Root/kernel/hrtimer.c

1/*
2 * linux/kernel/hrtimer.c
3 *
4 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
5 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
6 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
7 *
8 * High-resolution kernel timers
9 *
10 * In contrast to the low-resolution timeout API implemented in
11 * kernel/timer.c, hrtimers provide finer resolution and accuracy
12 * depending on system configuration and capabilities.
13 *
14 * These timers are currently used for:
15 * - itimers
16 * - POSIX timers
17 * - nanosleep
18 * - precise in-kernel timing
19 *
20 * Started by: Thomas Gleixner and Ingo Molnar
21 *
22 * Credits:
23 * based on kernel/timer.c
24 *
25 * Help, testing, suggestions, bugfixes, improvements were
26 * provided by:
27 *
28 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
29 * et. al.
30 *
31 * For licencing details see kernel-base/COPYING
32 */
33
34#include <linux/cpu.h>
35#include <linux/module.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/timer.h>
48
49#include <asm/uaccess.h>
50
51#include <trace/events/timer.h>
52
53/*
54 * The timer bases:
55 *
56 * There are more clockids then hrtimer bases. Thus, we index
57 * into the timer bases by the hrtimer_base_type enum. When trying
58 * to reach a base using a clockid, hrtimer_clockid_to_base()
59 * is used to convert from clockid to the proper hrtimer_base_type.
60 */
61DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
62{
63
64    .clock_base =
65    {
66        {
67            .index = HRTIMER_BASE_MONOTONIC,
68            .clockid = CLOCK_MONOTONIC,
69            .get_time = &ktime_get,
70            .resolution = KTIME_LOW_RES,
71        },
72        {
73            .index = HRTIMER_BASE_REALTIME,
74            .clockid = CLOCK_REALTIME,
75            .get_time = &ktime_get_real,
76            .resolution = KTIME_LOW_RES,
77        },
78        {
79            .index = HRTIMER_BASE_BOOTTIME,
80            .clockid = CLOCK_BOOTTIME,
81            .get_time = &ktime_get_boottime,
82            .resolution = KTIME_LOW_RES,
83        },
84    }
85};
86
87static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
88    [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
89    [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
90    [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
91};
92
93static inline int hrtimer_clockid_to_base(clockid_t clock_id)
94{
95    return hrtimer_clock_to_base_table[clock_id];
96}
97
98
99/*
100 * Get the coarse grained time at the softirq based on xtime and
101 * wall_to_monotonic.
102 */
103static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
104{
105    ktime_t xtim, mono, boot;
106    struct timespec xts, tom, slp;
107
108    get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp);
109
110    xtim = timespec_to_ktime(xts);
111    mono = ktime_add(xtim, timespec_to_ktime(tom));
112    boot = ktime_add(mono, timespec_to_ktime(slp));
113    base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
114    base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono;
115    base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot;
116}
117
118/*
119 * Functions and macros which are different for UP/SMP systems are kept in a
120 * single place
121 */
122#ifdef CONFIG_SMP
123
124/*
125 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
126 * means that all timers which are tied to this base via timer->base are
127 * locked, and the base itself is locked too.
128 *
129 * So __run_timers/migrate_timers can safely modify all timers which could
130 * be found on the lists/queues.
131 *
132 * When the timer's base is locked, and the timer removed from list, it is
133 * possible to set timer->base = NULL and drop the lock: the timer remains
134 * locked.
135 */
136static
137struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
138                         unsigned long *flags)
139{
140    struct hrtimer_clock_base *base;
141
142    for (;;) {
143        base = timer->base;
144        if (likely(base != NULL)) {
145            raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
146            if (likely(base == timer->base))
147                return base;
148            /* The timer has migrated to another CPU: */
149            raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
150        }
151        cpu_relax();
152    }
153}
154
155
156/*
157 * Get the preferred target CPU for NOHZ
158 */
159static int hrtimer_get_target(int this_cpu, int pinned)
160{
161#ifdef CONFIG_NO_HZ
162    if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu))
163        return get_nohz_timer_target();
164#endif
165    return this_cpu;
166}
167
168/*
169 * With HIGHRES=y we do not migrate the timer when it is expiring
170 * before the next event on the target cpu because we cannot reprogram
171 * the target cpu hardware and we would cause it to fire late.
172 *
173 * Called with cpu_base->lock of target cpu held.
174 */
175static int
176hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
177{
178#ifdef CONFIG_HIGH_RES_TIMERS
179    ktime_t expires;
180
181    if (!new_base->cpu_base->hres_active)
182        return 0;
183
184    expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
185    return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
186#else
187    return 0;
188#endif
189}
190
191/*
192 * Switch the timer base to the current CPU when possible.
193 */
194static inline struct hrtimer_clock_base *
195switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
196            int pinned)
197{
198    struct hrtimer_clock_base *new_base;
199    struct hrtimer_cpu_base *new_cpu_base;
200    int this_cpu = smp_processor_id();
201    int cpu = hrtimer_get_target(this_cpu, pinned);
202    int basenum = base->index;
203
204again:
205    new_cpu_base = &per_cpu(hrtimer_bases, cpu);
206    new_base = &new_cpu_base->clock_base[basenum];
207
208    if (base != new_base) {
209        /*
210         * We are trying to move timer to new_base.
211         * However we can't change timer's base while it is running,
212         * so we keep it on the same CPU. No hassle vs. reprogramming
213         * the event source in the high resolution case. The softirq
214         * code will take care of this when the timer function has
215         * completed. There is no conflict as we hold the lock until
216         * the timer is enqueued.
