2 * linux/kernel/hrtimer.c
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
8 * High-resolution kernel timers
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.
14 * These timers are currently used for:
18 * - precise in-kernel timing
20 * Started by: Thomas Gleixner and Ingo Molnar
23 * based on kernel/timer.c
25 * Help, testing, suggestions, bugfixes, improvements were
28 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
31 * For licencing details see kernel-base/COPYING
34 #include <linux/cpu.h>
35 #include <linux/export.h>
36 #include <linux/percpu.h>
37 #include <linux/hrtimer.h>
38 #include <linux/notifier.h>
39 #include <linux/syscalls.h>
40 #include <linux/kallsyms.h>
41 #include <linux/interrupt.h>
42 #include <linux/tick.h>
43 #include <linux/seq_file.h>
44 #include <linux/err.h>
45 #include <linux/debugobjects.h>
46 #include <linux/sched.h>
47 #include <linux/sched/sysctl.h>
48 #include <linux/sched/rt.h>
49 #include <linux/sched/deadline.h>
50 #include <linux/timer.h>
51 #include <linux/kthread.h>
52 #include <linux/freezer.h>
54 #include <asm/uaccess.h>
56 #include <trace/events/timer.h>
57 #include <trace/events/hist.h>
59 #include "tick-internal.h"
64 * There are more clockids than hrtimer bases. Thus, we index
65 * into the timer bases by the hrtimer_base_type enum. When trying
66 * to reach a base using a clockid, hrtimer_clockid_to_base()
67 * is used to convert from clockid to the proper hrtimer_base_type.
69 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
71 .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
72 .seq = SEQCNT_ZERO(hrtimer_bases.seq),
76 .index = HRTIMER_BASE_MONOTONIC,
77 .clockid = CLOCK_MONOTONIC,
78 .get_time = &ktime_get,
81 .index = HRTIMER_BASE_REALTIME,
82 .clockid = CLOCK_REALTIME,
83 .get_time = &ktime_get_real,
86 .index = HRTIMER_BASE_BOOTTIME,
87 .clockid = CLOCK_BOOTTIME,
88 .get_time = &ktime_get_boottime,
91 .index = HRTIMER_BASE_TAI,
93 .get_time = &ktime_get_clocktai,
98 static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
99 /* Make sure we catch unsupported clockids */
100 [0 ... MAX_CLOCKS - 1] = HRTIMER_MAX_CLOCK_BASES,
102 [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME,
103 [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC,
104 [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME,
105 [CLOCK_TAI] = HRTIMER_BASE_TAI,
108 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
110 int base = hrtimer_clock_to_base_table[clock_id];
111 BUG_ON(base == HRTIMER_MAX_CLOCK_BASES);
116 * Functions and macros which are different for UP/SMP systems are kept in a
122 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
123 * such that hrtimer_callback_running() can unconditionally dereference
124 * timer->base->cpu_base
126 static struct hrtimer_cpu_base migration_cpu_base = {
127 .seq = SEQCNT_ZERO(migration_cpu_base),
128 .clock_base = { { .cpu_base = &migration_cpu_base, }, },
131 #define migration_base migration_cpu_base.clock_base[0]
134 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
135 * means that all timers which are tied to this base via timer->base are
136 * locked, and the base itself is locked too.
138 * So __run_timers/migrate_timers can safely modify all timers which could
139 * be found on the lists/queues.
141 * When the timer's base is locked, and the timer removed from list, it is
142 * possible to set timer->base = &migration_base and drop the lock: the timer
146 struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
147 unsigned long *flags)
149 struct hrtimer_clock_base *base;
153 if (likely(base != &migration_base)) {
154 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
155 if (likely(base == timer->base))
157 /* The timer has migrated to another CPU: */
158 raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
165 * With HIGHRES=y we do not migrate the timer when it is expiring
166 * before the next event on the target cpu because we cannot reprogram
167 * the target cpu hardware and we would cause it to fire late.
169 * Called with cpu_base->lock of target cpu held.
172 hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
174 #ifdef CONFIG_HIGH_RES_TIMERS
177 if (!new_base->cpu_base->hres_active)
180 expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
181 return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
187 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
189 struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
192 if (pinned || !base->migration_enabled)
194 return &per_cpu(hrtimer_bases, get_nohz_timer_target());
198 struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
206 * We switch the timer base to a power-optimized selected CPU target,
208 * - NO_HZ_COMMON is enabled
209 * - timer migration is enabled
210 * - the timer callback is not running
211 * - the timer is not the first expiring timer on the new target
213 * If one of the above requirements is not fulfilled we move the timer
214 * to the current CPU or leave it on the previously assigned CPU if
215 * the timer callback is currently running.
217 static inline struct hrtimer_clock_base *
218 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
221 struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
222 struct hrtimer_clock_base *new_base;
223 int basenum = base->index;
225 this_cpu_base = this_cpu_ptr(&hrtimer_bases);
226 new_cpu_base = get_target_base(this_cpu_base, pinned);
228 new_base = &new_cpu_base->clock_base[basenum];
230 if (base != new_base) {
232 * We are trying to move timer to new_base.
233 * However we can't change timer's base while it is running,
234 * so we keep it on the same CPU. No hassle vs. reprogramming
235 * the event source in the high resolution case. The softirq
236 * code will take care of this when the timer function has
237 * completed. There is no conflict as we hold the lock until
238 * the timer is enqueued.
