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 * Note: If we want to add new timer bases, we have to skip the two
57 * clock ids captured by the cpu-timers. We do this by holding empty
58 * entries rather than doing math adjustment of the clock ids.
59 * This ensures that we capture erroneous accesses to these clock ids
60 * rather than moving them into the range of valid clock id's.
61 */
62DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
63{
64
65    .clock_base =
66    {
67        {
68            .index = CLOCK_REALTIME,
69            .get_time = &ktime_get_real,
70            .resolution = KTIME_LOW_RES,
71        },
72        {
73            .index = CLOCK_MONOTONIC,
74            .get_time = &ktime_get,
75            .resolution = KTIME_LOW_RES,
76        },
77    }
78};
79
80/*
81 * Get the coarse grained time at the softirq based on xtime and
82 * wall_to_monotonic.
83 */
84static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
85{
86    ktime_t xtim, tomono;
87    struct timespec xts, tom;
88    unsigned long seq;
89
90    do {
91        seq = read_seqbegin(&xtime_lock);
92        xts = current_kernel_time();
93        tom = wall_to_monotonic;
94    } while (read_seqretry(&xtime_lock, seq));
95
96    xtim = timespec_to_ktime(xts);
97    tomono = timespec_to_ktime(tom);
98    base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
99    base->clock_base[CLOCK_MONOTONIC].softirq_time =
100        ktime_add(xtim, tomono);
101}
102
103/*
104 * Functions and macros which are different for UP/SMP systems are kept in a
105 * single place
106 */
107#ifdef CONFIG_SMP
108
109/*
110 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
111 * means that all timers which are tied to this base via timer->base are
112 * locked, and the base itself is locked too.
113 *
114 * So __run_timers/migrate_timers can safely modify all timers which could
115 * be found on the lists/queues.
116 *
117 * When the timer's base is locked, and the timer removed from list, it is
118 * possible to set timer->base = NULL and drop the lock: the timer remains
119 * locked.
120 */
121static
122struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
123                         unsigned long *flags)
124{
125    struct hrtimer_clock_base *base;
126
127    for (;;) {
128        base = timer->base;
129        if (likely(base != NULL)) {
130            spin_lock_irqsave(&base->cpu_base->lock, *flags);
131            if (likely(base == timer->base))
132                return base;
133            /* The timer has migrated to another CPU: */
134            spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
135        }
136        cpu_relax();
137    }
138}
139
140
141/*
142 * Get the preferred target CPU for NOHZ
143 */
144static int hrtimer_get_target(int this_cpu, int pinned)
145{
146#ifdef CONFIG_NO_HZ
147    if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu)) {
148        int preferred_cpu = get_nohz_load_balancer();
149
150        if (preferred_cpu >= 0)
151            return preferred_cpu;
152    }
153#endif
154    return this_cpu;
155}
156
157/*
158 * With HIGHRES=y we do not migrate the timer when it is expiring
159 * before the next event on the target cpu because we cannot reprogram
160 * the target cpu hardware and we would cause it to fire late.
161 *
162 * Called with cpu_base->lock of target cpu held.
163 */
164static int
165hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
166{
167#ifdef CONFIG_HIGH_RES_TIMERS
168    ktime_t expires;
169
170    if (!new_base->cpu_base->hres_active)
171        return 0;
172
173    expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
174    return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
175#else
176    return 0;
177#endif
178}
179
180/*
181 * Switch the timer base to the current CPU when possible.
182 */
183static inline struct hrtimer_clock_base *
184switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
185            int pinned)
186{
187    struct hrtimer_clock_base *new_base;
188    struct hrtimer_cpu_base *new_cpu_base;
189    int this_cpu = smp_processor_id();
190    int cpu = hrtimer_get_target(this_cpu, pinned);
191
192again:
193    new_cpu_base = &per_cpu(hrtimer_bases, cpu);
194    new_base = &new_cpu_base->clock_base[base->index];
195
196    if (base != new_base) {
197        /*
198         * We are trying to move timer to new_base.
199         * However we can't change timer's base while it is running,
200         * so we keep it on the same CPU. No hassle vs. reprogramming
201         * the event source in the high resolution case. The softirq
202         * code will take care of this when the timer function has
203         * completed. There is no conflict as we hold the lock until
204         * the timer is enqueued.
205         */
206        if (unlikely(hrtimer_callback_running(timer)))
207            return base;
208
209        /* See the comment in lock_timer_base() */
210        timer->base = NULL;
211        spin_unlock(&base->cpu_base->lock);
212        spin_lock(&new_base->cpu_base->lock);
213
214        if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
215            cpu = this_cpu;
216            spin_unlock(&new_base->cpu_base->lock);
217            spin_lock(&base->cpu_base->lock);
218            timer->base = base;
219            goto again;
220        }
221        timer->base = new_base;
222    }
223    return new_base;
224}
225
226#else /* CONFIG_SMP */
227
228static inline struct hrtimer_clock_base *
229lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
230{
231    struct hrtimer_clock_base *base = timer->base;
232
233    spin_lock_irqsave(&base->cpu_base->lock, *flags);
234
235    return base;
236}
237
238# define switch_hrtimer_base(t, b, p) (b)
239
240#endif /* !CONFIG_SMP */
241
242/*
243 * Functions for the union type storage format of ktime_t which are
244 * too large for inlining:
245 */
246#if BITS_PER_LONG < 64
247# ifndef CONFIG_KTIME_SCALAR
248/**
249 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
250 * @kt: addend
251 * @nsec: the scalar nsec value to add
252 *
253 * Returns the sum of kt and nsec in ktime_t format
254 */
255ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
256{
257    ktime_t tmp;
258
259    if (likely(nsec < NSEC_PER_SEC)) {
260        tmp.tv64 = nsec;
261    } else {
262        unsigned long rem = do_div(nsec, NSEC_PER_SEC);
263
264        tmp = ktime_set((long)nsec, rem);
265    }
266
267    return ktime_add(kt, tmp);
268}
269
270EXPORT_SYMBOL_GPL(ktime_add_ns);
271
272/**
273 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
274 * @kt: minuend
275 * @nsec: the scalar nsec value to subtract
276 *
277 * Returns the subtraction of @nsec from @kt in ktime_t format
278 */
279ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
280{
281    ktime_t tmp;
282
283    if (likely(nsec < NSEC_PER_SEC)) {
284        tmp.tv64 = nsec;
285    } else {
286        unsigned long rem = do_div(nsec, NSEC_PER_SEC);
287
288        tmp = ktime_set((long)nsec, rem);
289    }
290
291    return ktime_sub(kt, tmp);
292}
293
294EXPORT_SYMBOL_GPL(ktime_sub_ns);
295# endif /* !CONFIG_KTIME_SCALAR */
296
297/*
298 * Divide a ktime value by a nanosecond value
299 */
300u64 ktime_divns(const ktime_t kt, s64 div)
301{
302    u64 dclc;
303    int sft = 0;
304
305    dclc = ktime_to_ns(kt);
306    /* Make sure the divisor is less than 2^32: */
307    while (div >> 32) {
308        sft++;
309        div >>= 1;
310    }
311    dclc >>= sft;
312    do_div(dclc, (unsigned long) div);
313
314    return dclc;
315}
316#endif /* BITS_PER_LONG >= 64 */
317
318/*
319 * Add two ktime values and do a safety check for overflow:
320 */
321ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
322{
323    ktime_t res = ktime_add(lhs, rhs);
324
325    /*
326     * We use KTIME_SEC_MAX here, the maximum timeout which we can
327     * return to user space in a timespec:
328     */
329    if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
330        res = ktime_set(KTIME_SEC_MAX, 0);
331
332    return res;
333}
334
335EXPORT_SYMBOL_GPL(ktime_add_safe);
336
337#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
338
339static struct debug_obj_descr hrtimer_debug_descr;
340
341/*
342 * fixup_init is called when:
343 * - an active object is initialized
344 */
345static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
346{
347    struct hrtimer *timer = addr;
348
349    switch (state) {
350    case ODEBUG_STATE_ACTIVE:
351        hrtimer_cancel(timer);
352        debug_object_init(timer, &hrtimer_debug_descr);
353        return 1;
354    default:
355        return 0;
356    }
357}
358
359/*
360 * fixup_activate is called when:
361 * - an active object is activated
362 * - an unknown object is activated (might be a statically initialized object)
363 */
364static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
365{
366    switch (state) {
367
368    case ODEBUG_STATE_NOTAVAILABLE:
369        WARN_ON_ONCE(1);
370        return 0;
371
372    case ODEBUG_STATE_ACTIVE:
373        WARN_ON(1);
374
375    default:
376        return 0;
377    }
378}
379
380/*
381 * fixup_free is called when:
382 * - an active object is freed
383 */
384static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
385{
386    struct hrtimer *timer = addr;
387
388    switch (state) {
389    case ODEBUG_STATE_ACTIVE:
390        hrtimer_cancel(timer);
391        debug_object_free(timer, &hrtimer_debug_descr);
392        return 1;
393    default:
394        return 0;
395    }
396}
397
398static struct debug_obj_descr hrtimer_debug_descr = {
399    .name = "hrtimer",
400    .fixup_init = hrtimer_fixup_init,
401    .fixup_activate = hrtimer_fixup_activate,
402    .fixup_free = hrtimer_fixup_free,
403};
404
405static inline void debug_hrtimer_init(struct hrtimer *timer)
406{
407    debug_object_init(timer, &hrtimer_debug_descr);
408}
409
410static inline void debug_hrtimer_activate(struct hrtimer *timer)
411{
412    debug_object_activate(timer, &hrtimer_debug_descr);
413}
414
415static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
416{
417    debug_object_deactivate(timer, &hrtimer_debug_descr);
418}
419
420static inline void debug_hrtimer_free(struct hrtimer *timer)
421{
422    debug_object_free(timer, &hrtimer_debug_descr);
423}
424
425static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
426               enum hrtimer_mode mode);
427
428void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
429               enum hrtimer_mode mode)
430{
431    debug_object_init_on_stack(timer, &hrtimer_debug_descr);
432    __hrtimer_init(timer, clock_id, mode);
433}
434EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
435
436void destroy_hrtimer_on_stack(struct hrtimer *timer)
437{
438    debug_object_free(timer, &hrtimer_debug_descr);
439}
440
441#else
442static inline void debug_hrtimer_init(struct hrtimer *timer) { }
443static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
444static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
445#endif
446
447static inline void
448debug_init(struct hrtimer *timer, clockid_t clockid,
449       enum hrtimer_mode mode)
450{
451    debug_hrtimer_init(timer);
452    trace_hrtimer_init(timer, clockid, mode);
453}
454
455static inline void debug_activate(struct hrtimer *timer)
456{
457    debug_hrtimer_activate(timer);
458    trace_hrtimer_start(timer);
459}
460
461static inline void debug_deactivate(struct hrtimer *timer)
462{
463    debug_hrtimer_deactivate(timer);
464    trace_hrtimer_cancel(timer);
465}
466
467/* High resolution timer related functions */
468#ifdef CONFIG_HIGH_RES_TIMERS
469
470/*
471 * High resolution timer enabled ?
472 */
473static int hrtimer_hres_enabled __read_mostly = 1;
474
475/*
476 * Enable / Disable high resolution mode
477 */
478static int __init setup_hrtimer_hres(char *str)
479{
480    if (!strcmp(str, "off"))
481        hrtimer_hres_enabled = 0;
482    else if (!strcmp(str, "on"))
483        hrtimer_hres_enabled = 1;
484    else
485        return 0;
486    return 1;
487}
488
489__setup("highres=", setup_hrtimer_hres);
490
491/*
492 * hrtimer_high_res_enabled - query, if the highres mode is enabled
493 */
494static inline int hrtimer_is_hres_enabled(void)
495{
496    return hrtimer_hres_enabled;
497}
498
499/*
500 * Is the high resolution mode active ?