217         */
218        if (unlikely(hrtimer_callback_running(timer)))
219            return base;
220
221        /* See the comment in lock_timer_base() */
222        timer->base = NULL;
223        raw_spin_unlock(&base->cpu_base->lock);
224        raw_spin_lock(&new_base->cpu_base->lock);
225
226        if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
227            cpu = this_cpu;
228            raw_spin_unlock(&new_base->cpu_base->lock);
229            raw_spin_lock(&base->cpu_base->lock);
230            timer->base = base;
231            goto again;
232        }
233        timer->base = new_base;
234    }
235    return new_base;
236}
237
238#else /* CONFIG_SMP */
239
240static inline struct hrtimer_clock_base *
241lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
242{
243    struct hrtimer_clock_base *base = timer->base;
244
245    raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
246
247    return base;
248}
249
250# define switch_hrtimer_base(t, b, p) (b)
251
252#endif /* !CONFIG_SMP */
253
254/*
255 * Functions for the union type storage format of ktime_t which are
256 * too large for inlining:
257 */
258#if BITS_PER_LONG < 64
259# ifndef CONFIG_KTIME_SCALAR
260/**
261 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
262 * @kt: addend
263 * @nsec: the scalar nsec value to add
264 *
265 * Returns the sum of kt and nsec in ktime_t format
266 */
267ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
268{
269    ktime_t tmp;
270
271    if (likely(nsec < NSEC_PER_SEC)) {
272        tmp.tv64 = nsec;
273    } else {
274        unsigned long rem = do_div(nsec, NSEC_PER_SEC);
275
276        tmp = ktime_set((long)nsec, rem);
277    }
278
279    return ktime_add(kt, tmp);
280}
281
282EXPORT_SYMBOL_GPL(ktime_add_ns);
283
284/**
285 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
286 * @kt: minuend
287 * @nsec: the scalar nsec value to subtract
288 *
289 * Returns the subtraction of @nsec from @kt in ktime_t format
290 */
291ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
292{
293    ktime_t tmp;
294
295    if (likely(nsec < NSEC_PER_SEC)) {
296        tmp.tv64 = nsec;
297    } else {
298        unsigned long rem = do_div(nsec, NSEC_PER_SEC);
299
300        tmp = ktime_set((long)nsec, rem);
301    }
302
303    return ktime_sub(kt, tmp);
304}
305
306EXPORT_SYMBOL_GPL(ktime_sub_ns);
307# endif /* !CONFIG_KTIME_SCALAR */
308
309/*
310 * Divide a ktime value by a nanosecond value
311 */
312u64 ktime_divns(const ktime_t kt, s64 div)
313{
314    u64 dclc;
315    int sft = 0;
316
317    dclc = ktime_to_ns(kt);
318    /* Make sure the divisor is less than 2^32: */
319    while (div >> 32) {
320        sft++;
321        div >>= 1;
322    }
323    dclc >>= sft;
324    do_div(dclc, (unsigned long) div);
325
326    return dclc;
327}
328#endif /* BITS_PER_LONG >= 64 */
329
330/*
331 * Add two ktime values and do a safety check for overflow:
332 */
333ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
334{
335    ktime_t res = ktime_add(lhs, rhs);
336
337    /*
338     * We use KTIME_SEC_MAX here, the maximum timeout which we can
339     * return to user space in a timespec:
340     */
341    if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
342        res = ktime_set(KTIME_SEC_MAX, 0);
343
344    return res;
345}
346
347EXPORT_SYMBOL_GPL(ktime_add_safe);
348
349#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
350
351static struct debug_obj_descr hrtimer_debug_descr;
352
353static void *hrtimer_debug_hint(void *addr)
354{
355    return ((struct hrtimer *) addr)->function;
356}
357
358/*
359 * fixup_init is called when:
360 * - an active object is initialized
361 */
362static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
363{
364    struct hrtimer *timer = addr;
365
366    switch (state) {
367    case ODEBUG_STATE_ACTIVE:
368        hrtimer_cancel(timer);
369        debug_object_init(timer, &hrtimer_debug_descr);
370        return 1;
371    default:
372        return 0;
373    }
374}
375
376/*
377 * fixup_activate is called when:
378 * - an active object is activated
379 * - an unknown object is activated (might be a statically initialized object)
380 */
381static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
382{
383    switch (state) {
384
385    case ODEBUG_STATE_NOTAVAILABLE:
386        WARN_ON_ONCE(1);
387        return 0;
388
389    case ODEBUG_STATE_ACTIVE:
390        WARN_ON(1);
391
392    default:
393        return 0;
394    }
395}
396
397/*
398 * fixup_free is called when:
399 * - an active object is freed
400 */
401static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
402{
403    struct hrtimer *timer = addr;
404
405    switch (state) {
406    case ODEBUG_STATE_ACTIVE:
407        hrtimer_cancel(timer);
408        debug_object_free(timer, &hrtimer_debug_descr);
409        return 1;
410    default:
411        return 0;
412    }
413}
414
415static struct debug_obj_descr hrtimer_debug_descr = {
416    .name = "hrtimer",
417    .debug_hint = hrtimer_debug_hint,
418    .fixup_init = hrtimer_fixup_init,
419    .fixup_activate = hrtimer_fixup_activate,
420    .fixup_free = hrtimer_fixup_free,
421};
422
423static inline void debug_hrtimer_init(struct hrtimer *timer)
424{
425    debug_object_init(timer, &hrtimer_debug_descr);
426}
427
428static inline void debug_hrtimer_activate(struct hrtimer *timer)
429{
430    debug_object_activate(timer, &hrtimer_debug_descr);
431}
432
433static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
434{
435    debug_object_deactivate(timer, &hrtimer_debug_descr);
436}
437
438static inline void debug_hrtimer_free(struct hrtimer *timer)
439{
440    debug_object_free(timer, &hrtimer_debug_descr);
441}
442
443static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
444               enum hrtimer_mode mode);
445
446void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
447               enum hrtimer_mode mode)
448{
449    debug_object_init_on_stack(timer, &hrtimer_debug_descr);
450    __hrtimer_init(timer, clock_id, mode);
451}
452EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
453
454void destroy_hrtimer_on_stack(struct hrtimer *timer)
455{
456    debug_object_free(timer, &hrtimer_debug_descr);
457}
458
459#else
460static inline void debug_hrtimer_init(struct hrtimer *timer) { }
461static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
462static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
463#endif
464
465static inline void
466debug_init(struct hrtimer *timer, clockid_t clockid,
467       enum hrtimer_mode mode)
468{
469    debug_hrtimer_init(timer);
470    trace_hrtimer_init(timer, clockid, mode);
471}
472
473static inline void debug_activate(struct hrtimer *timer)
474{
475    debug_hrtimer_activate(timer);
476    trace_hrtimer_start(timer);
477}
478
479static inline void debug_deactivate(struct hrtimer *timer)
480{
481    debug_hrtimer_deactivate(timer);
482    trace_hrtimer_cancel(timer);
483}
484
485/* High resolution timer related functions */
486#ifdef CONFIG_HIGH_RES_TIMERS
487
488/*
489 * High resolution timer enabled ?
490 */
491static int hrtimer_hres_enabled __read_mostly = 1;
492
493/*
494 * Enable / Disable high resolution mode
495 */
496static int __init setup_hrtimer_hres(char *str)
497{
498    if (!strcmp(str, "off"))
499        hrtimer_hres_enabled = 0;
500    else if (!strcmp(str, "on"))
501        hrtimer_hres_enabled = 1;
502    else
503        return 0;
504    return 1;
505}
506
507__setup("highres=", setup_hrtimer_hres);
508
509/*
510 * hrtimer_high_res_enabled - query, if the highres mode is enabled
511 */
512static inline int hrtimer_is_hres_enabled(void)
513{
514    return hrtimer_hres_enabled;
515}
516
517/*
518 * Is the high resolution mode active ?