240 if (unlikely(hrtimer_callback_running(timer)))
243 /* See the comment in lock_hrtimer_base() */
244 timer->base = &migration_base;
245 raw_spin_unlock(&base->cpu_base->lock);
246 raw_spin_lock(&new_base->cpu_base->lock);
248 if (new_cpu_base != this_cpu_base &&
249 hrtimer_check_target(timer, new_base)) {
250 raw_spin_unlock(&new_base->cpu_base->lock);
251 raw_spin_lock(&base->cpu_base->lock);
252 new_cpu_base = this_cpu_base;
256 timer->base = new_base;
258 if (new_cpu_base != this_cpu_base &&
259 hrtimer_check_target(timer, new_base)) {
260 new_cpu_base = this_cpu_base;
267 #else /* CONFIG_SMP */
269 static inline struct hrtimer_clock_base *
270 lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
272 struct hrtimer_clock_base *base = timer->base;
274 raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
279 # define switch_hrtimer_base(t, b, p) (b)
281 #endif /* !CONFIG_SMP */
284 * Functions for the union type storage format of ktime_t which are
285 * too large for inlining:
287 #if BITS_PER_LONG < 64
289 * Divide a ktime value by a nanosecond value
291 s64 __ktime_divns(const ktime_t kt, s64 div)
297 dclc = ktime_to_ns(kt);
298 tmp = dclc < 0 ? -dclc : dclc;
300 /* Make sure the divisor is less than 2^32: */
306 do_div(tmp, (unsigned long) div);
307 return dclc < 0 ? -tmp : tmp;
309 EXPORT_SYMBOL_GPL(__ktime_divns);
310 #endif /* BITS_PER_LONG >= 64 */
313 * Add two ktime values and do a safety check for overflow:
315 ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
317 ktime_t res = ktime_add(lhs, rhs);
320 * We use KTIME_SEC_MAX here, the maximum timeout which we can
321 * return to user space in a timespec:
323 if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
324 res = ktime_set(KTIME_SEC_MAX, 0);
329 EXPORT_SYMBOL_GPL(ktime_add_safe);
331 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
333 static struct debug_obj_descr hrtimer_debug_descr;
335 static void *hrtimer_debug_hint(void *addr)
337 return ((struct hrtimer *) addr)->function;
341 * fixup_init is called when:
342 * - an active object is initialized
344 static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
346 struct hrtimer *timer = addr;
349 case ODEBUG_STATE_ACTIVE:
350 hrtimer_cancel(timer);
351 debug_object_init(timer, &hrtimer_debug_descr);
359 * fixup_activate is called when:
360 * - an active object is activated
361 * - an unknown object is activated (might be a statically initialized object)
363 static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
367 case ODEBUG_STATE_NOTAVAILABLE:
371 case ODEBUG_STATE_ACTIVE:
380 * fixup_free is called when:
381 * - an active object is freed
383 static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
385 struct hrtimer *timer = addr;
388 case ODEBUG_STATE_ACTIVE:
389 hrtimer_cancel(timer);
390 debug_object_free(timer, &hrtimer_debug_descr);
397 static struct debug_obj_descr hrtimer_debug_descr = {
399 .debug_hint = hrtimer_debug_hint,
400 .fixup_init = hrtimer_fixup_init,
401 .fixup_activate = hrtimer_fixup_activate,
402 .fixup_free = hrtimer_fixup_free,
405 static inline void debug_hrtimer_init(struct hrtimer *timer)
407 debug_object_init(timer, &hrtimer_debug_descr);
410 static inline void debug_hrtimer_activate(struct hrtimer *timer)
412 debug_object_activate(timer, &hrtimer_debug_descr);
415 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
417 debug_object_deactivate(timer, &hrtimer_debug_descr);
420 static inline void debug_hrtimer_free(struct hrtimer *timer)
422 debug_object_free(timer, &hrtimer_debug_descr);
425 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
426 enum hrtimer_mode mode);
428 void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
429 enum hrtimer_mode mode)
431 debug_object_init_on_stack(timer, &hrtimer_debug_descr);
432 __hrtimer_init(timer, clock_id, mode);
434 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
436 void destroy_hrtimer_on_stack(struct hrtimer *timer)
438 debug_object_free(timer, &hrtimer_debug_descr);
442 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
443 static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
444 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
448 debug_init(struct hrtimer *timer, clockid_t clockid,
449 enum hrtimer_mode mode)
451 debug_hrtimer_init(timer);
452 trace_hrtimer_init(timer, clockid, mode);
455 static inline void debug_activate(struct hrtimer *timer)
457 debug_hrtimer_activate(timer);
458 trace_hrtimer_start(timer);
461 static inline void debug_deactivate(struct hrtimer *timer)
463 debug_hrtimer_deactivate(timer);
464 trace_hrtimer_cancel(timer);
467 #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS)
468 static inline void hrtimer_update_next_timer(struct hrtimer_cpu_base *cpu_base,
469 struct hrtimer *timer)
471 #ifdef CONFIG_HIGH_RES_TIMERS
472 cpu_base->next_timer = timer;
476 static ktime_t __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base)
478 struct hrtimer_clock_base *base = cpu_base->clock_base;
479 ktime_t expires, expires_next = { .tv64 = KTIME_MAX };
480 unsigned int active = cpu_base->active_bases;
482 hrtimer_update_next_timer(cpu_base, NULL);
483 for (; active; base++, active >>= 1) {
484 struct timerqueue_node *next;
485 struct hrtimer *timer;
487 if (!(active & 0x01))
490 next = timerqueue_getnext(&base->active);
491 timer = container_of(next, struct hrtimer, node);
492 expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
493 if (expires.tv64 < expires_next.tv64) {
494 expires_next = expires;
495 hrtimer_update_next_timer(cpu_base, timer);
499 * clock_was_set() might have changed base->offset of any of
500 * the clock bases so the result might be negative. Fix it up
501 * to prevent a false positive in clockevents_program_event().