501 */
502static inline int hrtimer_hres_active(void)
503{
504    return __get_cpu_var(hrtimer_bases).hres_active;
505}
506
507/*
508 * Reprogram the event source with checking both queues for the
509 * next event
510 * Called with interrupts disabled and base->lock held
511 */
512static void
513hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
514{
515    int i;
516    struct hrtimer_clock_base *base = cpu_base->clock_base;
517    ktime_t expires, expires_next;
518
519    expires_next.tv64 = KTIME_MAX;
520
521    for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
522        struct hrtimer *timer;
523
524        if (!base->first)
525            continue;
526        timer = rb_entry(base->first, struct hrtimer, node);
527        expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
528        /*
529         * clock_was_set() has changed base->offset so the
530         * result might be negative. Fix it up to prevent a
531         * false positive in clockevents_program_event()
532         */
533        if (expires.tv64 < 0)
534            expires.tv64 = 0;
535        if (expires.tv64 < expires_next.tv64)
536            expires_next = expires;
537    }
538
539    if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
540        return;
541
542    cpu_base->expires_next.tv64 = expires_next.tv64;
543
544    if (cpu_base->expires_next.tv64 != KTIME_MAX)
545        tick_program_event(cpu_base->expires_next, 1);
546}
547
548/*
549 * Shared reprogramming for clock_realtime and clock_monotonic
550 *
551 * When a timer is enqueued and expires earlier than the already enqueued
552 * timers, we have to check, whether it expires earlier than the timer for
553 * which the clock event device was armed.
554 *
555 * Called with interrupts disabled and base->cpu_base.lock held
556 */
557static int hrtimer_reprogram(struct hrtimer *timer,
558                 struct hrtimer_clock_base *base)
559{
560    ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next;
561    ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
562    int res;
563
564    WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
565
566    /*
567     * When the callback is running, we do not reprogram the clock event
568     * device. The timer callback is either running on a different CPU or
569     * the callback is executed in the hrtimer_interrupt context. The
570     * reprogramming is handled either by the softirq, which called the
571     * callback or at the end of the hrtimer_interrupt.
572     */
573    if (hrtimer_callback_running(timer))
574        return 0;
575
576    /*
577     * CLOCK_REALTIME timer might be requested with an absolute
578     * expiry time which is less than base->offset. Nothing wrong
579     * about that, just avoid to call into the tick code, which
580     * has now objections against negative expiry values.
581     */
582    if (expires.tv64 < 0)
583        return -ETIME;
584
585    if (expires.tv64 >= expires_next->tv64)
586        return 0;
587
588    /*
589     * Clockevents returns -ETIME, when the event was in the past.
590     */
591    res = tick_program_event(expires, 0);
592    if (!IS_ERR_VALUE(res))
593        *expires_next = expires;
594    return res;
595}
596
597
598/*
599 * Retrigger next event is called after clock was set
600 *
601 * Called with interrupts disabled via on_each_cpu()
602 */
603static void retrigger_next_event(void *arg)
604{
605    struct hrtimer_cpu_base *base;
606    struct timespec realtime_offset;
607    unsigned long seq;
608
609    if (!hrtimer_hres_active())
610        return;
611
612    do {
613        seq = read_seqbegin(&xtime_lock);
614        set_normalized_timespec(&realtime_offset,
615                    -wall_to_monotonic.tv_sec,
616                    -wall_to_monotonic.tv_nsec);
617    } while (read_seqretry(&xtime_lock, seq));
618
619    base = &__get_cpu_var(hrtimer_bases);
620
621    /* Adjust CLOCK_REALTIME offset */
622    spin_lock(&base->lock);
623    base->clock_base[CLOCK_REALTIME].offset =
624        timespec_to_ktime(realtime_offset);
625
626    hrtimer_force_reprogram(base, 0);
627    spin_unlock(&base->lock);
628}
629
630/*
631 * Clock realtime was set
632 *
633 * Change the offset of the realtime clock vs. the monotonic
634 * clock.
635 *
636 * We might have to reprogram the high resolution timer interrupt. On
637 * SMP we call the architecture specific code to retrigger _all_ high
638 * resolution timer interrupts. On UP we just disable interrupts and
639 * call the high resolution interrupt code.
640 */
641void clock_was_set(void)
642{
643    /* Retrigger the CPU local events everywhere */
644    on_each_cpu(retrigger_next_event, NULL, 1);
645}
646
647/*
648 * During resume we might have to reprogram the high resolution timer
649 * interrupt (on the local CPU):
650 */
651void hres_timers_resume(void)
652{
653    WARN_ONCE(!irqs_disabled(),
654          KERN_INFO "hres_timers_resume() called with IRQs enabled!");
655
656    retrigger_next_event(NULL);
657}
658
659/*
660 * Initialize the high resolution related parts of cpu_base
661 */
662static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
663{
664    base->expires_next.tv64 = KTIME_MAX;
665    base->hres_active = 0;
666}
667
668/*
669 * Initialize the high resolution related parts of a hrtimer
670 */
671static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
672{
673}
674
675
676/*
677 * When High resolution timers are active, try to reprogram. Note, that in case
678 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
679 * check happens. The timer gets enqueued into the rbtree. The reprogramming
680 * and expiry check is done in the hrtimer_interrupt or in the softirq.
681 */
682static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
683                        struct hrtimer_clock_base *base,
684                        int wakeup)
685{
686    if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
687        if (wakeup) {
688            spin_unlock(&base->cpu_base->lock);
689            raise_softirq_irqoff(HRTIMER_SOFTIRQ);
690            spin_lock(&base->cpu_base->lock);
691        } else
692            __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
693
694        return 1;
695    }
696
697    return 0;
698}
699
700/*
701 * Switch to high resolution mode
702 */
703static int hrtimer_switch_to_hres(void)
704{
705    int cpu = smp_processor_id();
706    struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
707    unsigned long flags;
708
709    if (base->hres_active)
710        return 1;
711
712    local_irq_save(flags);
713
714    if (tick_init_highres()) {
715        local_irq_restore(flags);
716        printk(KERN_WARNING "Could not switch to high resolution "
717                    "mode on CPU %d\n", cpu);
718        return 0;
719    }
720    base->hres_active = 1;
721    base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
722    base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES;
723
724    tick_setup_sched_timer();
725
726    /* "Retrigger" the interrupt to get things going */
727    retrigger_next_event(NULL);
728    local_irq_restore(flags);
729    return 1;
730}
731
732#else
733
734static inline int hrtimer_hres_active(void) { return 0; }
735static inline int hrtimer_is_hres_enabled(void) { return 0; }
736static inline int hrtimer_switch_to_hres(void) { return 0; }
737static inline void
738hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
739static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
740                        struct hrtimer_clock_base *base,
741                        int wakeup)
742{
743    return 0;
744}
745static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
746static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
747
748#endif /* CONFIG_HIGH_RES_TIMERS */
749
750#ifdef CONFIG_TIMER_STATS
751void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr)
752{
753    if (timer->start_site)
754        return;
755
756    timer->start_site = addr;
757    memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
758    timer->start_pid = current->pid;
759}
760#endif
761
762/*
763 * Counterpart to lock_hrtimer_base above:
764 */
765static inline
766void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
767{
768    spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
769}
770
771/**
772 * hrtimer_forward - forward the timer expiry
773 * @timer: hrtimer to forward
774 * @now: forward past this time
775 * @interval: the interval to forward
776 *
777 * Forward the timer expiry so it will expire in the future.
