ide: make void and rename ide_dma_timeout() method
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / kernel / hrtimer.c
1 /*
2 * linux/kernel/hrtimer.c
3 *
4 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
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 */
54 ktime_t ktime_get(void)
55 {
56 struct timespec now;
57
58 ktime_get_ts(&now);
59
60 return timespec_to_ktime(now);
61 }
62 EXPORT_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 */
69 ktime_t ktime_get_real(void)
70 {
71 struct timespec now;
72
73 getnstimeofday(&now);
74
75 return timespec_to_ktime(now);
76 }
77
78 EXPORT_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 */
89 DEFINE_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 */
115 void 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 }
130 EXPORT_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 */
136 static 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 #ifdef CONFIG_NO_HZ
145 getnstimeofday(&xts);
146 #else
147 xts = xtime;
148 #endif
149 tom = wall_to_monotonic;
150 } while (read_seqretry(&xtime_lock, seq));
151
152 xtim = timespec_to_ktime(xts);
153 tomono = timespec_to_ktime(tom);
154 base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
155 base->clock_base[CLOCK_MONOTONIC].softirq_time =
156 ktime_add(xtim, tomono);
157 }
158
159 /*
160 * Helper function to check, whether the timer is running the callback
161 * function
162 */
163 static inline int hrtimer_callback_running(struct hrtimer *timer)
164 {
165 return timer->state & HRTIMER_STATE_CALLBACK;
166 }
167
168 /*
169 * Functions and macros which are different for UP/SMP systems are kept in a
170 * single place
171 */
172 #ifdef CONFIG_SMP
173
174 /*
175 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
176 * means that all timers which are tied to this base via timer->base are
177 * locked, and the base itself is locked too.
178 *
179 * So __run_timers/migrate_timers can safely modify all timers which could
180 * be found on the lists/queues.
181 *
182 * When the timer's base is locked, and the timer removed from list, it is
183 * possible to set timer->base = NULL and drop the lock: the timer remains
184 * locked.
185 */
186 static
187 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
188 unsigned long *flags)
189 {
190 struct hrtimer_clock_base *base;
191
192 for (;;) {
193 base = timer->base;
194 if (likely(base != NULL)) {
195 spin_lock_irqsave(&base->cpu_base->lock, *flags);
196 if (likely(base == timer->base))
197 return base;
198 /* The timer has migrated to another CPU: */
199 spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
200 }
201 cpu_relax();
202 }
203 }
204
205 /*
206 * Switch the timer base to the current CPU when possible.
207 */
208 static inline struct hrtimer_clock_base *
209 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base)
210 {
211 struct hrtimer_clock_base *new_base;
212 struct hrtimer_cpu_base *new_cpu_base;
213
214 new_cpu_base = &__get_cpu_var(hrtimer_bases);
215 new_base = &new_cpu_base->clock_base[base->index];
216
217 if (base != new_base) {
218 /*
219 * We are trying to schedule the timer on the local CPU.
220 * However we can't change timer's base while it is running,
221 * so we keep it on the same CPU. No hassle vs. reprogramming
222 * the event source in the high resolution case. The softirq
223 * code will take care of this when the timer function has
224 * completed. There is no conflict as we hold the lock until
225 * the timer is enqueued.
226 */
227 if (unlikely(hrtimer_callback_running(timer)))
228 return base;
229
230 /* See the comment in lock_timer_base() */
231 timer->base = NULL;
232 spin_unlock(&base->cpu_base->lock);
233 spin_lock(&new_base->cpu_base->lock);
234 timer->base = new_base;
235 }
236 return new_base;
237 }
238
239 #else /* CONFIG_SMP */
240
241 static inline struct hrtimer_clock_base *
242 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
243 {
244 struct hrtimer_clock_base *base = timer->base;
245
246 spin_lock_irqsave(&base->cpu_base->lock, *flags);
247
248 return base;
249 }
250
251 # define switch_hrtimer_base(t, b) (b)
252
253 #endif /* !CONFIG_SMP */
254
255 /*
256 * Functions for the union type storage format of ktime_t which are
257 * too large for inlining:
258 */
259 #if BITS_PER_LONG < 64
260 # ifndef CONFIG_KTIME_SCALAR
261 /**
262 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
263 * @kt: addend
264 * @nsec: the scalar nsec value to add
265 *
266 * Returns the sum of kt and nsec in ktime_t format
267 */
268 ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
269 {
270 ktime_t tmp;
271
272 if (likely(nsec < NSEC_PER_SEC)) {
273 tmp.tv64 = nsec;
274 } else {
275 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
276
277 tmp = ktime_set((long)nsec, rem);
278 }
279
280 return ktime_add(kt, tmp);
281 }
282
283 EXPORT_SYMBOL_GPL(ktime_add_ns);
284 # endif /* !CONFIG_KTIME_SCALAR */
285
286 /*
287 * Divide a ktime value by a nanosecond value
288 */
289 unsigned long ktime_divns(const ktime_t kt, s64 div)
290 {
291 u64 dclc, inc, dns;
292 int sft = 0;
293
294 dclc = dns = ktime_to_ns(kt);
295 inc = div;
296 /* Make sure the divisor is less than 2^32: */
297 while (div >> 32) {
298 sft++;
299 div >>= 1;
300 }
301 dclc >>= sft;
302 do_div(dclc, (unsigned long) div);
303
304 return (unsigned long) dclc;
305 }
306 #endif /* BITS_PER_LONG >= 64 */
307
308 /* High resolution timer related functions */
309 #ifdef CONFIG_HIGH_RES_TIMERS
310
311 /*
312 * High resolution timer enabled ?
