projects/linux·blob v2.6.21-rc6 · refs

1428 lines · 34.8 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}
62
63/**
64 * ktime_get_real - get the real (wall-) time in ktime_t format
65 *
66 * returns the time in ktime_t format
67 */
68ktime_t ktime_get_real(void)
69{
70 struct timespec now;
71
72 getnstimeofday(&now);
73
74 return timespec_to_ktime(now);
75}
76
77EXPORT_SYMBOL_GPL(ktime_get_real);
78
79/*
80 * The timer bases:
81 *
82 * Note: If we want to add new timer bases, we have to skip the two
83 * clock ids captured by the cpu-timers. We do this by holding empty
84 * entries rather than doing math adjustment of the clock ids.
85 * This ensures that we capture erroneous accesses to these clock ids
86 * rather than moving them into the range of valid clock id's.
87 */
88DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
89{
90
91 .clock_base =
92 {
93 {
94 .index = CLOCK_REALTIME,
95 .get_time = &ktime_get_real,
96 .resolution = KTIME_LOW_RES,
97 },
98 {
99 .index = CLOCK_MONOTONIC,
100 .get_time = &ktime_get,
101 .resolution = KTIME_LOW_RES,
102 },
103 }
104};
105
106/**
107 * ktime_get_ts - get the monotonic clock in timespec format
108 * @ts: pointer to timespec variable
109 *
110 * The function calculates the monotonic clock from the realtime
111 * clock and the wall_to_monotonic offset and stores the result
112 * in normalized timespec format in the variable pointed to by @ts.
113 */
114void ktime_get_ts(struct timespec *ts)
115{
116 struct timespec tomono;
117 unsigned long seq;
118
119 do {
120 seq = read_seqbegin(&xtime_lock);
121 getnstimeofday(ts);
122 tomono = wall_to_monotonic;
123
124 } while (read_seqretry(&xtime_lock, seq));
125
126 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
127 ts->tv_nsec + tomono.tv_nsec);
128}
129EXPORT_SYMBOL_GPL(ktime_get_ts);
130
131/*
132 * Get the coarse grained time at the softirq based on xtime and
133 * wall_to_monotonic.
134 */
135static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
136{
137 ktime_t xtim, tomono;
138 struct timespec xts, tom;
139 unsigned long seq;
140
141 do {
142 seq = read_seqbegin(&xtime_lock);
143#ifdef CONFIG_NO_HZ
144 getnstimeofday(&xts);
145#else
146 xts = xtime;
147#endif
148 tom = wall_to_monotonic;
149 } while (read_seqretry(&xtime_lock, seq));
150
151 xtim = timespec_to_ktime(xts);
152 tomono = timespec_to_ktime(tom);
153 base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
154 base->clock_base[CLOCK_MONOTONIC].softirq_time =
155 ktime_add(xtim, tomono);
156}
157
158/*
159 * Helper function to check, whether the timer is running the callback
160 * function
161 */
162static inline int hrtimer_callback_running(struct hrtimer *timer)
163{
164 return timer->state & HRTIMER_STATE_CALLBACK;
165}
166
167/*
168 * Functions and macros which are different for UP/SMP systems are kept in a
169 * single place
170 */
171#ifdef CONFIG_SMP
172
173/*
174 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
175 * means that all timers which are tied to this base via timer->base are
176 * locked, and the base itself is locked too.
177 *
178 * So __run_timers/migrate_timers can safely modify all timers which could
179 * be found on the lists/queues.
180 *
181 * When the timer's base is locked, and the timer removed from list, it is
182 * possible to set timer->base = NULL and drop the lock: the timer remains
183 * locked.
184 */
185static
186struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
187 unsigned long *flags)
188{
189 struct hrtimer_clock_base *base;
190
191 for (;;) {
192 base = timer->base;
193 if (likely(base != NULL)) {
194 spin_lock_irqsave(&base->cpu_base->lock, *flags);
195 if (likely(base == timer->base))
196 return base;
197 /* The timer has migrated to another CPU: */
198 spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
199 }
200 cpu_relax();
201 }
202}
203
204/*
205 * Switch the timer base to the current CPU when possible.
206 */
207static inline struct hrtimer_clock_base *
208switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base)
209{
210 struct hrtimer_clock_base *new_base;
211 struct hrtimer_cpu_base *new_cpu_base;
212
213 new_cpu_base = &__get_cpu_var(hrtimer_bases);
214 new_base = &new_cpu_base->clock_base[base->index];
215
216 if (base != new_base) {
217 /*
218 * We are trying to schedule the timer on the local CPU.
219 * However we can't change timer's base while it is running,
220 * so we keep it on the same CPU. No hassle vs. reprogramming
221 * the event source in the high resolution case. The softirq
222 * code will take care of this when the timer function has
223 * completed. There is no conflict as we hold the lock until
224 * the timer is enqueued.
