projects/linux·blob v4.1 · refs

1852 lines · 46.9 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#include "tick-internal.h"
58
59/*
60 * The timer bases:
61 *
62 * There are more clockids then hrtimer bases. Thus, we index
63 * into the timer bases by the hrtimer_base_type enum. When trying
64 * to reach a base using a clockid, hrtimer_clockid_to_base()
65 * is used to convert from clockid to the proper hrtimer_base_type.
66 */
67DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
68{
69
70 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
71 .clock_base =
72 {
73 {
74 .index = HRTIMER_BASE_MONOTONIC,
75 .clockid = CLOCK_MONOTONIC,
76 .get_time = &ktime_get,
77 .resolution = KTIME_LOW_RES,
78 },
79 {
80 .index = HRTIMER_BASE_REALTIME,
81 .clockid = CLOCK_REALTIME,
82 .get_time = &ktime_get_real,
83 .resolution = KTIME_LOW_RES,
84 },
85 {
86 .index = HRTIMER_BASE_BOOTTIME,
87 .clockid = CLOCK_BOOTTIME,
88 .get_time = &ktime_get_boottime,
89 .resolution = KTIME_LOW_RES,
90 },
91 {
92 .index = HRTIMER_BASE_TAI,
93 .clockid = CLOCK_TAI,
94 .get_time = &ktime_get_clocktai,
95 .resolution = KTIME_LOW_RES,
96 },
97 }
98};
99
100static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
101 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
102 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
103 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
104 [CLOCK_TAI] = HRTIMER_BASE_TAI,
105};
106
107static inline int hrtimer_clockid_to_base(clockid_t clock_id)
108{
109 return hrtimer_clock_to_base_table[clock_id];
110}
111
112
113/*
114 * Get the coarse grained time at the softirq based on xtime and
115 * wall_to_monotonic.
116 */
117static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
118{
119 ktime_t xtim, mono, boot, tai;
120 ktime_t off_real, off_boot, off_tai;
121
122 mono = ktime_get_update_offsets_tick(&off_real, &off_boot, &off_tai);
123 boot = ktime_add(mono, off_boot);
124 xtim = ktime_add(mono, off_real);
125 tai = ktime_add(mono, off_tai);
126
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 = tai;
131}
132
133/*
134 * Functions and macros which are different for UP/SMP systems are kept in a
135 * single place
136 */
137#ifdef CONFIG_SMP
138
139/*
140 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
141 * means that all timers which are tied to this base via timer->base are
142 * locked, and the base itself is locked too.
143 *
144 * So __run_timers/migrate_timers can safely modify all timers which could
145 * be found on the lists/queues.
146 *
147 * When the timer's base is locked, and the timer removed from list, it is
148 * possible to set timer->base = NULL and drop the lock: the timer remains
149 * locked.
150 */
151static
152struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
153 unsigned long *flags)
154{
155 struct hrtimer_clock_base *base;
156
157 for (;;) {
158 base = timer->base;
159 if (likely(base != NULL)) {
160 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
161 if (likely(base == timer->base))
162 return base;
163 /* The timer has migrated to another CPU: */
164 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
165 }
166 cpu_relax();
167 }
168}
169
170/*
171 * With HIGHRES=y we do not migrate the timer when it is expiring
172 * before the next event on the target cpu because we cannot reprogram
173 * the target cpu hardware and we would cause it to fire late.
174 *
175 * Called with cpu_base->lock of target cpu held.
176 */
177static int
178hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
179{
180#ifdef CONFIG_HIGH_RES_TIMERS
181 ktime_t expires;
182
183 if (!new_base->cpu_base->hres_active)
184 return 0;
185
186 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
187 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
188#else
189 return 0;
190#endif
191}
192
193/*
194 * Switch the timer base to the current CPU when possible.
195 */
196static inline struct hrtimer_clock_base *
197switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
198 int pinned)
199{
200 struct hrtimer_clock_base *new_base;
201 struct hrtimer_cpu_base *new_cpu_base;
202 int this_cpu = smp_processor_id();
203 int cpu = get_nohz_timer_target(pinned);
204 int basenum = base->index;
205
206again:
207 new_cpu_base = &per_cpu(hrtimer_bases, cpu);
208 new_base = &new_cpu_base->clock_base[basenum];
209
210 if (base != new_base) {
211 /*
212 * We are trying to move timer to new_base.
213 * However we can't change timer's base while it is running,
214 * so we keep it on the same CPU. No hassle vs. reprogramming
215 * the event source in the high resolution case. The softirq
216 * code will take care of this when the timer function has
217 * completed. There is no conflict as we hold the lock until
218 * the timer is enqueued.