313 */
314 static int hrtimer_hres_enabled __read_mostly = 1;
315
316 /*
317 * Enable / Disable high resolution mode
318 */
319 static int __init setup_hrtimer_hres(char *str)
320 {
321 if (!strcmp(str, "off"))
322 hrtimer_hres_enabled = 0;
323 else if (!strcmp(str, "on"))
324 hrtimer_hres_enabled = 1;
325 else
326 return 0;
327 return 1;
328 }
329
330 __setup("highres=", setup_hrtimer_hres);
331
332 /*
333 * hrtimer_high_res_enabled - query, if the highres mode is enabled
334 */
335 static inline int hrtimer_is_hres_enabled(void)
336 {
337 return hrtimer_hres_enabled;
338 }
339
340 /*
341 * Is the high resolution mode active ?
342 */
343 static inline int hrtimer_hres_active(void)
344 {
345 return __get_cpu_var(hrtimer_bases).hres_active;
346 }
347
348 /*
349 * Reprogram the event source with checking both queues for the
350 * next event
351 * Called with interrupts disabled and base->lock held
352 */
353 static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base)
354 {
355 int i;
356 struct hrtimer_clock_base *base = cpu_base->clock_base;
357 ktime_t expires;
358
359 cpu_base->expires_next.tv64 = KTIME_MAX;
360
361 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
362 struct hrtimer *timer;
363
364 if (!base->first)
365 continue;
366 timer = rb_entry(base->first, struct hrtimer, node);
367 expires = ktime_sub(timer->expires, base->offset);
368 if (expires.tv64 < cpu_base->expires_next.tv64)
369 cpu_base->expires_next = expires;
370 }
371
372 if (cpu_base->expires_next.tv64 != KTIME_MAX)
373 tick_program_event(cpu_base->expires_next, 1);
374 }
375
376 /*
377 * Shared reprogramming for clock_realtime and clock_monotonic
378 *
379 * When a timer is enqueued and expires earlier than the already enqueued
380 * timers, we have to check, whether it expires earlier than the timer for
381 * which the clock event device was armed.
382 *
383 * Called with interrupts disabled and base->cpu_base.lock held
384 */
385 static int hrtimer_reprogram(struct hrtimer *timer,
386 struct hrtimer_clock_base *base)
387 {
388 ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next;
389 ktime_t expires = ktime_sub(timer->expires, base->offset);
390 int res;
391
392 /*
393 * When the callback is running, we do not reprogram the clock event
394 * device. The timer callback is either running on a different CPU or
395 * the callback is executed in the hrtimer_interupt context. The
396 * reprogramming is handled either by the softirq, which called the
397 * callback or at the end of the hrtimer_interrupt.
398 */
399 if (hrtimer_callback_running(timer))
400 return 0;
401
402 if (expires.tv64 >= expires_next->tv64)
403 return 0;
404
405 /*
406 * Clockevents returns -ETIME, when the event was in the past.
407 */
408 res = tick_program_event(expires, 0);
409 if (!IS_ERR_VALUE(res))
410 *expires_next = expires;
411 return res;
412 }
413
414
415 /*
416 * Retrigger next event is called after clock was set
417 *
418 * Called with interrupts disabled via on_each_cpu()
419 */
420 static void retrigger_next_event(void *arg)
421 {
422 struct hrtimer_cpu_base *base;
423 struct timespec realtime_offset;
424 unsigned long seq;
425
426 if (!hrtimer_hres_active())
427 return;
428
429 do {
430 seq = read_seqbegin(&xtime_lock);
431 set_normalized_timespec(&realtime_offset,
432 -wall_to_monotonic.tv_sec,
433 -wall_to_monotonic.tv_nsec);
434 } while (read_seqretry(&xtime_lock, seq));
435
436 base = &__get_cpu_var(hrtimer_bases);
437
438 /* Adjust CLOCK_REALTIME offset */
439 spin_lock(&base->lock);
440 base->clock_base[CLOCK_REALTIME].offset =
441 timespec_to_ktime(realtime_offset);
442
443 hrtimer_force_reprogram(base);
444 spin_unlock(&base->lock);
445 }
446
447 /*
448 * Clock realtime was set
449 *
450 * Change the offset of the realtime clock vs. the monotonic
451 * clock.
452 *
453 * We might have to reprogram the high resolution timer interrupt. On
454 * SMP we call the architecture specific code to retrigger _all_ high
455 * resolution timer interrupts. On UP we just disable interrupts and
456 * call the high resolution interrupt code.