225 */
226 if (unlikely(hrtimer_callback_running(timer)))
227 return base;
228
229 /* See the comment in lock_timer_base() */
230 timer->base = NULL;
231 spin_unlock(&base->cpu_base->lock);
232 spin_lock(&new_base->cpu_base->lock);
233 timer->base = new_base;
234 }
235 return new_base;
236}
237
238#else /* CONFIG_SMP */
239
240static inline struct hrtimer_clock_base *
241lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
242{
243 struct hrtimer_clock_base *base = timer->base;
244
245 spin_lock_irqsave(&base->cpu_base->lock, *flags);
246
247 return base;
248}
249
250# define switch_hrtimer_base(t, b) (b)
251
252#endif /* !CONFIG_SMP */
253
254/*
255 * Functions for the union type storage format of ktime_t which are
256 * too large for inlining:
257 */
258#if BITS_PER_LONG < 64
259# ifndef CONFIG_KTIME_SCALAR
260/**
261 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
262 * @kt: addend
263 * @nsec: the scalar nsec value to add
264 *
265 * Returns the sum of kt and nsec in ktime_t format
266 */
267ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
268{
269 ktime_t tmp;
270
271 if (likely(nsec < NSEC_PER_SEC)) {
272 tmp.tv64 = nsec;
273 } else {
274 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
275
276 tmp = ktime_set((long)nsec, rem);
277 }
278
279 return ktime_add(kt, tmp);
280}
281# endif /* !CONFIG_KTIME_SCALAR */
282
283/*
284 * Divide a ktime value by a nanosecond value
285 */
286unsigned long ktime_divns(const ktime_t kt, s64 div)
287{
288 u64 dclc, inc, dns;
289 int sft = 0;
290
291 dclc = dns = ktime_to_ns(kt);
292 inc = div;
293 /* Make sure the divisor is less than 2^32: */
294 while (div >> 32) {
295 sft++;
296 div >>= 1;
297 }
298 dclc >>= sft;
299 do_div(dclc, (unsigned long) div);
300
301 return (unsigned long) dclc;
302}
303#endif /* BITS_PER_LONG >= 64 */
304
305/* High resolution timer related functions */
306#ifdef CONFIG_HIGH_RES_TIMERS
307
308/*
309 * High resolution timer enabled ?
310 */
311static int hrtimer_hres_enabled __read_mostly = 1;
312
313/*
314 * Enable / Disable high resolution mode
315 */
316static int __init setup_hrtimer_hres(char *str)
317{
318 if (!strcmp(str, "off"))
319 hrtimer_hres_enabled = 0;
320 else if (!strcmp(str, "on"))
321 hrtimer_hres_enabled = 1;
322 else
323 return 0;
324 return 1;
325}
326
327__setup("highres=", setup_hrtimer_hres);
328
329/*
330 * hrtimer_high_res_enabled - query, if the highres mode is enabled
331 */
332static inline int hrtimer_is_hres_enabled(void)
333{
334 return hrtimer_hres_enabled;
335}
336
337/*
338 * Is the high resolution mode active ?
339 */
340static inline int hrtimer_hres_active(void)
341{
342 return __get_cpu_var(hrtimer_bases).hres_active;
343}
344
345/*
346 * Reprogram the event source with checking both queues for the
347 * next event
348 * Called with interrupts disabled and base->lock held
349 */
350static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base)
351{
352 int i;
353 struct hrtimer_clock_base *base = cpu_base->clock_base;
354 ktime_t expires;
355
356 cpu_base->expires_next.tv64 = KTIME_MAX;
357
358 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
359 struct hrtimer *timer;
360
361 if (!base->first)
362 continue;
363 timer = rb_entry(base->first, struct hrtimer, node);
364 expires = ktime_sub(timer->expires, base->offset);
365 if (expires.tv64 < cpu_base->expires_next.tv64)
366 cpu_base->expires_next = expires;
367 }
368
369 if (cpu_base->expires_next.tv64 != KTIME_MAX)
370 tick_program_event(cpu_base->expires_next, 1);
371}
372
373/*
374 * Shared reprogramming for clock_realtime and clock_monotonic
375 *
376 * When a timer is enqueued and expires earlier than the already enqueued
377 * timers, we have to check, whether it expires earlier than the timer for
378 * which the clock event device was armed.
379 *
380 * Called with interrupts disabled and base->cpu_base.lock held
381 */
382static int hrtimer_reprogram(struct hrtimer *timer,
383 struct hrtimer_clock_base *base)
384{
385 ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next;
386 ktime_t expires = ktime_sub(timer->expires, base->offset);
387 int res;
388
389 /*
390 * When the callback is running, we do not reprogram the clock event
391 * device. The timer callback is either running on a different CPU or
392 * the callback is executed in the hrtimer_interupt context. The
393 * reprogramming is handled either by the softirq, which called the
394 * callback or at the end of the hrtimer_interrupt.
395 */
396 if (hrtimer_callback_running(timer))
397 return 0;
398
399 if (expires.tv64 >= expires_next->tv64)
400 return 0;
401
402 /*
403 * Clockevents returns -ETIME, when the event was in the past.