503 if (expires_next.tv64 < 0)
504 expires_next.tv64 = 0;
509 static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
511 ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
512 ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
513 ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
515 return ktime_get_update_offsets_now(&base->clock_was_set_seq,
516 offs_real, offs_boot, offs_tai);
519 /* High resolution timer related functions */
520 #ifdef CONFIG_HIGH_RES_TIMERS
523 * High resolution timer enabled ?
525 static int hrtimer_hres_enabled __read_mostly = 1;
526 unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
527 EXPORT_SYMBOL_GPL(hrtimer_resolution);
530 * Enable / Disable high resolution mode
532 static int __init setup_hrtimer_hres(char *str)
534 if (!strcmp(str, "off"))
535 hrtimer_hres_enabled = 0;
536 else if (!strcmp(str, "on"))
537 hrtimer_hres_enabled = 1;
543 __setup("highres=", setup_hrtimer_hres);
546 * hrtimer_high_res_enabled - query, if the highres mode is enabled
548 static inline int hrtimer_is_hres_enabled(void)
550 return hrtimer_hres_enabled;
554 * Is the high resolution mode active ?
556 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
558 return cpu_base->hres_active;
561 static inline int hrtimer_hres_active(void)
563 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
567 * Reprogram the event source with checking both queues for the
569 * Called with interrupts disabled and base->lock held
572 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
574 ktime_t expires_next;
576 if (!cpu_base->hres_active)
579 expires_next = __hrtimer_get_next_event(cpu_base);
581 if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
584 cpu_base->expires_next.tv64 = expires_next.tv64;
587 * If a hang was detected in the last timer interrupt then we
588 * leave the hang delay active in the hardware. We want the
589 * system to make progress. That also prevents the following
591 * T1 expires 50ms from now
592 * T2 expires 5s from now
594 * T1 is removed, so this code is called and would reprogram
595 * the hardware to 5s from now. Any hrtimer_start after that
596 * will not reprogram the hardware due to hang_detected being
597 * set. So we'd effectivly block all timers until the T2 event
600 if (cpu_base->hang_detected)
603 tick_program_event(cpu_base->expires_next, 1);
607 * When a timer is enqueued and expires earlier than the already enqueued
608 * timers, we have to check, whether it expires earlier than the timer for
609 * which the clock event device was armed.
611 * Called with interrupts disabled and base->cpu_base.lock held
613 static void hrtimer_reprogram(struct hrtimer *timer,
614 struct hrtimer_clock_base *base)
616 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
617 ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
619 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
622 * If the timer is not on the current cpu, we cannot reprogram
623 * the other cpus clock event device.
625 if (base->cpu_base != cpu_base)
629 * If the hrtimer interrupt is running, then it will
630 * reevaluate the clock bases and reprogram the clock event
631 * device. The callbacks are always executed in hard interrupt
632 * context so we don't need an extra check for a running
635 if (cpu_base->in_hrtirq)
638 if (base->cpu_base != cpu_base)
641 if (cpu_base->in_hrtirq)
645 * CLOCK_REALTIME timer might be requested with an absolute
646 * expiry time which is less than base->offset. Set it to 0.
648 if (expires.tv64 < 0)
651 if (expires.tv64 >= cpu_base->expires_next.tv64)
654 /* Update the pointer to the next expiring timer */
655 cpu_base->next_timer = timer;
658 * If a hang was detected in the last timer interrupt then we
659 * do not schedule a timer which is earlier than the expiry
660 * which we enforced in the hang detection. We want the system
663 if (cpu_base->hang_detected)
666 cpu_base->expires_next = expires;
668 * Program the timer hardware. We enforce the expiry for
669 * events which are already in the past.
671 tick_program_event(expires, 1);
675 * Initialize the high resolution related parts of cpu_base
677 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
679 base->expires_next.tv64 = KTIME_MAX;
680 base->hres_active = 0;
684 * Retrigger next event is called after clock was set
686 * Called with interrupts disabled via on_each_cpu()
688 static void retrigger_next_event(void *arg)
690 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
692 if (!base->hres_active)
695 raw_spin_lock(&base->lock);
696 hrtimer_update_base(base);
697 hrtimer_force_reprogram(base, 0);
698 raw_spin_unlock(&base->lock);
702 * Switch to high resolution mode
704 static void hrtimer_switch_to_hres(void)
706 struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
708 if (tick_init_highres()) {
709 printk(KERN_WARNING "Could not switch to high resolution "
710 "mode on CPU %d\n", base->cpu);
713 base->hres_active = 1;
714 hrtimer_resolution = HIGH_RES_NSEC;
716 tick_setup_sched_timer();
717 /* "Retrigger" the interrupt to get things going */
718 retrigger_next_event(NULL);
721 static void clock_was_set_work(struct work_struct *work)
726 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
728 #ifdef CONFIG_PREEMPT_RT_FULL
730 * RT can not call schedule_work from real interrupt context.