778 * Returns the number of overruns.
779 */
780u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
781{
782    u64 orun = 1;
783    ktime_t delta;
784
785    delta = ktime_sub(now, hrtimer_get_expires(timer));
786
787    if (delta.tv64 < 0)
788        return 0;
789
790    if (interval.tv64 < timer->base->resolution.tv64)
791        interval.tv64 = timer->base->resolution.tv64;
792
793    if (unlikely(delta.tv64 >= interval.tv64)) {
794        s64 incr = ktime_to_ns(interval);
795
796        orun = ktime_divns(delta, incr);
797        hrtimer_add_expires_ns(timer, incr * orun);
798        if (hrtimer_get_expires_tv64(timer) > now.tv64)
799            return orun;
800        /*
801         * This (and the ktime_add() below) is the
802         * correction for exact:
803         */
804        orun++;
805    }
806    hrtimer_add_expires(timer, interval);
807
808    return orun;
809}
810EXPORT_SYMBOL_GPL(hrtimer_forward);
811
812/*
813 * enqueue_hrtimer - internal function to (re)start a timer
814 *
815 * The timer is inserted in expiry order. Insertion into the
816 * red black tree is O(log(n)). Must hold the base lock.
817 *
818 * Returns 1 when the new timer is the leftmost timer in the tree.
819 */
820static int enqueue_hrtimer(struct hrtimer *timer,
821               struct hrtimer_clock_base *base)
822{
823    struct rb_node **link = &base->active.rb_node;
824    struct rb_node *parent = NULL;
825    struct hrtimer *entry;
826    int leftmost = 1;
827
828    debug_activate(timer);
829
830    /*
831     * Find the right place in the rbtree:
832     */
833    while (*link) {
834        parent = *link;
835        entry = rb_entry(parent, struct hrtimer, node);
836        /*
837         * We dont care about collisions. Nodes with
838         * the same expiry time stay together.
839         */
840        if (hrtimer_get_expires_tv64(timer) <
841                hrtimer_get_expires_tv64(entry)) {
842            link = &(*link)->rb_left;
843        } else {
844            link = &(*link)->rb_right;
845            leftmost = 0;
846        }
847    }
848
849    /*
850     * Insert the timer to the rbtree and check whether it
851     * replaces the first pending timer
852     */
853    if (leftmost)
854        base->first = &timer->node;
855
856    rb_link_node(&timer->node, parent, link);
857    rb_insert_color(&timer->node, &base->active);
858    /*
859     * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
860     * state of a possibly running callback.
861     */
862    timer->state |= HRTIMER_STATE_ENQUEUED;
863
864    return leftmost;
865}
866
867/*
868 * __remove_hrtimer - internal function to remove a timer
869 *
870 * Caller must hold the base lock.
871 *
872 * High resolution timer mode reprograms the clock event device when the
873 * timer is the one which expires next. The caller can disable this by setting
874 * reprogram to zero. This is useful, when the context does a reprogramming
875 * anyway (e.g. timer interrupt)
876 */
877static void __remove_hrtimer(struct hrtimer *timer,
878                 struct hrtimer_clock_base *base,
879                 unsigned long newstate, int reprogram)
880{
881    if (!(timer->state & HRTIMER_STATE_ENQUEUED))
882        goto out;
883
884    /*
885     * Remove the timer from the rbtree and replace the first
886     * entry pointer if necessary.
887     */
888    if (base->first == &timer->node) {
889        base->first = rb_next(&timer->node);
890#ifdef CONFIG_HIGH_RES_TIMERS
891        /* Reprogram the clock event device. if enabled */
892        if (reprogram && hrtimer_hres_active()) {
893            ktime_t expires;
894
895            expires = ktime_sub(hrtimer_get_expires(timer),
896                        base->offset);
897            if (base->cpu_base->expires_next.tv64 == expires.tv64)
898                hrtimer_force_reprogram(base->cpu_base, 1);
899        }
900#endif
901    }
902    rb_erase(&timer->node, &base->active);
903out:
904    timer->state = newstate;
905}
906
907/*
908 * remove hrtimer, called with base lock held
909 */
910static inline int
911remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
912{
913    if (hrtimer_is_queued(timer)) {
914        int reprogram;
915
916        /*
917         * Remove the timer and force reprogramming when high
918         * resolution mode is active and the timer is on the current
919         * CPU. If we remove a timer on another CPU, reprogramming is
920         * skipped. The interrupt event on this CPU is fired and
921         * reprogramming happens in the interrupt handler. This is a
922         * rare case and less expensive than a smp call.
923         */
924        debug_deactivate(timer);
925        timer_stats_hrtimer_clear_start_info(timer);
926        reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
927        __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
928                 reprogram);
929        return 1;
930    }
931    return 0;
932}
933
934int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
935        unsigned long delta_ns, const enum hrtimer_mode mode,
936        int wakeup)
937{
938    struct hrtimer_clock_base *base, *new_base;
939    unsigned long flags;
940    int ret, leftmost;
941
942    base = lock_hrtimer_base(timer, &flags);
943
944    /* Remove an active timer from the queue: */
945    ret = remove_hrtimer(timer, base);
946
947    /* Switch the timer base, if necessary: */
948    new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
949
950    if (mode & HRTIMER_MODE_REL) {
951        tim = ktime_add_safe(tim, new_base->get_time());
952        /*
953         * CONFIG_TIME_LOW_RES is a temporary way for architectures
954         * to signal that they simply return xtime in
955         * do_gettimeoffset(). In this case we want to round up by
956         * resolution when starting a relative timer, to avoid short
957         * timeouts. This will go away with the GTOD framework.
958         */
959#ifdef CONFIG_TIME_LOW_RES
960        tim = ktime_add_safe(tim, base->resolution);
961#endif
962    }
963
964    hrtimer_set_expires_range_ns(timer, tim, delta_ns);
965
966    timer_stats_hrtimer_set_start_info(timer);
967
968    leftmost = enqueue_hrtimer(timer, new_base);
969
970    /*
971     * Only allow reprogramming if the new base is on this CPU.