404 */
405 res = tick_program_event(expires, 0);
406 if (!IS_ERR_VALUE(res))
407 *expires_next = expires;
408 return res;
409}
410
411
412/*
413 * Retrigger next event is called after clock was set
414 *
415 * Called with interrupts disabled via on_each_cpu()
416 */
417static void retrigger_next_event(void *arg)
418{
419 struct hrtimer_cpu_base *base;
420 struct timespec realtime_offset;
421 unsigned long seq;
422
423 if (!hrtimer_hres_active())
424 return;
425
426 do {
427 seq = read_seqbegin(&xtime_lock);
428 set_normalized_timespec(&realtime_offset,
429 -wall_to_monotonic.tv_sec,
430 -wall_to_monotonic.tv_nsec);
431 } while (read_seqretry(&xtime_lock, seq));
432
433 base = &__get_cpu_var(hrtimer_bases);
434
435 /* Adjust CLOCK_REALTIME offset */
436 spin_lock(&base->lock);
437 base->clock_base[CLOCK_REALTIME].offset =
438 timespec_to_ktime(realtime_offset);
439
440 hrtimer_force_reprogram(base);
441 spin_unlock(&base->lock);
442}
443
444/*
445 * Clock realtime was set
446 *
447 * Change the offset of the realtime clock vs. the monotonic
448 * clock.
449 *
450 * We might have to reprogram the high resolution timer interrupt. On
451 * SMP we call the architecture specific code to retrigger _all_ high
452 * resolution timer interrupts. On UP we just disable interrupts and
453 * call the high resolution interrupt code.
454 */
455void clock_was_set(void)
456{
457 /* Retrigger the CPU local events everywhere */
458 on_each_cpu(retrigger_next_event, NULL, 0, 1);
459}
460
461/*
462 * Check, whether the timer is on the callback pending list
463 */
464static inline int hrtimer_cb_pending(const struct hrtimer *timer)
465{
466 return timer->state & HRTIMER_STATE_PENDING;
467}
468
469/*
470 * Remove a timer from the callback pending list
471 */
472static inline void hrtimer_remove_cb_pending(struct hrtimer *timer)
473{
474 list_del_init(&timer->cb_entry);
475}
476
477/*
478 * Initialize the high resolution related parts of cpu_base
479 */
480static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
481{
482 base->expires_next.tv64 = KTIME_MAX;
483 base->hres_active = 0;
484 INIT_LIST_HEAD(&base->cb_pending);
485}
486
487/*
488 * Initialize the high resolution related parts of a hrtimer
489 */
490static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
491{
492 INIT_LIST_HEAD(&timer->cb_entry);
493}
494
495/*
496 * When High resolution timers are active, try to reprogram. Note, that in case
497 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
498 * check happens. The timer gets enqueued into the rbtree. The reprogramming
499 * and expiry check is done in the hrtimer_interrupt or in the softirq.
500 */
501static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
502 struct hrtimer_clock_base *base)
503{
504 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
505
506 /* Timer is expired, act upon the callback mode */
507 switch(timer->cb_mode) {
508 case HRTIMER_CB_IRQSAFE_NO_RESTART:
509 /*
510 * We can call the callback from here. No restart
511 * happens, so no danger of recursion
512 */
513 BUG_ON(timer->function(timer) != HRTIMER_NORESTART);
514 return 1;
515 case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ:
516 /*
517 * This is solely for the sched tick emulation with
518 * dynamic tick support to ensure that we do not
519 * restart the tick right on the edge and end up with
520 * the tick timer in the softirq ! The calling site
521 * takes care of this.
522 */
523 return 1;
524 case HRTIMER_CB_IRQSAFE:
525 case HRTIMER_CB_SOFTIRQ:
526 /*
527 * Move everything else into the softirq pending list !
528 */
529 list_add_tail(&timer->cb_entry,
530 &base->cpu_base->cb_pending);
531 timer->state = HRTIMER_STATE_PENDING;
532 raise_softirq(HRTIMER_SOFTIRQ);
533 return 1;
534 default:
535 BUG();
536 }
537 }
538 return 0;
539}
540
541/*
542 * Switch to high resolution mode
543 */
544static int hrtimer_switch_to_hres(void)
545{
546 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
547 unsigned long flags;
548
549 if (base->hres_active)
550 return 1;
551
552 local_irq_save(flags);
553
554 if (tick_init_highres()) {
555 local_irq_restore(flags);
556 return 0;
557 }
558 base->hres_active = 1;
559 base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
560 base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
561
562 tick_setup_sched_timer();
563
564 /* "Retrigger" the interrupt to get things going */
565 retrigger_next_event(NULL);
566 local_irq_restore(flags);
567 printk(KERN_INFO "Switched to high resolution mode on CPU %d\n",
568 smp_processor_id());
569 return 1;
570}
571
572#else
573
574static inline int hrtimer_hres_active(void) { return 0; }
575static inline int hrtimer_is_hres_enabled(void) { return 0; }
576static inline int hrtimer_switch_to_hres(void) { return 0; }
577static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { }
578static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
579 struct hrtimer_clock_base *base)
580{
581 return 0;
582}
583static inline int hrtimer_cb_pending(struct hrtimer *timer) { return 0; }
584static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) { }
585static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
586static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
587
588#endif /* CONFIG_HIGH_RES_TIMERS */
589
590#ifdef CONFIG_TIMER_STATS
591void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr)
592{
593 if (timer->start_site)
594 return;
595
596 timer->start_site = addr;
597 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
598 timer->start_pid = current->pid;
599}
600#endif
601
602/*
603 * Counterpart to lock_timer_base above:
604 */
605static inline
606void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
607{
608 spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
609}
610
611/**
612 * hrtimer_forward - forward the timer expiry
613 * @timer: hrtimer to forward
614 * @now: forward past this time
615 * @interval: the interval to forward
616 *
617 * Forward the timer expiry so it will expire in the future.