219 */
220 if (unlikely(hrtimer_callback_running(timer)))
221 return base;
222
223 /* See the comment in lock_timer_base() */
224 timer->base = NULL;
225 raw_spin_unlock(&base->cpu_base->lock);
226 raw_spin_lock(&new_base->cpu_base->lock);
227
228 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
229 cpu = this_cpu;
230 raw_spin_unlock(&new_base->cpu_base->lock);
231 raw_spin_lock(&base->cpu_base->lock);
232 timer->base = base;
233 goto again;
234 }
235 timer->base = new_base;
236 } else {
237 if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
238 cpu = this_cpu;
239 goto again;
240 }
241 }
242 return new_base;
243}
244
245#else /* CONFIG_SMP */
246
247static inline struct hrtimer_clock_base *
248lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
249{
250 struct hrtimer_clock_base *base = timer->base;
251
252 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
253
254 return base;
255}
256
257# define switch_hrtimer_base(t, b, p) (b)
258
259#endif /* !CONFIG_SMP */
260
261/*
262 * Functions for the union type storage format of ktime_t which are
263 * too large for inlining:
264 */
265#if BITS_PER_LONG < 64
266/*
267 * Divide a ktime value by a nanosecond value
268 */
269s64 __ktime_divns(const ktime_t kt, s64 div)
270{
271 int sft = 0;
272 s64 dclc;
273 u64 tmp;
274
275 dclc = ktime_to_ns(kt);
276 tmp = dclc < 0 ? -dclc : dclc;
277
278 /* Make sure the divisor is less than 2^32: */
279 while (div >> 32) {
280 sft++;
281 div >>= 1;
282 }
283 tmp >>= sft;
284 do_div(tmp, (unsigned long) div);
285 return dclc < 0 ? -tmp : tmp;
286}
287EXPORT_SYMBOL_GPL(__ktime_divns);
288#endif /* BITS_PER_LONG >= 64 */
289
290/*
291 * Add two ktime values and do a safety check for overflow:
292 */
293ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
294{
295 ktime_t res = ktime_add(lhs, rhs);
296
297 /*
298 * We use KTIME_SEC_MAX here, the maximum timeout which we can
299 * return to user space in a timespec:
300 */
301 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
302 res = ktime_set(KTIME_SEC_MAX, 0);
303
304 return res;
305}
306
307EXPORT_SYMBOL_GPL(ktime_add_safe);
308
309#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
310
311static struct debug_obj_descr hrtimer_debug_descr;
312
313static void *hrtimer_debug_hint(void *addr)
314{
315 return ((struct hrtimer *) addr)->function;
316}
317
318/*
319 * fixup_init is called when:
320 * - an active object is initialized
321 */
322static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
323{
324 struct hrtimer *timer = addr;
325
326 switch (state) {
327 case ODEBUG_STATE_ACTIVE:
328 hrtimer_cancel(timer);
329 debug_object_init(timer, &hrtimer_debug_descr);
330 return 1;
331 default:
332 return 0;
333 }
334}
335
336/*
337 * fixup_activate is called when:
338 * - an active object is activated
339 * - an unknown object is activated (might be a statically initialized object)
340 */
341static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
342{
343 switch (state) {
344
345 case ODEBUG_STATE_NOTAVAILABLE:
346 WARN_ON_ONCE(1);
347 return 0;
348
349 case ODEBUG_STATE_ACTIVE:
350 WARN_ON(1);
351
352 default:
353 return 0;
354 }
355}
356
357/*
358 * fixup_free is called when:
359 * - an active object is freed
360 */
361static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
362{
363 struct hrtimer *timer = addr;
364
365 switch (state) {
366 case ODEBUG_STATE_ACTIVE:
367 hrtimer_cancel(timer);
368 debug_object_free(timer, &hrtimer_debug_descr);
369 return 1;
370 default:
371 return 0;
372 }
373}
374
375static struct debug_obj_descr hrtimer_debug_descr = {
376 .name = "hrtimer",
377 .debug_hint = hrtimer_debug_hint,
378 .fixup_init = hrtimer_fixup_init,
379 .fixup_activate = hrtimer_fixup_activate,
380 .fixup_free = hrtimer_fixup_free,
381};
382
383static inline void debug_hrtimer_init(struct hrtimer *timer)
384{
385 debug_object_init(timer, &hrtimer_debug_descr);
386}
387
388static inline void debug_hrtimer_activate(struct hrtimer *timer)
389{
390 debug_object_activate(timer, &hrtimer_debug_descr);
391}
392
393static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
394{
395 debug_object_deactivate(timer, &hrtimer_debug_descr);
396}
397
398static inline void debug_hrtimer_free(struct hrtimer *timer)
399{
400 debug_object_free(timer, &hrtimer_debug_descr);
401}
402
403static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
404 enum hrtimer_mode mode);
405
406void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
407 enum hrtimer_mode mode)
408{
409 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
410 __hrtimer_init(timer, clock_id, mode);
411}
412EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
413
414void destroy_hrtimer_on_stack(struct hrtimer *timer)
415{
416 debug_object_free(timer, &hrtimer_debug_descr);
417}
418
419#else
420static inline void debug_hrtimer_init(struct hrtimer *timer) { }
421static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
422static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
423#endif
424
425static inline void
426debug_init(struct hrtimer *timer, clockid_t clockid,
427 enum hrtimer_mode mode)
428{
429 debug_hrtimer_init(timer);
430 trace_hrtimer_init(timer, clockid, mode);
431}
432
433static inline void debug_activate(struct hrtimer *timer)
434{
435 debug_hrtimer_activate(timer);
436 trace_hrtimer_start(timer);
437}
438
439static inline void debug_deactivate(struct hrtimer *timer)
440{
441 debug_hrtimer_deactivate(timer);
442 trace_hrtimer_cancel(timer);
443}
444
445#if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS)
446static ktime_t __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base)
447{
448 struct hrtimer_clock_base *base = cpu_base->clock_base;
449 ktime_t expires, expires_next = { .tv64 = KTIME_MAX };
450 int i;
451
452 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
453 struct timerqueue_node *next;
454 struct hrtimer *timer;
455
456 next = timerqueue_getnext(&base->active);
457 if (!next)
458 continue;
459
460 timer = container_of(next, struct hrtimer, node);
461 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
462 if (expires.tv64 < expires_next.tv64)
463 expires_next = expires;
464 }
465 /*
466 * clock_was_set() might have changed base->offset of any of
467 * the clock bases so the result might be negative. Fix it up
468 * to prevent a false positive in clockevents_program_event().
469 */
470 if (expires_next.tv64 < 0)
471 expires_next.tv64 = 0;
472 return expires_next;
473}
474#endif
475
476/* High resolution timer related functions */
477#ifdef CONFIG_HIGH_RES_TIMERS
478
479/*
480 * High resolution timer enabled ?
481 */
482static int hrtimer_hres_enabled __read_mostly = 1;
483
484/*
485 * Enable / Disable high resolution mode
486 */
487static int __init setup_hrtimer_hres(char *str)
488{
489 if (!strcmp(str, "off"))
490 hrtimer_hres_enabled = 0;
491 else if (!strcmp(str, "on"))
492 hrtimer_hres_enabled = 1;
493 else
494 return 0;
495 return 1;
496}
497
498__setup("highres=", setup_hrtimer_hres);
499
500/*
501 * hrtimer_high_res_enabled - query, if the highres mode is enabled
502 */
503static inline int hrtimer_is_hres_enabled(void)
504{
505 return hrtimer_hres_enabled;
506}
507
508/*
509 * Is the high resolution mode active ?
510 */
511static inline int hrtimer_hres_active(void)
512{
513 return __this_cpu_read(hrtimer_bases.hres_active);
514}
515
516/*
517 * Reprogram the event source with checking both queues for the
518 * next event
519 * Called with interrupts disabled and base->lock held
520 */
521static void
522hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
523{
524 ktime_t expires_next = __hrtimer_get_next_event(cpu_base);
525
526 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
527 return;
528
529 cpu_base->expires_next.tv64 = expires_next.tv64;
530
531 /*
532 * If a hang was detected in the last timer interrupt then we
533 * leave the hang delay active in the hardware. We want the
534 * system to make progress. That also prevents the following
535 * scenario:
536 * T1 expires 50ms from now
537 * T2 expires 5s from now
538 *
539 * T1 is removed, so this code is called and would reprogram
540 * the hardware to 5s from now. Any hrtimer_start after that
541 * will not reprogram the hardware due to hang_detected being
542 * set. So we'd effectivly block all timers until the T2 event
543 * fires.