972     * (it might still be on another CPU if the timer was pending)
973     *
974     * XXX send_remote_softirq() ?
975     */
976    if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases))
977        hrtimer_enqueue_reprogram(timer, new_base, wakeup);
978
979    unlock_hrtimer_base(timer, &flags);
980
981    return ret;
982}
983
984/**
985 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
986 * @timer: the timer to be added
987 * @tim: expiry time
988 * @delta_ns: "slack" range for the timer
989 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
990 *
991 * Returns:
992 * 0 on success
993 * 1 when the timer was active
994 */
995int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
996        unsigned long delta_ns, const enum hrtimer_mode mode)
997{
998    return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
999}
1000EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1001
1002/**
1003 * hrtimer_start - (re)start an hrtimer on the current CPU
1004 * @timer: the timer to be added
1005 * @tim: expiry time
1006 * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1007 *
1008 * Returns:
1009 * 0 on success
1010 * 1 when the timer was active
1011 */
1012int
1013hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1014{
1015    return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1016}
1017EXPORT_SYMBOL_GPL(hrtimer_start);
1018
1019
1020/**
1021 * hrtimer_try_to_cancel - try to deactivate a timer
1022 * @timer: hrtimer to stop
1023 *
1024 * Returns:
1025 * 0 when the timer was not active
1026 * 1 when the timer was active
1027 * -1 when the timer is currently excuting the callback function and
1028 * cannot be stopped
1029 */
1030int hrtimer_try_to_cancel(struct hrtimer *timer)
1031{
1032    struct hrtimer_clock_base *base;
1033    unsigned long flags;
1034    int ret = -1;
1035
1036    base = lock_hrtimer_base(timer, &flags);
1037
1038    if (!hrtimer_callback_running(timer))
1039        ret = remove_hrtimer(timer, base);
1040
1041    unlock_hrtimer_base(timer, &flags);
1042
1043    return ret;
1044
1045}
1046EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1047
1048/**
1049 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1050 * @timer: the timer to be cancelled
1051 *
1052 * Returns:
1053 * 0 when the timer was not active
1054 * 1 when the timer was active
1055 */
1056int hrtimer_cancel(struct hrtimer *timer)
1057{
1058    for (;;) {
1059        int ret = hrtimer_try_to_cancel(timer);
1060
1061        if (ret >= 0)
1062            return ret;
1063        cpu_relax();
1064    }
1065}
1066EXPORT_SYMBOL_GPL(hrtimer_cancel);
1067
1068/**
1069 * hrtimer_get_remaining - get remaining time for the timer
1070 * @timer: the timer to read
1071 */
1072ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1073{
1074    struct hrtimer_clock_base *base;
1075    unsigned long flags;
1076    ktime_t rem;
1077
1078    base = lock_hrtimer_base(timer, &flags);
1079    rem = hrtimer_expires_remaining(timer);
1080    unlock_hrtimer_base(timer, &flags);
1081
1082    return rem;
1083}
1084EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1085
1086#ifdef CONFIG_NO_HZ
1087/**
1088 * hrtimer_get_next_event - get the time until next expiry event
1089 *
1090 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1091 * is pending.
1092 */
1093ktime_t hrtimer_get_next_event(void)
1094{
1095    struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1096    struct hrtimer_clock_base *base = cpu_base->clock_base;
1097    ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1098    unsigned long flags;
1099    int i;
1100
1101    spin_lock_irqsave(&cpu_base->lock, flags);
1102
1103    if (!hrtimer_hres_active()) {
1104        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1105            struct hrtimer *timer;
1106
1107            if (!base->first)
1108                continue;
1109
1110            timer = rb_entry(base->first, struct hrtimer, node);
1111            delta.tv64 = hrtimer_get_expires_tv64(timer);
1112            delta = ktime_sub(delta, base->get_time());
1113            if (delta.tv64 < mindelta.tv64)
1114                mindelta.tv64 = delta.tv64;
1115        }
1116    }
1117
1118    spin_unlock_irqrestore(&cpu_base->lock, flags);
1119
1120    if (mindelta.tv64 < 0)
1121        mindelta.tv64 = 0;
1122    return mindelta;
1123}
1124#endif
1125
1126static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1127               enum hrtimer_mode mode)
1128{
1129    struct hrtimer_cpu_base *cpu_base;
1130
1131    memset(timer, 0, sizeof(struct hrtimer));
1132
1133    cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1134
1135    if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1136        clock_id = CLOCK_MONOTONIC;
1137
1138    timer->base = &cpu_base->clock_base[clock_id];
1139    hrtimer_init_timer_hres(timer);
1140
1141#ifdef CONFIG_TIMER_STATS
1142    timer->start_site = NULL;
1143    timer->start_pid = -1;
1144    memset(timer->start_comm, 0, TASK_COMM_LEN);
1145#endif
1146}
1147
1148/**
1149 * hrtimer_init - initialize a timer to the given clock
1150 * @timer: the timer to be initialized
1151 * @clock_id: the clock to be used
1152 * @mode: timer mode abs/rel
1153 */
1154void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1155          enum hrtimer_mode mode)
1156{
1157    debug_init(timer, clock_id, mode);
1158    __hrtimer_init(timer, clock_id, mode);
1159}
1160EXPORT_SYMBOL_GPL(hrtimer_init);
1161
1162/**
1163 * hrtimer_get_res - get the timer resolution for a clock
1164 * @which_clock: which clock to query
1165 * @tp: pointer to timespec variable to store the resolution
1166 *
1167 * Store the resolution of the clock selected by @which_clock in the
1168 * variable pointed to by @tp.
1169 */
1170int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1171{
1172    struct hrtimer_cpu_base *cpu_base;
1173
1174    cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1175    *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution);
1176
1177    return 0;
1178}
1179EXPORT_SYMBOL_GPL(hrtimer_get_res);
1180
1181static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1182{
1183    struct hrtimer_clock_base *base = timer->base;
1184    struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1185    enum hrtimer_restart (*fn)(struct hrtimer *);
1186    int restart;
1187
1188    WARN_ON(!irqs_disabled());
1189
1190    debug_deactivate(timer);
1191    __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1192    timer_stats_account_hrtimer(timer);
1193    fn = timer->function;
1194
1195    /*
1196     * Because we run timers from hardirq context, there is no chance
1197     * they get migrated to another cpu, therefore its safe to unlock
1198     * the timer base.