457 */
458 void clock_was_set(void)
459 {
460 /* Retrigger the CPU local events everywhere */
461 on_each_cpu(retrigger_next_event, NULL, 0, 1);
462 }
463
464 /*
465 * During resume we might have to reprogram the high resolution timer
466 * interrupt (on the local CPU):
467 */
468 void hres_timers_resume(void)
469 {
470 WARN_ON_ONCE(num_online_cpus() > 1);
471
472 /* Retrigger the CPU local events: */
473 retrigger_next_event(NULL);
474 }
475
476 /*
477 * Check, whether the timer is on the callback pending list
478 */
479 static inline int hrtimer_cb_pending(const struct hrtimer *timer)
480 {
481 return timer->state & HRTIMER_STATE_PENDING;
482 }
483
484 /*
485 * Remove a timer from the callback pending list
486 */
487 static inline void hrtimer_remove_cb_pending(struct hrtimer *timer)
488 {
489 list_del_init(&timer->cb_entry);
490 }
491
492 /*
493 * Initialize the high resolution related parts of cpu_base
494 */
495 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
496 {
497 base->expires_next.tv64 = KTIME_MAX;
498 base->hres_active = 0;
499 INIT_LIST_HEAD(&base->cb_pending);
500 }
501
502 /*
503 * Initialize the high resolution related parts of a hrtimer
504 */
505 static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
506 {
507 INIT_LIST_HEAD(&timer->cb_entry);
508 }
509
510 /*
511 * When High resolution timers are active, try to reprogram. Note, that in case
512 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
513 * check happens. The timer gets enqueued into the rbtree. The reprogramming
514 * and expiry check is done in the hrtimer_interrupt or in the softirq.
515 */
516 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
517 struct hrtimer_clock_base *base)
518 {
519 if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
520
521 /* Timer is expired, act upon the callback mode */
522 switch(timer->cb_mode) {
523 case HRTIMER_CB_IRQSAFE_NO_RESTART:
524 /*
525 * We can call the callback from here. No restart
526 * happens, so no danger of recursion
527 */
528 BUG_ON(timer->function(timer) != HRTIMER_NORESTART);
529 return 1;
530 case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ:
531 /*
532 * This is solely for the sched tick emulation with
533 * dynamic tick support to ensure that we do not
534 * restart the tick right on the edge and end up with
535 * the tick timer in the softirq ! The calling site
536 * takes care of this.
537 */
538 return 1;
539 case HRTIMER_CB_IRQSAFE:
540 case HRTIMER_CB_SOFTIRQ:
541 /*
542 * Move everything else into the softirq pending list !
543 */
544 list_add_tail(&timer->cb_entry,
545 &base->cpu_base->cb_pending);
546 timer->state = HRTIMER_STATE_PENDING;
547 raise_softirq(HRTIMER_SOFTIRQ);
548 return 1;
549 default:
550 BUG();
551 }
552 }
553 return 0;
554 }
555
556 /*
557 * Switch to high resolution mode
558 */
559 static int hrtimer_switch_to_hres(void)
560 {
561 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
562 unsigned long flags;
563
564 if (base->hres_active)
565 return 1;
566
567 local_irq_save(flags);
568
569 if (tick_init_highres()) {
570 local_irq_restore(flags);
571 return 0;
572 }
573 base->hres_active = 1;
574 base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
575 base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
576
577 tick_setup_sched_timer();
578
579 /* "Retrigger" the interrupt to get things going */
580 retrigger_next_event(NULL);
581 local_irq_restore(flags);
582 printk(KERN_INFO "Switched to high resolution mode on CPU %d\n",
583 smp_processor_id());
584 return 1;
585 }
586
587 #else
588
589 static inline int hrtimer_hres_active(void) { return 0; }
590 static inline int hrtimer_is_hres_enabled(void) { return 0; }
591 static inline int hrtimer_switch_to_hres(void) { return 0; }
592 static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { }
593 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
594 struct hrtimer_clock_base *base)
595 {
596 return 0;
597 }
598 static inline int hrtimer_cb_pending(struct hrtimer *timer) { return 0; }
599 static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) { }
600 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
601 static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
602
603 #endif /* CONFIG_HIGH_RES_TIMERS */
604
605 #ifdef CONFIG_TIMER_STATS
606 void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr)
607 {
608 if (timer->start_site)
609 return;
610
611 timer->start_site = addr;
612 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
613 timer->start_pid = current->pid;
614 }
615 #endif
616
617 /*
618 * Counterpart to lock_timer_base above:
619 */
620 static inline
621 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
622 {
623 spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
624 }
625
626 /**
627 * hrtimer_forward - forward the timer expiry
628 * @timer: hrtimer to forward
629 * @now: forward past this time
630 * @interval: the interval to forward
631 *
632 * Forward the timer expiry so it will expire in the future.
633 * Returns the number of overruns.