731 * Need to make a thread to do the real work.
733 static struct task_struct *clock_set_delay_thread;
734 static bool do_clock_set_delay;
736 static int run_clock_set_delay(void *ignore)
738 while (!kthread_should_stop()) {
739 set_current_state(TASK_INTERRUPTIBLE);
740 if (do_clock_set_delay) {
741 do_clock_set_delay = false;
742 schedule_work(&hrtimer_work);
746 __set_current_state(TASK_RUNNING);
750 void clock_was_set_delayed(void)
752 do_clock_set_delay = true;
753 /* Make visible before waking up process */
755 wake_up_process(clock_set_delay_thread);
758 static __init int create_clock_set_delay_thread(void)
760 clock_set_delay_thread = kthread_run(run_clock_set_delay, NULL, "kclksetdelayd");
761 BUG_ON(!clock_set_delay_thread);
764 early_initcall(create_clock_set_delay_thread);
765 #else /* PREEMPT_RT_FULL */
767 * Called from timekeeping and resume code to reprogramm the hrtimer
768 * interrupt device on all cpus.
770 void clock_was_set_delayed(void)
772 schedule_work(&hrtimer_work);
778 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *b) { return 0; }
779 static inline int hrtimer_hres_active(void) { return 0; }
780 static inline int hrtimer_is_hres_enabled(void) { return 0; }
781 static inline void hrtimer_switch_to_hres(void) { }
783 hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
784 static inline void hrtimer_reprogram(struct hrtimer *timer,
785 struct hrtimer_clock_base *base) { }
786 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
787 static inline void retrigger_next_event(void *arg) { }
789 #endif /* CONFIG_HIGH_RES_TIMERS */
792 * Clock realtime was set
794 * Change the offset of the realtime clock vs. the monotonic
797 * We might have to reprogram the high resolution timer interrupt. On
798 * SMP we call the architecture specific code to retrigger _all_ high
799 * resolution timer interrupts. On UP we just disable interrupts and
800 * call the high resolution interrupt code.
802 void clock_was_set(void)
804 #ifdef CONFIG_HIGH_RES_TIMERS
805 /* Retrigger the CPU local events everywhere */
806 on_each_cpu(retrigger_next_event, NULL, 1);
808 timerfd_clock_was_set();
812 * During resume we might have to reprogram the high resolution timer
813 * interrupt on all online CPUs. However, all other CPUs will be
814 * stopped with IRQs interrupts disabled so the clock_was_set() call
817 void hrtimers_resume(void)
819 WARN_ONCE(!irqs_disabled(),
820 KERN_INFO "hrtimers_resume() called with IRQs enabled!");
822 /* Retrigger on the local CPU */
823 retrigger_next_event(NULL);
824 /* And schedule a retrigger for all others */
825 clock_was_set_delayed();
828 static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
830 #ifdef CONFIG_TIMER_STATS
831 if (timer->start_site)
833 timer->start_site = __builtin_return_address(0);
834 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
835 timer->start_pid = current->pid;
839 static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
841 #ifdef CONFIG_TIMER_STATS
842 timer->start_site = NULL;
846 static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
848 #ifdef CONFIG_TIMER_STATS
849 if (likely(!timer_stats_active))
851 timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
852 timer->function, timer->start_comm, 0);
857 * Counterpart to lock_hrtimer_base above:
860 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
862 raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
866 * hrtimer_forward - forward the timer expiry
867 * @timer: hrtimer to forward
868 * @now: forward past this time
869 * @interval: the interval to forward
871 * Forward the timer expiry so it will expire in the future.
872 * Returns the number of overruns.
874 * Can be safely called from the callback function of @timer. If
875 * called from other contexts @timer must neither be enqueued nor
876 * running the callback and the caller needs to take care of
879 * Note: This only updates the timer expiry value and does not requeue
882 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
887 delta = ktime_sub(now, hrtimer_get_expires(timer));
892 if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
895 if (interval.tv64 < hrtimer_resolution)
896 interval.tv64 = hrtimer_resolution;
898 if (unlikely(delta.tv64 >= interval.tv64)) {
899 s64 incr = ktime_to_ns(interval);
901 orun = ktime_divns(delta, incr);
902 hrtimer_add_expires_ns(timer, incr * orun);
903 if (hrtimer_get_expires_tv64(timer) > now.tv64)
906 * This (and the ktime_add() below) is the
907 * correction for exact:
911 hrtimer_add_expires(timer, interval);
915 EXPORT_SYMBOL_GPL(hrtimer_forward);
917 #ifdef CONFIG_PREEMPT_RT_BASE
918 # define wake_up_timer_waiters(b) wake_up(&(b)->wait)
921 * hrtimer_wait_for_timer - Wait for a running timer
923 * @timer: timer to wait for
925 * The function waits in case the timers callback function is
926 * currently executed on the waitqueue of the timer base. The
927 * waitqueue is woken up after the timer callback function has
928 * finished execution.
930 void hrtimer_wait_for_timer(const struct hrtimer *timer)
932 struct hrtimer_clock_base *base = timer->base;
934 if (base && base->cpu_base && !timer->irqsafe)
935 wait_event(base->cpu_base->wait,
936 !(hrtimer_callback_running(timer)));
940 # define wake_up_timer_waiters(b) do { } while (0)
944 * enqueue_hrtimer - internal function to (re)start a timer
946 * The timer is inserted in expiry order. Insertion into the
947 * red black tree is O(log(n)). Must hold the base lock.