1199     */
1200    spin_unlock(&cpu_base->lock);
1201    trace_hrtimer_expire_entry(timer, now);
1202    restart = fn(timer);
1203    trace_hrtimer_expire_exit(timer);
1204    spin_lock(&cpu_base->lock);
1205
1206    /*
1207     * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1208     * we do not reprogramm the event hardware. Happens either in
1209     * hrtimer_start_range_ns() or in hrtimer_interrupt()
1210     */
1211    if (restart != HRTIMER_NORESTART) {
1212        BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1213        enqueue_hrtimer(timer, base);
1214    }
1215    timer->state &= ~HRTIMER_STATE_CALLBACK;
1216}
1217
1218#ifdef CONFIG_HIGH_RES_TIMERS
1219
1220static int force_clock_reprogram;
1221
1222/*
1223 * After 5 iteration's attempts, we consider that hrtimer_interrupt()
1224 * is hanging, which could happen with something that slows the interrupt
1225 * such as the tracing. Then we force the clock reprogramming for each future
1226 * hrtimer interrupts to avoid infinite loops and use the min_delta_ns
1227 * threshold that we will overwrite.
1228 * The next tick event will be scheduled to 3 times we currently spend on
1229 * hrtimer_interrupt(). This gives a good compromise, the cpus will spend
1230 * 1/4 of their time to process the hrtimer interrupts. This is enough to
1231 * let it running without serious starvation.
1232 */
1233
1234static inline void
1235hrtimer_interrupt_hanging(struct clock_event_device *dev,
1236            ktime_t try_time)
1237{
1238    force_clock_reprogram = 1;
1239    dev->min_delta_ns = (unsigned long)try_time.tv64 * 3;
1240    printk(KERN_WARNING "hrtimer: interrupt too slow, "
1241        "forcing clock min delta to %lu ns\n", dev->min_delta_ns);
1242}
1243/*
1244 * High resolution timer interrupt
1245 * Called with interrupts disabled
1246 */
1247void hrtimer_interrupt(struct clock_event_device *dev)
1248{
1249    struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1250    struct hrtimer_clock_base *base;
1251    ktime_t expires_next, now;
1252    int nr_retries = 0;
1253    int i;
1254
1255    BUG_ON(!cpu_base->hres_active);
1256    cpu_base->nr_events++;
1257    dev->next_event.tv64 = KTIME_MAX;
1258
1259 retry:
1260    /* 5 retries is enough to notice a hang */
1261    if (!(++nr_retries % 5))
1262        hrtimer_interrupt_hanging(dev, ktime_sub(ktime_get(), now));
1263
1264    now = ktime_get();
1265
1266    expires_next.tv64 = KTIME_MAX;
1267
1268    spin_lock(&cpu_base->lock);
1269    /*
1270     * We set expires_next to KTIME_MAX here with cpu_base->lock
1271     * held to prevent that a timer is enqueued in our queue via
1272     * the migration code. This does not affect enqueueing of
1273     * timers which run their callback and need to be requeued on
1274     * this CPU.
1275     */
1276    cpu_base->expires_next.tv64 = KTIME_MAX;
1277
1278    base = cpu_base->clock_base;
1279
1280    for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1281        ktime_t basenow;
1282        struct rb_node *node;
1283
1284        basenow = ktime_add(now, base->offset);
1285
1286        while ((node = base->first)) {
1287            struct hrtimer *timer;
1288
1289            timer = rb_entry(node, struct hrtimer, node);
1290
1291            /*
1292             * The immediate goal for using the softexpires is
1293             * minimizing wakeups, not running timers at the
1294             * earliest interrupt after their soft expiration.
1295             * This allows us to avoid using a Priority Search
1296             * Tree, which can answer a stabbing querry for
1297             * overlapping intervals and instead use the simple
1298             * BST we already have.
1299             * We don't add extra wakeups by delaying timers that
1300             * are right-of a not yet expired timer, because that
1301             * timer will have to trigger a wakeup anyway.
1302             */
1303
1304            if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1305                ktime_t expires;
1306
1307                expires = ktime_sub(hrtimer_get_expires(timer),
1308                            base->offset);
1309                if (expires.tv64 < expires_next.tv64)
1310                    expires_next = expires;
1311                break;
1312            }
1313
1314            __run_hrtimer(timer, &basenow);
1315        }
1316        base++;
1317    }
1318
1319    /*
1320     * Store the new expiry value so the migration code can verify
1321     * against it.
1322     */
1323    cpu_base->expires_next = expires_next;
1324    spin_unlock(&cpu_base->lock);
1325
1326    /* Reprogramming necessary ? */
1327    if (expires_next.tv64 != KTIME_MAX) {
1328        if (tick_program_event(expires_next, force_clock_reprogram))
1329            goto retry;
1330    }
1331}
1332
1333/*
1334 * local version of hrtimer_peek_ahead_timers() called with interrupts
1335 * disabled.
1336 */
1337static void __hrtimer_peek_ahead_timers(void)
1338{
1339    struct tick_device *td;
1340
1341    if (!hrtimer_hres_active())
1342        return;
1343
1344    td = &__get_cpu_var(tick_cpu_device);
1345    if (td && td->evtdev)
1346        hrtimer_interrupt(td->evtdev);
1347}
1348
1349/**
1350 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1351 *
1352 * hrtimer_peek_ahead_timers will peek at the timer queue of
1353 * the current cpu and check if there are any timers for which
1354 * the soft expires time has passed. If any such timers exist,
1355 * they are run immediately and then removed from the timer queue.
1356 *
1357 */
1358void hrtimer_peek_ahead_timers(void)
1359{
1360    unsigned long flags;
1361
1362    local_irq_save(flags);
1363    __hrtimer_peek_ahead_timers();
1364    local_irq_restore(flags);
1365}
1366
1367static void run_hrtimer_softirq(struct softirq_action *h)
1368{
1369    hrtimer_peek_ahead_timers();
1370}
1371
1372#else /* CONFIG_HIGH_RES_TIMERS */
1373
1374static inline void __hrtimer_peek_ahead_timers(void) { }
1375
1376#endif /* !CONFIG_HIGH_RES_TIMERS */
1377
1378/*
1379 * Called from timer softirq every jiffy, expire hrtimers:
1380 *
1381 * For HRT its the fall back code to run the softirq in the timer
1382 * softirq context in case the hrtimer initialization failed or has
1383 * not been done yet.
