projects/linux·blob v3.16 · refs

1915 lines · 48.2 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/export.h>
36#include <linux/percpu.h>
37#include <linux/hrtimer.h>
38#include <linux/notifier.h>
39#include <linux/syscalls.h>
40#include <linux/kallsyms.h>
41#include <linux/interrupt.h>
42#include <linux/tick.h>
43#include <linux/seq_file.h>
44#include <linux/err.h>
45#include <linux/debugobjects.h>
46#include <linux/sched.h>
47#include <linux/sched/sysctl.h>
48#include <linux/sched/rt.h>
49#include <linux/sched/deadline.h>
50#include <linux/timer.h>
51#include <linux/freezer.h>
52
53#include <asm/uaccess.h>
54
55#include <trace/events/timer.h>
56
57/*
58 * The timer bases:
59 *
60 * There are more clockids then hrtimer bases. Thus, we index
61 * into the timer bases by the hrtimer_base_type enum. When trying
62 * to reach a base using a clockid, hrtimer_clockid_to_base()
63 * is used to convert from clockid to the proper hrtimer_base_type.
64 */
65DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
66{
67
68 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
69 .clock_base =
70 {
71 {
72 .index = HRTIMER_BASE_MONOTONIC,
73 .clockid = CLOCK_MONOTONIC,
74 .get_time = &ktime_get,
75 .resolution = KTIME_LOW_RES,
76 },
77 {
78 .index = HRTIMER_BASE_REALTIME,
79 .clockid = CLOCK_REALTIME,
80 .get_time = &ktime_get_real,
81 .resolution = KTIME_LOW_RES,
82 },
83 {
84 .index = HRTIMER_BASE_BOOTTIME,
85 .clockid = CLOCK_BOOTTIME,
86 .get_time = &ktime_get_boottime,
87 .resolution = KTIME_LOW_RES,
88 },
89 {
90 .index = HRTIMER_BASE_TAI,
91 .clockid = CLOCK_TAI,
92 .get_time = &ktime_get_clocktai,
93 .resolution = KTIME_LOW_RES,
94 },
95 }
96};
97
98static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
99 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
100 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
101 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
102 [CLOCK_TAI] = HRTIMER_BASE_TAI,
103};
104
105static inline int hrtimer_clockid_to_base(clockid_t clock_id)
106{
107 return hrtimer_clock_to_base_table[clock_id];
108}
109
110
111/*
112 * Get the coarse grained time at the softirq based on xtime and
113 * wall_to_monotonic.
114 */
115static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
116{
117 ktime_t xtim, mono, boot;
118 struct timespec xts, tom, slp;
119 s32 tai_offset;
120
121 get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp);
122 tai_offset = timekeeping_get_tai_offset();
123
124 xtim = timespec_to_ktime(xts);
125 mono = ktime_add(xtim, timespec_to_ktime(tom));
126 boot = ktime_add(mono, timespec_to_ktime(slp));
127 base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
128 base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono;
129 base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot;
130 base->clock_base[HRTIMER_BASE_TAI].softirq_time =
131 ktime_add(xtim, ktime_set(tai_offset, 0));
132}
133
134/*
135 * Functions and macros which are different for UP/SMP systems are kept in a
136 * single place
137 */
138#ifdef CONFIG_SMP
139
140/*
141 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
142 * means that all timers which are tied to this base via timer->base are
143 * locked, and the base itself is locked too.
144 *
145 * So __run_timers/migrate_timers can safely modify all timers which could
146 * be found on the lists/queues.
147 *
148 * When the timer's base is locked, and the timer removed from list, it is
149 * possible to set timer->base = NULL and drop the lock: the timer remains
150 * locked.
151 */
152static
153struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
154 unsigned long *flags)
155{
156 struct hrtimer_clock_base *base;
157
158 for (;;) {
159 base = timer->base;
160 if (likely(base != NULL)) {
161 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
162 if (likely(base == timer->base))
163 return base;
164 /* The timer has migrated to another CPU: */
165 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
166 }
167 cpu_relax();
168 }
169}
170
171/*
172 * With HIGHRES=y we do not migrate the timer when it is expiring
173 * before the next event on the target cpu because we cannot reprogram
174 * the target cpu hardware and we would cause it to fire late.
175 *
176 * Called with cpu_base->lock of target cpu held.
177 */
178static int
179hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
180{
181#ifdef CONFIG_HIGH_RES_TIMERS
182 ktime_t expires;
183
184 if (!new_base->cpu_base->hres_active)
185 return 0;
186
187 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
188 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
189#else
190 return 0;
191#endif
192}
193
194/*
195 * Switch the timer base to the current CPU when possible.
196 */
197static inline struct hrtimer_clock_base *
198switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
199 int pinned)
200{
201 struct hrtimer_clock_base *new_base;
202 struct hrtimer_cpu_base *new_cpu_base;
203 int this_cpu = smp_processor_id();
204 int cpu = get_nohz_timer_target(pinned);
205 int basenum = base->index;
206
207again:
208 new_cpu_base = &per_cpu(hrtimer_bases, cpu);
209 new_base = &new_cpu_base->clock_base[basenum];
210
211 if (base != new_base) {
212 /*
213 * We are trying to move timer to new_base.
214 * However we can't change timer's base while it is running,
215 * so we keep it on the same CPU. No hassle vs. reprogramming
216 * the event source in the high resolution case. The softirq
217 * code will take care of this when the timer function has
218 * completed. There is no conflict as we hold the lock until
219 * the timer is enqueued.