634 */
635 unsigned long
636 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
637 {
638 unsigned long orun = 1;
639 ktime_t delta;
640
641 delta = ktime_sub(now, timer->expires);
642
643 if (delta.tv64 < 0)
644 return 0;
645
646 if (interval.tv64 < timer->base->resolution.tv64)
647 interval.tv64 = timer->base->resolution.tv64;
648
649 if (unlikely(delta.tv64 >= interval.tv64)) {
650 s64 incr = ktime_to_ns(interval);
651
652 orun = ktime_divns(delta, incr);
653 timer->expires = ktime_add_ns(timer->expires, incr * orun);
654 if (timer->expires.tv64 > now.tv64)
655 return orun;
656 /*
657 * This (and the ktime_add() below) is the
658 * correction for exact:
659 */
660 orun++;
661 }
662 timer->expires = ktime_add(timer->expires, interval);
663 /*
664 * Make sure, that the result did not wrap with a very large
665 * interval.
666 */
667 if (timer->expires.tv64 < 0)
668 timer->expires = ktime_set(KTIME_SEC_MAX, 0);
669
670 return orun;
671 }
672 EXPORT_SYMBOL_GPL(hrtimer_forward);
673
674 /*
675 * enqueue_hrtimer - internal function to (re)start a timer
676 *
677 * The timer is inserted in expiry order. Insertion into the
678 * red black tree is O(log(n)). Must hold the base lock.
679 */
680 static void enqueue_hrtimer(struct hrtimer *timer,
681 struct hrtimer_clock_base *base, int reprogram)
682 {
683 struct rb_node **link = &base->active.rb_node;
684 struct rb_node *parent = NULL;
685 struct hrtimer *entry;
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 }
702
703 /*
704 * Insert the timer to the rbtree and check whether it
705 * replaces the first pending timer
706 */
707 if (!base->first || timer->expires.tv64 <
708 rb_entry(base->first, struct hrtimer, node)->expires.tv64) {
709 /*
710 * Reprogram the clock event device. When the timer is already
711 * expired hrtimer_enqueue_reprogram has either called the
712 * callback or added it to the pending list and raised the
713 * softirq.
714 *
715 * This is a NOP for !HIGHRES
716 */
717 if (reprogram && hrtimer_enqueue_reprogram(timer, base))
718 return;
719
720 base->first = &timer->node;
721 }
722
723 rb_link_node(&timer->node, parent, link);
724 rb_insert_color(&timer->node, &base->active);
725 /*
726 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
727 * state of a possibly running callback.
728 */
729 timer->state |= HRTIMER_STATE_ENQUEUED;
730 }
731
732 /*
733 * __remove_hrtimer - internal function to remove a timer
734 *
735 * Caller must hold the base lock.
736 *
737 * High resolution timer mode reprograms the clock event device when the
738 * timer is the one which expires next. The caller can disable this by setting
739 * reprogram to zero. This is useful, when the context does a reprogramming
740 * anyway (e.g. timer interrupt)
741 */
742 static void __remove_hrtimer(struct hrtimer *timer,
743 struct hrtimer_clock_base *base,
744 unsigned long newstate, int reprogram)
745 {
746 /* High res. callback list. NOP for !HIGHRES */
747 if (hrtimer_cb_pending(timer))
748 hrtimer_remove_cb_pending(timer);
749 else {
750 /*
751 * Remove the timer from the rbtree and replace the
752 * first entry pointer if necessary.
753 */
754 if (base->first == &timer->node) {
755 base->first = rb_next(&timer->node);
756 /* Reprogram the clock event device. if enabled */
757 if (reprogram && hrtimer_hres_active())
758 hrtimer_force_reprogram(base->cpu_base);
759 }
760 rb_erase(&timer->node, &base->active);
761 }
762 timer->state = newstate;
763 }
764
765 /*
766 * remove hrtimer, called with base lock held
767 */
768 static inline int
769 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
770 {
771 if (hrtimer_is_queued(timer)) {
772 int reprogram;
773
774 /*
775 * Remove the timer and force reprogramming when high
776 * resolution mode is active and the timer is on the current
777 * CPU. If we remove a timer on another CPU, reprogramming is
778 * skipped. The interrupt event on this CPU is fired and
779 * reprogramming happens in the interrupt handler. This is a
780 * rare case and less expensive than a smp call.
781 */
782 timer_stats_hrtimer_clear_start_info(timer);
783 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
784 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
785 reprogram);
786 return 1;
787 }
788 return 0;
789 }
790
791 /**
792 * hrtimer_start - (re)start an relative timer on the current CPU
793 * @timer: the timer to be added
794 * @tim: expiry time
795 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
796 *
797 * Returns:
798 * 0 on success
799 * 1 when the timer was active
800 */
801 int
802 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
803 {
804 struct hrtimer_clock_base *base, *new_base;
805 unsigned long flags;
806 int ret;
807
808 base = lock_hrtimer_base(timer, &flags);
809
810 /* Remove an active timer from the queue: */
811 ret = remove_hrtimer(timer, base);
812
813 /* Switch the timer base, if necessary: */
814 new_base = switch_hrtimer_base(timer, base);
815
816 if (mode == HRTIMER_MODE_REL) {
817 tim = ktime_add(tim, new_base->get_time());
818 /*
819 * CONFIG_TIME_LOW_RES is a temporary way for architectures
820 * to signal that they simply return xtime in
821 * do_gettimeoffset(). In this case we want to round up by
822 * resolution when starting a relative timer, to avoid short
823 * timeouts. This will go away with the GTOD framework.