949 * Returns 1 when the new timer is the leftmost timer in the tree.
951 static int enqueue_hrtimer(struct hrtimer *timer,
952 struct hrtimer_clock_base *base)
954 debug_activate(timer);
956 base->cpu_base->active_bases |= 1 << base->index;
958 timer->state = HRTIMER_STATE_ENQUEUED;
960 return timerqueue_add(&base->active, &timer->node);
964 * __remove_hrtimer - internal function to remove a timer
966 * Caller must hold the base lock.
968 * High resolution timer mode reprograms the clock event device when the
969 * timer is the one which expires next. The caller can disable this by setting
970 * reprogram to zero. This is useful, when the context does a reprogramming
971 * anyway (e.g. timer interrupt)
973 static void __remove_hrtimer(struct hrtimer *timer,
974 struct hrtimer_clock_base *base,
975 u8 newstate, int reprogram)
977 struct hrtimer_cpu_base *cpu_base = base->cpu_base;
978 u8 state = timer->state;
980 timer->state = newstate;
981 if (!(state & HRTIMER_STATE_ENQUEUED))
984 if (unlikely(!list_empty(&timer->cb_entry))) {
985 list_del_init(&timer->cb_entry);
989 if (!timerqueue_del(&base->active, &timer->node))
990 cpu_base->active_bases &= ~(1 << base->index);
992 #ifdef CONFIG_HIGH_RES_TIMERS
994 * Note: If reprogram is false we do not update
995 * cpu_base->next_timer. This happens when we remove the first
996 * timer on a remote cpu. No harm as we never dereference
997 * cpu_base->next_timer. So the worst thing what can happen is
998 * an superflous call to hrtimer_force_reprogram() on the
999 * remote cpu later on if the same timer gets enqueued again.
1001 if (reprogram && timer == cpu_base->next_timer)
1002 hrtimer_force_reprogram(cpu_base, 1);
1007 * remove hrtimer, called with base lock held
1010 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
1012 if (hrtimer_is_queued(timer)) {
1013 u8 state = timer->state;
1017 * Remove the timer and force reprogramming when high
1018 * resolution mode is active and the timer is on the current
1019 * CPU. If we remove a timer on another CPU, reprogramming is
1020 * skipped. The interrupt event on this CPU is fired and
1021 * reprogramming happens in the interrupt handler. This is a
1022 * rare case and less expensive than a smp call.
1024 debug_deactivate(timer);
1025 timer_stats_hrtimer_clear_start_info(timer);
1026 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1029 state = HRTIMER_STATE_INACTIVE;
1031 __remove_hrtimer(timer, base, state, reprogram);
1037 static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
1038 const enum hrtimer_mode mode)
1040 #ifdef CONFIG_TIME_LOW_RES
1042 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1043 * granular time values. For relative timers we add hrtimer_resolution
1044 * (i.e. one jiffie) to prevent short timeouts.
1046 timer->is_rel = mode & HRTIMER_MODE_REL;
1048 tim = ktime_add_safe(tim, ktime_set(0, hrtimer_resolution));
1054 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1055 * @timer: the timer to be added
1057 * @delta_ns: "slack" range for the timer
1058 * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or
1059 * relative (HRTIMER_MODE_REL)
1061 void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1062 unsigned long delta_ns, const enum hrtimer_mode mode)
1064 struct hrtimer_clock_base *base, *new_base;
1065 unsigned long flags;
1068 base = lock_hrtimer_base(timer, &flags);
1070 /* Remove an active timer from the queue: */
1071 remove_hrtimer(timer, base, true);
1073 if (mode & HRTIMER_MODE_REL)
1074 tim = ktime_add_safe(tim, base->get_time());
1076 tim = hrtimer_update_lowres(timer, tim, mode);
1078 hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1080 /* Switch the timer base, if necessary: */
1081 new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
1083 timer_stats_hrtimer_set_start_info(timer);
1084 #ifdef CONFIG_MISSED_TIMER_OFFSETS_HIST
1086 ktime_t now = new_base->get_time();
1088 if (ktime_to_ns(tim) < ktime_to_ns(now))
1089 timer->praecox = now;
1091 timer->praecox = ktime_set(0, 0);
1094 leftmost = enqueue_hrtimer(timer, new_base);
1098 if (!hrtimer_is_hres_active(timer)) {
1100 * Kick to reschedule the next tick to handle the new timer
1101 * on dynticks target.
1103 if (new_base->cpu_base->nohz_active)
1104 wake_up_nohz_cpu(new_base->cpu_base->cpu);
1106 hrtimer_reprogram(timer, new_base);
1109 unlock_hrtimer_base(timer, &flags);
1111 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1114 * hrtimer_try_to_cancel - try to deactivate a timer
1115 * @timer: hrtimer to stop
1118 * 0 when the timer was not active
1119 * 1 when the timer was active
1120 * -1 when the timer is currently excuting the callback function and
1123 int hrtimer_try_to_cancel(struct hrtimer *timer)
1125 struct hrtimer_clock_base *base;
1126 unsigned long flags;
1130 * Check lockless first. If the timer is not active (neither
1131 * enqueued nor running the callback, nothing to do here. The
1132 * base lock does not serialize against a concurrent enqueue,
1133 * so we can avoid taking it.