220 */
221 if (unlikely(hrtimer_callback_running(timer)))
222 return base;
223
224 /* See the comment in lock_timer_base() */
225 timer->base = NULL;
226 raw_spin_unlock(&base->cpu_base->lock);
227 raw_spin_lock(&new_base->cpu_base->lock);
228
229 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
230 cpu = this_cpu;
231 raw_spin_unlock(&new_base->cpu_base->lock);
232 raw_spin_lock(&base->cpu_base->lock);
233 timer->base = base;
234 goto again;
235 }
236 timer->base = new_base;
237 } else {
238 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
239 cpu = this_cpu;
240 goto again;
241 }
242 }
243 return new_base;
244}
245
246#else /* CONFIG_SMP */
247
248static inline struct hrtimer_clock_base *
249lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
250{
251 struct hrtimer_clock_base *base = timer->base;
252
253 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
254
255 return base;
256}
257
258# define switch_hrtimer_base(t, b, p) (b)
259
260#endif /* !CONFIG_SMP */
261
262/*
263 * Functions for the union type storage format of ktime_t which are
264 * too large for inlining:
265 */
266#if BITS_PER_LONG < 64
267# ifndef CONFIG_KTIME_SCALAR
268/**
269 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
270 * @kt: addend
271 * @nsec: the scalar nsec value to add
272 *
273 * Returns the sum of kt and nsec in ktime_t format
274 */
275ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
276{
277 ktime_t tmp;
278
279 if (likely(nsec < NSEC_PER_SEC)) {
280 tmp.tv64 = nsec;
281 } else {
282 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
283
284 /* Make sure nsec fits into long */
285 if (unlikely(nsec > KTIME_SEC_MAX))
286 return (ktime_t){ .tv64 = KTIME_MAX };
287
288 tmp = ktime_set((long)nsec, rem);
289 }
290
291 return ktime_add(kt, tmp);
292}
293
294EXPORT_SYMBOL_GPL(ktime_add_ns);
295
296/**
297 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
298 * @kt: minuend
299 * @nsec: the scalar nsec value to subtract
300 *
301 * Returns the subtraction of @nsec from @kt in ktime_t format
302 */
303ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
304{
305 ktime_t tmp;
306
307 if (likely(nsec < NSEC_PER_SEC)) {
308 tmp.tv64 = nsec;
309 } else {
310 unsigned long rem = do_div(nsec, NSEC_PER_SEC);
311
312 tmp = ktime_set((long)nsec, rem);
313 }
314
315 return ktime_sub(kt, tmp);
316}
317
318EXPORT_SYMBOL_GPL(ktime_sub_ns);
319# endif /* !CONFIG_KTIME_SCALAR */
320
321/*
322 * Divide a ktime value by a nanosecond value
323 */
324u64 ktime_divns(const ktime_t kt, s64 div)
325{
326 u64 dclc;
327 int sft = 0;
328
329 dclc = ktime_to_ns(kt);
330 /* Make sure the divisor is less than 2^32: */
331 while (div >> 32) {
332 sft++;
333 div >>= 1;
334 }
335 dclc >>= sft;
336 do_div(dclc, (unsigned long) div);
337
338 return dclc;
339}
340#endif /* BITS_PER_LONG >= 64 */
341
342/*
343 * Add two ktime values and do a safety check for overflow:
344 */
345ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
346{
347 ktime_t res = ktime_add(lhs, rhs);
348
349 /*
350 * We use KTIME_SEC_MAX here, the maximum timeout which we can
351 * return to user space in a timespec:
352 */
353 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
354 res = ktime_set(KTIME_SEC_MAX, 0);
355
356 return res;
357}
358
359EXPORT_SYMBOL_GPL(ktime_add_safe);
360
361#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
362
363static struct debug_obj_descr hrtimer_debug_descr;
364
365static void *hrtimer_debug_hint(void *addr)
366{
367 return ((struct hrtimer *) addr)->function;
368}
369
370/*
371 * fixup_init is called when:
372 * - an active object is initialized
373 */
374static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
375{
376 struct hrtimer *timer = addr;
377
378 switch (state) {
379 case ODEBUG_STATE_ACTIVE:
380 hrtimer_cancel(timer);
381 debug_object_init(timer, &hrtimer_debug_descr);
382 return 1;
383 default:
384 return 0;
385 }
386}
387
388/*
389 * fixup_activate is called when:
390 * - an active object is activated
391 * - an unknown object is activated (might be a statically initialized object)
392 */
393static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
394{
395 switch (state) {
396
397 case ODEBUG_STATE_NOTAVAILABLE:
398 WARN_ON_ONCE(1);
399 return 0;
400
401 case ODEBUG_STATE_ACTIVE:
402 WARN_ON(1);
403
404 default:
405 return 0;
406 }
407}
408
409/*
410 * fixup_free is called when:
411 * - an active object is freed
412 */
413static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
414{
415 struct hrtimer *timer = addr;
416
417 switch (state) {
418 case ODEBUG_STATE_ACTIVE:
419 hrtimer_cancel(timer);
420 debug_object_free(timer, &hrtimer_debug_descr);
421 return 1;
422 default:
423 return 0;
424 }
425}
426
427static struct debug_obj_descr hrtimer_debug_descr = {
428 .name = "hrtimer",
429 .debug_hint = hrtimer_debug_hint,
430 .fixup_init = hrtimer_fixup_init,
431 .fixup_activate = hrtimer_fixup_activate,
432 .fixup_free = hrtimer_fixup_free,
433};
434
435static inline void debug_hrtimer_init(struct hrtimer *timer)
436{
437 debug_object_init(timer, &hrtimer_debug_descr);
438}
439
440static inline void debug_hrtimer_activate(struct hrtimer *timer)
441{
442 debug_object_activate(timer, &hrtimer_debug_descr);
443}
444
445static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
446{
447 debug_object_deactivate(timer, &hrtimer_debug_descr);
448}
449
450static inline void debug_hrtimer_free(struct hrtimer *timer)
451{
452 debug_object_free(timer, &hrtimer_debug_descr);
453}
454
455static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
456 enum hrtimer_mode mode);
457
458void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
459 enum hrtimer_mode mode)
460{
461 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
462 __hrtimer_init(timer, clock_id, mode);
463}
464EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
465
466void destroy_hrtimer_on_stack(struct hrtimer *timer)
467{
468 debug_object_free(timer, &hrtimer_debug_descr);
469}
470
471#else
472static inline void debug_hrtimer_init(struct hrtimer *timer) { }
473static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
474static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
475#endif
476
477static inline void
478debug_init(struct hrtimer *timer, clockid_t clockid,
479 enum hrtimer_mode mode)
480{
481 debug_hrtimer_init(timer);
482 trace_hrtimer_init(timer, clockid, mode);
483}
484
485static inline void debug_activate(struct hrtimer *timer)
486{
487 debug_hrtimer_activate(timer);
488 trace_hrtimer_start(timer);
489}
490
491static inline void debug_deactivate(struct hrtimer *timer)
492{
493 debug_hrtimer_deactivate(timer);
494 trace_hrtimer_cancel(timer);
495}
496
497/* High resolution timer related functions */
498#ifdef CONFIG_HIGH_RES_TIMERS
499
500/*
501 * High resolution timer enabled ?
502 */
503static int hrtimer_hres_enabled __read_mostly = 1;
504
505/*
506 * Enable / Disable high resolution mode
507 */
508static int __init setup_hrtimer_hres(char *str)
509{
510 if (!strcmp(str, "off"))
511 hrtimer_hres_enabled = 0;
512 else if (!strcmp(str, "on"))
513 hrtimer_hres_enabled = 1;
514 else
515 return 0;
516 return 1;
517}
518
519__setup("highres=", setup_hrtimer_hres);
520
521/*
522 * hrtimer_high_res_enabled - query, if the highres mode is enabled
523 */
524static inline int hrtimer_is_hres_enabled(void)
525{
526 return hrtimer_hres_enabled;
527}
528
529/*
530 * Is the high resolution mode active ?