519 */
520static inline int hrtimer_hres_active(void)
521{
522    return __this_cpu_read(hrtimer_bases.hres_active);
523}
524
525/*
526 * Reprogram the event source with checking both queues for the
527 * next event
528 * Called with interrupts disabled and base->lock held
529 */
530static void
531hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
532{
533    int i;
534    struct hrtimer_clock_base *base = cpu_base->clock_base;
535    ktime_t expires, expires_next;
536
537    expires_next.tv64 = KTIME_MAX;
538
539    for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
540        struct hrtimer *timer;
541        struct timerqueue_node *next;
542
543        next = timerqueue_getnext(&base->active);
544        if (!next)
545            continue;
546        timer = container_of(next, struct hrtimer, node);
547
548        expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
549        /*
550         * clock_was_set() has changed base->offset so the
551         * result might be negative. Fix it up to prevent a
552         * false positive in clockevents_program_event()
553         */
554        if (expires.tv64 < 0)
555            expires.tv64 = 0;
556        if (expires.tv64 < expires_next.tv64)
557            expires_next = expires;
558    }
559
560    if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
561        return;
562
563    cpu_base->expires_next.tv64 = expires_next.tv64;
564
565    if (cpu_base->expires_next.tv64 != KTIME_MAX)
566        tick_program_event(cpu_base->expires_next, 1);
567}
568
569/*
570 * Shared reprogramming for clock_realtime and clock_monotonic
571 *
572 * When a timer is enqueued and expires earlier than the already enqueued
573 * timers, we have to check, whether it expires earlier than the timer for
574 * which the clock event device was armed.
575 *
576 * Called with interrupts disabled and base->cpu_base.lock held
577 */
578static int hrtimer_reprogram(struct hrtimer *timer,
579                 struct hrtimer_clock_base *base)
580{
581    struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
582    ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
583    int res;
584
585    WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
586
587    /*
588     * When the callback is running, we do not reprogram the clock event
589     * device. The timer callback is either running on a different CPU or
590     * the callback is executed in the hrtimer_interrupt context. The
591     * reprogramming is handled either by the softirq, which called the
592     * callback or at the end of the hrtimer_interrupt.
593     */
594    if (hrtimer_callback_running(timer))
595        return 0;
596
597    /*
598     * CLOCK_REALTIME timer might be requested with an absolute
599     * expiry time which is less than base->offset. Nothing wrong
600     * about that, just avoid to call into the tick code, which
601     * has now objections against negative expiry values.
602     */
603    if (expires.tv64 < 0)
604        return -ETIME;
605
606    if (expires.tv64 >= cpu_base->expires_next.tv64)
607        return 0;
608
609    /*
610     * If a hang was detected in the last timer interrupt then we
611     * do not schedule a timer which is earlier than the expiry
612     * which we enforced in the hang detection. We want the system
613     * to make progress.
614     */
615    if (cpu_base->hang_detected)
616        return 0;
617
618    /*
619     * Clockevents returns -ETIME, when the event was in the past.
620     */
621    res = tick_program_event(expires, 0);
622    if (!IS_ERR_VALUE(res))
623        cpu_base->expires_next = expires;
624    return res;
625}
626
627/*
628 * Initialize the high resolution related parts of cpu_base
629 */
630static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
631{
632    base->expires_next.tv64 = KTIME_MAX;
633    base->hres_active = 0;
634}
635
636/*
637 * When High resolution timers are active, try to reprogram. Note, that in case
638 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
639 * check happens. The timer gets enqueued into the rbtree. The reprogramming
640 * and expiry check is done in the hrtimer_interrupt or in the softirq.
641 */
642static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
643                        struct hrtimer_clock_base *base,
644                        int wakeup)
645{
646    if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
647        if (wakeup) {
648            raw_spin_unlock(&base->cpu_base->lock);
649            raise_softirq_irqoff(HRTIMER_SOFTIRQ);
650            raw_spin_lock(&base->cpu_base->lock);
651        } else
652            __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
653
654        return 1;
655    }
656
657    return 0;
658}
659
660/*
661 * Retrigger next event is called after clock was set
662 *
663 * Called with interrupts disabled via on_each_cpu()
664 */
665static void retrigger_next_event(void *arg)
666{
667    struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
668    struct timespec realtime_offset, xtim, wtm, sleep;
669
670    if (!hrtimer_hres_active())
671        return;
672
673    /* Optimized out for !HIGH_RES */
674    get_xtime_and_monotonic_and_sleep_offset(&xtim, &wtm, &sleep);
675    set_normalized_timespec(&realtime_offset, -wtm.tv_sec, -wtm.tv_nsec);
676
677    /* Adjust CLOCK_REALTIME offset */
678    raw_spin_lock(&base->lock);
679    base->clock_base[HRTIMER_BASE_REALTIME].offset =
680        timespec_to_ktime(realtime_offset);
681    base->clock_base[HRTIMER_BASE_BOOTTIME].offset =
682        timespec_to_ktime(sleep);
683
684    hrtimer_force_reprogram(base, 0);
685    raw_spin_unlock(&base->lock);
686}
687
688/*
689 * Switch to high resolution mode
690 */
691static int hrtimer_switch_to_hres(void)
692{
693    int i, cpu = smp_processor_id();
694    struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
695    unsigned long flags;
696
697    if (base->hres_active)
698        return 1;
699
700    local_irq_save(flags);
701
702    if (tick_init_highres()) {
703        local_irq_restore(flags);
704        printk(KERN_WARNING "Could not switch to high resolution "
705                    "mode on CPU %d\n", cpu);
706        return 0;
707    }
708    base->hres_active = 1;
709    for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
710        base->clock_base[i].resolution = KTIME_HIGH_RES;
711
712    tick_setup_sched_timer();
713
714    /* "Retrigger" the interrupt to get things going */
715    retrigger_next_event(NULL);
716    local_irq_restore(flags);
717    return 1;
718}
719
720#else
721
722static inline int hrtimer_hres_active(void) { return 0; }
723static inline int hrtimer_is_hres_enabled(void) { return 0; }
724static inline int hrtimer_switch_to_hres(void) { return 0; }
725static inline void
726hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
727static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
728                        struct hrtimer_clock_base *base,
729                        int wakeup)
730{
731    return 0;
732}
733static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
734static inline void retrigger_next_event(void *arg) { }
735
736#endif /* CONFIG_HIGH_RES_TIMERS */
737
738/*
739 * Clock realtime was set
740 *
741 * Change the offset of the realtime clock vs. the monotonic
742 * clock.