544 */
545 if (cpu_base->hang_detected)
546 return;
547
548 if (cpu_base->expires_next.tv64 != KTIME_MAX)
549 tick_program_event(cpu_base->expires_next, 1);
550}
551
552/*
553 * Shared reprogramming for clock_realtime and clock_monotonic
554 *
555 * When a timer is enqueued and expires earlier than the already enqueued
556 * timers, we have to check, whether it expires earlier than the timer for
557 * which the clock event device was armed.
558 *
559 * Note, that in case the state has HRTIMER_STATE_CALLBACK set, no reprogramming
560 * and no expiry check happens. The timer gets enqueued into the rbtree. The
561 * reprogramming and expiry check is done in the hrtimer_interrupt or in the
562 * softirq.
563 *
564 * Called with interrupts disabled and base->cpu_base.lock held
565 */
566static int hrtimer_reprogram(struct hrtimer *timer,
567 struct hrtimer_clock_base *base)
568{
569 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
570 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
571 int res;
572
573 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
574
575 /*
576 * When the callback is running, we do not reprogram the clock event
577 * device. The timer callback is either running on a different CPU or
578 * the callback is executed in the hrtimer_interrupt context. The
579 * reprogramming is handled either by the softirq, which called the
580 * callback or at the end of the hrtimer_interrupt.
581 */
582 if (hrtimer_callback_running(timer))
583 return 0;
584
585 /*
586 * CLOCK_REALTIME timer might be requested with an absolute
587 * expiry time which is less than base->offset. Nothing wrong
588 * about that, just avoid to call into the tick code, which
589 * has now objections against negative expiry values.
590 */
591 if (expires.tv64 < 0)
592 return -ETIME;
593
594 if (expires.tv64 >= cpu_base->expires_next.tv64)
595 return 0;
596
597 /*
598 * When the target cpu of the timer is currently executing
599 * hrtimer_interrupt(), then we do not touch the clock event
600 * device. hrtimer_interrupt() will reevaluate all clock bases
601 * before reprogramming the device.
602 */
603 if (cpu_base->in_hrtirq)
604 return 0;
605
606 /*
607 * If a hang was detected in the last timer interrupt then we
608 * do not schedule a timer which is earlier than the expiry
609 * which we enforced in the hang detection. We want the system
610 * to make progress.
611 */
612 if (cpu_base->hang_detected)
613 return 0;
614
615 /*
616 * Clockevents returns -ETIME, when the event was in the past.
617 */
618 res = tick_program_event(expires, 0);
619 if (!IS_ERR_VALUE(res))
620 cpu_base->expires_next = expires;
621 return res;
622}
623
624/*
625 * Initialize the high resolution related parts of cpu_base
626 */
627static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
628{
629 base->expires_next.tv64 = KTIME_MAX;
630 base->hres_active = 0;
631}
632
633static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
634{
635 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
636 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
637 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
638
639 return ktime_get_update_offsets_now(offs_real, offs_boot, offs_tai);
640}
641
642/*
643 * Retrigger next event is called after clock was set
644 *
645 * Called with interrupts disabled via on_each_cpu()
646 */
647static void retrigger_next_event(void *arg)
648{
649 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
650
651 if (!hrtimer_hres_active())
652 return;
653
654 raw_spin_lock(&base->lock);
655 hrtimer_update_base(base);
656 hrtimer_force_reprogram(base, 0);
657 raw_spin_unlock(&base->lock);
658}
659
660/*
661 * Switch to high resolution mode
662 */
663static int hrtimer_switch_to_hres(void)
664{
665 int i, cpu = smp_processor_id();
666 struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
667 unsigned long flags;
668
669 if (base->hres_active)
670 return 1;
671
672 local_irq_save(flags);
673
674 if (tick_init_highres()) {
675 local_irq_restore(flags);
676 printk(KERN_WARNING "Could not switch to high resolution "
677 "mode on CPU %d\n", cpu);
678 return 0;
679 }
680 base->hres_active = 1;
681 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
682 base->clock_base[i].resolution = KTIME_HIGH_RES;
683
684 tick_setup_sched_timer();
685 /* "Retrigger" the interrupt to get things going */
686 retrigger_next_event(NULL);
687 local_irq_restore(flags);
688 return 1;
689}
690
691static void clock_was_set_work(struct work_struct *work)
692{
693 clock_was_set();
694}
695
696static DECLARE_WORK(hrtimer_work, clock_was_set_work);
697
698/*
699 * Called from timekeeping and resume code to reprogramm the hrtimer
700 * interrupt device on all cpus.
701 */
702void clock_was_set_delayed(void)
703{
704 schedule_work(&hrtimer_work);
705}
706
707#else
708
709static inline int hrtimer_hres_active(void) { return 0; }
710static inline int hrtimer_is_hres_enabled(void) { return 0; }
711static inline int hrtimer_switch_to_hres(void) { return 0; }
712static inline void
713hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
714static inline int hrtimer_reprogram(struct hrtimer *timer,
715 struct hrtimer_clock_base *base)
716{
717 return 0;
718}
719static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
720static inline void retrigger_next_event(void *arg) { }
721
722#endif /* CONFIG_HIGH_RES_TIMERS */
723
724/*
725 * Clock realtime was set
726 *
727 * Change the offset of the realtime clock vs. the monotonic
728 * clock.
729 *
730 * We might have to reprogram the high resolution timer interrupt. On
731 * SMP we call the architecture specific code to retrigger _all_ high
732 * resolution timer interrupts. On UP we just disable interrupts and
733 * call the high resolution interrupt code.
734 */
735void clock_was_set(void)
736{
737#ifdef CONFIG_HIGH_RES_TIMERS
738 /* Retrigger the CPU local events everywhere */
739 on_each_cpu(retrigger_next_event, NULL, 1);
740#endif
741 timerfd_clock_was_set();
742}
743
744/*
745 * During resume we might have to reprogram the high resolution timer
746 * interrupt on all online CPUs. However, all other CPUs will be
747 * stopped with IRQs interrupts disabled so the clock_was_set() call
748 * must be deferred.