531 */
532static inline int hrtimer_hres_active(void)
533{
534 return __this_cpu_read(hrtimer_bases.hres_active);
535}
536
537/*
538 * Reprogram the event source with checking both queues for the
539 * next event
540 * Called with interrupts disabled and base->lock held
541 */
542static void
543hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
544{
545 int i;
546 struct hrtimer_clock_base *base = cpu_base->clock_base;
547 ktime_t expires, expires_next;
548
549 expires_next.tv64 = KTIME_MAX;
550
551 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
552 struct hrtimer *timer;
553 struct timerqueue_node *next;
554
555 next = timerqueue_getnext(&base->active);
556 if (!next)
557 continue;
558 timer = container_of(next, struct hrtimer, node);
559
560 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
561 /*
562 * clock_was_set() has changed base->offset so the
563 * result might be negative. Fix it up to prevent a
564 * false positive in clockevents_program_event()
565 */
566 if (expires.tv64 < 0)
567 expires.tv64 = 0;
568 if (expires.tv64 < expires_next.tv64)
569 expires_next = expires;
570 }
571
572 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
573 return;
574
575 cpu_base->expires_next.tv64 = expires_next.tv64;
576
577 /*
578 * If a hang was detected in the last timer interrupt then we
579 * leave the hang delay active in the hardware. We want the
580 * system to make progress. That also prevents the following
581 * scenario:
582 * T1 expires 50ms from now
583 * T2 expires 5s from now
584 *
585 * T1 is removed, so this code is called and would reprogram
586 * the hardware to 5s from now. Any hrtimer_start after that
587 * will not reprogram the hardware due to hang_detected being
588 * set. So we'd effectivly block all timers until the T2 event
589 * fires.
590 */
591 if (cpu_base->hang_detected)
592 return;
593
594 if (cpu_base->expires_next.tv64 != KTIME_MAX)
595 tick_program_event(cpu_base->expires_next, 1);
596}
597
598/*
599 * Shared reprogramming for clock_realtime and clock_monotonic
600 *
601 * When a timer is enqueued and expires earlier than the already enqueued
602 * timers, we have to check, whether it expires earlier than the timer for
603 * which the clock event device was armed.
604 *
605 * Called with interrupts disabled and base->cpu_base.lock held
606 */
607static int hrtimer_reprogram(struct hrtimer *timer,
608 struct hrtimer_clock_base *base)
609{
610 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
611 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
612 int res;
613
614 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
615
616 /*
617 * When the callback is running, we do not reprogram the clock event
618 * device. The timer callback is either running on a different CPU or
619 * the callback is executed in the hrtimer_interrupt context. The
620 * reprogramming is handled either by the softirq, which called the
621 * callback or at the end of the hrtimer_interrupt.
622 */
623 if (hrtimer_callback_running(timer))
624 return 0;
625
626 /*
627 * CLOCK_REALTIME timer might be requested with an absolute
628 * expiry time which is less than base->offset. Nothing wrong
629 * about that, just avoid to call into the tick code, which
630 * has now objections against negative expiry values.
631 */
632 if (expires.tv64 < 0)
633 return -ETIME;
634
635 if (expires.tv64 >= cpu_base->expires_next.tv64)
636 return 0;
637
638 /*
639 * If a hang was detected in the last timer interrupt then we
640 * do not schedule a timer which is earlier than the expiry
641 * which we enforced in the hang detection. We want the system
642 * to make progress.
643 */
644 if (cpu_base->hang_detected)
645 return 0;
646
647 /*
648 * Clockevents returns -ETIME, when the event was in the past.
649 */
650 res = tick_program_event(expires, 0);
651 if (!IS_ERR_VALUE(res))
652 cpu_base->expires_next = expires;
653 return res;
654}
655
656/*
657 * Initialize the high resolution related parts of cpu_base
658 */
659static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
660{
661 base->expires_next.tv64 = KTIME_MAX;
662 base->hres_active = 0;
663}
664
665/*
666 * When High resolution timers are active, try to reprogram. Note, that in case
667 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
668 * check happens. The timer gets enqueued into the rbtree. The reprogramming
669 * and expiry check is done in the hrtimer_interrupt or in the softirq.
670 */
671static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
672 struct hrtimer_clock_base *base)
673{
674 return base->cpu_base->hres_active && hrtimer_reprogram(timer, base);
675}
676
677static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
678{
679 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
680 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
681 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
682
683 return ktime_get_update_offsets(offs_real, offs_boot, offs_tai);
684}
685
686/*
687 * Retrigger next event is called after clock was set
688 *
689 * Called with interrupts disabled via on_each_cpu()
690 */
691static void retrigger_next_event(void *arg)
692{
693 struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
694
695 if (!hrtimer_hres_active())
696 return;
697
698 raw_spin_lock(&base->lock);
699 hrtimer_update_base(base);
700 hrtimer_force_reprogram(base, 0);
701 raw_spin_unlock(&base->lock);
702}
703
704/*
705 * Switch to high resolution mode
706 */
707static int hrtimer_switch_to_hres(void)
708{
709 int i, cpu = smp_processor_id();
710 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
711 unsigned long flags;
712
713 if (base->hres_active)
714 return 1;
715
716 local_irq_save(flags);
717
718 if (tick_init_highres()) {
719 local_irq_restore(flags);
720 printk(KERN_WARNING "Could not switch to high resolution "
721 "mode on CPU %d\n", cpu);
722 return 0;
723 }
724 base->hres_active = 1;
725 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
726 base->clock_base[i].resolution = KTIME_HIGH_RES;
727
728 tick_setup_sched_timer();
729 /* "Retrigger" the interrupt to get things going */
730 retrigger_next_event(NULL);
731 local_irq_restore(flags);
732 return 1;
733}
734
735static void clock_was_set_work(struct work_struct *work)
736{
737 clock_was_set();
738}
739
740static DECLARE_WORK(hrtimer_work, clock_was_set_work);
741
742/*
743 * Called from timekeeping and resume code to reprogramm the hrtimer
744 * interrupt device on all cpus.
745 */
746void clock_was_set_delayed(void)
747{
748 schedule_work(&hrtimer_work);
749}
750
751#else
752
753static inline int hrtimer_hres_active(void) { return 0; }
754static inline int hrtimer_is_hres_enabled(void) { return 0; }
755static inline int hrtimer_switch_to_hres(void) { return 0; }
756static inline void
757hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
758static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
759 struct hrtimer_clock_base *base)
760{
761 return 0;
762}
763static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
764static inline void retrigger_next_event(void *arg) { }
765
766#endif /* CONFIG_HIGH_RES_TIMERS */
767
768/*
769 * Clock realtime was set
770 *
771 * Change the offset of the realtime clock vs. the monotonic
772 * clock.
773 *
774 * We might have to reprogram the high resolution timer interrupt. On
775 * SMP we call the architecture specific code to retrigger _all_ high
776 * resolution timer interrupts. On UP we just disable interrupts and
777 * call the high resolution interrupt code.