743 *
744 * We might have to reprogram the high resolution timer interrupt. On
745 * SMP we call the architecture specific code to retrigger _all_ high
746 * resolution timer interrupts. On UP we just disable interrupts and
747 * call the high resolution interrupt code.
748 */
749void clock_was_set(void)
750{
751#ifdef CONFIG_HIGH_RES_TIMERS
752    /* Retrigger the CPU local events everywhere */
753    on_each_cpu(retrigger_next_event, NULL, 1);
754#endif
755    timerfd_clock_was_set();
756}
757
758/*
759 * During resume we might have to reprogram the high resolution timer
760 * interrupt (on the local CPU):
761 */
762void hrtimers_resume(void)
763{
764    WARN_ONCE(!irqs_disabled(),
765          KERN_INFO "hrtimers_resume() called with IRQs enabled!");
766
767    retrigger_next_event(NULL);
768    timerfd_clock_was_set();
769}
770
771static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
772{
773#ifdef CONFIG_TIMER_STATS
774    if (timer->start_site)
775        return;
776    timer->start_site = __builtin_return_address(0);
777    memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
778    timer->start_pid = current->pid;
779#endif
780}
781
782static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
783{
784#ifdef CONFIG_TIMER_STATS
785    timer->start_site = NULL;
786#endif
787}
788
789static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
790{
791#ifdef CONFIG_TIMER_STATS
792    if (likely(!timer_stats_active))
793        return;
794    timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
795                 timer->function, timer->start_comm, 0);
796#endif
797}
798
799/*
800 * Counterpart to lock_hrtimer_base above:
801 */
802static inline
803void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
804{
805    raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
806}
807
808/**
809 * hrtimer_forward - forward the timer expiry
810 * @timer: hrtimer to forward
811 * @now: forward past this time
812 * @interval: the interval to forward
813 *
814 * Forward the timer expiry so it will expire in the future.
815 * Returns the number of overruns.
816 */
817u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
818{
819    u64 orun = 1;
820    ktime_t delta;
821
822    delta = ktime_sub(now, hrtimer_get_expires(timer));
823
824    if (delta.tv64 < 0)
825        return 0;
826
827    if (interval.tv64 < timer->base->resolution.tv64)
828        interval.tv64 = timer->base->resolution.tv64;
829
830    if (unlikely(delta.tv64 >= interval.tv64)) {
831        s64 incr = ktime_to_ns(interval);
832
833        orun = ktime_divns(delta, incr);
834        hrtimer_add_expires_ns(timer, incr * orun);
835        if (hrtimer_get_expires_tv64(timer) > now.tv64)
836            return orun;
837        /*
838         * This (and the ktime_add() below) is the
839         * correction for exact:
840         */
841        orun++;
842    }
843    hrtimer_add_expires(timer, interval);
844
845    return orun;
846}
847EXPORT_SYMBOL_GPL(hrtimer_forward);
848
849/*
850 * enqueue_hrtimer - internal function to (re)start a timer
851 *
852 * The timer is inserted in expiry order. Insertion into the
853 * red black tree is O(log(n)). Must hold the base lock.
854 *
855 * Returns 1 when the new timer is the leftmost timer in the tree.
856 */
857static int enqueue_hrtimer(struct hrtimer *timer,
858               struct hrtimer_clock_base *base)
859{
860    debug_activate(timer);
861
862    timerqueue_add(&base->active, &timer->node);
863    base->cpu_base->active_bases |= 1 << base->index;
864
865    /*
866     * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
867     * state of a possibly running callback.
868     */
869    timer->state |= HRTIMER_STATE_ENQUEUED;
870
871    return (&timer->node == base->active.next);
872}
873
874/*
875 * __remove_hrtimer - internal function to remove a timer
876 *
877 * Caller must hold the base lock.
878 *
879 * High resolution timer mode reprograms the clock event device when the
880 * timer is the one which expires next. The caller can disable this by setting
881 * reprogram to zero. This is useful, when the context does a reprogramming
882 * anyway (e.g. timer interrupt)
883 */
884static void __remove_hrtimer(struct hrtimer *timer,
885                 struct hrtimer_clock_base *base,
886                 unsigned long newstate, int reprogram)
887{
888    if (!(timer->state & HRTIMER_STATE_ENQUEUED))
889        goto out;
890
891    if (&timer->node == timerqueue_getnext(&base->active)) {
892#ifdef CONFIG_HIGH_RES_TIMERS
893        /* Reprogram the clock event device. if enabled */
894        if (reprogram && hrtimer_hres_active()) {
895            ktime_t expires;
896
897            expires = ktime_sub(hrtimer_get_expires(timer),
898                        base->offset);
899            if (base->cpu_base->expires_next.tv64 == expires.tv64)
900                hrtimer_force_reprogram(base->cpu_base, 1);
901        }
902#endif
903    }
904    timerqueue_del(&base->active, &timer->node);
905    if (!timerqueue_getnext(&base->active))
906        base->cpu_base->active_bases &= ~(1 << base->index);
907out:
908    timer->state = newstate;
909}
910
911/*
912 * remove hrtimer, called with base lock held
913 */
914static inline int
915remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
916{
917    if (hrtimer_is_queued(timer)) {
918        unsigned long state;
919        int reprogram;
920
921        /*
922         * Remove the timer and force reprogramming when high
923         * resolution mode is active and the timer is on the current
924         * CPU. If we remove a timer on another CPU, reprogramming is
925         * skipped. The interrupt event on this CPU is fired and
926         * reprogramming happens in the interrupt handler. This is a
927         * rare case and less expensive than a smp call.
928         */
929        debug_deactivate(timer);
930        timer_stats_hrtimer_clear_start_info(timer);
931        reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
932        /*
933         * We must preserve the CALLBACK state flag here,
934         * otherwise we could move the timer base in
935         * switch_hrtimer_base.
936         */
937        state = timer->state & HRTIMER_STATE_CALLBACK;
938        __remove_hrtimer(timer, base, state, reprogram);
939        return 1;
940    }
941    return 0;
942}
943
944int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
945        unsigned long delta_ns, const enum hrtimer_mode mode,
946        int wakeup)
947{
948    struct hrtimer_clock_base *base, *new_base;
949    unsigned long flags;
950    int ret, leftmost;
951
952    base = lock_hrtimer_base(timer, &flags);
953
954    /* Remove an active timer from the queue: */
955    ret = remove_hrtimer(timer, base);
956
957    /* Switch the timer base, if necessary: */
958    new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
959
960    if (mode & HRTIMER_MODE_REL) {
961        tim = ktime_add_safe(tim, new_base->get_time());
962        /*
963         * CONFIG_TIME_LOW_RES is a temporary way for architectures
964         * to signal that they simply return xtime in
965         * do_gettimeoffset(). In this case we want to round up by
966         * resolution when starting a relative timer, to avoid short
967         * timeouts. This will go away with the GTOD framework.