749 */
750void hrtimers_resume(void)
751{
752 WARN_ONCE(!irqs_disabled(),
753 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
754
755 /* Retrigger on the local CPU */
756 retrigger_next_event(NULL);
757 /* And schedule a retrigger for all others */
758 clock_was_set_delayed();
759}
760
761static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
762{
763#ifdef CONFIG_TIMER_STATS
764 if (timer->start_site)
765 return;
766 timer->start_site = __builtin_return_address(0);
767 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
768 timer->start_pid = current->pid;
769#endif
770}
771
772static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
773{
774#ifdef CONFIG_TIMER_STATS
775 timer->start_site = NULL;
776#endif
777}
778
779static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
780{
781#ifdef CONFIG_TIMER_STATS
782 if (likely(!timer_stats_active))
783 return;
784 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
785 timer->function, timer->start_comm, 0);
786#endif
787}
788
789/*
790 * Counterpart to lock_hrtimer_base above:
791 */
792static inline
793void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
794{
795 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
796}
797
798/**
799 * hrtimer_forward - forward the timer expiry
800 * @timer: hrtimer to forward
801 * @now: forward past this time
802 * @interval: the interval to forward
803 *
804 * Forward the timer expiry so it will expire in the future.
805 * Returns the number of overruns.
806 */
807u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
808{
809 u64 orun = 1;
810 ktime_t delta;
811
812 delta = ktime_sub(now, hrtimer_get_expires(timer));
813
814 if (delta.tv64 < 0)
815 return 0;
816
817 if (interval.tv64 < timer->base->resolution.tv64)
818 interval.tv64 = timer->base->resolution.tv64;
819
820 if (unlikely(delta.tv64 >= interval.tv64)) {
821 s64 incr = ktime_to_ns(interval);
822
823 orun = ktime_divns(delta, incr);
824 hrtimer_add_expires_ns(timer, incr * orun);
825 if (hrtimer_get_expires_tv64(timer) > now.tv64)
826 return orun;
827 /*
828 * This (and the ktime_add() below) is the
829 * correction for exact:
830 */
831 orun++;
832 }
833 hrtimer_add_expires(timer, interval);
834
835 return orun;
836}
837EXPORT_SYMBOL_GPL(hrtimer_forward);
838
839/*
840 * enqueue_hrtimer - internal function to (re)start a timer
841 *
842 * The timer is inserted in expiry order. Insertion into the
843 * red black tree is O(log(n)). Must hold the base lock.
844 *
845 * Returns 1 when the new timer is the leftmost timer in the tree.
846 */
847static int enqueue_hrtimer(struct hrtimer *timer,
848 struct hrtimer_clock_base *base)
849{
850 debug_activate(timer);
851
852 timerqueue_add(&base->active, &timer->node);
853 base->cpu_base->active_bases |= 1 << base->index;
854
855 /*
856 * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
857 * state of a possibly running callback.
858 */
859 timer->state |= HRTIMER_STATE_ENQUEUED;
860
861 return (&timer->node == base->active.next);
862}
863
864/*
865 * __remove_hrtimer - internal function to remove a timer
866 *
867 * Caller must hold the base lock.
868 *
869 * High resolution timer mode reprograms the clock event device when the
870 * timer is the one which expires next. The caller can disable this by setting
871 * reprogram to zero. This is useful, when the context does a reprogramming
872 * anyway (e.g. timer interrupt)
873 */
874static void __remove_hrtimer(struct hrtimer *timer,
875 struct hrtimer_clock_base *base,
876 unsigned long newstate, int reprogram)
877{
878 struct timerqueue_node *next_timer;
879 if (!(timer->state & HRTIMER_STATE_ENQUEUED))
880 goto out;
881
882 next_timer = timerqueue_getnext(&base->active);
883 timerqueue_del(&base->active, &timer->node);
884 if (&timer->node == next_timer) {
885#ifdef CONFIG_HIGH_RES_TIMERS
886 /* Reprogram the clock event device. if enabled */
887 if (reprogram && hrtimer_hres_active()) {
888 ktime_t expires;
889
890 expires = ktime_sub(hrtimer_get_expires(timer),
891 base->offset);
892 if (base->cpu_base->expires_next.tv64 == expires.tv64)
893 hrtimer_force_reprogram(base->cpu_base, 1);
894 }
895#endif
896 }
897 if (!timerqueue_getnext(&base->active))
898 base->cpu_base->active_bases &= ~(1 << base->index);
899out:
900 timer->state = newstate;
901}
902
903/*
904 * remove hrtimer, called with base lock held
905 */
906static inline int
907remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
908{
909 if (hrtimer_is_queued(timer)) {
910 unsigned long state;
911 int reprogram;
912
913 /*
914 * Remove the timer and force reprogramming when high
915 * resolution mode is active and the timer is on the current
916 * CPU. If we remove a timer on another CPU, reprogramming is
917 * skipped. The interrupt event on this CPU is fired and
918 * reprogramming happens in the interrupt handler. This is a
919 * rare case and less expensive than a smp call.
920 */
921 debug_deactivate(timer);
922 timer_stats_hrtimer_clear_start_info(timer);
923 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
924 /*
925 * We must preserve the CALLBACK state flag here,
926 * otherwise we could move the timer base in
927 * switch_hrtimer_base.
928 */
929 state = timer->state & HRTIMER_STATE_CALLBACK;
930 __remove_hrtimer(timer, base, state, reprogram);
931 return 1;
932 }
933 return 0;
934}
935
936int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
937 unsigned long delta_ns, const enum hrtimer_mode mode,
938 int wakeup)
939{
940 struct hrtimer_clock_base *base, *new_base;
941 unsigned long flags;
942 int ret, leftmost;
943
944 base = lock_hrtimer_base(timer, &flags);
945
946 /* Remove an active timer from the queue: */
947 ret = remove_hrtimer(timer, base);
948
949 if (mode & HRTIMER_MODE_REL) {
950 tim = ktime_add_safe(tim, base->get_time());
951 /*
952 * CONFIG_TIME_LOW_RES is a temporary way for architectures
953 * to signal that they simply return xtime in
954 * do_gettimeoffset(). In this case we want to round up by
955 * resolution when starting a relative timer, to avoid short
956 * timeouts. This will go away with the GTOD framework.
957 */
958#ifdef CONFIG_TIME_LOW_RES
959 tim = ktime_add_safe(tim, base->resolution);
960#endif
961 }
962
963 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
964
965 /* Switch the timer base, if necessary: */
966 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
967
968 timer_stats_hrtimer_set_start_info(timer);
969
970 leftmost = enqueue_hrtimer(timer, new_base);
971
972 if (!leftmost) {
973 unlock_hrtimer_base(timer, &flags);
974 return ret;
975 }
976
977 if (!hrtimer_is_hres_active(timer)) {
978 /*
979 * Kick to reschedule the next tick to handle the new timer
980 * on dynticks target.