778 */
779void clock_was_set(void)
780{
781#ifdef CONFIG_HIGH_RES_TIMERS
782 /* Retrigger the CPU local events everywhere */
783 on_each_cpu(retrigger_next_event, NULL, 1);
784#endif
785 timerfd_clock_was_set();
786}
787
788/*
789 * During resume we might have to reprogram the high resolution timer
790 * interrupt on all online CPUs. However, all other CPUs will be
791 * stopped with IRQs interrupts disabled so the clock_was_set() call
792 * must be deferred.
793 */
794void hrtimers_resume(void)
795{
796 WARN_ONCE(!irqs_disabled(),
797 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
798
799 /* Retrigger on the local CPU */
800 retrigger_next_event(NULL);
801 /* And schedule a retrigger for all others */
802 clock_was_set_delayed();
803}
804
805static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
806{
807#ifdef CONFIG_TIMER_STATS
808 if (timer->start_site)
809 return;
810 timer->start_site = __builtin_return_address(0);
811 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
812 timer->start_pid = current->pid;
813#endif
814}
815
816static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
817{
818#ifdef CONFIG_TIMER_STATS
819 timer->start_site = NULL;
820#endif
821}
822
823static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
824{
825#ifdef CONFIG_TIMER_STATS
826 if (likely(!timer_stats_active))
827 return;
828 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
829 timer->function, timer->start_comm, 0);
830#endif
831}
832
833/*
834 * Counterpart to lock_hrtimer_base above:
835 */
836static inline
837void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
838{
839 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
840}
841
842/**
843 * hrtimer_forward - forward the timer expiry
844 * @timer: hrtimer to forward
845 * @now: forward past this time
846 * @interval: the interval to forward
847 *
848 * Forward the timer expiry so it will expire in the future.
849 * Returns the number of overruns.
850 */
851u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
852{
853 u64 orun = 1;
854 ktime_t delta;
855
856 delta = ktime_sub(now, hrtimer_get_expires(timer));
857
858 if (delta.tv64 < 0)
859 return 0;
860
861 if (interval.tv64 < timer->base->resolution.tv64)
862 interval.tv64 = timer->base->resolution.tv64;
863
864 if (unlikely(delta.tv64 >= interval.tv64)) {
865 s64 incr = ktime_to_ns(interval);
866
867 orun = ktime_divns(delta, incr);
868 hrtimer_add_expires_ns(timer, incr * orun);
869 if (hrtimer_get_expires_tv64(timer) > now.tv64)
870 return orun;
871 /*
872 * This (and the ktime_add() below) is the
873 * correction for exact:
874 */
875 orun++;
876 }
877 hrtimer_add_expires(timer, interval);
878
879 return orun;
880}
881EXPORT_SYMBOL_GPL(hrtimer_forward);
882
883/*
884 * enqueue_hrtimer - internal function to (re)start a timer
885 *
886 * The timer is inserted in expiry order. Insertion into the
887 * red black tree is O(log(n)). Must hold the base lock.
888 *
889 * Returns 1 when the new timer is the leftmost timer in the tree.
890 */
891static int enqueue_hrtimer(struct hrtimer *timer,
892 struct hrtimer_clock_base *base)
893{
894 debug_activate(timer);
895
896 timerqueue_add(&base->active, &timer->node);
897 base->cpu_base->active_bases |= 1 << base->index;
898
899 /*
900 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
901 * state of a possibly running callback.
902 */
903 timer->state |= HRTIMER_STATE_ENQUEUED;
904
905 return (&timer->node == base->active.next);
906}
907
908/*
909 * __remove_hrtimer - internal function to remove a timer
910 *
911 * Caller must hold the base lock.
912 *
913 * High resolution timer mode reprograms the clock event device when the
914 * timer is the one which expires next. The caller can disable this by setting
915 * reprogram to zero. This is useful, when the context does a reprogramming
916 * anyway (e.g. timer interrupt)
917 */
918static void __remove_hrtimer(struct hrtimer *timer,
919 struct hrtimer_clock_base *base,
920 unsigned long newstate, int reprogram)
921{
922 struct timerqueue_node *next_timer;
923 if (!(timer->state & HRTIMER_STATE_ENQUEUED))
924 goto out;
925
926 next_timer = timerqueue_getnext(&base->active);
927 timerqueue_del(&base->active, &timer->node);
928 if (&timer->node == next_timer) {
929#ifdef CONFIG_HIGH_RES_TIMERS
930 /* Reprogram the clock event device. if enabled */
931 if (reprogram && hrtimer_hres_active()) {
932 ktime_t expires;
933
934 expires = ktime_sub(hrtimer_get_expires(timer),
935 base->offset);
936 if (base->cpu_base->expires_next.tv64 == expires.tv64)
937 hrtimer_force_reprogram(base->cpu_base, 1);
938 }
939#endif
940 }
941 if (!timerqueue_getnext(&base->active))
942 base->cpu_base->active_bases &= ~(1 << base->index);
943out:
944 timer->state = newstate;
945}
946
947/*
948 * remove hrtimer, called with base lock held
949 */
950static inline int
951remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
952{
953 if (hrtimer_is_queued(timer)) {
954 unsigned long state;
955 int reprogram;
956
957 /*
958 * Remove the timer and force reprogramming when high
959 * resolution mode is active and the timer is on the current
960 * CPU. If we remove a timer on another CPU, reprogramming is
961 * skipped. The interrupt event on this CPU is fired and
962 * reprogramming happens in the interrupt handler. This is a
963 * rare case and less expensive than a smp call.
964 */
965 debug_deactivate(timer);
966 timer_stats_hrtimer_clear_start_info(timer);
967 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
968 /*
969 * We must preserve the CALLBACK state flag here,
970 * otherwise we could move the timer base in
971 * switch_hrtimer_base.
972 */
973 state = timer->state & HRTIMER_STATE_CALLBACK;
974 __remove_hrtimer(timer, base, state, reprogram);
975 return 1;
976 }
977 return 0;
978}
979
980int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
981 unsigned long delta_ns, const enum hrtimer_mode mode,
982 int wakeup)
983{
984 struct hrtimer_clock_base *base, *new_base;
985 unsigned long flags;
986 int ret, leftmost;
987
988 base = lock_hrtimer_base(timer, &flags);
989
990 /* Remove an active timer from the queue: */
991 ret = remove_hrtimer(timer, base);
992
993 if (mode & HRTIMER_MODE_REL) {
994 tim = ktime_add_safe(tim, base->get_time());
995 /*
996 * CONFIG_TIME_LOW_RES is a temporary way for architectures
997 * to signal that they simply return xtime in
998 * do_gettimeoffset(). In this case we want to round up by
999 * resolution when starting a relative timer, to avoid short
1000 * timeouts. This will go away with the GTOD framework.