618 * Returns the number of overruns.
619 */
620unsigned long
621hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
622{
623 unsigned long orun = 1;
624 ktime_t delta;
625
626 delta = ktime_sub(now, timer->expires);
627
628 if (delta.tv64 < 0)
629 return 0;
630
631 if (interval.tv64 < timer->base->resolution.tv64)
632 interval.tv64 = timer->base->resolution.tv64;
633
634 if (unlikely(delta.tv64 >= interval.tv64)) {
635 s64 incr = ktime_to_ns(interval);
636
637 orun = ktime_divns(delta, incr);
638 timer->expires = ktime_add_ns(timer->expires, incr * orun);
639 if (timer->expires.tv64 > now.tv64)
640 return orun;
641 /*
642 * This (and the ktime_add() below) is the
643 * correction for exact:
644 */
645 orun++;
646 }
647 timer->expires = ktime_add(timer->expires, interval);
648 /*
649 * Make sure, that the result did not wrap with a very large
650 * interval.
651 */
652 if (timer->expires.tv64 < 0)
653 timer->expires = ktime_set(KTIME_SEC_MAX, 0);
654
655 return orun;
656}
657
658/*
659 * enqueue_hrtimer - internal function to (re)start a timer
660 *
661 * The timer is inserted in expiry order. Insertion into the
662 * red black tree is O(log(n)). Must hold the base lock.
663 */
664static void enqueue_hrtimer(struct hrtimer *timer,
665 struct hrtimer_clock_base *base, int reprogram)
666{
667 struct rb_node **link = &base->active.rb_node;
668 struct rb_node *parent = NULL;
669 struct hrtimer *entry;
670
671 /*
672 * Find the right place in the rbtree:
673 */
674 while (*link) {
675 parent = *link;
676 entry = rb_entry(parent, struct hrtimer, node);
677 /*
678 * We dont care about collisions. Nodes with
679 * the same expiry time stay together.
680 */
681 if (timer->expires.tv64 < entry->expires.tv64)
682 link = &(*link)->rb_left;
683 else
684 link = &(*link)->rb_right;
685 }
686
687 /*
688 * Insert the timer to the rbtree and check whether it
689 * replaces the first pending timer
690 */
691 if (!base->first || timer->expires.tv64 <
692 rb_entry(base->first, struct hrtimer, node)->expires.tv64) {
693 /*
694 * Reprogram the clock event device. When the timer is already
695 * expired hrtimer_enqueue_reprogram has either called the
696 * callback or added it to the pending list and raised the
697 * softirq.
698 *
699 * This is a NOP for !HIGHRES
700 */
701 if (reprogram && hrtimer_enqueue_reprogram(timer, base))
702 return;
703
704 base->first = &timer->node;
705 }
706
707 rb_link_node(&timer->node, parent, link);
708 rb_insert_color(&timer->node, &base->active);
709 /*
710 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
711 * state of a possibly running callback.
712 */
713 timer->state |= HRTIMER_STATE_ENQUEUED;
714}
715
716/*
717 * __remove_hrtimer - internal function to remove a timer
718 *
719 * Caller must hold the base lock.
720 *
721 * High resolution timer mode reprograms the clock event device when the
722 * timer is the one which expires next. The caller can disable this by setting
723 * reprogram to zero. This is useful, when the context does a reprogramming
724 * anyway (e.g. timer interrupt)
725 */
726static void __remove_hrtimer(struct hrtimer *timer,
727 struct hrtimer_clock_base *base,
728 unsigned long newstate, int reprogram)
729{
730 /* High res. callback list. NOP for !HIGHRES */
731 if (hrtimer_cb_pending(timer))
732 hrtimer_remove_cb_pending(timer);
733 else {
734 /*
735 * Remove the timer from the rbtree and replace the
736 * first entry pointer if necessary.
737 */
738 if (base->first == &timer->node) {
739 base->first = rb_next(&timer->node);
740 /* Reprogram the clock event device. if enabled */
741 if (reprogram && hrtimer_hres_active())
742 hrtimer_force_reprogram(base->cpu_base);
743 }
744 rb_erase(&timer->node, &base->active);
745 }
746 timer->state = newstate;
747}
748
749/*
750 * remove hrtimer, called with base lock held
751 */
752static inline int
753remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
754{
755 if (hrtimer_is_queued(timer)) {
756 int reprogram;
757
758 /*
759 * Remove the timer and force reprogramming when high
760 * resolution mode is active and the timer is on the current
761 * CPU. If we remove a timer on another CPU, reprogramming is
762 * skipped. The interrupt event on this CPU is fired and
763 * reprogramming happens in the interrupt handler. This is a
764 * rare case and less expensive than a smp call.