981 */
982 wake_up_nohz_cpu(new_base->cpu_base->cpu);
983 } else if (new_base->cpu_base == this_cpu_ptr(&hrtimer_bases) &&
984 hrtimer_reprogram(timer, new_base)) {
985 /*
986 * Only allow reprogramming if the new base is on this CPU.
987 * (it might still be on another CPU if the timer was pending)
988 *
989 * XXX send_remote_softirq() ?
990 */
991 if (wakeup) {
992 /*
993 * We need to drop cpu_base->lock to avoid a
994 * lock ordering issue vs. rq->lock.
995 */
996 raw_spin_unlock(&new_base->cpu_base->lock);
997 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
998 local_irq_restore(flags);
999 return ret;
1000 } else {
1001 __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1002 }
1003 }
1004
1005 unlock_hrtimer_base(timer, &flags);
1006
1007 return ret;
1008}
1009EXPORT_SYMBOL_GPL(__hrtimer_start_range_ns);
1010
1011/**
1012 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1013 * @timer: the timer to be added
1014 * @tim: expiry time
1015 * @delta_ns: "slack" range for the timer
1016 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1017 * relative (HRTIMER_MODE_REL)
1018 *
1019 * Returns:
1020 * 0 on success
1021 * 1 when the timer was active
1022 */
1023int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1024 unsigned long delta_ns, const enum hrtimer_mode mode)
1025{
1026 return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1027}
1028EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1029
1030/**
1031 * hrtimer_start - (re)start an hrtimer on the current CPU
1032 * @timer: the timer to be added
1033 * @tim: expiry time
1034 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1035 * relative (HRTIMER_MODE_REL)
1036 *
1037 * Returns:
1038 * 0 on success
1039 * 1 when the timer was active
1040 */
1041int
1042hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1043{
1044 return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1045}
1046EXPORT_SYMBOL_GPL(hrtimer_start);
1047
1048
1049/**
1050 * hrtimer_try_to_cancel - try to deactivate a timer
1051 * @timer: hrtimer to stop
1052 *
1053 * Returns:
1054 * 0 when the timer was not active
1055 * 1 when the timer was active
1056 * -1 when the timer is currently excuting the callback function and
1057 * cannot be stopped
1058 */
1059int hrtimer_try_to_cancel(struct hrtimer *timer)
1060{
1061 struct hrtimer_clock_base *base;
1062 unsigned long flags;
1063 int ret = -1;
1064
1065 base = lock_hrtimer_base(timer, &flags);
1066
1067 if (!hrtimer_callback_running(timer))
1068 ret = remove_hrtimer(timer, base);
1069
1070 unlock_hrtimer_base(timer, &flags);
1071
1072 return ret;
1073
1074}
1075EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1076
1077/**
1078 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1079 * @timer: the timer to be cancelled
1080 *
1081 * Returns:
1082 * 0 when the timer was not active
1083 * 1 when the timer was active
1084 */
1085int hrtimer_cancel(struct hrtimer *timer)
1086{
1087 for (;;) {
1088 int ret = hrtimer_try_to_cancel(timer);
1089
1090 if (ret >= 0)
1091 return ret;
1092 cpu_relax();
1093 }
1094}
1095EXPORT_SYMBOL_GPL(hrtimer_cancel);
1096
1097/**
1098 * hrtimer_get_remaining - get remaining time for the timer
1099 * @timer: the timer to read
1100 */
1101ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1102{
1103 unsigned long flags;
1104 ktime_t rem;
1105
1106 lock_hrtimer_base(timer, &flags);
1107 rem = hrtimer_expires_remaining(timer);
1108 unlock_hrtimer_base(timer, &flags);
1109
1110 return rem;
1111}
1112EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1113
1114#ifdef CONFIG_NO_HZ_COMMON
1115/**
1116 * hrtimer_get_next_event - get the time until next expiry event
1117 *
1118 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1119 * is pending.
1120 */
1121ktime_t hrtimer_get_next_event(void)
1122{
1123 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1124 ktime_t mindelta = { .tv64 = KTIME_MAX };
1125 unsigned long flags;
1126
1127 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1128
1129 if (!hrtimer_hres_active())
1130 mindelta = ktime_sub(__hrtimer_get_next_event(cpu_base),
1131 ktime_get());
1132
1133 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1134
1135 if (mindelta.tv64 < 0)
1136 mindelta.tv64 = 0;
1137 return mindelta;
1138}
1139#endif
1140
1141static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1142 enum hrtimer_mode mode)
1143{
1144 struct hrtimer_cpu_base *cpu_base;
1145 int base;
1146
1147 memset(timer, 0, sizeof(struct hrtimer));
1148
1149 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1150
1151 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1152 clock_id = CLOCK_MONOTONIC;
1153
1154 base = hrtimer_clockid_to_base(clock_id);
1155 timer->base = &cpu_base->clock_base[base];
1156 timerqueue_init(&timer->node);
1157
1158#ifdef CONFIG_TIMER_STATS
1159 timer->start_site = NULL;
1160 timer->start_pid = -1;
1161 memset(timer->start_comm, 0, TASK_COMM_LEN);
1162#endif
1163}
1164
1165/**
1166 * hrtimer_init - initialize a timer to the given clock
1167 * @timer: the timer to be initialized
1168 * @clock_id: the clock to be used
1169 * @mode: timer mode abs/rel
1170 */
1171void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1172 enum hrtimer_mode mode)
1173{
1174 debug_init(timer, clock_id, mode);
1175 __hrtimer_init(timer, clock_id, mode);
1176}
1177EXPORT_SYMBOL_GPL(hrtimer_init);
1178
1179/**
1180 * hrtimer_get_res - get the timer resolution for a clock
1181 * @which_clock: which clock to query
1182 * @tp: pointer to timespec variable to store the resolution
1183 *
1184 * Store the resolution of the clock selected by @which_clock in the
1185 * variable pointed to by @tp.
1186 */
1187int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1188{
1189 struct hrtimer_cpu_base *cpu_base;
1190 int base = hrtimer_clockid_to_base(which_clock);
1191
1192 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1193 *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1194
1195 return 0;
1196}
1197EXPORT_SYMBOL_GPL(hrtimer_get_res);
1198
1199static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1200{
1201 struct hrtimer_clock_base *base = timer->base;
1202 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1203 enum hrtimer_restart (*fn)(struct hrtimer *);
1204 int restart;
1205
1206 WARN_ON(!irqs_disabled());
1207
1208 debug_deactivate(timer);
1209 __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1210 timer_stats_account_hrtimer(timer);
1211 fn = timer->function;
1212
1213 /*
1214 * Because we run timers from hardirq context, there is no chance
1215 * they get migrated to another cpu, therefore its safe to unlock
1216 * the timer base.