824 */
825 #ifdef CONFIG_TIME_LOW_RES
826 tim = ktime_add(tim, base->resolution);
827 #endif
828 }
829 timer->expires = tim;
830
831 timer_stats_hrtimer_set_start_info(timer);
832
833 /*
834 * Only allow reprogramming if the new base is on this CPU.
835 * (it might still be on another CPU if the timer was pending)
836 */
837 enqueue_hrtimer(timer, new_base,
838 new_base->cpu_base == &__get_cpu_var(hrtimer_bases));
839
840 unlock_hrtimer_base(timer, &flags);
841
842 return ret;
843 }
844 EXPORT_SYMBOL_GPL(hrtimer_start);
845
846 /**
847 * hrtimer_try_to_cancel - try to deactivate a timer
848 * @timer: hrtimer to stop
849 *
850 * Returns:
851 * 0 when the timer was not active
852 * 1 when the timer was active
853 * -1 when the timer is currently excuting the callback function and
854 * cannot be stopped
855 */
856 int hrtimer_try_to_cancel(struct hrtimer *timer)
857 {
858 struct hrtimer_clock_base *base;
859 unsigned long flags;
860 int ret = -1;
861
862 base = lock_hrtimer_base(timer, &flags);
863
864 if (!hrtimer_callback_running(timer))
865 ret = remove_hrtimer(timer, base);
866
867 unlock_hrtimer_base(timer, &flags);
868
869 return ret;
870
871 }
872 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
873
874 /**
875 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
876 * @timer: the timer to be cancelled
877 *
878 * Returns:
879 * 0 when the timer was not active
880 * 1 when the timer was active
881 */
882 int hrtimer_cancel(struct hrtimer *timer)
883 {
884 for (;;) {
885 int ret = hrtimer_try_to_cancel(timer);
886
887 if (ret >= 0)
888 return ret;
889 cpu_relax();
890 }
891 }
892 EXPORT_SYMBOL_GPL(hrtimer_cancel);
893
894 /**
895 * hrtimer_get_remaining - get remaining time for the timer
896 * @timer: the timer to read
897 */
898 ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
899 {
900 struct hrtimer_clock_base *base;
901 unsigned long flags;
902 ktime_t rem;
903
904 base = lock_hrtimer_base(timer, &flags);
905 rem = ktime_sub(timer->expires, base->get_time());
906 unlock_hrtimer_base(timer, &flags);
907
908 return rem;
909 }
910 EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
911
912 #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
913 /**
914 * hrtimer_get_next_event - get the time until next expiry event
915 *
916 * Returns the delta to the next expiry event or KTIME_MAX if no timer
917 * is pending.
918 */
919 ktime_t hrtimer_get_next_event(void)
920 {
921 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
922 struct hrtimer_clock_base *base = cpu_base->clock_base;
923 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
924 unsigned long flags;
925 int i;
926
927 spin_lock_irqsave(&cpu_base->lock, flags);
928
929 if (!hrtimer_hres_active()) {
930 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
931 struct hrtimer *timer;
932
933 if (!base->first)
934 continue;
935
936 timer = rb_entry(base->first, struct hrtimer, node);
937 delta.tv64 = timer->expires.tv64;
938 delta = ktime_sub(delta, base->get_time());
939 if (delta.tv64 < mindelta.tv64)
940 mindelta.tv64 = delta.tv64;
941 }
942 }
943
944 spin_unlock_irqrestore(&cpu_base->lock, flags);
945
946 if (mindelta.tv64 < 0)
947 mindelta.tv64 = 0;
948 return mindelta;
949 }
950 #endif
951
952 /**
953 * hrtimer_init - initialize a timer to the given clock
954 * @timer: the timer to be initialized
955 * @clock_id: the clock to be used
956 * @mode: timer mode abs/rel
957 */
958 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
959 enum hrtimer_mode mode)
960 {
961 struct hrtimer_cpu_base *cpu_base;
962
963 memset(timer, 0, sizeof(struct hrtimer));
964
965 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
966
967 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
968 clock_id = CLOCK_MONOTONIC;
969
970 timer->base = &cpu_base->clock_base[clock_id];
971 hrtimer_init_timer_hres(timer);
972
973 #ifdef CONFIG_TIMER_STATS
974 timer->start_site = NULL;
975 timer->start_pid = -1;
976 memset(timer->start_comm, 0, TASK_COMM_LEN);
977 #endif
978 }
979 EXPORT_SYMBOL_GPL(hrtimer_init);
980
981 /**
982 * hrtimer_get_res - get the timer resolution for a clock
983 * @which_clock: which clock to query
984 * @tp: pointer to timespec variable to store the resolution
985 *
986 * Store the resolution of the clock selected by @which_clock in the
987 * variable pointed to by @tp.