765 */
766 timer_stats_hrtimer_clear_start_info(timer);
767 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
768 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
769 reprogram);
770 return 1;
771 }
772 return 0;
773}
774
775/**
776 * hrtimer_start - (re)start an relative timer on the current CPU
777 * @timer: the timer to be added
778 * @tim: expiry time
779 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
780 *
781 * Returns:
782 * 0 on success
783 * 1 when the timer was active
784 */
785int
786hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
787{
788 struct hrtimer_clock_base *base, *new_base;
789 unsigned long flags;
790 int ret;
791
792 base = lock_hrtimer_base(timer, &flags);
793
794 /* Remove an active timer from the queue: */
795 ret = remove_hrtimer(timer, base);
796
797 /* Switch the timer base, if necessary: */
798 new_base = switch_hrtimer_base(timer, base);
799
800 if (mode == HRTIMER_MODE_REL) {
801 tim = ktime_add(tim, new_base->get_time());
802 /*
803 * CONFIG_TIME_LOW_RES is a temporary way for architectures
804 * to signal that they simply return xtime in
805 * do_gettimeoffset(). In this case we want to round up by
806 * resolution when starting a relative timer, to avoid short
807 * timeouts. This will go away with the GTOD framework.
808 */
809#ifdef CONFIG_TIME_LOW_RES
810 tim = ktime_add(tim, base->resolution);
811#endif
812 }
813 timer->expires = tim;
814
815 timer_stats_hrtimer_set_start_info(timer);
816
817 /*
818 * Only allow reprogramming if the new base is on this CPU.
819 * (it might still be on another CPU if the timer was pending)
820 */
821 enqueue_hrtimer(timer, new_base,
822 new_base->cpu_base == &__get_cpu_var(hrtimer_bases));
823
824 unlock_hrtimer_base(timer, &flags);
825
826 return ret;
827}
828EXPORT_SYMBOL_GPL(hrtimer_start);
829
830/**
831 * hrtimer_try_to_cancel - try to deactivate a timer
832 * @timer: hrtimer to stop
833 *
834 * Returns:
835 * 0 when the timer was not active
836 * 1 when the timer was active
837 * -1 when the timer is currently excuting the callback function and
838 * cannot be stopped
839 */
840int hrtimer_try_to_cancel(struct hrtimer *timer)
841{
842 struct hrtimer_clock_base *base;
843 unsigned long flags;
844 int ret = -1;
845
846 base = lock_hrtimer_base(timer, &flags);
847
848 if (!hrtimer_callback_running(timer))
849 ret = remove_hrtimer(timer, base);
850
851 unlock_hrtimer_base(timer, &flags);
852
853 return ret;
854
855}
856EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
857
858/**
859 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
860 * @timer: the timer to be cancelled
861 *
862 * Returns:
863 * 0 when the timer was not active
864 * 1 when the timer was active
865 */
866int hrtimer_cancel(struct hrtimer *timer)
867{
868 for (;;) {
869 int ret = hrtimer_try_to_cancel(timer);
870
871 if (ret >= 0)
872 return ret;
873 cpu_relax();
874 }
875}
876EXPORT_SYMBOL_GPL(hrtimer_cancel);
877
878/**
879 * hrtimer_get_remaining - get remaining time for the timer
880 * @timer: the timer to read
881 */
882ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
883{
884 struct hrtimer_clock_base *base;
885 unsigned long flags;
886 ktime_t rem;
887
888 base = lock_hrtimer_base(timer, &flags);
889 rem = ktime_sub(timer->expires, base->get_time());
890 unlock_hrtimer_base(timer, &flags);
891
892 return rem;
893}
894EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
895
896#if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
897/**
898 * hrtimer_get_next_event - get the time until next expiry event
899 *
900 * Returns the delta to the next expiry event or KTIME_MAX if no timer
901 * is pending.
902 */
903ktime_t hrtimer_get_next_event(void)
904{
905 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
906 struct hrtimer_clock_base *base = cpu_base->clock_base;
907 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
908 unsigned long flags;
909 int i;
910
911 spin_lock_irqsave(&cpu_base->lock, flags);
912
913 if (!hrtimer_hres_active()) {
914 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
915 struct hrtimer *timer;
916
917 if (!base->first)
918 continue;
919
920 timer = rb_entry(base->first, struct hrtimer, node);
921 delta.tv64 = timer->expires.tv64;
922 delta = ktime_sub(delta, base->get_time());
923 if (delta.tv64 < mindelta.tv64)
924 mindelta.tv64 = delta.tv64;
925 }
926 }
927
928 spin_unlock_irqrestore(&cpu_base->lock, flags);
929
930 if (mindelta.tv64 < 0)
931 mindelta.tv64 = 0;
932 return mindelta;
933}
934#endif
935
936/**
937 * hrtimer_init - initialize a timer to the given clock
938 * @timer: the timer to be initialized
939 * @clock_id: the clock to be used
940 * @mode: timer mode abs/rel
941 */
942void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
943 enum hrtimer_mode mode)
944{
945 struct hrtimer_cpu_base *cpu_base;
946
947 memset(timer, 0, sizeof(struct hrtimer));
948
949 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
950
951 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
952 clock_id = CLOCK_MONOTONIC;
953
954 timer->base = &cpu_base->clock_base[clock_id];
955 hrtimer_init_timer_hres(timer);
956
957#ifdef CONFIG_TIMER_STATS
958 timer->start_site = NULL;
959 timer->start_pid = -1;
960 memset(timer->start_comm, 0, TASK_COMM_LEN);
961#endif
962}
963EXPORT_SYMBOL_GPL(hrtimer_init);
964
965/**
966 * hrtimer_get_res - get the timer resolution for a clock
967 * @which_clock: which clock to query
968 * @tp: pointer to timespec variable to store the resolution
969 *
970 * Store the resolution of the clock selected by @which_clock in the
971 * variable pointed to by @tp.