1135 if (!hrtimer_active(timer))
1138 base = lock_hrtimer_base(timer, &flags);
1140 if (!hrtimer_callback_running(timer))
1141 ret = remove_hrtimer(timer, base, false);
1143 unlock_hrtimer_base(timer, &flags);
1148 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1151 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1152 * @timer: the timer to be cancelled
1155 * 0 when the timer was not active
1156 * 1 when the timer was active
1158 int hrtimer_cancel(struct hrtimer *timer)
1161 int ret = hrtimer_try_to_cancel(timer);
1165 hrtimer_wait_for_timer(timer);
1168 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1171 * hrtimer_get_remaining - get remaining time for the timer
1172 * @timer: the timer to read
1173 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1175 ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1177 unsigned long flags;
1180 lock_hrtimer_base(timer, &flags);
1181 if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1182 rem = hrtimer_expires_remaining_adjusted(timer);
1184 rem = hrtimer_expires_remaining(timer);
1185 unlock_hrtimer_base(timer, &flags);
1189 EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1191 #ifdef CONFIG_NO_HZ_COMMON
1193 * hrtimer_get_next_event - get the time until next expiry event
1195 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1197 u64 hrtimer_get_next_event(void)
1199 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1200 u64 expires = KTIME_MAX;
1201 unsigned long flags;
1203 raw_spin_lock_irqsave(&cpu_base->lock, flags);
1205 if (!__hrtimer_hres_active(cpu_base))
1206 expires = __hrtimer_get_next_event(cpu_base).tv64;
1208 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1214 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1215 enum hrtimer_mode mode)
1217 struct hrtimer_cpu_base *cpu_base;
1220 memset(timer, 0, sizeof(struct hrtimer));
1222 cpu_base = raw_cpu_ptr(&hrtimer_bases);
1224 if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1225 clock_id = CLOCK_MONOTONIC;
1227 base = hrtimer_clockid_to_base(clock_id);
1228 timer->base = &cpu_base->clock_base[base];
1229 INIT_LIST_HEAD(&timer->cb_entry);
1230 timerqueue_init(&timer->node);
1232 #ifdef CONFIG_TIMER_STATS
1233 timer->start_site = NULL;
1234 timer->start_pid = -1;
1235 memset(timer->start_comm, 0, TASK_COMM_LEN);
1240 * hrtimer_init - initialize a timer to the given clock
1241 * @timer: the timer to be initialized
1242 * @clock_id: the clock to be used
1243 * @mode: timer mode abs/rel
1245 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1246 enum hrtimer_mode mode)
1248 debug_init(timer, clock_id, mode);
1249 __hrtimer_init(timer, clock_id, mode);
1251 EXPORT_SYMBOL_GPL(hrtimer_init);
1254 * A timer is active, when it is enqueued into the rbtree or the
1255 * callback function is running or it's in the state of being migrated
1258 * It is important for this function to not return a false negative.
1260 bool hrtimer_active(const struct hrtimer *timer)
1262 struct hrtimer_cpu_base *cpu_base;
1266 cpu_base = READ_ONCE(timer->base->cpu_base);
1267 seq = raw_read_seqcount_begin(&cpu_base->seq);
1269 if (timer->state != HRTIMER_STATE_INACTIVE ||
1270 cpu_base->running_soft == timer ||
1271 cpu_base->running == timer)
1274 } while (read_seqcount_retry(&cpu_base->seq, seq) ||
1275 cpu_base != READ_ONCE(timer->base->cpu_base));
1279 EXPORT_SYMBOL_GPL(hrtimer_active);
1282 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1283 * distinct sections:
1285 * - queued: the timer is queued
1286 * - callback: the timer is being ran
1287 * - post: the timer is inactive or (re)queued
1289 * On the read side we ensure we observe timer->state and cpu_base->running
1290 * from the same section, if anything changed while we looked at it, we retry.
1291 * This includes timer->base changing because sequence numbers alone are
1292 * insufficient for that.
1294 * The sequence numbers are required because otherwise we could still observe
1295 * a false negative if the read side got smeared over multiple consequtive
1296 * __run_hrtimer() invocations.
1299 static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1300 struct hrtimer_clock_base *base,
1301 struct hrtimer *timer, ktime_t *now)
1303 enum hrtimer_restart (*fn)(struct hrtimer *);
1306 lockdep_assert_held(&cpu_base->lock);
1308 debug_deactivate(timer);
1309 cpu_base->running = timer;
1312 * Separate the ->running assignment from the ->state assignment.
1314 * As with a regular write barrier, this ensures the read side in
1315 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1316 * timer->state == INACTIVE.
1318 raw_write_seqcount_barrier(&cpu_base->seq);
1320 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1321 timer_stats_account_hrtimer(timer);
1322 fn = timer->function;
1325 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1326 * timer is restarted with a period then it becomes an absolute
1327 * timer. If its not restarted it does not matter.
1329 if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1330 timer->is_rel = false;
1333 * Because we run timers from hardirq context, there is no chance
1334 * they get migrated to another cpu, therefore its safe to unlock
1337 raw_spin_unlock(&cpu_base->lock);
1338 trace_hrtimer_expire_entry(timer, now);
1339 restart = fn(timer);
1340 trace_hrtimer_expire_exit(timer);
1341 raw_spin_lock(&cpu_base->lock);
1344 * Note: We clear the running state after enqueue_hrtimer and
1345 * we do not reprogramm the event hardware. Happens either in
1346 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1348 * Note: Because we dropped the cpu_base->lock above,
1349 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1352 if (restart != HRTIMER_NORESTART &&
1353 !(timer->state & HRTIMER_STATE_ENQUEUED))
1354 enqueue_hrtimer(timer, base);
1357 * Separate the ->running assignment from the ->state assignment.
