 * (it might still be on another CPU if the timer was pending)
837 */
838 enqueue_hrtimer(timer, new_base,
839 new_base->cpu_base == &__get_cpu_var(hrtimer_bases));
840
841 unlock_hrtimer_base(timer, &flags);
842
843 return ret;
844}
845EXPORT_SYMBOL_GPL(hrtimer_start);
846
847/**
848 * hrtimer_try_to_cancel - try to deactivate a timer
849 * @timer: hrtimer to stop
850 *
851 * Returns:
852 * 0 when the timer was not active
853 * 1 when the timer was active
854 * -1 when the timer is currently excuting the callback function and
855 * cannot be stopped
856 */
857int hrtimer_try_to_cancel(struct hrtimer *timer)
858{
859 struct hrtimer_clock_base *base;
860 unsigned long flags;
861 int ret = -1;
862
863 base = lock_hrtimer_base(timer, &flags);
864
865 if (!hrtimer_callback_running(timer))
866 ret = remove_hrtimer(timer, base);
867
868 unlock_hrtimer_base(timer, &flags);
869
870 return ret;
871
872}
873EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
874
875/**
876 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
877 * @timer: the timer to be cancelled
878 *
879 * Returns:
880 * 0 when the timer was not active
881 * 1 when the timer was active
882 */
883int hrtimer_cancel(struct hrtimer *timer)
884{
885 for (;;) {
886 int ret = hrtimer_try_to_cancel(timer);
887
888 if (ret >= 0)
889 return ret;
890 cpu_relax();
891 }
892}
893EXPORT_SYMBOL_GPL(hrtimer_cancel);
894
895/**
896 * hrtimer_get_remaining - get remaining time for the timer
897 * @timer: the timer to read
898 */
899ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
900{
901 struct hrtimer_clock_base *base;
902 unsigned long flags;
903 ktime_t rem;
904
905 base = lock_hrtimer_base(timer, &flags);
906 rem = ktime_sub(timer->expires, base->get_time());
907 unlock_hrtimer_base(timer, &flags);
908
909 return rem;
910}
911EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
912
913#if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
914/**
915 * hrtimer_get_next_event - get the time until next expiry event
916 *
917 * Returns the delta to the next expiry event or KTIME_MAX if no timer
918 * is pending.
919 */
920ktime_t hrtimer_get_next_event(void)
921{
922 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
923 struct hrtimer_clock_base *base = cpu_base->clock_base;
924 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
925 unsigned long flags;
926 int i;
927
928 spin_lock_irqsave(&cpu_base->lock, flags);
929
930 if (!hrtimer_hres_active()) {
931 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
932 struct hrtimer *timer;
933
934 if (!base->first)
935 continue;
936
937 timer = rb_entry(base->first, struct hrtimer, node);
938 delta.tv64 = timer->expires.tv64;
939 delta = ktime_sub(delta, base->get_time());
940 if (delta.tv64 < mindelta.tv64)
941 mindelta.tv64 = delta.tv64;
942 }
943 }
944
945 spin_unlock_irqrestore(&cpu_base->lock, flags);
946
947 if (mindelta.tv64 < 0)
948 mindelta.tv64 = 0;
949 return mindelta;
950}
951#endif
952
953/**
954 * hrtimer_init - initialize a timer to the given clock
955 * @timer: the timer to be initialized
956 * @clock_id: the clock to be used
957 * @mode: timer mode abs/rel
958 */
959void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
960 enum hrtimer_mode mode)
961{
962 struct hrtimer_cpu_base *cpu_base;
963
964 memset(timer, 0, sizeof(struct hrtimer));
965
966 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
967
968 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
969 clock_id = CLOCK_MONOTONIC;
970
971 timer->base = &cpu_base->clock_base[clock_id];
972 hrtimer_init_timer_hres(timer);
973
974#ifdef CONFIG_TIMER_STATS
975 timer->start_site = NULL;
976 timer->start_pid = -1;
977 memset(timer->start_comm, 0, TASK_COMM_LEN);
978#endif
979}
980EXPORT_SYMBOL_GPL(hrtimer_init);
981
982/**
983 * hrtimer_get_res - get the timer resolution for a clock
984 * @which_clock: which clock to query
985 * @tp: pointer to timespec variable to store the resolution
986 *
987 * Store the resolution of the clock selected by @which_clock in the
988 * variable pointed to by @tp.