972 */
973int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
974{
975 struct hrtimer_cpu_base *cpu_base;
976
977 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
978 *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
979
980 return 0;
981}
982EXPORT_SYMBOL_GPL(hrtimer_get_res);
983
984#ifdef CONFIG_HIGH_RES_TIMERS
985
986/*
987 * High resolution timer interrupt
988 * Called with interrupts disabled
989 */
990void hrtimer_interrupt(struct clock_event_device *dev)
991{
992 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
993 struct hrtimer_clock_base *base;
994 ktime_t expires_next, now;
995 int i, raise = 0;
996
997 BUG_ON(!cpu_base->hres_active);
998 cpu_base->nr_events++;
999 dev->next_event.tv64 = KTIME_MAX;
1000
1001 retry:
1002 now = ktime_get();
1003
1004 expires_next.tv64 = KTIME_MAX;
1005
1006 base = cpu_base->clock_base;
1007
1008 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1009 ktime_t basenow;
1010 struct rb_node *node;
1011
1012 spin_lock(&cpu_base->lock);
1013
1014 basenow = ktime_add(now, base->offset);
1015
1016 while ((node = base->first)) {
1017 struct hrtimer *timer;
1018
1019 timer = rb_entry(node, struct hrtimer, node);
1020
1021 if (basenow.tv64 < timer->expires.tv64) {
1022 ktime_t expires;
1023
1024 expires = ktime_sub(timer->expires,
1025 base->offset);
1026 if (expires.tv64 < expires_next.tv64)
1027 expires_next = expires;
1028 break;
1029 }
1030
1031 /* Move softirq callbacks to the pending list */
1032 if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) {
1033 __remove_hrtimer(timer, base,
1034 HRTIMER_STATE_PENDING, 0);
1035 list_add_tail(&timer->cb_entry,
1036 &base->cpu_base->cb_pending);
1037 raise = 1;
1038 continue;
1039 }
1040
1041 __remove_hrtimer(timer, base,
1042 HRTIMER_STATE_CALLBACK, 0);
1043 timer_stats_account_hrtimer(timer);
1044
1045 /*
1046 * Note: We clear the CALLBACK bit after
1047 * enqueue_hrtimer to avoid reprogramming of
1048 * the event hardware. This happens at the end
1049 * of this function anyway.
1050 */
1051 if (timer->function(timer) != HRTIMER_NORESTART) {
1052 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1053 enqueue_hrtimer(timer, base, 0);
1054 }
1055 timer->state &= ~HRTIMER_STATE_CALLBACK;
1056 }
1057 spin_unlock(&cpu_base->lock);
1058 base++;
1059 }
1060
1061 cpu_base->expires_next = expires_next;
1062
1063 /* Reprogramming necessary ? */
1064 if (expires_next.tv64 != KTIME_MAX) {
1065 if (tick_program_event(expires_next, 0))
1066 goto retry;
1067 }
1068
1069 /* Raise softirq ? */
1070 if (raise)
1071 raise_softirq(HRTIMER_SOFTIRQ);
1072}
1073
1074static void run_hrtimer_softirq(struct softirq_action *h)
1075{
1076 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1077
1078 spin_lock_irq(&cpu_base->lock);
1079
1080 while (!list_empty(&cpu_base->cb_pending)) {
1081 enum hrtimer_restart (*fn)(struct hrtimer *);
1082 struct hrtimer *timer;
1083 int restart;
1084
1085 timer = list_entry(cpu_base->cb_pending.next,
1086 struct hrtimer, cb_entry);
1087
1088 timer_stats_account_hrtimer(timer);
1089
1090 fn = timer->function;
1091 __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0);
1092 spin_unlock_irq(&cpu_base->lock);
1093
1094 restart = fn(timer);
1095
1096 spin_lock_irq(&cpu_base->lock);
1097
1098 timer->state &= ~HRTIMER_STATE_CALLBACK;
1099 if (restart == HRTIMER_RESTART) {
1100 BUG_ON(hrtimer_active(timer));
1101 /*
1102 * Enqueue the timer, allow reprogramming of the event
1103 * device
1104 */
1105 enqueue_hrtimer(timer, timer->base, 1);
1106 } else if (hrtimer_active(timer)) {
1107 /*
1108 * If the timer was rearmed on another CPU, reprogram
1109 * the event device.