1359 * As with a regular write barrier, this ensures the read side in
1360 * hrtimer_active() cannot observe cpu_base->running == NULL &&
1361 * timer->state == INACTIVE.
1363 raw_write_seqcount_barrier(&cpu_base->seq);
1365 WARN_ON_ONCE(cpu_base->running != timer);
1366 cpu_base->running = NULL;
1369 #ifdef CONFIG_PREEMPT_RT_BASE
1370 static void hrtimer_rt_reprogram(int restart, struct hrtimer *timer,
1371 struct hrtimer_clock_base *base)
1375 if (restart != HRTIMER_NORESTART &&
1376 !(timer->state & HRTIMER_STATE_ENQUEUED)) {
1378 leftmost = enqueue_hrtimer(timer, base);
1381 #ifdef CONFIG_HIGH_RES_TIMERS
1382 if (!hrtimer_is_hres_active(timer)) {
1384 * Kick to reschedule the next tick to handle the new timer
1385 * on dynticks target.
1387 if (base->cpu_base->nohz_active)
1388 wake_up_nohz_cpu(base->cpu_base->cpu);
1391 hrtimer_reprogram(timer, base);
1398 * The changes in mainline which removed the callback modes from
1399 * hrtimer are not yet working with -rt. The non wakeup_process()
1400 * based callbacks which involve sleeping locks need to be treated
1403 static void hrtimer_rt_run_pending(void)
1405 enum hrtimer_restart (*fn)(struct hrtimer *);
1406 struct hrtimer_cpu_base *cpu_base;
1407 struct hrtimer_clock_base *base;
1408 struct hrtimer *timer;
1411 local_irq_disable();
1412 cpu_base = &per_cpu(hrtimer_bases, smp_processor_id());
1414 raw_spin_lock(&cpu_base->lock);
1416 for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1417 base = &cpu_base->clock_base[index];
1419 while (!list_empty(&base->expired)) {
1420 timer = list_first_entry(&base->expired,
1421 struct hrtimer, cb_entry);
1424 * Same as the above __run_hrtimer function
1425 * just we run with interrupts enabled.
1427 debug_deactivate(timer);
1428 cpu_base->running_soft = timer;
1429 raw_write_seqcount_barrier(&cpu_base->seq);
1431 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1432 timer_stats_account_hrtimer(timer);
1433 fn = timer->function;
1435 raw_spin_unlock_irq(&cpu_base->lock);
1436 restart = fn(timer);
1437 raw_spin_lock_irq(&cpu_base->lock);
1439 hrtimer_rt_reprogram(restart, timer, base);
1440 raw_write_seqcount_barrier(&cpu_base->seq);
1442 WARN_ON_ONCE(cpu_base->running_soft != timer);
1443 cpu_base->running_soft = NULL;
1447 raw_spin_unlock_irq(&cpu_base->lock);
1449 wake_up_timer_waiters(cpu_base);
1452 static int hrtimer_rt_defer(struct hrtimer *timer)
1457 __remove_hrtimer(timer, timer->base, timer->state, 0);
1458 list_add_tail(&timer->cb_entry, &timer->base->expired);
1464 static inline int hrtimer_rt_defer(struct hrtimer *timer) { return 0; }
1468 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer);
1470 static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now)
1472 struct hrtimer_clock_base *base = cpu_base->clock_base;
1473 unsigned int active = cpu_base->active_bases;
1476 for (; active; base++, active >>= 1) {
1477 struct timerqueue_node *node;
1480 if (!(active & 0x01))
1483 basenow = ktime_add(now, base->offset);
1485 while ((node = timerqueue_getnext(&base->active))) {
1486 struct hrtimer *timer;
1488 timer = container_of(node, struct hrtimer, node);
1490 trace_hrtimer_interrupt(raw_smp_processor_id(),
1491 ktime_to_ns(ktime_sub(ktime_to_ns(timer->praecox) ?
1492 timer->praecox : hrtimer_get_expires(timer),
1495 timer->function == hrtimer_wakeup ?
1496 container_of(timer, struct hrtimer_sleeper,
1497 timer)->task : NULL);
1500 * The immediate goal for using the softexpires is
1501 * minimizing wakeups, not running timers at the
1502 * earliest interrupt after their soft expiration.
1503 * This allows us to avoid using a Priority Search
1504 * Tree, which can answer a stabbing querry for
1505 * overlapping intervals and instead use the simple
1506 * BST we already have.
1507 * We don't add extra wakeups by delaying timers that
1508 * are right-of a not yet expired timer, because that
1509 * timer will have to trigger a wakeup anyway.