968         */
969#ifdef CONFIG_TIME_LOW_RES
970        tim = ktime_add_safe(tim, base->resolution);
971#endif
972    }
973
974    hrtimer_set_expires_range_ns(timer, tim, delta_ns);
975
976    timer_stats_hrtimer_set_start_info(timer);
977
978    leftmost = enqueue_hrtimer(timer, new_base);
979
980    /*
981     * Only allow reprogramming if the new base is on this CPU.
982     * (it might still be on another CPU if the timer was pending)
983     *
984     * XXX send_remote_softirq() ?
985     */
986    if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases))
987        hrtimer_enqueue_reprogram(timer, new_base, wakeup);
988
989    unlock_hrtimer_base(timer, &flags);
990
991    return ret;
992}
993
994/**
995 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
996 * @timer: the timer to be added
997 * @tim: expiry time
998 * @delta_ns: "slack" range for the timer
999 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1000 *
1001 * Returns:
1002 * 0 on success
1003 * 1 when the timer was active
1004 */
1005int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1006        unsigned long delta_ns, const enum hrtimer_mode mode)
1007{
1008    return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1009}
1010EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1011
1012/**
1013 * hrtimer_start - (re)start an hrtimer on the current CPU
1014 * @timer: the timer to be added
1015 * @tim: expiry time
1016 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1017 *
1018 * Returns:
1019 * 0 on success
1020 * 1 when the timer was active
1021 */
1022int
1023hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1024{
1025    return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1026}
1027EXPORT_SYMBOL_GPL(hrtimer_start);
1028
1029
1030/**
1031 * hrtimer_try_to_cancel - try to deactivate a timer
1032 * @timer: hrtimer to stop
1033 *
1034 * Returns:
1035 * 0 when the timer was not active
1036 * 1 when the timer was active
1037 * -1 when the timer is currently excuting the callback function and
1038 * cannot be stopped
1039 */
1040int hrtimer_try_to_cancel(struct hrtimer *timer)
1041{
1042    struct hrtimer_clock_base *base;
1043    unsigned long flags;
1044    int ret = -1;
1045
1046    base = lock_hrtimer_base(timer, &flags);
1047
1048    if (!hrtimer_callback_running(timer))
1049        ret = remove_hrtimer(timer, base);
1050
1051    unlock_hrtimer_base(timer, &flags);
1052
1053    return ret;
1054
1055}
1056EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1057
1058/**
1059 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1060 * @timer: the timer to be cancelled
1061 *
1062 * Returns:
1063 * 0 when the timer was not active
1064 * 1 when the timer was active
1065 */
1066int hrtimer_cancel(struct hrtimer *timer)
1067{
1068    for (;;) {
1069        int ret = hrtimer_try_to_cancel(timer);
1070
1071        if (ret >= 0)
1072            return ret;
1073        cpu_relax();
1074    }
1075}
1076EXPORT_SYMBOL_GPL(hrtimer_cancel);
1077
1078/**
1079 * hrtimer_get_remaining - get remaining time for the timer
1080 * @timer: the timer to read
1081 */
1082ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1083{
1084    unsigned long flags;
1085    ktime_t rem;
1086
1087    lock_hrtimer_base(timer, &flags);
1088    rem = hrtimer_expires_remaining(timer);
1089    unlock_hrtimer_base(timer, &flags);
1090
1091    return rem;
1092}
1093EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1094
1095#ifdef CONFIG_NO_HZ
1096/**
1097 * hrtimer_get_next_event - get the time until next expiry event
1098 *
1099 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1100 * is pending.
1101 */
1102ktime_t hrtimer_get_next_event(void)
1103{
1104    struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1105    struct hrtimer_clock_base *base = cpu_base->clock_base;
1106    ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1107    unsigned long flags;
1108    int i;
1109
1110    raw_spin_lock_irqsave(&cpu_base->lock, flags);
1111
1112    if (!hrtimer_hres_active()) {
1113        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1114            struct hrtimer *timer;
1115            struct timerqueue_node *next;
1116
1117            next = timerqueue_getnext(&base->active);
1118            if (!next)
1119                continue;
1120
1121            timer = container_of(next, struct hrtimer, node);
1122            delta.tv64 = hrtimer_get_expires_tv64(timer);
1123            delta = ktime_sub(delta, base->get_time());
1124            if (delta.tv64 < mindelta.tv64)
1125                mindelta.tv64 = delta.tv64;
1126        }
1127    }
1128
1129    raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1130
1131    if (mindelta.tv64 < 0)
1132        mindelta.tv64 = 0;
1133    return mindelta;
1134}
1135#endif
1136
1137static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1138               enum hrtimer_mode mode)
1139{
1140    struct hrtimer_cpu_base *cpu_base;
1141    int base;
1142
1143    memset(timer, 0, sizeof(struct hrtimer));
1144
1145    cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1146
1147    if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1148        clock_id = CLOCK_MONOTONIC;
1149
1150    base = hrtimer_clockid_to_base(clock_id);
1151    timer->base = &cpu_base->clock_base[base];
1152    timerqueue_init(&timer->node);
1153
1154#ifdef CONFIG_TIMER_STATS
1155    timer->start_site = NULL;
1156    timer->start_pid = -1;
1157    memset(timer->start_comm, 0, TASK_COMM_LEN);
1158#endif
1159}
1160
1161/**
1162 * hrtimer_init - initialize a timer to the given clock
1163 * @timer: the timer to be initialized
1164 * @clock_id: the clock to be used
1165 * @mode: timer mode abs/rel
1166 */
1167void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1168          enum hrtimer_mode mode)
1169{
1170    debug_init(timer, clock_id, mode);
1171    __hrtimer_init(timer, clock_id, mode);
1172}
1173EXPORT_SYMBOL_GPL(hrtimer_init);
1174
1175/**
1176 * hrtimer_get_res - get the timer resolution for a clock
1177 * @which_clock: which clock to query
1178 * @tp: pointer to timespec variable to store the resolution
1179 *
1180 * Store the resolution of the clock selected by @which_clock in the
1181 * variable pointed to by @tp.
1182 */
1183int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1184{
1185    struct hrtimer_cpu_base *cpu_base;
1186    int base = hrtimer_clockid_to_base(which_clock);
1187
1188    cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1189    *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1190
1191    return 0;
1192}
1193EXPORT_SYMBOL_GPL(hrtimer_get_res);
1194
1195static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1196{
1197    struct hrtimer_clock_base *base = timer->base;
1198    struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1199    enum hrtimer_restart (*fn)(struct hrtimer *);
1200    int restart;
1201
1202    WARN_ON(!irqs_disabled());
1203
1204    debug_deactivate(timer);
1205    __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1206    timer_stats_account_hrtimer(timer);
1207    fn = timer->function;
1208
1209    /*
1210     * Because we run timers from hardirq context, there is no chance
1211     * they get migrated to another cpu, therefore its safe to unlock
1212     * the timer base.