1001 */
1002#ifdef CONFIG_TIME_LOW_RES
1003 tim = ktime_add_safe(tim, base->resolution);
1004#endif
1005 }
1006
1007 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1008
1009 /* Switch the timer base, if necessary: */
1010 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
1011
1012 timer_stats_hrtimer_set_start_info(timer);
1013
1014 leftmost = enqueue_hrtimer(timer, new_base);
1015
1016 /*
1017 * Only allow reprogramming if the new base is on this CPU.
1018 * (it might still be on another CPU if the timer was pending)
1019 *
1020 * XXX send_remote_softirq() ?
1021 */
1022 if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)
1023 && hrtimer_enqueue_reprogram(timer, new_base)) {
1024 if (wakeup) {
1025 /*
1026 * We need to drop cpu_base->lock to avoid a
1027 * lock ordering issue vs. rq->lock.
1028 */
1029 raw_spin_unlock(&new_base->cpu_base->lock);
1030 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1031 local_irq_restore(flags);
1032 return ret;
1033 } else {
1034 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1035 }
1036 }
1037
1038 unlock_hrtimer_base(timer, &flags);
1039
1040 return ret;
1041}
1042EXPORT_SYMBOL_GPL(__hrtimer_start_range_ns);
1043
1044/**
1045 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1046 * @timer: the timer to be added
1047 * @tim: expiry time
1048 * @delta_ns: "slack" range for the timer
1049 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1050 * relative (HRTIMER_MODE_REL)
1051 *
1052 * Returns:
1053 * 0 on success
1054 * 1 when the timer was active
1055 */
1056int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1057 unsigned long delta_ns, const enum hrtimer_mode mode)
1058{
1059 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1060}
1061EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1062
1063/**
1064 * hrtimer_start - (re)start an hrtimer on the current CPU
1065 * @timer: the timer to be added
1066 * @tim: expiry time
1067 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1068 * relative (HRTIMER_MODE_REL)
1069 *
1070 * Returns:
1071 * 0 on success
1072 * 1 when the timer was active
1073 */
1074int
1075hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1076{
1077 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1078}
1079EXPORT_SYMBOL_GPL(hrtimer_start);
1080
1081
1082/**
1083 * hrtimer_try_to_cancel - try to deactivate a timer
1084 * @timer: hrtimer to stop
1085 *
1086 * Returns:
1087 * 0 when the timer was not active
1088 * 1 when the timer was active
1089 * -1 when the timer is currently excuting the callback function and
1090 * cannot be stopped
1091 */
1092int hrtimer_try_to_cancel(struct hrtimer *timer)
1093{
1094 struct hrtimer_clock_base *base;
1095 unsigned long flags;
1096 int ret = -1;
1097
1098 base = lock_hrtimer_base(timer, &flags);
1099
1100 if (!hrtimer_callback_running(timer))
1101 ret = remove_hrtimer(timer, base);
1102
1103 unlock_hrtimer_base(timer, &flags);
1104
1105 return ret;
1106
1107}
1108EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1109
1110/**
1111 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1112 * @timer: the timer to be cancelled
1113 *
1114 * Returns:
1115 * 0 when the timer was not active
1116 * 1 when the timer was active
1117 */
1118int hrtimer_cancel(struct hrtimer *timer)
1119{
1120 for (;;) {
1121 int ret = hrtimer_try_to_cancel(timer);
1122
1123 if (ret >= 0)
1124 return ret;
1125 cpu_relax();
1126 }
1127}
1128EXPORT_SYMBOL_GPL(hrtimer_cancel);
1129
1130/**
1131 * hrtimer_get_remaining - get remaining time for the timer
1132 * @timer: the timer to read
1133 */
1134ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1135{
1136 unsigned long flags;
1137 ktime_t rem;
1138
1139 lock_hrtimer_base(timer, &flags);
1140 rem = hrtimer_expires_remaining(timer);
1141 unlock_hrtimer_base(timer, &flags);
1142
1143 return rem;
1144}
1145EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1146
1147#ifdef CONFIG_NO_HZ_COMMON
1148/**
1149 * hrtimer_get_next_event - get the time until next expiry event
1150 *
1151 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1152 * is pending.
1153 */
1154ktime_t hrtimer_get_next_event(void)
1155{
1156 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1157 struct hrtimer_clock_base *base = cpu_base->clock_base;
1158 ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1159 unsigned long flags;
1160 int i;
1161
1162 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1163
1164 if (!hrtimer_hres_active()) {
1165 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1166 struct hrtimer *timer;
1167 struct timerqueue_node *next;
1168
1169 next = timerqueue_getnext(&base->active);
1170 if (!next)
1171 continue;
1172
1173 timer = container_of(next, struct hrtimer, node);
1174 delta.tv64 = hrtimer_get_expires_tv64(timer);
1175 delta = ktime_sub(delta, base->get_time());
1176 if (delta.tv64 < mindelta.tv64)
1177 mindelta.tv64 = delta.tv64;
1178 }
1179 }
1180
1181 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1182
1183 if (mindelta.tv64 < 0)
1184 mindelta.tv64 = 0;
1185 return mindelta;
1186}
1187#endif
1188
1189static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1190 enum hrtimer_mode mode)
1191{
1192 struct hrtimer_cpu_base *cpu_base;
1193 int base;
1194
1195 memset(timer, 0, sizeof(struct hrtimer));
1196
1197 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1198
1199 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1200 clock_id = CLOCK_MONOTONIC;
1201
1202 base = hrtimer_clockid_to_base(clock_id);
1203 timer->base = &cpu_base->clock_base[base];
1204 timerqueue_init(&timer->node);
1205
1206#ifdef CONFIG_TIMER_STATS
1207 timer->start_site = NULL;
1208 timer->start_pid = -1;
1209 memset(timer->start_comm, 0, TASK_COMM_LEN);
1210#endif
1211}
1212
1213/**
1214 * hrtimer_init - initialize a timer to the given clock
1215 * @timer: the timer to be initialized
1216 * @clock_id: the clock to be used
1217 * @mode: timer mode abs/rel
1218 */
1219void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1220 enum hrtimer_mode mode)
1221{
1222 debug_init(timer, clock_id, mode);
1223 __hrtimer_init(timer, clock_id, mode);
1224}
1225EXPORT_SYMBOL_GPL(hrtimer_init);
1226
1227/**
1228 * hrtimer_get_res - get the timer resolution for a clock
1229 * @which_clock: which clock to query
1230 * @tp: pointer to timespec variable to store the resolution
1231 *
1232 * Store the resolution of the clock selected by @which_clock in the
1233 * variable pointed to by @tp.