1217 */
1218 raw_spin_unlock(&cpu_base->lock);
1219 trace_hrtimer_expire_entry(timer, now);
1220 restart = fn(timer);
1221 trace_hrtimer_expire_exit(timer);
1222 raw_spin_lock(&cpu_base->lock);
1223
1224 /*
1225 * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1226 * we do not reprogramm the event hardware. Happens either in
1227 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1228 */
1229 if (restart != HRTIMER_NORESTART) {
1230 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1231 enqueue_hrtimer(timer, base);
1232 }
1233
1234 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1235
1236 timer->state &= ~HRTIMER_STATE_CALLBACK;
1237}
1238
1239#ifdef CONFIG_HIGH_RES_TIMERS
1240
1241/*
1242 * High resolution timer interrupt
1243 * Called with interrupts disabled
1244 */
1245void hrtimer_interrupt(struct clock_event_device *dev)
1246{
1247 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1248 ktime_t expires_next, now, entry_time, delta;
1249 int i, retries = 0;
1250
1251 BUG_ON(!cpu_base->hres_active);
1252 cpu_base->nr_events++;
1253 dev->next_event.tv64 = KTIME_MAX;
1254
1255 raw_spin_lock(&cpu_base->lock);
1256 entry_time = now = hrtimer_update_base(cpu_base);
1257retry:
1258 cpu_base->in_hrtirq = 1;
1259 /*
1260 * We set expires_next to KTIME_MAX here with cpu_base->lock
1261 * held to prevent that a timer is enqueued in our queue via
1262 * the migration code. This does not affect enqueueing of
1263 * timers which run their callback and need to be requeued on
1264 * this CPU.
1265 */
1266 cpu_base->expires_next.tv64 = KTIME_MAX;
1267
1268 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1269 struct hrtimer_clock_base *base;
1270 struct timerqueue_node *node;
1271 ktime_t basenow;
1272
1273 if (!(cpu_base->active_bases & (1 << i)))
1274 continue;
1275
1276 base = cpu_base->clock_base + i;
1277 basenow = ktime_add(now, base->offset);
1278
1279 while ((node = timerqueue_getnext(&base->active))) {
1280 struct hrtimer *timer;
1281
1282 timer = container_of(node, struct hrtimer, node);
1283
1284 /*
1285 * The immediate goal for using the softexpires is
1286 * minimizing wakeups, not running timers at the
1287 * earliest interrupt after their soft expiration.
1288 * This allows us to avoid using a Priority Search
1289 * Tree, which can answer a stabbing querry for
1290 * overlapping intervals and instead use the simple
1291 * BST we already have.
1292 * We don't add extra wakeups by delaying timers that
1293 * are right-of a not yet expired timer, because that
1294 * timer will have to trigger a wakeup anyway.
1295 */
1296 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer))
1297 break;
1298
1299 __run_hrtimer(timer, &basenow);
1300 }
1301 }
1302 /* Reevaluate the clock bases for the next expiry */
1303 expires_next = __hrtimer_get_next_event(cpu_base);
1304 /*
1305 * Store the new expiry value so the migration code can verify
1306 * against it.
1307 */
1308 cpu_base->expires_next = expires_next;
1309 cpu_base->in_hrtirq = 0;
1310 raw_spin_unlock(&cpu_base->lock);
1311
1312 /* Reprogramming necessary ? */
1313 if (expires_next.tv64 == KTIME_MAX ||
1314 !tick_program_event(expires_next, 0)) {
1315 cpu_base->hang_detected = 0;
1316 return;
1317 }
1318
1319 /*
1320 * The next timer was already expired due to:
1321 * - tracing
1322 * - long lasting callbacks
1323 * - being scheduled away when running in a VM
1324 *
1325 * We need to prevent that we loop forever in the hrtimer
1326 * interrupt routine. We give it 3 attempts to avoid
1327 * overreacting on some spurious event.
1328 *
1329 * Acquire base lock for updating the offsets and retrieving
1330 * the current time.
1331 */
1332 raw_spin_lock(&cpu_base->lock);
1333 now = hrtimer_update_base(cpu_base);
1334 cpu_base->nr_retries++;
1335 if (++retries < 3)
1336 goto retry;
1337 /*
1338 * Give the system a chance to do something else than looping
1339 * here. We stored the entry time, so we know exactly how long
1340 * we spent here. We schedule the next event this amount of
1341 * time away.
1342 */
1343 cpu_base->nr_hangs++;
1344 cpu_base->hang_detected = 1;
1345 raw_spin_unlock(&cpu_base->lock);
1346 delta = ktime_sub(now, entry_time);
1347 if (delta.tv64 > cpu_base->max_hang_time.tv64)
1348 cpu_base->max_hang_time = delta;
1349 /*
1350 * Limit it to a sensible value as we enforce a longer
1351 * delay. Give the CPU at least 100ms to catch up.
1352 */
1353 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1354 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1355 else
1356 expires_next = ktime_add(now, delta);
1357 tick_program_event(expires_next, 1);
1358 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1359 ktime_to_ns(delta));
1360}
1361
1362/*
1363 * local version of hrtimer_peek_ahead_timers() called with interrupts
1364 * disabled.
1365 */
1366static void __hrtimer_peek_ahead_timers(void)
1367{
1368 struct tick_device *td;
1369
1370 if (!hrtimer_hres_active())
1371 return;
1372
1373 td = this_cpu_ptr(&tick_cpu_device);
1374 if (td && td->evtdev)
1375 hrtimer_interrupt(td->evtdev);
1376}
1377
1378/**
1379 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1380 *
1381 * hrtimer_peek_ahead_timers will peek at the timer queue of
1382 * the current cpu and check if there are any timers for which
1383 * the soft expires time has passed. If any such timers exist,
1384 * they are run immediately and then removed from the timer queue.
1385 *
1386 */
1387void hrtimer_peek_ahead_timers(void)
1388{
1389 unsigned long flags;
1390
1391 local_irq_save(flags);
1392 __hrtimer_peek_ahead_timers();
1393 local_irq_restore(flags);
1394}
1395
1396static void run_hrtimer_softirq(struct softirq_action *h)
1397{
1398 hrtimer_peek_ahead_timers();
1399}
1400
1401#else /* CONFIG_HIGH_RES_TIMERS */
1402
1403static inline void __hrtimer_peek_ahead_timers(void) { }
1404
1405#endif /* !CONFIG_HIGH_RES_TIMERS */
1406
1407/*
1408 * Called from timer softirq every jiffy, expire hrtimers:
1409 *
1410 * For HRT its the fall back code to run the softirq in the timer
1411 * softirq context in case the hrtimer initialization failed or has
1412 * not been done yet.