1384 */
1385void hrtimer_run_pending(void)
1386{
1387    if (hrtimer_hres_active())
1388        return;
1389
1390    /*
1391     * This _is_ ugly: We have to check in the softirq context,
1392     * whether we can switch to highres and / or nohz mode. The
1393     * clocksource switch happens in the timer interrupt with
1394     * xtime_lock held. Notification from there only sets the
1395     * check bit in the tick_oneshot code, otherwise we might
1396     * deadlock vs. xtime_lock.
1397     */
1398    if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1399        hrtimer_switch_to_hres();
1400}
1401
1402/*
1403 * Called from hardirq context every jiffy
1404 */
1405void hrtimer_run_queues(void)
1406{
1407    struct rb_node *node;
1408    struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1409    struct hrtimer_clock_base *base;
1410    int index, gettime = 1;
1411
1412    if (hrtimer_hres_active())
1413        return;
1414
1415    for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1416        base = &cpu_base->clock_base[index];
1417
1418        if (!base->first)
1419            continue;
1420
1421        if (gettime) {
1422            hrtimer_get_softirq_time(cpu_base);
1423            gettime = 0;
1424        }
1425
1426        spin_lock(&cpu_base->lock);
1427
1428        while ((node = base->first)) {
1429            struct hrtimer *timer;
1430
1431            timer = rb_entry(node, struct hrtimer, node);
1432            if (base->softirq_time.tv64 <=
1433                    hrtimer_get_expires_tv64(timer))
1434                break;
1435
1436            __run_hrtimer(timer, &base->softirq_time);
1437        }
1438        spin_unlock(&cpu_base->lock);
1439    }
1440}
1441
1442/*
1443 * Sleep related functions:
1444 */
1445static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1446{
1447    struct hrtimer_sleeper *t =
1448        container_of(timer, struct hrtimer_sleeper, timer);
1449    struct task_struct *task = t->task;
1450
1451    t->task = NULL;
1452    if (task)
1453        wake_up_process(task);
1454
1455    return HRTIMER_NORESTART;
1456}
1457
1458void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1459{
1460    sl->timer.function = hrtimer_wakeup;
1461    sl->task = task;
1462}
1463EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1464
1465static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1466{
1467    hrtimer_init_sleeper(t, current);
1468
1469    do {
1470        set_current_state(TASK_INTERRUPTIBLE);
1471        hrtimer_start_expires(&t->timer, mode);
1472        if (!hrtimer_active(&t->timer))
1473            t->task = NULL;
1474
1475        if (likely(t->task))
1476            schedule();
1477
1478        hrtimer_cancel(&t->timer);
1479        mode = HRTIMER_MODE_ABS;
1480
1481    } while (t->task && !signal_pending(current));
1482
1483    __set_current_state(TASK_RUNNING);
1484
1485    return t->task == NULL;
1486}
1487
1488static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1489{
1490    struct timespec rmt;
1491    ktime_t rem;
1492
1493    rem = hrtimer_expires_remaining(timer);
1494    if (rem.tv64 <= 0)
1495        return 0;
1496    rmt = ktime_to_timespec(rem);
1497
1498    if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1499        return -EFAULT;
1500
1501    return 1;
1502}
1503
1504long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1505{
1506    struct hrtimer_sleeper t;
1507    struct timespec __user *rmtp;
1508    int ret = 0;
1509
1510    hrtimer_init_on_stack(&t.timer, restart->nanosleep.index,
1511                HRTIMER_MODE_ABS);
1512    hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1513
1514    if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1515        goto out;
1516
1517    rmtp = restart->nanosleep.rmtp;
1518    if (rmtp) {
1519        ret = update_rmtp(&t.timer, rmtp);
1520        if (ret <= 0)
1521            goto out;
1522    }
1523
1524    /* The other values in restart are already filled in */
1525    ret = -ERESTART_RESTARTBLOCK;
1526out:
1527    destroy_hrtimer_on_stack(&t.timer);
1528    return ret;
1529}
1530
1531long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1532               const enum hrtimer_mode mode, const clockid_t clockid)
1533{
1534    struct restart_block *restart;
1535    struct hrtimer_sleeper t;
1536    int ret = 0;
1537    unsigned long slack;
1538
1539    slack = current->timer_slack_ns;
1540    if (rt_task(current))
1541        slack = 0;
1542
1543    hrtimer_init_on_stack(&t.timer, clockid, mode);
1544    hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1545    if (do_nanosleep(&t, mode))
1546        goto out;
1547
1548    /* Absolute timers do not update the rmtp value and restart: */
1549    if (mode == HRTIMER_MODE_ABS) {
1550        ret = -ERESTARTNOHAND;
1551        goto out;
1552    }
1553
1554    if (rmtp) {
1555        ret = update_rmtp(&t.timer, rmtp);
1556        if (ret <= 0)
1557            goto out;
1558    }
1559
1560    restart = &current_thread_info()->restart_block;
1561    restart->fn = hrtimer_nanosleep_restart;
1562    restart->nanosleep.index = t.timer.base->index;
1563    restart->nanosleep.rmtp = rmtp;
1564    restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1565
1566    ret = -ERESTART_RESTARTBLOCK;
1567out:
1568    destroy_hrtimer_on_stack(&t.timer);
1569    return ret;
1570}
1571
1572SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1573        struct timespec __user *, rmtp)
1574{
1575    struct timespec tu;
1576
1577    if (copy_from_user(&tu, rqtp, sizeof(tu)))
1578        return -EFAULT;
1579
1580    if (!timespec_valid(&tu))
1581        return -EINVAL;
1582
1583    return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1584}
1585
1586/*
1587 * Functions related to boot-time initialization:
1588 */
1589static void __cpuinit init_hrtimers_cpu(int cpu)
1590{
1591    struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1592    int i;
1593
1594    spin_lock_init(&cpu_base->lock);
1595
1596    for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
1597        cpu_base->clock_base[i].cpu_base = cpu_base;
1598
1599    hrtimer_init_hres(cpu_base);
1600}
1601
1602#ifdef CONFIG_HOTPLUG_CPU
1603
1604static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1605                struct hrtimer_clock_base *new_base)
1606{
1607    struct hrtimer *timer;
1608    struct rb_node *node;
1609
1610    while ((node = rb_first(&old_base->active))) {
1611        timer = rb_entry(node, struct hrtimer, node);
1612        BUG_ON(hrtimer_callback_running(timer));
1613        debug_deactivate(timer);
1614
1615        /*
1616         * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1617         * timer could be seen as !active and just vanish away
1618         * under us on another CPU
1619         */
1620        __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1621        timer->base = new_base;
1622        /*
1623         * Enqueue the timers on the new cpu. This does not
1624         * reprogram the event device in case the timer
1625         * expires before the earliest on this CPU, but we run
1626         * hrtimer_interrupt after we migrated everything to
1627         * sort out already expired timers and reprogram the
1628         * event device.