1234 */
1235int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1236{
1237 struct hrtimer_cpu_base *cpu_base;
1238 int base = hrtimer_clockid_to_base(which_clock);
1239
1240 cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1241 *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1242
1243 return 0;
1244}
1245EXPORT_SYMBOL_GPL(hrtimer_get_res);
1246
1247static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1248{
1249 struct hrtimer_clock_base *base = timer->base;
1250 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1251 enum hrtimer_restart (*fn)(struct hrtimer *);
1252 int restart;
1253
1254 WARN_ON(!irqs_disabled());
1255
1256 debug_deactivate(timer);
1257 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1258 timer_stats_account_hrtimer(timer);
1259 fn = timer->function;
1260
1261 /*
1262 * Because we run timers from hardirq context, there is no chance
1263 * they get migrated to another cpu, therefore its safe to unlock
1264 * the timer base.
1265 */
1266 raw_spin_unlock(&cpu_base->lock);
1267 trace_hrtimer_expire_entry(timer, now);
1268 restart = fn(timer);
1269 trace_hrtimer_expire_exit(timer);
1270 raw_spin_lock(&cpu_base->lock);
1271
1272 /*
1273 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1274 * we do not reprogramm the event hardware. Happens either in
1275 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1276 */
1277 if (restart != HRTIMER_NORESTART) {
1278 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1279 enqueue_hrtimer(timer, base);
1280 }
1281
1282 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1283
1284 timer->state &= ~HRTIMER_STATE_CALLBACK;
1285}
1286
1287#ifdef CONFIG_HIGH_RES_TIMERS
1288
1289/*
1290 * High resolution timer interrupt
1291 * Called with interrupts disabled
1292 */
1293void hrtimer_interrupt(struct clock_event_device *dev)
1294{
1295 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1296 ktime_t expires_next, now, entry_time, delta;
1297 int i, retries = 0;
1298
1299 BUG_ON(!cpu_base->hres_active);
1300 cpu_base->nr_events++;
1301 dev->next_event.tv64 = KTIME_MAX;
1302
1303 raw_spin_lock(&cpu_base->lock);
1304 entry_time = now = hrtimer_update_base(cpu_base);
1305retry:
1306 expires_next.tv64 = KTIME_MAX;
1307 /*
1308 * We set expires_next to KTIME_MAX here with cpu_base->lock
1309 * held to prevent that a timer is enqueued in our queue via
1310 * the migration code. This does not affect enqueueing of
1311 * timers which run their callback and need to be requeued on
1312 * this CPU.
1313 */
1314 cpu_base->expires_next.tv64 = KTIME_MAX;
1315
1316 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1317 struct hrtimer_clock_base *base;
1318 struct timerqueue_node *node;
1319 ktime_t basenow;
1320
1321 if (!(cpu_base->active_bases & (1 << i)))
1322 continue;
1323
1324 base = cpu_base->clock_base + i;
1325 basenow = ktime_add(now, base->offset);
1326
1327 while ((node = timerqueue_getnext(&base->active))) {
1328 struct hrtimer *timer;
1329
1330 timer = container_of(node, struct hrtimer, node);
1331
1332 /*
1333 * The immediate goal for using the softexpires is
1334 * minimizing wakeups, not running timers at the
1335 * earliest interrupt after their soft expiration.
1336 * This allows us to avoid using a Priority Search
1337 * Tree, which can answer a stabbing querry for
1338 * overlapping intervals and instead use the simple
1339 * BST we already have.
1340 * We don't add extra wakeups by delaying timers that
1341 * are right-of a not yet expired timer, because that
1342 * timer will have to trigger a wakeup anyway.
1343 */
1344
1345 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1346 ktime_t expires;
1347
1348 expires = ktime_sub(hrtimer_get_expires(timer),
1349 base->offset);
1350 if (expires.tv64 < 0)
1351 expires.tv64 = KTIME_MAX;
1352 if (expires.tv64 < expires_next.tv64)
1353 expires_next = expires;
1354 break;
1355 }
1356
1357 __run_hrtimer(timer, &basenow);
1358 }
1359 }
1360
1361 /*
1362 * Store the new expiry value so the migration code can verify
1363 * against it.
1364 */
1365 cpu_base->expires_next = expires_next;
1366 raw_spin_unlock(&cpu_base->lock);
1367
1368 /* Reprogramming necessary ? */
1369 if (expires_next.tv64 == KTIME_MAX ||
1370 !tick_program_event(expires_next, 0)) {
1371 cpu_base->hang_detected = 0;
1372 return;
1373 }
1374
1375 /*
1376 * The next timer was already expired due to:
1377 * - tracing
1378 * - long lasting callbacks
1379 * - being scheduled away when running in a VM
1380 *
1381 * We need to prevent that we loop forever in the hrtimer
1382 * interrupt routine. We give it 3 attempts to avoid
1383 * overreacting on some spurious event.
1384 *
1385 * Acquire base lock for updating the offsets and retrieving
1386 * the current time.
1387 */
1388 raw_spin_lock(&cpu_base->lock);
1389 now = hrtimer_update_base(cpu_base);
1390 cpu_base->nr_retries++;
1391 if (++retries < 3)
1392 goto retry;
1393 /*
1394 * Give the system a chance to do something else than looping
1395 * here. We stored the entry time, so we know exactly how long
1396 * we spent here. We schedule the next event this amount of
1397 * time away.
1398 */
1399 cpu_base->nr_hangs++;
1400 cpu_base->hang_detected = 1;
1401 raw_spin_unlock(&cpu_base->lock);
1402 delta = ktime_sub(now, entry_time);
1403 if (delta.tv64 > cpu_base->max_hang_time.tv64)
1404 cpu_base->max_hang_time = delta;
1405 /*
1406 * Limit it to a sensible value as we enforce a longer
1407 * delay. Give the CPU at least 100ms to catch up.
1408 */
1409 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1410 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1411 else
1412 expires_next = ktime_add(now, delta);
1413 tick_program_event(expires_next, 1);
1414 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1415 ktime_to_ns(delta));
1416}
1417
1418/*
1419 * local version of hrtimer_peek_ahead_timers() called with interrupts
1420 * disabled.
1421 */
1422static void __hrtimer_peek_ahead_timers(void)
1423{
1424 struct tick_device *td;
1425
1426 if (!hrtimer_hres_active())
1427 return;
1428
1429 td = &__get_cpu_var(tick_cpu_device);
1430 if (td && td->evtdev)
1431 hrtimer_interrupt(td->evtdev);
1432}
1433
1434/**
1435 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1436 *
1437 * hrtimer_peek_ahead_timers will peek at the timer queue of
1438 * the current cpu and check if there are any timers for which
1439 * the soft expires time has passed. If any such timers exist,
1440 * they are run immediately and then removed from the timer queue.