1110 */
1111 if (timer->base->first == &timer->node)
1112 hrtimer_reprogram(timer, timer->base);
1113 }
1114 }
1115 spin_unlock_irq(&cpu_base->lock);
1116}
1117
1118#endif /* CONFIG_HIGH_RES_TIMERS */
1119
1120/*
1121 * Expire the per base hrtimer-queue:
1122 */
1123static inline void run_hrtimer_queue(struct hrtimer_cpu_base *cpu_base,
1124 int index)
1125{
1126 struct rb_node *node;
1127 struct hrtimer_clock_base *base = &cpu_base->clock_base[index];
1128
1129 if (!base->first)
1130 return;
1131
1132 if (base->get_softirq_time)
1133 base->softirq_time = base->get_softirq_time();
1134
1135 spin_lock_irq(&cpu_base->lock);
1136
1137 while ((node = base->first)) {
1138 struct hrtimer *timer;
1139 enum hrtimer_restart (*fn)(struct hrtimer *);
1140 int restart;
1141
1142 timer = rb_entry(node, struct hrtimer, node);
1143 if (base->softirq_time.tv64 <= timer->expires.tv64)
1144 break;
1145
1146#ifdef CONFIG_HIGH_RES_TIMERS
1147 WARN_ON_ONCE(timer->cb_mode == HRTIMER_CB_IRQSAFE_NO_SOFTIRQ);
1148#endif
1149 timer_stats_account_hrtimer(timer);
1150
1151 fn = timer->function;
1152 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1153 spin_unlock_irq(&cpu_base->lock);
1154
1155 restart = fn(timer);
1156
1157 spin_lock_irq(&cpu_base->lock);
1158
1159 timer->state &= ~HRTIMER_STATE_CALLBACK;
1160 if (restart != HRTIMER_NORESTART) {
1161 BUG_ON(hrtimer_active(timer));
1162 enqueue_hrtimer(timer, base, 0);
1163 }
1164 }
1165 spin_unlock_irq(&cpu_base->lock);
1166}
1167
1168/*
1169 * Called from timer softirq every jiffy, expire hrtimers:
1170 *
1171 * For HRT its the fall back code to run the softirq in the timer
1172 * softirq context in case the hrtimer initialization failed or has
1173 * not been done yet.
1174 */
1175void hrtimer_run_queues(void)
1176{
1177 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1178 int i;
1179
1180 if (hrtimer_hres_active())
1181 return;
1182
1183 /*
1184 * This _is_ ugly: We have to check in the softirq context,
1185 * whether we can switch to highres and / or nohz mode. The
1186 * clocksource switch happens in the timer interrupt with
1187 * xtime_lock held. Notification from there only sets the
1188 * check bit in the tick_oneshot code, otherwise we might
1189 * deadlock vs. xtime_lock.
1190 */
1191 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1192 if (hrtimer_switch_to_hres())
1193 return;
1194
1195 hrtimer_get_softirq_time(cpu_base);
1196
1197 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1198 run_hrtimer_queue(cpu_base, i);
1199}
1200
1201/*
1202 * Sleep related functions:
1203 */
1204static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1205{
1206 struct hrtimer_sleeper *t =
1207 container_of(timer, struct hrtimer_sleeper, timer);
1208 struct task_struct *task = t->task;
1209
1210 t->task = NULL;
1211 if (task)
1212 wake_up_process(task);
1213
1214 return HRTIMER_NORESTART;
1215}
1216
1217void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1218{
1219 sl->timer.function = hrtimer_wakeup;
1220 sl->task = task;
1221#ifdef CONFIG_HIGH_RES_TIMERS
1222 sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_RESTART;
1223#endif
1224}
1225
1226static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1227{
1228 hrtimer_init_sleeper(t, current);
1229
1230 do {
1231 set_current_state(TASK_INTERRUPTIBLE);
1232 hrtimer_start(&t->timer, t->timer.expires, mode);
1233
1234 if (likely(t->task))
1235 schedule();
1236
1237 hrtimer_cancel(&t->timer);
1238 mode = HRTIMER_MODE_ABS;
1239
1240 } while (t->task && !signal_pending(current));
1241
1242 return t->task == NULL;
1243}
1244
1245long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1246{
1247 struct hrtimer_sleeper t;
1248 struct timespec __user *rmtp;
1249 struct timespec tu;
1250 ktime_t time;
1251
1252 restart->fn = do_no_restart_syscall;
1253
1254 hrtimer_init(&t.timer, restart->arg0, HRTIMER_MODE_ABS);
1255 t.timer.expires.tv64 = ((u64)restart->arg3 << 32) | (u64) restart->arg2;
1256
1257 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1258 return 0;
1259
1260 rmtp = (struct timespec __user *) restart->arg1;
1261 if (rmtp) {
1262 time = ktime_sub(t.timer.expires, t.timer.base->get_time());
1263 if (time.tv64 <= 0)
1264 return 0;
1265 tu = ktime_to_timespec(time);
1266 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1267 return -EFAULT;
1268 }