1629         */
1630        enqueue_hrtimer(timer, new_base);
1631
1632        /* Clear the migration state bit */
1633        timer->state &= ~HRTIMER_STATE_MIGRATE;
1634    }
1635}
1636
1637static void migrate_hrtimers(int scpu)
1638{
1639    struct hrtimer_cpu_base *old_base, *new_base;
1640    int i;
1641
1642    BUG_ON(cpu_online(scpu));
1643    tick_cancel_sched_timer(scpu);
1644
1645    local_irq_disable();
1646    old_base = &per_cpu(hrtimer_bases, scpu);
1647    new_base = &__get_cpu_var(hrtimer_bases);
1648    /*
1649     * The caller is globally serialized and nobody else
1650     * takes two locks at once, deadlock is not possible.
1651     */
1652    spin_lock(&new_base->lock);
1653    spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1654
1655    for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1656        migrate_hrtimer_list(&old_base->clock_base[i],
1657                     &new_base->clock_base[i]);
1658    }
1659
1660    spin_unlock(&old_base->lock);
1661    spin_unlock(&new_base->lock);
1662
1663    /* Check, if we got expired work to do */
1664    __hrtimer_peek_ahead_timers();
1665    local_irq_enable();
1666}
1667
1668#endif /* CONFIG_HOTPLUG_CPU */
1669
1670static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1671                    unsigned long action, void *hcpu)
1672{
1673    int scpu = (long)hcpu;
1674
1675    switch (action) {
1676
1677    case CPU_UP_PREPARE:
1678    case CPU_UP_PREPARE_FROZEN:
1679        init_hrtimers_cpu(scpu);
1680        break;
1681
1682#ifdef CONFIG_HOTPLUG_CPU
1683    case CPU_DYING:
1684    case CPU_DYING_FROZEN:
1685        clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1686        break;
1687    case CPU_DEAD:
1688    case CPU_DEAD_FROZEN:
1689    {
1690        clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1691        migrate_hrtimers(scpu);
1692        break;
1693    }
1694#endif
1695
1696    default:
1697        break;
1698    }
1699
1700    return NOTIFY_OK;
1701}
1702
1703static struct notifier_block __cpuinitdata hrtimers_nb = {
1704    .notifier_call = hrtimer_cpu_notify,
1705};
1706
1707void __init hrtimers_init(void)
1708{
1709    hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1710              (void *)(long)smp_processor_id());
1711    register_cpu_notifier(&hrtimers_nb);
1712#ifdef CONFIG_HIGH_RES_TIMERS
1713    open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1714#endif
1715}
1716
1717/**
1718 * schedule_hrtimeout_range - sleep until timeout
1719 * @expires: timeout value (ktime_t)
1720 * @delta: slack in expires timeout (ktime_t)
1721 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1722 *
1723 * Make the current task sleep until the given expiry time has
1724 * elapsed. The routine will return immediately unless
1725 * the current task state has been set (see set_current_state()).
1726 *
1727 * The @delta argument gives the kernel the freedom to schedule the
1728 * actual wakeup to a time that is both power and performance friendly.
1729 * The kernel give the normal best effort behavior for "@expires+@delta",
1730 * but may decide to fire the timer earlier, but no earlier than @expires.
1731 *
1732 * You can set the task state as follows -
1733 *
1734 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1735 * pass before the routine returns.
1736 *
1737 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1738 * delivered to the current task.
1739 *
1740 * The current task state is guaranteed to be TASK_RUNNING when this
1741 * routine returns.
1742 *
1743 * Returns 0 when the timer has expired otherwise -EINTR
1744 */
1745int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1746                   const enum hrtimer_mode mode)
1747{
1748    struct hrtimer_sleeper t;
1749
1750    /*
1751     * Optimize when a zero timeout value is given. It does not
1752     * matter whether this is an absolute or a relative time.
1753     */
1754    if (expires && !expires->tv64) {
1755        __set_current_state(TASK_RUNNING);
1756        return 0;
1757    }
1758
1759    /*
1760     * A NULL parameter means "inifinte"
1761     */
1762    if (!expires) {
1763        schedule();
1764        __set_current_state(TASK_RUNNING);
1765        return -EINTR;
1766    }
1767
1768    hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, mode);
1769    hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1770
1771    hrtimer_init_sleeper(&t, current);
1772
1773    hrtimer_start_expires(&t.timer, mode);
1774    if (!hrtimer_active(&t.timer))
1775        t.task = NULL;
1776
1777    if (likely(t.task))
1778        schedule();
1779
1780    hrtimer_cancel(&t.timer);
1781    destroy_hrtimer_on_stack(&t.timer);
1782
1783    __set_current_state(TASK_RUNNING);
1784
1785    return !t.task ? 0 : -EINTR;
1786}
1787EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1788
1789/**
1790 * schedule_hrtimeout - sleep until timeout
1791 * @expires: timeout value (ktime_t)
1792 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1793 *
1794 * Make the current task sleep until the given expiry time has
1795 * elapsed. The routine will return immediately unless
1796 * the current task state has been set (see set_current_state()).
1797 *
1798 * You can set the task state as follows -
1799 *
1800 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1801 * pass before the routine returns.
1802 *
1803 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1804 * delivered to the current task.
1805 *
1806 * The current task state is guaranteed to be TASK_RUNNING when this
1807 * routine returns.
1808 *
1809 * Returns 0 when the timer has expired otherwise -EINTR
1810 */
1811int __sched schedule_hrtimeout(ktime_t *expires,
1812                   const enum hrtimer_mode mode)
1813{
1814    return schedule_hrtimeout_range(expires, 0, mode);
1815}
1816EXPORT_SYMBOL_GPL(schedule_hrtimeout);
1817

Archive Download this file

Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master

Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9



interactive