1441 *
1442 */
1443void hrtimer_peek_ahead_timers(void)
1444{
1445 unsigned long flags;
1446
1447 local_irq_save(flags);
1448 __hrtimer_peek_ahead_timers();
1449 local_irq_restore(flags);
1450}
1451
1452static void run_hrtimer_softirq(struct softirq_action *h)
1453{
1454 hrtimer_peek_ahead_timers();
1455}
1456
1457#else /* CONFIG_HIGH_RES_TIMERS */
1458
1459static inline void __hrtimer_peek_ahead_timers(void) { }
1460
1461#endif /* !CONFIG_HIGH_RES_TIMERS */
1462
1463/*
1464 * Called from timer softirq every jiffy, expire hrtimers:
1465 *
1466 * For HRT its the fall back code to run the softirq in the timer
1467 * softirq context in case the hrtimer initialization failed or has
1468 * not been done yet.
1469 */
1470void hrtimer_run_pending(void)
1471{
1472 if (hrtimer_hres_active())
1473 return;
1474
1475 /*
1476 * This _is_ ugly: We have to check in the softirq context,
1477 * whether we can switch to highres and / or nohz mode. The
1478 * clocksource switch happens in the timer interrupt with
1479 * xtime_lock held. Notification from there only sets the
1480 * check bit in the tick_oneshot code, otherwise we might
1481 * deadlock vs. xtime_lock.
1482 */
1483 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1484 hrtimer_switch_to_hres();
1485}
1486
1487/*
1488 * Called from hardirq context every jiffy
1489 */
1490void hrtimer_run_queues(void)
1491{
1492 struct timerqueue_node *node;
1493 struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1494 struct hrtimer_clock_base *base;
1495 int index, gettime = 1;
1496
1497 if (hrtimer_hres_active())
1498 return;
1499
1500 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1501 base = &cpu_base->clock_base[index];
1502 if (!timerqueue_getnext(&base->active))
1503 continue;
1504
1505 if (gettime) {
1506 hrtimer_get_softirq_time(cpu_base);
1507 gettime = 0;
1508 }
1509
1510 raw_spin_lock(&cpu_base->lock);
1511
1512 while ((node = timerqueue_getnext(&base->active))) {
1513 struct hrtimer *timer;
1514
1515 timer = container_of(node, struct hrtimer, node);
1516 if (base->softirq_time.tv64 <=
1517 hrtimer_get_expires_tv64(timer))
1518 break;
1519
1520 __run_hrtimer(timer, &base->softirq_time);
1521 }
1522 raw_spin_unlock(&cpu_base->lock);
1523 }
1524}
1525
1526/*
1527 * Sleep related functions:
1528 */
1529static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1530{
1531 struct hrtimer_sleeper *t =
1532 container_of(timer, struct hrtimer_sleeper, timer);
1533 struct task_struct *task = t->task;
1534
1535 t->task = NULL;
1536 if (task)
1537 wake_up_process(task);
1538
1539 return HRTIMER_NORESTART;
1540}
1541
1542void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1543{
1544 sl->timer.function = hrtimer_wakeup;
1545 sl->task = task;
1546}
1547EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1548
1549static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1550{
1551 hrtimer_init_sleeper(t, current);
1552
1553 do {
1554 set_current_state(TASK_INTERRUPTIBLE);
1555 hrtimer_start_expires(&t->timer, mode);
1556 if (!hrtimer_active(&t->timer))
1557 t->task = NULL;
1558
1559 if (likely(t->task))
1560 freezable_schedule();
1561
1562 hrtimer_cancel(&t->timer);
1563 mode = HRTIMER_MODE_ABS;
1564
1565 } while (t->task && !signal_pending(current));
1566
1567 __set_current_state(TASK_RUNNING);
1568
1569 return t->task == NULL;
1570}
1571
1572static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1573{
1574 struct timespec rmt;
1575 ktime_t rem;
1576
1577 rem = hrtimer_expires_remaining(timer);
1578 if (rem.tv64 <= 0)
1579 return 0;
1580 rmt = ktime_to_timespec(rem);
1581
1582 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1583 return -EFAULT;
1584
1585 return 1;
1586}
1587
1588long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1589{
1590 struct hrtimer_sleeper t;
1591 struct timespec __user *rmtp;
1592 int ret = 0;
1593
1594 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1595 HRTIMER_MODE_ABS);
1596 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1597
1598 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1599 goto out;
1600
1601 rmtp = restart->nanosleep.rmtp;
1602 if (rmtp) {
1603 ret = update_rmtp(&t.timer, rmtp);
1604 if (ret <= 0)
1605 goto out;
1606 }
1607
1608 /* The other values in restart are already filled in */
1609 ret = -ERESTART_RESTARTBLOCK;
1610out:
1611 destroy_hrtimer_on_stack(&t.timer);
1612 return ret;
1613}
1614
1615long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1616 const enum hrtimer_mode mode, const clockid_t clockid)
1617{
1618 struct restart_block *restart;
1619 struct hrtimer_sleeper t;
1620 int ret = 0;
1621 unsigned long slack;
1622
1623 slack = current->timer_slack_ns;
1624 if (dl_task(current) || rt_task(current))
1625 slack = 0;
1626
1627 hrtimer_init_on_stack(&t.timer, clockid, mode);
1628 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1629 if (do_nanosleep(&t, mode))
1630 goto out;
1631
1632 /* Absolute timers do not update the rmtp value and restart: */
1633 if (mode == HRTIMER_MODE_ABS) {
1634 ret = -ERESTARTNOHAND;
1635 goto out;
1636 }
1637
1638 if (rmtp) {
1639 ret = update_rmtp(&t.timer, rmtp);
1640 if (ret <= 0)
1641 goto out;
1642 }
1643
1644 restart = &current_thread_info()->restart_block;
1645 restart->fn = hrtimer_nanosleep_restart;
1646 restart->nanosleep.clockid = t.timer.base->clockid;
1647 restart->nanosleep.rmtp = rmtp;
1648 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1649
1650 ret = -ERESTART_RESTARTBLOCK;
1651out:
1652 destroy_hrtimer_on_stack(&t.timer);
1653 return ret;
1654}
1655
1656SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1657 struct timespec __user *, rmtp)
1658{
1659 struct timespec tu;
1660
1661 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1662 return -EFAULT;
1663
1664 if (!timespec_valid(&tu))
1665 return -EINVAL;
1666
1667 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1668}
1669
1670/*
1671 * Functions related to boot-time initialization:
1672 */
1673static void init_hrtimers_cpu(int cpu)
1674{
1675 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1676 int i;
1677
1678 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1679 cpu_base->clock_base[i].cpu_base = cpu_base;
1680 timerqueue_init_head(&cpu_base->clock_base[i].active);
1681 }
1682
1683 hrtimer_init_hres(cpu_base);
1684}
1685
1686#ifdef CONFIG_HOTPLUG_CPU
1687
1688static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1689 struct hrtimer_clock_base *new_base)
1690{
1691 struct hrtimer *timer;
1692 struct timerqueue_node *node;
1693
1694 while ((node = timerqueue_getnext(&old_base->active))) {
1695 timer = container_of(node, struct hrtimer, node);
1696 BUG_ON(hrtimer_callback_running(timer));
1697 debug_deactivate(timer);
1698
1699 /*
1700 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1701 * timer could be seen as !active and just vanish away
1702 * under us on another CPU
1703 */
1704 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1705 timer->base = new_base;
1706 /*
1707 * Enqueue the timers on the new cpu. This does not
1708 * reprogram the event device in case the timer
1709 * expires before the earliest on this CPU, but we run
1710 * hrtimer_interrupt after we migrated everything to
1711 * sort out already expired timers and reprogram the
1712 * event device.