1213     */
1214    raw_spin_unlock(&cpu_base->lock);
1215    trace_hrtimer_expire_entry(timer, now);
1216    restart = fn(timer);
1217    trace_hrtimer_expire_exit(timer);
1218    raw_spin_lock(&cpu_base->lock);
1219
1220    /*
1221     * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1222     * we do not reprogramm the event hardware. Happens either in
1223     * hrtimer_start_range_ns() or in hrtimer_interrupt()
1224     */
1225    if (restart != HRTIMER_NORESTART) {
1226        BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1227        enqueue_hrtimer(timer, base);
1228    }
1229
1230    WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1231
1232    timer->state &= ~HRTIMER_STATE_CALLBACK;
1233}
1234
1235#ifdef CONFIG_HIGH_RES_TIMERS
1236
1237/*
1238 * High resolution timer interrupt
1239 * Called with interrupts disabled
1240 */
1241void hrtimer_interrupt(struct clock_event_device *dev)
1242{
1243    struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1244    ktime_t expires_next, now, entry_time, delta;
1245    int i, retries = 0;
1246
1247    BUG_ON(!cpu_base->hres_active);
1248    cpu_base->nr_events++;
1249    dev->next_event.tv64 = KTIME_MAX;
1250
1251    entry_time = now = ktime_get();
1252retry:
1253    expires_next.tv64 = KTIME_MAX;
1254
1255    raw_spin_lock(&cpu_base->lock);
1256    /*
1257     * We set expires_next to KTIME_MAX here with cpu_base->lock
1258     * held to prevent that a timer is enqueued in our queue via
1259     * the migration code. This does not affect enqueueing of
1260     * timers which run their callback and need to be requeued on
1261     * this CPU.
1262     */
1263    cpu_base->expires_next.tv64 = KTIME_MAX;
1264
1265    for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1266        struct hrtimer_clock_base *base;
1267        struct timerqueue_node *node;
1268        ktime_t basenow;
1269
1270        if (!(cpu_base->active_bases & (1 << i)))
1271            continue;
1272
1273        base = cpu_base->clock_base + i;
1274        basenow = ktime_add(now, base->offset);
1275
1276        while ((node = timerqueue_getnext(&base->active))) {
1277            struct hrtimer *timer;
1278
1279            timer = container_of(node, struct hrtimer, node);
1280
1281            /*
1282             * The immediate goal for using the softexpires is
1283             * minimizing wakeups, not running timers at the
1284             * earliest interrupt after their soft expiration.
1285             * This allows us to avoid using a Priority Search
1286             * Tree, which can answer a stabbing querry for
1287             * overlapping intervals and instead use the simple
1288             * BST we already have.
1289             * We don't add extra wakeups by delaying timers that
1290             * are right-of a not yet expired timer, because that
1291             * timer will have to trigger a wakeup anyway.
1292             */
1293
1294            if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1295                ktime_t expires;
1296
1297                expires = ktime_sub(hrtimer_get_expires(timer),
1298                            base->offset);
1299                if (expires.tv64 < expires_next.tv64)
1300                    expires_next = expires;
1301                break;
1302            }
1303
1304            __run_hrtimer(timer, &basenow);
1305        }
1306    }
1307
1308    /*
1309     * Store the new expiry value so the migration code can verify
1310     * against it.
1311     */
1312    cpu_base->expires_next = expires_next;
1313    raw_spin_unlock(&cpu_base->lock);
1314
1315    /* Reprogramming necessary ? */
1316    if (expires_next.tv64 == KTIME_MAX ||
1317        !tick_program_event(expires_next, 0)) {
1318        cpu_base->hang_detected = 0;
1319        return;
1320    }
1321
1322    /*
1323     * The next timer was already expired due to:
1324     * - tracing
1325     * - long lasting callbacks
1326     * - being scheduled away when running in a VM
1327     *
1328     * We need to prevent that we loop forever in the hrtimer
1329     * interrupt routine. We give it 3 attempts to avoid
1330     * overreacting on some spurious event.
1331     */
1332    now = ktime_get();
1333    cpu_base->nr_retries++;
1334    if (++retries < 3)
1335        goto retry;
1336    /*
1337     * Give the system a chance to do something else than looping
1338     * here. We stored the entry time, so we know exactly how long
1339     * we spent here. We schedule the next event this amount of
1340     * time away.
1341     */
1342    cpu_base->nr_hangs++;
1343    cpu_base->hang_detected = 1;
1344    delta = ktime_sub(now, entry_time);
1345    if (delta.tv64 > cpu_base->max_hang_time.tv64)
1346        cpu_base->max_hang_time = delta;
1347    /*
1348     * Limit it to a sensible value as we enforce a longer
1349     * delay. Give the CPU at least 100ms to catch up.
1350     */
1351    if (delta.tv64 > 100 * NSEC_PER_MSEC)
1352        expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1353    else
1354        expires_next = ktime_add(now, delta);
1355    tick_program_event(expires_next, 1);
1356    printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1357            ktime_to_ns(delta));
1358}
1359
1360/*
1361 * local version of hrtimer_peek_ahead_timers() called with interrupts
1362 * disabled.
1363 */
1364static void __hrtimer_peek_ahead_timers(void)
1365{
1366    struct tick_device *td;
1367
1368    if (!hrtimer_hres_active())
1369        return;
1370
1371    td = &__get_cpu_var(tick_cpu_device);
1372    if (td && td->evtdev)
1373        hrtimer_interrupt(td->evtdev);
1374}
1375
1376/**
1377 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1378 *
1379 * hrtimer_peek_ahead_timers will peek at the timer queue of
1380 * the current cpu and check if there are any timers for which
1381 * the soft expires time has passed. If any such timers exist,
1382 * they are run immediately and then removed from the timer queue.
1383 *
1384 */
1385void hrtimer_peek_ahead_timers(void)
1386{
1387    unsigned long flags;
1388
1389    local_irq_save(flags);
1390    __hrtimer_peek_ahead_timers();
1391    local_irq_restore(flags);
1392}
1393
1394static void run_hrtimer_softirq(struct softirq_action *h)
1395{
1396    hrtimer_peek_ahead_timers();
1397}
1398
1399#else /* CONFIG_HIGH_RES_TIMERS */
1400
1401static inline void __hrtimer_peek_ahead_timers(void) { }
1402
1403#endif /* !CONFIG_HIGH_RES_TIMERS */
1404
1405/*
1406 * Called from timer softirq every jiffy, expire hrtimers:
1407 *
1408 * For HRT its the fall back code to run the softirq in the timer
1409 * softirq context in case the hrtimer initialization failed or has
1410 * not been done yet.