989 */
990int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
991{
992 struct hrtimer_cpu_base *cpu_base;
993
994 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
995 *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
996
997 return 0;
998}
999EXPORT_SYMBOL_GPL(hrtimer_get_res);
1000
1001#ifdef CONFIG_HIGH_RES_TIMERS
1002
1003/*
1004 * High resolution timer interrupt
1005 * Called with interrupts disabled
1006 */
1007void hrtimer_interrupt(struct clock_event_device *dev)
1008{
1009 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1010 struct hrtimer_clock_base *base;
1011 ktime_t expires_next, now;
1012 int i, raise = 0;
1013
1014 BUG_ON(!cpu_base->hres_active);
1015 cpu_base->nr_events++;
1016 dev->next_event.tv64 = KTIME_MAX;
1017
1018 retry:
1019 now = ktime_get();
1020
1021 expires_next.tv64 = KTIME_MAX;
1022
1023 base = cpu_base->clock_base;
1024
1025 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1026 ktime_t basenow;
1027 struct rb_node *node;
1028
1029 spin_lock(&cpu_base->lock);
1030
1031 basenow = ktime_add(now, base->offset);
1032
1033 while ((node = base->first)) {
1034 struct hrtimer *timer;
1035
1036 timer = rb_entry(node, struct hrtimer, node);
1037
1038 if (basenow.tv64 < timer->expires.tv64) {
1039 ktime_t expires;
1040
1041 expires = ktime_sub(timer->expires,
1042 base->offset);
1043 if (expires.tv64 < expires_next.tv64)
1044 expires_next = expires;
1045 break;
1046 }
1047
1048 /* Move softirq callbacks to the pending list */
1049 if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) {
1050 __remove_hrtimer(timer, base,
1051 HRTIMER_STATE_PENDING, 0);
1052 list_add_tail(&timer->cb_entry,
1053 &base->cpu_base->cb_pending);
1054 raise = 1;
1055 continue;
1056 }
1057
1058 __remove_hrtimer(timer, base,
1059 HRTIMER_STATE_CALLBACK, 0);
1060 timer_stats_account_hrtimer(timer);
1061
1062 /*
1063 * Note: We clear the CALLBACK bit after
1064 * enqueue_hrtimer to avoid reprogramming of
1065 * the event hardware. This happens at the end
1066 * of this function anyway.
1067 */
1068 if (timer->function(timer) != HRTIMER_NORESTART) {
1069 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1070 enqueue_hrtimer(timer, base, 0);
1071 }
1072 timer->state &= ~HRTIMER_STATE_CALLBACK;
1073 }
1074 spin_unlock(&cpu_base->lock);
1075 base++;
1076 }
1077
1078 cpu_base->expires_next = expires_next;
1079
1080 /* Reprogramming necessary ? */
1081 if (expires_next.tv64 != KTIME_MAX) {
1082 if (tick_program_event(expires_next, 0))
1083 goto retry;
1084 }
1085
1086 /* Raise softirq ? */
1087 if (raise)
1088 raise_softirq(HRTIMER_SOFTIRQ);
1089}
1090
1091static void run_hrtimer_softirq(struct softirq_action *h)
1092{
1093 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1094
1095 spin_lock_irq(&cpu_base->lock);
1096
1097 while (!list_empty(&cpu_base->cb_pending)) {
1098 enum hrtimer_restart (*fn)(struct hrtimer *);
1099 struct hrtimer *timer;
1100 int restart;
1101
1102 timer = list_entry(cpu_base->cb_pending.next,
1103 struct hrtimer, cb_entry);
1104
1105 timer_stats_account_hrtimer(timer);
1106
1107 fn = timer->function;
1108 __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0);
1109 spin_unlock_irq(&cpu_base->lock);
1110
1111 restart = fn(timer);
1112
1113 spin_lock_irq(&cpu_base->lock);
1114
1115 timer->state &= ~HRTIMER_STATE_CALLBACK;
1116 if (restart == HRTIMER_RESTART) {
1117 BUG_ON(hrtimer_active(timer));
1118 /*
1119 * Enqueue the timer, allow reprogramming of the event
1120 * device
1121 */
1122 enqueue_hrtimer(timer, timer->base, 1);
1123 } else if (hrtimer_active(timer)) {
1124 /*
1125 * If the timer was rearmed on another CPU, reprogram
1126 * the event device.