1713 */
1714 enqueue_hrtimer(timer, new_base);
1715
1716 /* Clear the migration state bit */
1717 timer->state &= ~HRTIMER_STATE_MIGRATE;
1718 }
1719}
1720
1721static void migrate_hrtimers(int scpu)
1722{
1723 struct hrtimer_cpu_base *old_base, *new_base;
1724 int i;
1725
1726 BUG_ON(cpu_online(scpu));
1727 tick_cancel_sched_timer(scpu);
1728
1729 local_irq_disable();
1730 old_base = &per_cpu(hrtimer_bases, scpu);
1731 new_base = &__get_cpu_var(hrtimer_bases);
1732 /*
1733 * The caller is globally serialized and nobody else
1734 * takes two locks at once, deadlock is not possible.
1735 */
1736 raw_spin_lock(&new_base->lock);
1737 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1738
1739 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1740 migrate_hrtimer_list(&old_base->clock_base[i],
1741 &new_base->clock_base[i]);
1742 }
1743
1744 raw_spin_unlock(&old_base->lock);
1745 raw_spin_unlock(&new_base->lock);
1746
1747 /* Check, if we got expired work to do */
1748 __hrtimer_peek_ahead_timers();
1749 local_irq_enable();
1750}
1751
1752#endif /* CONFIG_HOTPLUG_CPU */
1753
1754static int hrtimer_cpu_notify(struct notifier_block *self,
1755 unsigned long action, void *hcpu)
1756{
1757 int scpu = (long)hcpu;
1758
1759 switch (action) {
1760
1761 case CPU_UP_PREPARE:
1762 case CPU_UP_PREPARE_FROZEN:
1763 init_hrtimers_cpu(scpu);
1764 break;
1765
1766#ifdef CONFIG_HOTPLUG_CPU
1767 case CPU_DYING:
1768 case CPU_DYING_FROZEN:
1769 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1770 break;
1771 case CPU_DEAD:
1772 case CPU_DEAD_FROZEN:
1773 {
1774 clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1775 migrate_hrtimers(scpu);
1776 break;
1777 }
1778#endif
1779
1780 default:
1781 break;
1782 }
1783
1784 return NOTIFY_OK;
1785}
1786
1787static struct notifier_block hrtimers_nb = {
1788 .notifier_call = hrtimer_cpu_notify,
1789};
1790
1791void __init hrtimers_init(void)
1792{
1793 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1794 (void *)(long)smp_processor_id());
1795 register_cpu_notifier(&hrtimers_nb);
1796#ifdef CONFIG_HIGH_RES_TIMERS
1797 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1798#endif
1799}
1800
1801/**
1802 * schedule_hrtimeout_range_clock - sleep until timeout
1803 * @expires: timeout value (ktime_t)
1804 * @delta: slack in expires timeout (ktime_t)
1805 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1806 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1807 */
1808int __sched
1809schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1810 const enum hrtimer_mode mode, int clock)
1811{
1812 struct hrtimer_sleeper t;
1813
1814 /*
1815 * Optimize when a zero timeout value is given. It does not
1816 * matter whether this is an absolute or a relative time.
1817 */
1818 if (expires && !expires->tv64) {
1819 __set_current_state(TASK_RUNNING);
1820 return 0;
1821 }
1822
1823 /*
1824 * A NULL parameter means "infinite"
1825 */
1826 if (!expires) {
1827 schedule();
1828 __set_current_state(TASK_RUNNING);
1829 return -EINTR;
1830 }
1831
1832 hrtimer_init_on_stack(&t.timer, clock, mode);
1833 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1834
1835 hrtimer_init_sleeper(&t, current);
1836
1837 hrtimer_start_expires(&t.timer, mode);
1838 if (!hrtimer_active(&t.timer))
1839 t.task = NULL;
1840
1841 if (likely(t.task))
1842 schedule();
1843
1844 hrtimer_cancel(&t.timer);
1845 destroy_hrtimer_on_stack(&t.timer);
1846
1847 __set_current_state(TASK_RUNNING);
1848
1849 return !t.task ? 0 : -EINTR;
1850}
1851
1852/**
1853 * schedule_hrtimeout_range - sleep until timeout
1854 * @expires: timeout value (ktime_t)
1855 * @delta: slack in expires timeout (ktime_t)
1856 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1857 *
1858 * Make the current task sleep until the given expiry time has
1859 * elapsed. The routine will return immediately unless
1860 * the current task state has been set (see set_current_state()).
1861 *
1862 * The @delta argument gives the kernel the freedom to schedule the
1863 * actual wakeup to a time that is both power and performance friendly.
1864 * The kernel give the normal best effort behavior for "@expires+@delta",
1865 * but may decide to fire the timer earlier, but no earlier than @expires.
1866 *
1867 * You can set the task state as follows -
1868 *
1869 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1870 * pass before the routine returns.
1871 *
1872 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1873 * delivered to the current task.
1874 *
1875 * The current task state is guaranteed to be TASK_RUNNING when this
1876 * routine returns.
1877 *
1878 * Returns 0 when the timer has expired otherwise -EINTR
1879 */
1880int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1881 const enum hrtimer_mode mode)
1882{
1883 return schedule_hrtimeout_range_clock(expires, delta, mode,
1884 CLOCK_MONOTONIC);
1885}
1886EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1887
1888/**
1889 * schedule_hrtimeout - sleep until timeout
1890 * @expires: timeout value (ktime_t)
1891 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1892 *
1893 * Make the current task sleep until the given expiry time has
1894 * elapsed. The routine will return immediately unless
1895 * the current task state has been set (see set_current_state()).
1896 *
1897 * You can set the task state as follows -
1898 *
1899 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1900 * pass before the routine returns.
1901 *
1902 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1903 * delivered to the current task.
1904 *
1905 * The current task state is guaranteed to be TASK_RUNNING when this
1906 * routine returns.
1907 *
1908 * Returns 0 when the timer has expired otherwise -EINTR
1909 */
1910int __sched schedule_hrtimeout(ktime_t *expires,
1911 const enum hrtimer_mode mode)
1912{
1913 return schedule_hrtimeout_range(expires, 0, mode);
1914}
1915EXPORT_SYMBOL_GPL(schedule_hrtimeout);