988 */
989 int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
990 {
991 struct hrtimer_cpu_base *cpu_base;
992
993 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
994 *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
995
996 return 0;
997 }
998 EXPORT_SYMBOL_GPL(hrtimer_get_res);
999
1000 #ifdef CONFIG_HIGH_RES_TIMERS
1001
1002 /*
1003 * High resolution timer interrupt
1004 * Called with interrupts disabled
1005 */
1006 void hrtimer_interrupt(struct clock_event_device *dev)
1007 {
1008 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1009 struct hrtimer_clock_base *base;
1010 ktime_t expires_next, now;
1011 int i, raise = 0;
1012
1013 BUG_ON(!cpu_base->hres_active);
1014 cpu_base->nr_events++;
1015 dev->next_event.tv64 = KTIME_MAX;
1016
1017 retry:
1018 now = ktime_get();
1019
1020 expires_next.tv64 = KTIME_MAX;
1021
1022 base = cpu_base->clock_base;
1023
1024 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1025 ktime_t basenow;
1026 struct rb_node *node;
1027
1028 spin_lock(&cpu_base->lock);
1029
1030 basenow = ktime_add(now, base->offset);
1031
1032 while ((node = base->first)) {
1033 struct hrtimer *timer;
1034
1035 timer = rb_entry(node, struct hrtimer, node);
1036
1037 if (basenow.tv64 < timer->expires.tv64) {
1038 ktime_t expires;
1039
1040 expires = ktime_sub(timer->expires,
1041 base->offset);
1042 if (expires.tv64 < expires_next.tv64)
1043 expires_next = expires;
1044 break;
1045 }
1046
1047 /* Move softirq callbacks to the pending list */
1048 if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) {
1049 __remove_hrtimer(timer, base,
1050 HRTIMER_STATE_PENDING, 0);
1051 list_add_tail(&timer->cb_entry,
1052 &base->cpu_base->cb_pending);
1053 raise = 1;
1054 continue;
1055 }
1056
1057 __remove_hrtimer(timer, base,
1058 HRTIMER_STATE_CALLBACK, 0);
1059 timer_stats_account_hrtimer(timer);
1060
1061 /*
1062 * Note: We clear the CALLBACK bit after
1063 * enqueue_hrtimer to avoid reprogramming of
1064 * the event hardware. This happens at the end
1065 * of this function anyway.
1066 */
1067 if (timer->function(timer) != HRTIMER_NORESTART) {
1068 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1069 enqueue_hrtimer(timer, base, 0);
1070 }
1071 timer->state &= ~HRTIMER_STATE_CALLBACK;
1072 }
1073 spin_unlock(&cpu_base->lock);
1074 base++;
1075 }
1076
1077 cpu_base->expires_next = expires_next;
1078
1079 /* Reprogramming necessary ? */
1080 if (expires_next.tv64 != KTIME_MAX) {
1081 if (tick_program_event(expires_next, 0))
1082 goto retry;
1083 }
1084
1085 /* Raise softirq ? */
1086 if (raise)
1087 raise_softirq(HRTIMER_SOFTIRQ);
1088 }
1089
1090 static void run_hrtimer_softirq(struct softirq_action *h)
1091 {
1092 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1093
1094 spin_lock_irq(&cpu_base->lock);
1095
1096 while (!list_empty(&cpu_base->cb_pending)) {
1097 enum hrtimer_restart (*fn)(struct hrtimer *);
1098 struct hrtimer *timer;
1099 int restart;
1100
1101 timer = list_entry(cpu_base->cb_pending.next,
1102 struct hrtimer, cb_entry);
1103
1104 timer_stats_account_hrtimer(timer);
1105
1106 fn = timer->function;
1107 __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0);
1108 spin_unlock_irq(&cpu_base->lock);
1109
1110 restart = fn(timer);
1111
1112 spin_lock_irq(&cpu_base->lock);
1113
1114 timer->state &= ~HRTIMER_STATE_CALLBACK;
1115 if (restart == HRTIMER_RESTART) {
1116 BUG_ON(hrtimer_active(timer));
1117 /*
1118 * Enqueue the timer, allow reprogramming of the event
1119 * device
1120 */
1121 enqueue_hrtimer(timer, timer->base, 1);
1122 } else if (hrtimer_active(timer)) {
1123 /*
1124 * If the timer was rearmed on another CPU, reprogram
1125 * the event device.