1127 */
1128 if (timer->base->first == &timer->node)
1129 hrtimer_reprogram(timer, timer->base);
1130 }
1131 }
1132 spin_unlock_irq(&cpu_base->lock);
1133}
1134
1135#endif /* CONFIG_HIGH_RES_TIMERS */
1136
1137/*
1138 * Expire the per base hrtimer-queue:
1139 */
1140static inline void run_hrtimer_queue(struct hrtimer_cpu_base *cpu_base,
1141 int index)
1142{
1143 struct rb_node *node;
1144 struct hrtimer_clock_base *base = &cpu_base->clock_base[index];
1145
1146 if (!base->first)
1147 return;
1148
1149 if (base->get_softirq_time)
1150 base->softirq_time = base->get_softirq_time();
1151
1152 spin_lock_irq(&cpu_base->lock);
1153
1154 while ((node = base->first)) {
1155 struct hrtimer *timer;
1156 enum hrtimer_restart (*fn)(struct hrtimer *);
1157 int restart;
1158
1159 timer = rb_entry(node, struct hrtimer, node);
1160 if (base->softirq_time.tv64 <= timer->expires.tv64)
1161 break;
1162
1163#ifdef CONFIG_HIGH_RES_TIMERS
1164 WARN_ON_ONCE(timer->cb_mode == HRTIMER_CB_IRQSAFE_NO_SOFTIRQ);
1165#endif
1166 timer_stats_account_hrtimer(timer);
1167
1168 fn = timer->function;
1169 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1170 spin_unlock_irq(&cpu_base->lock);
1171
1172 restart = fn(timer);
1173
1174 spin_lock_irq(&cpu_base->lock);
1175
1176 timer->state &= ~HRTIMER_STATE_CALLBACK;
1177 if (restart != HRTIMER_NORESTART) {
1178 BUG_ON(hrtimer_active(timer));
1179 enqueue_hrtimer(timer, base, 0);
1180 }
1181 }
1182 spin_unlock_irq(&cpu_base->lock);
1183}
1184
1185/*
1186 * Called from timer softirq every jiffy, expire hrtimers:
1187 *
1188 * For HRT its the fall back code to run the softirq in the timer
1189 * softirq context in case the hrtimer initialization failed or has
1190 * not been done yet.
1191 */
1192void hrtimer_run_queues(void)
1193{
1194 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1195 int i;
1196
1197 if (hrtimer_hres_active())
1198 return;
1199
1200 /*
1201 * This _is_ ugly: We have to check in the softirq context,
1202 * whether we can switch to highres and / or nohz mode. The
1203 * clocksource switch happens in the timer interrupt with
1204 * xtime_lock held. Notification from there only sets the
1205 * check bit in the tick_oneshot code, otherwise we might
1206 * deadlock vs. xtime_lock.
1207 */
1208 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1209 if (hrtimer_switch_to_hres())
1210 return;
1211
1212 hrtimer_get_softirq_time(cpu_base);
1213
1214 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1215 run_hrtimer_queue(cpu_base, i);
1216}
1217
1218/*
1219 * Sleep related functions:
1220 */
1221static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1222{
1223 struct hrtimer_sleeper *t =
1224 container_of(timer, struct hrtimer_sleeper, timer);
1225 struct task_struct *task = t->task;
1226
1227 t->task = NULL;
1228 if (task)
1229 wake_up_process(task);
1230
1231 return HRTIMER_NORESTART;
1232}
1233
1234void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1235{
1236 sl->timer.function = hrtimer_wakeup;
1237 sl->task = task;
1238#ifdef CONFIG_HIGH_RES_TIMERS
1239 sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_RESTART;
1240#endif
1241}
1242
1243static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1244{
1245 hrtimer_init_sleeper(t, current);
1246
1247 do {
1248 set_current_state(TASK_INTERRUPTIBLE);
1249 hrtimer_start(&t->timer, t->timer.expires, mode);
1250
1251 if (likely(t->task))
1252 schedule();
1253
1254 hrtimer_cancel(&t->timer);
1255 mode = HRTIMER_MODE_ABS;
1256
1257 } while (t->task && !signal_pending(current));
1258
1259 return t->task == NULL;
1260}
1261
1262long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1263{
1264 struct hrtimer_sleeper t;
1265 struct timespec __user *rmtp;
1266 struct timespec tu;
1267 ktime_t time;
1268
1269 restart->fn = do_no_restart_syscall;
1270
1271 hrtimer_init(&t.timer, restart->arg0, HRTIMER_MODE_ABS);
1272 t.timer.expires.tv64 = ((u64)restart->arg3 << 32) | (u64) restart->arg2;
1273
1274 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1275 return 0;
1276
1277 rmtp = (struct timespec __user *) restart->arg1;
1278 if (rmtp) {
1279 time = ktime_sub(t.timer.expires, t.timer.base->get_time());
1280 if (time.tv64 <= 0)
1281 return 0;
1282 tu = ktime_to_timespec(time);
1283 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1284 return -EFAULT;
1285 }
1286
1287 restart->fn = hrtimer_nanosleep_restart;