1411 */
1412void hrtimer_run_pending(void)
1413{
1414    if (hrtimer_hres_active())
1415        return;
1416
1417    /*
1418     * This _is_ ugly: We have to check in the softirq context,
1419     * whether we can switch to highres and / or nohz mode. The
1420     * clocksource switch happens in the timer interrupt with
1421     * xtime_lock held. Notification from there only sets the
1422     * check bit in the tick_oneshot code, otherwise we might
1423     * deadlock vs. xtime_lock.
1424     */
1425    if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1426        hrtimer_switch_to_hres();
1427}
1428
1429/*
1430 * Called from hardirq context every jiffy
1431 */
1432void hrtimer_run_queues(void)
1433{
1434    struct timerqueue_node *node;
1435    struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1436    struct hrtimer_clock_base *base;
1437    int index, gettime = 1;
1438
1439    if (hrtimer_hres_active())
1440        return;
1441
1442    for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1443        base = &cpu_base->clock_base[index];
1444        if (!timerqueue_getnext(&base->active))
1445            continue;
1446
1447        if (gettime) {
1448            hrtimer_get_softirq_time(cpu_base);
1449            gettime = 0;
1450        }
1451
1452        raw_spin_lock(&cpu_base->lock);
1453
1454        while ((node = timerqueue_getnext(&base->active))) {
1455            struct hrtimer *timer;
1456
1457            timer = container_of(node, struct hrtimer, node);
1458            if (base->softirq_time.tv64 <=
1459                    hrtimer_get_expires_tv64(timer))
1460                break;
1461
1462            __run_hrtimer(timer, &base->softirq_time);
1463        }
1464        raw_spin_unlock(&cpu_base->lock);
1465    }
1466}
1467
1468/*
1469 * Sleep related functions:
1470 */
1471static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1472{
1473    struct hrtimer_sleeper *t =
1474        container_of(timer, struct hrtimer_sleeper, timer);
1475    struct task_struct *task = t->task;
1476
1477    t->task = NULL;
1478    if (task)
1479        wake_up_process(task);
1480
1481    return HRTIMER_NORESTART;
1482}
1483
1484void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1485{
1486    sl->timer.function = hrtimer_wakeup;
1487    sl->task = task;
1488}
1489EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1490
1491static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1492{
1493    hrtimer_init_sleeper(t, current);
1494
1495    do {
1496        set_current_state(TASK_INTERRUPTIBLE);
1497        hrtimer_start_expires(&t->timer, mode);
1498        if (!hrtimer_active(&t->timer))
1499            t->task = NULL;
1500
1501        if (likely(t->task))
1502            schedule();
1503
1504        hrtimer_cancel(&t->timer);
1505        mode = HRTIMER_MODE_ABS;
1506
1507    } while (t->task && !signal_pending(current));
1508
1509    __set_current_state(TASK_RUNNING);
1510
1511    return t->task == NULL;
1512}
1513
1514static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1515{
1516    struct timespec rmt;
1517    ktime_t rem;
1518
1519    rem = hrtimer_expires_remaining(timer);
1520    if (rem.tv64 <= 0)
1521        return 0;
1522    rmt = ktime_to_timespec(rem);
1523
1524    if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1525        return -EFAULT;
1526
1527    return 1;
1528}
1529
1530long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1531{
1532    struct hrtimer_sleeper t;
1533    struct timespec __user *rmtp;
1534    int ret = 0;
1535
1536    hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1537                HRTIMER_MODE_ABS);
1538    hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1539
1540    if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1541        goto out;
1542
1543    rmtp = restart->nanosleep.rmtp;
1544    if (rmtp) {
1545        ret = update_rmtp(&t.timer, rmtp);
1546        if (ret <= 0)
1547            goto out;
1548    }
1549
1550    /* The other values in restart are already filled in */
1551    ret = -ERESTART_RESTARTBLOCK;
1552out:
1553    destroy_hrtimer_on_stack(&t.timer);
1554    return ret;
1555}
1556
1557long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1558               const enum hrtimer_mode mode, const clockid_t clockid)
1559{
1560    struct restart_block *restart;
1561    struct hrtimer_sleeper t;
1562    int ret = 0;
1563    unsigned long slack;
1564
1565    slack = current->timer_slack_ns;
1566    if (rt_task(current))
1567        slack = 0;
1568
1569    hrtimer_init_on_stack(&t.timer, clockid, mode);
1570    hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1571    if (do_nanosleep(&t, mode))
1572        goto out;
1573
1574    /* Absolute timers do not update the rmtp value and restart: */
1575    if (mode == HRTIMER_MODE_ABS) {
1576        ret = -ERESTARTNOHAND;
1577        goto out;
1578    }
1579
1580    if (rmtp) {
1581        ret = update_rmtp(&t.timer, rmtp);
1582        if (ret <= 0)
1583            goto out;
1584    }
1585
1586    restart = &current_thread_info()->restart_block;
1587    restart->fn = hrtimer_nanosleep_restart;
1588    restart->nanosleep.clockid = t.timer.base->clockid;
1589    restart->nanosleep.rmtp = rmtp;
1590    restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1591
1592    ret = -ERESTART_RESTARTBLOCK;
1593out:
1594    destroy_hrtimer_on_stack(&t.timer);
1595    return ret;
1596}
1597
1598SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1599        struct timespec __user *, rmtp)
1600{
1601    struct timespec tu;
1602
1603    if (copy_from_user(&tu, rqtp, sizeof(tu)))
1604        return -EFAULT;
1605
1606    if (!timespec_valid(&tu))
1607        return -EINVAL;
1608
1609    return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1610}
1611
1612/*
1613 * Functions related to boot-time initialization:
1614 */
1615static void __cpuinit init_hrtimers_cpu(int cpu)
1616{
1617    struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1618    int i;
1619
1620    raw_spin_lock_init(&cpu_base->lock);
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    hrtimer_init_hres(cpu_base);
1628}
1629
1630#ifdef CONFIG_HOTPLUG_CPU
1631
1632static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1633                struct hrtimer_clock_base *new_base)
1634{
1635    struct hrtimer *timer;
1636    struct timerqueue_node *node;
1637
1638    while ((node = timerqueue_getnext(&old_base->active))) {
1639        timer = container_of(node, struct hrtimer, node);
1640        BUG_ON(hrtimer_callback_running(timer));
1641        debug_deactivate(timer);
1642
1643        /*
1644         * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1645         * timer could be seen as !active and just vanish away
1646         * under us on another CPU
1647         */
1648        __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1649        timer->base = new_base;
1650        /*
1651         * Enqueue the timers on the new cpu. This does not
1652         * reprogram the event device in case the timer
1653         * expires before the earliest on this CPU, but we run
1654         * hrtimer_interrupt after we migrated everything to
1655         * sort out already expired timers and reprogram the
1656         * event device.