1126 */
1127 if (timer->base->first == &timer->node)
1128 hrtimer_reprogram(timer, timer->base);
1129 }
1130 }
1131 spin_unlock_irq(&cpu_base->lock);
1132 }
1133
1134 #endif /* CONFIG_HIGH_RES_TIMERS */
1135
1136 /*
1137 * Expire the per base hrtimer-queue:
1138 */
1139 static inline void run_hrtimer_queue(struct hrtimer_cpu_base *cpu_base,
1140 int index)
1141 {
1142 struct rb_node *node;
1143 struct hrtimer_clock_base *base = &cpu_base->clock_base[index];
1144
1145 if (!base->first)
1146 return;
1147
1148 if (base->get_softirq_time)
1149 base->softirq_time = base->get_softirq_time();
1150
1151 spin_lock_irq(&cpu_base->lock);
1152
1153 while ((node = base->first)) {
1154 struct hrtimer *timer;
1155 enum hrtimer_restart (*fn)(struct hrtimer *);
1156 int restart;
1157
1158 timer = rb_entry(node, struct hrtimer, node);
1159 if (base->softirq_time.tv64 <= timer->expires.tv64)
1160 break;
1161
1162 #ifdef CONFIG_HIGH_RES_TIMERS
1163 WARN_ON_ONCE(timer->cb_mode == HRTIMER_CB_IRQSAFE_NO_SOFTIRQ);
1164 #endif
1165 timer_stats_account_hrtimer(timer);
1166
1167 fn = timer->function;
1168 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1169 spin_unlock_irq(&cpu_base->lock);
1170
1171 restart = fn(timer);
1172
1173 spin_lock_irq(&cpu_base->lock);
1174
1175 timer->state &= ~HRTIMER_STATE_CALLBACK;
1176 if (restart != HRTIMER_NORESTART) {
1177 BUG_ON(hrtimer_active(timer));
1178 enqueue_hrtimer(timer, base, 0);
1179 }
1180 }
1181 spin_unlock_irq(&cpu_base->lock);
1182 }
1183
1184 /*
1185 * Called from timer softirq every jiffy, expire hrtimers:
1186 *
1187 * For HRT its the fall back code to run the softirq in the timer
1188 * softirq context in case the hrtimer initialization failed or has
1189 * not been done yet.
1190 */
1191 void hrtimer_run_queues(void)
1192 {
1193 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1194 int i;
1195
1196 if (hrtimer_hres_active())
1197 return;
1198
1199 /*
1200 * This _is_ ugly: We have to check in the softirq context,
1201 * whether we can switch to highres and / or nohz mode. The
1202 * clocksource switch happens in the timer interrupt with
1203 * xtime_lock held. Notification from there only sets the
1204 * check bit in the tick_oneshot code, otherwise we might
1205 * deadlock vs. xtime_lock.
1206 */
1207 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1208 if (hrtimer_switch_to_hres())
1209 return;
1210
1211 hrtimer_get_softirq_time(cpu_base);
1212
1213 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1214 run_hrtimer_queue(cpu_base, i);
1215 }
1216
1217 /*
1218 * Sleep related functions:
1219 */
1220 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1221 {
1222 struct hrtimer_sleeper *t =
1223 container_of(timer, struct hrtimer_sleeper, timer);
1224 struct task_struct *task = t->task;
1225
1226 t->task = NULL;
1227 if (task)
1228 wake_up_process(task);
1229
1230 return HRTIMER_NORESTART;
1231 }
1232
1233 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1234 {
1235 sl->timer.function = hrtimer_wakeup;
1236 sl->task = task;
1237 #ifdef CONFIG_HIGH_RES_TIMERS
1238 sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_RESTART;
1239 #endif
1240 }
1241
1242 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1243 {
1244 hrtimer_init_sleeper(t, current);
1245
1246 do {
1247 set_current_state(TASK_INTERRUPTIBLE);
1248 hrtimer_start(&t->timer, t->timer.expires, mode);
1249
1250 if (likely(t->task))
1251 schedule();
1252
1253 hrtimer_cancel(&t->timer);
1254 mode = HRTIMER_MODE_ABS;
1255
1256 } while (t->task && !signal_pending(current));
1257
1258 return t->task == NULL;
1259 }
1260
1261 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1262 {
1263 struct hrtimer_sleeper t;
1264 struct timespec __user *rmtp;
1265 struct timespec tu;
1266 ktime_t time;
1267
1268 restart->fn = do_no_restart_syscall;
1269
1270 hrtimer_init(&t.timer, restart->arg0, HRTIMER_MODE_ABS);
1271 t.timer.expires.tv64 = ((u64)restart->arg3 << 32) | (u64) restart->arg2;
1272
1273 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1274 return 0;
1275
1276 rmtp = (struct timespec __user *) restart->arg1;
1277 if (rmtp) {
1278 time = ktime_sub(t.timer.expires, t.timer.base->get_time());
1279 if (time.tv64 <= 0)
1280 return 0;
1281 tu = ktime_to_timespec(time);
1282 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1283 return -EFAULT;
1284 }
1285