1413 */
1414void hrtimer_run_pending(void)
1415{
1416 if (hrtimer_hres_active())
1417 return;
1418
1419 /*
1420 * This _is_ ugly: We have to check in the softirq context,
1421 * whether we can switch to highres and / or nohz mode. The
1422 * clocksource switch happens in the timer interrupt with
1423 * xtime_lock held. Notification from there only sets the
1424 * check bit in the tick_oneshot code, otherwise we might
1425 * deadlock vs. xtime_lock.
1426 */
1427 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1428 hrtimer_switch_to_hres();
1429}
1430
1431/*
1432 * Called from hardirq context every jiffy
1433 */
1434void hrtimer_run_queues(void)
1435{
1436 struct timerqueue_node *node;
1437 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1438 struct hrtimer_clock_base *base;
1439 int index, gettime = 1;
1440
1441 if (hrtimer_hres_active())
1442 return;
1443
1444 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1445 base = &cpu_base->clock_base[index];
1446 if (!timerqueue_getnext(&base->active))
1447 continue;
1448
1449 if (gettime) {
1450 hrtimer_get_softirq_time(cpu_base);
1451 gettime = 0;
1452 }
1453
1454 raw_spin_lock(&cpu_base->lock);
1455
1456 while ((node = timerqueue_getnext(&base->active))) {
1457 struct hrtimer *timer;
1458
1459 timer = container_of(node, struct hrtimer, node);
1460 if (base->softirq_time.tv64 <=
1461 hrtimer_get_expires_tv64(timer))
1462 break;
1463
1464 __run_hrtimer(timer, &base->softirq_time);
1465 }
1466 raw_spin_unlock(&cpu_base->lock);
1467 }
1468}
1469
1470/*
1471 * Sleep related functions:
1472 */
1473static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1474{
1475 struct hrtimer_sleeper *t =
1476 container_of(timer, struct hrtimer_sleeper, timer);
1477 struct task_struct *task = t->task;
1478
1479 t->task = NULL;
1480 if (task)
1481 wake_up_process(task);
1482
1483 return HRTIMER_NORESTART;
1484}
1485
1486void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1487{
1488 sl->timer.function = hrtimer_wakeup;
1489 sl->task = task;
1490}
1491EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1492
1493static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1494{
1495 hrtimer_init_sleeper(t, current);
1496
1497 do {
1498 set_current_state(TASK_INTERRUPTIBLE);
1499 hrtimer_start_expires(&t->timer, mode);
1500 if (!hrtimer_active(&t->timer))
1501 t->task = NULL;
1502
1503 if (likely(t->task))
1504 freezable_schedule();
1505
1506 hrtimer_cancel(&t->timer);
1507 mode = HRTIMER_MODE_ABS;
1508
1509 } while (t->task && !signal_pending(current));
1510
1511 __set_current_state(TASK_RUNNING);
1512
1513 return t->task == NULL;
1514}
1515
1516static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1517{
1518 struct timespec rmt;
1519 ktime_t rem;
1520
1521 rem = hrtimer_expires_remaining(timer);
1522 if (rem.tv64 <= 0)
1523 return 0;
1524 rmt = ktime_to_timespec(rem);
1525
1526 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1527 return -EFAULT;
1528
1529 return 1;
1530}
1531
1532long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1533{
1534 struct hrtimer_sleeper t;
1535 struct timespec __user *rmtp;
1536 int ret = 0;
1537
1538 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1539 HRTIMER_MODE_ABS);
1540 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1541
1542 if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1543 goto out;
1544
1545 rmtp = restart->nanosleep.rmtp;
1546 if (rmtp) {
1547 ret = update_rmtp(&t.timer, rmtp);
1548 if (ret <= 0)
1549 goto out;
1550 }
1551
1552 /* The other values in restart are already filled in */
1553 ret = -ERESTART_RESTARTBLOCK;
1554out:
1555 destroy_hrtimer_on_stack(&t.timer);
1556 return ret;
1557}
1558
1559long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1560 const enum hrtimer_mode mode, const clockid_t clockid)
1561{
1562 struct restart_block *restart;
1563 struct hrtimer_sleeper t;
1564 int ret = 0;
1565 unsigned long slack;
1566
1567 slack = current->timer_slack_ns;
1568 if (dl_task(current) || rt_task(current))
1569 slack = 0;
1570
1571 hrtimer_init_on_stack(&t.timer, clockid, mode);
1572 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1573 if (do_nanosleep(&t, mode))
1574 goto out;
1575
1576 /* Absolute timers do not update the rmtp value and restart: */
1577 if (mode == HRTIMER_MODE_ABS) {
1578 ret = -ERESTARTNOHAND;
1579 goto out;
1580 }
1581
1582 if (rmtp) {
1583 ret = update_rmtp(&t.timer, rmtp);
1584 if (ret <= 0)
1585 goto out;
1586 }
1587
1588 restart = &current->restart_block;
1589 restart->fn = hrtimer_nanosleep_restart;
1590 restart->nanosleep.clockid = t.timer.base->clockid;
1591 restart->nanosleep.rmtp = rmtp;
1592 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1593
1594 ret = -ERESTART_RESTARTBLOCK;
1595out:
1596 destroy_hrtimer_on_stack(&t.timer);
1597 return ret;
1598}
1599
1600SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1601 struct timespec __user *, rmtp)
1602{
1603 struct timespec tu;
1604
1605 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1606 return -EFAULT;
1607
1608 if (!timespec_valid(&tu))
1609 return -EINVAL;
1610
1611 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1612}
1613
1614/*
1615 * Functions related to boot-time initialization:
1616 */
1617static void init_hrtimers_cpu(int cpu)
1618{
1619 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1620 int i;
1621
1622 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1623 cpu_base->clock_base[i].cpu_base = cpu_base;
1624 timerqueue_init_head(&cpu_base->clock_base[i].active);
1625 }
1626
1627 cpu_base->cpu = cpu;
1628 hrtimer_init_hres(cpu_base);
1629}
1630
1631#ifdef CONFIG_HOTPLUG_CPU
1632
1633static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1634 struct hrtimer_clock_base *new_base)
1635{
1636 struct hrtimer *timer;
1637 struct timerqueue_node *node;
1638
1639 while ((node = timerqueue_getnext(&old_base->active))) {
1640 timer = container_of(node, struct hrtimer, node);
1641 BUG_ON(hrtimer_callback_running(timer));
1642 debug_deactivate(timer);
1643
1644 /*
1645 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1646 * timer could be seen as !active and just vanish away
1647 * under us on another CPU
1648 */
1649 __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1650 timer->base = new_base;
1651 /*
1652 * Enqueue the timers on the new cpu. This does not
1653 * reprogram the event device in case the timer
1654 * expires before the earliest on this CPU, but we run
1655 * hrtimer_interrupt after we migrated everything to
1656 * sort out already expired timers and reprogram the
1657 * event device.