1657         */
1658        enqueue_hrtimer(timer, new_base);
1659
1660        /* Clear the migration state bit */
1661        timer->state &= ~HRTIMER_STATE_MIGRATE;
1662    }
1663}
1664
1665static void migrate_hrtimers(int scpu)
1666{
1667    struct hrtimer_cpu_base *old_base, *new_base;
1668    int i;
1669
1670    BUG_ON(cpu_online(scpu));
1671    tick_cancel_sched_timer(scpu);
1672
1673    local_irq_disable();
1674    old_base = &per_cpu(hrtimer_bases, scpu);
1675    new_base = &__get_cpu_var(hrtimer_bases);
1676    /*
1677     * The caller is globally serialized and nobody else
1678     * takes two locks at once, deadlock is not possible.
1679     */
1680    raw_spin_lock(&new_base->lock);
1681    raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1682
1683    for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1684        migrate_hrtimer_list(&old_base->clock_base[i],
1685                     &new_base->clock_base[i]);
1686    }
1687
1688    raw_spin_unlock(&old_base->lock);
1689    raw_spin_unlock(&new_base->lock);
1690
1691    /* Check, if we got expired work to do */
1692    __hrtimer_peek_ahead_timers();
1693    local_irq_enable();
1694}
1695
1696#endif /* CONFIG_HOTPLUG_CPU */
1697
1698static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1699                    unsigned long action, void *hcpu)
1700{
1701    int scpu = (long)hcpu;
1702
1703    switch (action) {
1704
1705    case CPU_UP_PREPARE:
1706    case CPU_UP_PREPARE_FROZEN:
1707        init_hrtimers_cpu(scpu);
1708        break;
1709
1710#ifdef CONFIG_HOTPLUG_CPU
1711    case CPU_DYING:
1712    case CPU_DYING_FROZEN:
1713        clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1714        break;
1715    case CPU_DEAD:
1716    case CPU_DEAD_FROZEN:
1717    {
1718        clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1719        migrate_hrtimers(scpu);
1720        break;
1721    }
1722#endif
1723
1724    default:
1725        break;
1726    }
1727
1728    return NOTIFY_OK;
1729}
1730
1731static struct notifier_block __cpuinitdata hrtimers_nb = {
1732    .notifier_call = hrtimer_cpu_notify,
1733};
1734
1735void __init hrtimers_init(void)
1736{
1737    hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1738              (void *)(long)smp_processor_id());
1739    register_cpu_notifier(&hrtimers_nb);
1740#ifdef CONFIG_HIGH_RES_TIMERS
1741    open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1742#endif
1743}
1744
1745/**
1746 * schedule_hrtimeout_range_clock - sleep until timeout
1747 * @expires: timeout value (ktime_t)
1748 * @delta: slack in expires timeout (ktime_t)
1749 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1750 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1751 */
1752int __sched
1753schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1754                   const enum hrtimer_mode mode, int clock)
1755{
1756    struct hrtimer_sleeper t;
1757
1758    /*
1759     * Optimize when a zero timeout value is given. It does not
1760     * matter whether this is an absolute or a relative time.
1761     */
1762    if (expires && !expires->tv64) {
1763        __set_current_state(TASK_RUNNING);
1764        return 0;
1765    }
1766
1767    /*
1768     * A NULL parameter means "infinite"
1769     */
1770    if (!expires) {
1771        schedule();
1772        __set_current_state(TASK_RUNNING);
1773        return -EINTR;
1774    }
1775
1776    hrtimer_init_on_stack(&t.timer, clock, mode);
1777    hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1778
1779    hrtimer_init_sleeper(&t, current);
1780
1781    hrtimer_start_expires(&t.timer, mode);
1782    if (!hrtimer_active(&t.timer))
1783        t.task = NULL;
1784
1785    if (likely(t.task))
1786        schedule();
1787
1788    hrtimer_cancel(&t.timer);
1789    destroy_hrtimer_on_stack(&t.timer);
1790
1791    __set_current_state(TASK_RUNNING);
1792
1793    return !t.task ? 0 : -EINTR;
1794}
1795
1796/**
1797 * schedule_hrtimeout_range - sleep until timeout
1798 * @expires: timeout value (ktime_t)
1799 * @delta: slack in expires timeout (ktime_t)
1800 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1801 *
1802 * Make the current task sleep until the given expiry time has
1803 * elapsed. The routine will return immediately unless
1804 * the current task state has been set (see set_current_state()).
1805 *
1806 * The @delta argument gives the kernel the freedom to schedule the
1807 * actual wakeup to a time that is both power and performance friendly.
1808 * The kernel give the normal best effort behavior for "@expires+@delta",
1809 * but may decide to fire the timer earlier, but no earlier than @expires.
1810 *
1811 * You can set the task state as follows -
1812 *
1813 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1814 * pass before the routine returns.
1815 *
1816 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1817 * delivered to the current task.
1818 *
1819 * The current task state is guaranteed to be TASK_RUNNING when this
1820 * routine returns.
1821 *
1822 * Returns 0 when the timer has expired otherwise -EINTR
1823 */
1824int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1825                     const enum hrtimer_mode mode)
1826{
1827    return schedule_hrtimeout_range_clock(expires, delta, mode,
1828                          CLOCK_MONOTONIC);
1829}
1830EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1831
1832/**
1833 * schedule_hrtimeout - sleep until timeout
1834 * @expires: timeout value (ktime_t)
1835 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1836 *
1837 * Make the current task sleep until the given expiry time has
1838 * elapsed. The routine will return immediately unless
1839 * the current task state has been set (see set_current_state()).
1840 *
1841 * You can set the task state as follows -
1842 *
1843 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1844 * pass before the routine returns.
1845 *
1846 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1847 * delivered to the current task.
1848 *
1849 * The current task state is guaranteed to be TASK_RUNNING when this
1850 * routine returns.
1851 *
1852 * Returns 0 when the timer has expired otherwise -EINTR
1853 */
1854int __sched schedule_hrtimeout(ktime_t *expires,
1855                   const enum hrtimer_mode mode)
1856{
1857    return schedule_hrtimeout_range(expires, 0, mode);
1858}
1859EXPORT_SYMBOL_GPL(schedule_hrtimeout);
1860

Archive Download this file



interactive