1286 restart->fn = hrtimer_nanosleep_restart;
1287
1288 /* The other values in restart are already filled in */
1289 return -ERESTART_RESTARTBLOCK;
1290 }
1291
1292 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1293 const enum hrtimer_mode mode, const clockid_t clockid)
1294 {
1295 struct restart_block *restart;
1296 struct hrtimer_sleeper t;
1297 struct timespec tu;
1298 ktime_t rem;
1299
1300 hrtimer_init(&t.timer, clockid, mode);
1301 t.timer.expires = timespec_to_ktime(*rqtp);
1302 if (do_nanosleep(&t, mode))
1303 return 0;
1304
1305 /* Absolute timers do not update the rmtp value and restart: */
1306 if (mode == HRTIMER_MODE_ABS)
1307 return -ERESTARTNOHAND;
1308
1309 if (rmtp) {
1310 rem = ktime_sub(t.timer.expires, t.timer.base->get_time());
1311 if (rem.tv64 <= 0)
1312 return 0;
1313 tu = ktime_to_timespec(rem);
1314 if (copy_to_user(rmtp, &tu, sizeof(tu)))
1315 return -EFAULT;
1316 }
1317
1318 restart = &current_thread_info()->restart_block;
1319 restart->fn = hrtimer_nanosleep_restart;
1320 restart->arg0 = (unsigned long) t.timer.base->index;
1321 restart->arg1 = (unsigned long) rmtp;
1322 restart->arg2 = t.timer.expires.tv64 & 0xFFFFFFFF;
1323 restart->arg3 = t.timer.expires.tv64 >> 32;
1324
1325 return -ERESTART_RESTARTBLOCK;
1326 }
1327
1328 asmlinkage long
1329 sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1330 {
1331 struct timespec tu;
1332
1333 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1334 return -EFAULT;
1335
1336 if (!timespec_valid(&tu))
1337 return -EINVAL;
1338
1339 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1340 }
1341
1342 /*
1343 * Functions related to boot-time initialization:
1344 */
1345 static void __devinit init_hrtimers_cpu(int cpu)
1346 {
1347 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1348 int i;
1349
1350 spin_lock_init(&cpu_base->lock);
1351 lockdep_set_class(&cpu_base->lock, &cpu_base->lock_key);
1352
1353 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1354 cpu_base->clock_base[i].cpu_base = cpu_base;
1355
1356 hrtimer_init_hres(cpu_base);
1357 }
1358
1359 #ifdef CONFIG_HOTPLUG_CPU
1360
1361 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1362 struct hrtimer_clock_base *new_base)
1363 {
1364 struct hrtimer *timer;
1365 struct rb_node *node;
1366
1367 while ((node = rb_first(&old_base->active))) {
1368 timer = rb_entry(node, struct hrtimer, node);
1369 BUG_ON(hrtimer_callback_running(timer));
1370 __remove_hrtimer(timer, old_base, HRTIMER_STATE_INACTIVE, 0);
1371 timer->base = new_base;
1372 /*
1373 * Enqueue the timer. Allow reprogramming of the event device
1374 */
1375 enqueue_hrtimer(timer, new_base, 1);
1376 }
1377 }
1378
1379 static void migrate_hrtimers(int cpu)
1380 {
1381 struct hrtimer_cpu_base *old_base, *new_base;
1382 int i;
1383
1384 BUG_ON(cpu_online(cpu));
1385 old_base = &per_cpu(hrtimer_bases, cpu);
1386 new_base = &get_cpu_var(hrtimer_bases);
1387
1388 tick_cancel_sched_timer(cpu);
1389
1390 local_irq_disable();
1391 double_spin_lock(&new_base->lock, &old_base->lock,
1392 smp_processor_id() < cpu);
1393
1394 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1395 migrate_hrtimer_list(&old_base->clock_base[i],
1396 &new_base->clock_base[i]);
1397 }
1398
1399 double_spin_unlock(&new_base->lock, &old_base->lock,
1400 smp_processor_id() < cpu);
1401 local_irq_enable();
1402 put_cpu_var(hrtimer_bases);
1403 }
1404 #endif /* CONFIG_HOTPLUG_CPU */
1405
1406 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1407 unsigned long action, void *hcpu)
1408 {
1409 long cpu = (long)hcpu;
1410
1411 switch (action) {
1412
1413 case CPU_UP_PREPARE:
1414 case CPU_UP_PREPARE_FROZEN:
1415 init_hrtimers_cpu(cpu);
1416 break;
1417
1418 #ifdef CONFIG_HOTPLUG_CPU
1419 case CPU_DEAD:
1420 case CPU_DEAD_FROZEN:
1421 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu);
1422 migrate_hrtimers(cpu);
1423 break;
1424 #endif
1425
1426 default:
1427 break;
1428 }
1429
1430 return NOTIFY_OK;
1431 }
1432
1433 static struct notifier_block __cpuinitdata hrtimers_nb = {
1434 .notifier_call = hrtimer_cpu_notify,
1435 };
1436
1437 void __init hrtimers_init(void)
1438 {
1439 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1440 (void *)(long)smp_processor_id());
1441 register_cpu_notifier(&hrtimers_nb);
1442 #ifdef CONFIG_HIGH_RES_TIMERS
1443 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq, NULL);
1444 #endif
1445 }
1446