1658 */
1659 enqueue_hrtimer(timer, new_base);
1660
1661 /* Clear the migration state bit */
1662 timer->state &= ~HRTIMER_STATE_MIGRATE;
1663 }
1664}
1665
1666static void migrate_hrtimers(int scpu)
1667{
1668 struct hrtimer_cpu_base *old_base, *new_base;
1669 int i;
1670
1671 BUG_ON(cpu_online(scpu));
1672 tick_cancel_sched_timer(scpu);
1673
1674 local_irq_disable();
1675 old_base = &per_cpu(hrtimer_bases, scpu);
1676 new_base = this_cpu_ptr(&hrtimer_bases);
1677 /*
1678 * The caller is globally serialized and nobody else
1679 * takes two locks at once, deadlock is not possible.
1680 */
1681 raw_spin_lock(&new_base->lock);
1682 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1683
1684 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1685 migrate_hrtimer_list(&old_base->clock_base[i],
1686 &new_base->clock_base[i]);
1687 }
1688
1689 raw_spin_unlock(&old_base->lock);
1690 raw_spin_unlock(&new_base->lock);
1691
1692 /* Check, if we got expired work to do */
1693 __hrtimer_peek_ahead_timers();
1694 local_irq_enable();
1695}
1696
1697#endif /* CONFIG_HOTPLUG_CPU */
1698
1699static int hrtimer_cpu_notify(struct notifier_block *self,
1700 unsigned long action, void *hcpu)
1701{
1702 int scpu = (long)hcpu;
1703
1704 switch (action) {
1705
1706 case CPU_UP_PREPARE:
1707 case CPU_UP_PREPARE_FROZEN:
1708 init_hrtimers_cpu(scpu);
1709 break;
1710
1711#ifdef CONFIG_HOTPLUG_CPU
1712 case CPU_DEAD:
1713 case CPU_DEAD_FROZEN:
1714 migrate_hrtimers(scpu);
1715 break;
1716#endif
1717
1718 default:
1719 break;
1720 }
1721
1722 return NOTIFY_OK;
1723}
1724
1725static struct notifier_block hrtimers_nb = {
1726 .notifier_call = hrtimer_cpu_notify,
1727};
1728
1729void __init hrtimers_init(void)
1730{
1731 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1732 (void *)(long)smp_processor_id());
1733 register_cpu_notifier(&hrtimers_nb);
1734#ifdef CONFIG_HIGH_RES_TIMERS
1735 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1736#endif
1737}
1738
1739/**
1740 * schedule_hrtimeout_range_clock - sleep until timeout
1741 * @expires: timeout value (ktime_t)
1742 * @delta: slack in expires timeout (ktime_t)
1743 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1744 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1745 */
1746int __sched
1747schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1748 const enum hrtimer_mode mode, int clock)
1749{
1750 struct hrtimer_sleeper t;
1751
1752 /*
1753 * Optimize when a zero timeout value is given. It does not
1754 * matter whether this is an absolute or a relative time.
1755 */
1756 if (expires && !expires->tv64) {
1757 __set_current_state(TASK_RUNNING);
1758 return 0;
1759 }
1760
1761 /*
1762 * A NULL parameter means "infinite"
1763 */
1764 if (!expires) {
1765 schedule();
1766 return -EINTR;
1767 }
1768
1769 hrtimer_init_on_stack(&t.timer, clock, mode);
1770 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1771
1772 hrtimer_init_sleeper(&t, current);
1773
1774 hrtimer_start_expires(&t.timer, mode);
1775 if (!hrtimer_active(&t.timer))
1776 t.task = NULL;
1777
1778 if (likely(t.task))
1779 schedule();
1780
1781 hrtimer_cancel(&t.timer);
1782 destroy_hrtimer_on_stack(&t.timer);
1783
1784 __set_current_state(TASK_RUNNING);
1785
1786 return !t.task ? 0 : -EINTR;
1787}
1788
1789/**
1790 * schedule_hrtimeout_range - sleep until timeout
1791 * @expires: timeout value (ktime_t)
1792 * @delta: slack in expires timeout (ktime_t)
1793 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1794 *
1795 * Make the current task sleep until the given expiry time has
1796 * elapsed. The routine will return immediately unless
1797 * the current task state has been set (see set_current_state()).
1798 *
1799 * The @delta argument gives the kernel the freedom to schedule the
1800 * actual wakeup to a time that is both power and performance friendly.
1801 * The kernel give the normal best effort behavior for "@expires+@delta",
1802 * but may decide to fire the timer earlier, but no earlier than @expires.
1803 *
1804 * You can set the task state as follows -
1805 *
1806 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1807 * pass before the routine returns.
1808 *
1809 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1810 * delivered to the current task.
1811 *
1812 * The current task state is guaranteed to be TASK_RUNNING when this
1813 * routine returns.
1814 *
1815 * Returns 0 when the timer has expired otherwise -EINTR
1816 */
1817int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1818 const enum hrtimer_mode mode)
1819{
1820 return schedule_hrtimeout_range_clock(expires, delta, mode,
1821 CLOCK_MONOTONIC);
1822}
1823EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1824
1825/**
1826 * schedule_hrtimeout - sleep until timeout
1827 * @expires: timeout value (ktime_t)
1828 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1829 *
1830 * Make the current task sleep until the given expiry time has
1831 * elapsed. The routine will return immediately unless
1832 * the current task state has been set (see set_current_state()).
1833 *
1834 * You can set the task state as follows -
1835 *
1836 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1837 * pass before the routine returns.
1838 *
1839 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1840 * delivered to the current task.
1841 *
1842 * The current task state is guaranteed to be TASK_RUNNING when this
1843 * routine returns.
1844 *
1845 * Returns 0 when the timer has expired otherwise -EINTR
1846 */
1847int __sched schedule_hrtimeout(ktime_t *expires,
1848 const enum hrtimer_mode mode)
1849{
1850 return schedule_hrtimeout_range(expires, 0, mode);
1851}
1852EXPORT_SYMBOL_GPL(schedule_hrtimeout);