1269
1270 restart->fn = hrtimer_nanosleep_restart;
1271
1272 /* The other values in restart are already filled in */
1273 return -ERESTART_RESTARTBLOCK;
1274}
1275
1276long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1277 const enum hrtimer_mode mode, const clockid_t clockid)
1278{
1279 struct restart_block *restart;
1280 struct hrtimer_sleeper t;
1281 struct timespec tu;
1282 ktime_t rem;
1283
1284 hrtimer_init(&t.timer, clockid, mode);
1285 t.timer.expires = timespec_to_ktime(*rqtp);
1286 if (do_nanosleep(&t, mode))
1287 return 0;
1288
1289 /* Absolute timers do not update the rmtp value and restart: */
1290 if (mode == HRTIMER_MODE_ABS)
1291 return -ERESTARTNOHAND;
1292
1293 if (rmtp) {
1294 rem = ktime_sub(t.timer.expires, t.timer.base->get_time());
1295 if (rem.tv64 <= 0)
1296 return 0;
1297 tu = ktime_to_timespec(rem);
1298 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1299 return -EFAULT;
1300 }
1301
1302 restart = &current_thread_info()->restart_block;
1303 restart->fn = hrtimer_nanosleep_restart;
1304 restart->arg0 = (unsigned long) t.timer.base->index;
1305 restart->arg1 = (unsigned long) rmtp;
1306 restart->arg2 = t.timer.expires.tv64 & 0xFFFFFFFF;
1307 restart->arg3 = t.timer.expires.tv64 >> 32;
1308
1309 return -ERESTART_RESTARTBLOCK;
1310}
1311
1312asmlinkage long
1313sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1314{
1315 struct timespec tu;
1316
1317 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1318 return -EFAULT;
1319
1320 if (!timespec_valid(&tu))
1321 return -EINVAL;
1322
1323 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1324}
1325
1326/*
1327 * Functions related to boot-time initialization:
1328 */
1329static void __devinit init_hrtimers_cpu(int cpu)
1330{
1331 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1332 int i;
1333
1334 spin_lock_init(&cpu_base->lock);
1335 lockdep_set_class(&cpu_base->lock, &cpu_base->lock_key);
1336
1337 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1338 cpu_base->clock_base[i].cpu_base = cpu_base;
1339
1340 hrtimer_init_hres(cpu_base);
1341}
1342
1343#ifdef CONFIG_HOTPLUG_CPU
1344
1345static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1346 struct hrtimer_clock_base *new_base)
1347{
1348 struct hrtimer *timer;
1349 struct rb_node *node;
1350
1351 while ((node = rb_first(&old_base->active))) {
1352 timer = rb_entry(node, struct hrtimer, node);
1353 BUG_ON(hrtimer_callback_running(timer));
1354 __remove_hrtimer(timer, old_base, HRTIMER_STATE_INACTIVE, 0);
1355 timer->base = new_base;
1356 /*
1357 * Enqueue the timer. Allow reprogramming of the event device
1358 */
1359 enqueue_hrtimer(timer, new_base, 1);
1360 }
1361}
1362
1363static void migrate_hrtimers(int cpu)
1364{
1365 struct hrtimer_cpu_base *old_base, *new_base;
1366 int i;
1367
1368 BUG_ON(cpu_online(cpu));
1369 old_base = &per_cpu(hrtimer_bases, cpu);
1370 new_base = &get_cpu_var(hrtimer_bases);
1371
1372 tick_cancel_sched_timer(cpu);
1373
1374 local_irq_disable();
1375 double_spin_lock(&new_base->lock, &old_base->lock,
1376 smp_processor_id() < cpu);
1377
1378 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1379 migrate_hrtimer_list(&old_base->clock_base[i],
1380 &new_base->clock_base[i]);
1381 }
1382
1383 double_spin_unlock(&new_base->lock, &old_base->lock,
1384 smp_processor_id() < cpu);
1385 local_irq_enable();
1386 put_cpu_var(hrtimer_bases);
1387}
1388#endif /* CONFIG_HOTPLUG_CPU */
1389
1390static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1391 unsigned long action, void *hcpu)
1392{
1393 long cpu = (long)hcpu;
1394
1395 switch (action) {
1396
1397 case CPU_UP_PREPARE:
1398 init_hrtimers_cpu(cpu);
1399 break;
1400
1401#ifdef CONFIG_HOTPLUG_CPU
1402 case CPU_DEAD:
1403 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu);
1404 migrate_hrtimers(cpu);
1405 break;
1406#endif
1407
1408 default:
1409 break;
1410 }
1411
1412 return NOTIFY_OK;
1413}
1414
1415static struct notifier_block __cpuinitdata hrtimers_nb = {
1416 .notifier_call = hrtimer_cpu_notify,
1417};
1418
1419void __init hrtimers_init(void)
1420{
1421 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1422 (void *)(long)smp_processor_id());
1423 register_cpu_notifier(&hrtimers_nb);
1424#ifdef CONFIG_HIGH_RES_TIMERS
1425 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq, NULL);
1426#endif
1427}
1428