1288
1289 /* The other values in restart are already filled in */
1290 return -ERESTART_RESTARTBLOCK;
1291}
1292
1293long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1294 const enum hrtimer_mode mode, const clockid_t clockid)
1295{
1296 struct restart_block *restart;
1297 struct hrtimer_sleeper t;
1298 struct timespec tu;
1299 ktime_t rem;
1300
1301 hrtimer_init(&t.timer, clockid, mode);
1302 t.timer.expires = timespec_to_ktime(*rqtp);
1303 if (do_nanosleep(&t, mode))
1304 return 0;
1305
1306 /* Absolute timers do not update the rmtp value and restart: */
1307 if (mode == HRTIMER_MODE_ABS)
1308 return -ERESTARTNOHAND;
1309
1310 if (rmtp) {
1311 rem = ktime_sub(t.timer.expires, t.timer.base->get_time());
1312 if (rem.tv64 <= 0)
1313 return 0;
1314 tu = ktime_to_timespec(rem);
1315 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1316 return -EFAULT;
1317 }
1318
1319 restart = &current_thread_info()->restart_block;
1320 restart->fn = hrtimer_nanosleep_restart;
1321 restart->arg0 = (unsigned long) t.timer.base->index;
1322 restart->arg1 = (unsigned long) rmtp;
1323 restart->arg2 = t.timer.expires.tv64 & 0xFFFFFFFF;
1324 restart->arg3 = t.timer.expires.tv64 >> 32;
1325
1326 return -ERESTART_RESTARTBLOCK;
1327}
1328
1329asmlinkage long
1330sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1331{
1332 struct timespec tu;
1333
1334 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1335 return -EFAULT;
1336
1337 if (!timespec_valid(&tu))
1338 return -EINVAL;
1339
1340 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1341}
1342
1343/*
1344 * Functions related to boot-time initialization:
1345 */
1346static void __devinit init_hrtimers_cpu(int cpu)
1347{
1348 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1349 int i;
1350
1351 spin_lock_init(&cpu_base->lock);
1352 lockdep_set_class(&cpu_base->lock, &cpu_base->lock_key);
1353
1354 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1355 cpu_base->clock_base[i].cpu_base = cpu_base;
1356
1357 hrtimer_init_hres(cpu_base);
1358}
1359
1360#ifdef CONFIG_HOTPLUG_CPU
1361
1362static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1363 struct hrtimer_clock_base *new_base)
1364{
1365 struct hrtimer *timer;
1366 struct rb_node *node;
1367
1368 while ((node = rb_first(&old_base->active))) {
1369 timer = rb_entry(node, struct hrtimer, node);
1370 BUG_ON(hrtimer_callback_running(timer));
1371 __remove_hrtimer(timer, old_base, HRTIMER_STATE_INACTIVE, 0);
1372 timer->base = new_base;
1373 /*
1374 * Enqueue the timer. Allow reprogramming of the event device
1375 */
1376 enqueue_hrtimer(timer, new_base, 1);
1377 }
1378}
1379
1380static void migrate_hrtimers(int cpu)
1381{
1382 struct hrtimer_cpu_base *old_base, *new_base;
1383 int i;
1384
1385 BUG_ON(cpu_online(cpu));
1386 old_base = &per_cpu(hrtimer_bases, cpu);
1387 new_base = &get_cpu_var(hrtimer_bases);
1388
1389 tick_cancel_sched_timer(cpu);
1390
1391 local_irq_disable();
1392 double_spin_lock(&new_base->lock, &old_base->lock,
1393 smp_processor_id() < cpu);
1394
1395 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1396 migrate_hrtimer_list(&old_base->clock_base[i],
1397 &new_base->clock_base[i]);
1398 }
1399
1400 double_spin_unlock(&new_base->lock, &old_base->lock,
1401 smp_processor_id() < cpu);
1402 local_irq_enable();
1403 put_cpu_var(hrtimer_bases);
1404}
1405#endif /* CONFIG_HOTPLUG_CPU */
1406
1407static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1408 unsigned long action, void *hcpu)
1409{
1410 unsigned int cpu = (long)hcpu;
1411
1412 switch (action) {
1413
1414 case CPU_UP_PREPARE:
1415 case CPU_UP_PREPARE_FROZEN:
1416 init_hrtimers_cpu(cpu);
1417 break;
1418
1419#ifdef CONFIG_HOTPLUG_CPU
1420 case CPU_DEAD:
1421 case CPU_DEAD_FROZEN:
1422 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu);
1423 migrate_hrtimers(cpu);
1424 break;
1425#endif
1426
1427 default:
1428 break;
1429 }
1430
1431 return NOTIFY_OK;
1432}
1433
1434static struct notifier_block __cpuinitdata hrtimers_nb = {
1435 .notifier_call = hrtimer_cpu_notify,
1436};
1437
1438void __init hrtimers_init(void)
1439{
1440 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1441 (void *)(long)smp_processor_id());
1442 register_cpu_notifier(&hrtimers_nb);
1443#ifdef CONFIG_HIGH_RES_TIMERS
1444 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq, NULL);
1445#endif
1446}
1447