1511 if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer))
1514 if (!hrtimer_rt_defer(timer))
1515 __run_hrtimer(cpu_base, base, timer, &basenow);
1521 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1524 #ifdef CONFIG_HIGH_RES_TIMERS
1527 * High resolution timer interrupt
1528 * Called with interrupts disabled
1530 void hrtimer_interrupt(struct clock_event_device *dev)
1532 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1533 ktime_t expires_next, now, entry_time, delta;
1536 BUG_ON(!cpu_base->hres_active);
1537 cpu_base->nr_events++;
1538 dev->next_event.tv64 = KTIME_MAX;
1540 raw_spin_lock(&cpu_base->lock);
1541 entry_time = now = hrtimer_update_base(cpu_base);
1543 cpu_base->in_hrtirq = 1;
1545 * We set expires_next to KTIME_MAX here with cpu_base->lock
1546 * held to prevent that a timer is enqueued in our queue via
1547 * the migration code. This does not affect enqueueing of
1548 * timers which run their callback and need to be requeued on
1551 cpu_base->expires_next.tv64 = KTIME_MAX;
1553 __hrtimer_run_queues(cpu_base, now);
1555 /* Reevaluate the clock bases for the next expiry */
1556 expires_next = __hrtimer_get_next_event(cpu_base);
1558 * Store the new expiry value so the migration code can verify
1561 cpu_base->expires_next = expires_next;
1562 cpu_base->in_hrtirq = 0;
1563 raw_spin_unlock(&cpu_base->lock);
1565 /* Reprogramming necessary ? */
1566 if (!tick_program_event(expires_next, 0)) {
1567 cpu_base->hang_detected = 0;
1572 * The next timer was already expired due to:
1574 * - long lasting callbacks
1575 * - being scheduled away when running in a VM
1577 * We need to prevent that we loop forever in the hrtimer
1578 * interrupt routine. We give it 3 attempts to avoid
1579 * overreacting on some spurious event.
1581 * Acquire base lock for updating the offsets and retrieving
1584 raw_spin_lock(&cpu_base->lock);
1585 now = hrtimer_update_base(cpu_base);
1586 cpu_base->nr_retries++;
1590 * Give the system a chance to do something else than looping
1591 * here. We stored the entry time, so we know exactly how long
1592 * we spent here. We schedule the next event this amount of
1595 cpu_base->nr_hangs++;
1596 cpu_base->hang_detected = 1;
1597 raw_spin_unlock(&cpu_base->lock);
1598 delta = ktime_sub(now, entry_time);
1599 if ((unsigned int)delta.tv64 > cpu_base->max_hang_time)
1600 cpu_base->max_hang_time = (unsigned int) delta.tv64;
1602 * Limit it to a sensible value as we enforce a longer
1603 * delay. Give the CPU at least 100ms to catch up.
1605 if (delta.tv64 > 100 * NSEC_PER_MSEC)
1606 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1608 expires_next = ktime_add(now, delta);
1609 tick_program_event(expires_next, 1);
1610 printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1611 ktime_to_ns(delta));
1615 * local version of hrtimer_peek_ahead_timers() called with interrupts
1618 static inline void __hrtimer_peek_ahead_timers(void)
1620 struct tick_device *td;
1622 if (!hrtimer_hres_active())
1625 td = this_cpu_ptr(&tick_cpu_device);
1626 if (td && td->evtdev)
1627 hrtimer_interrupt(td->evtdev);
1630 #else /* CONFIG_HIGH_RES_TIMERS */
1632 static inline void __hrtimer_peek_ahead_timers(void) { }
1634 #endif /* !CONFIG_HIGH_RES_TIMERS */
1637 * Called from run_local_timers in hardirq context every jiffy
1639 void hrtimer_run_queues(void)
1641 struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1644 if (__hrtimer_hres_active(cpu_base))
1648 * This _is_ ugly: We have to check periodically, whether we
1649 * can switch to highres and / or nohz mode. The clocksource
1650 * switch happens with xtime_lock held. Notification from
1651 * there only sets the check bit in the tick_oneshot code,
1652 * otherwise we might deadlock vs. xtime_lock.
1654 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1655 hrtimer_switch_to_hres();
1659 raw_spin_lock(&cpu_base->lock);
1660 now = hrtimer_update_base(cpu_base);
1661 __hrtimer_run_queues(cpu_base, now);
1662 raw_spin_unlock(&cpu_base->lock);
1666 * Sleep related functions:
1668 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1670 struct hrtimer_sleeper *t =
1671 container_of(timer, struct hrtimer_sleeper, timer);
1672 struct task_struct *task = t->task;
1676 wake_up_process(task);
1678 return HRTIMER_NORESTART;
1681 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1683 sl->timer.function = hrtimer_wakeup;
1684 sl->timer.irqsafe = 1;
1687 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1689 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode,
1690 unsigned long state)
1692 hrtimer_init_sleeper(t, current);
1695 set_current_state(state);
1696 hrtimer_start_expires(&t->timer, mode);
1698 if (likely(t->task))
1699 freezable_schedule();
1701 hrtimer_cancel(&t->timer);
1702 mode = HRTIMER_MODE_ABS;
1704 } while (t->task && !signal_pending(current));
1706 __set_current_state(TASK_RUNNING);
1708 return t->task == NULL;
1711 static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1713 struct timespec rmt;
1716 rem = hrtimer_expires_remaining(timer);
1719 rmt = ktime_to_timespec(rem);
1721 if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1727 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1729 struct hrtimer_sleeper t;
1730 struct timespec __user *rmtp;
1733 hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1735 hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1737 /* cpu_chill() does not care about restart state. */
1738 if (do_nanosleep(&t, HRTIMER_MODE_ABS, TASK_INTERRUPTIBLE))
1741 rmtp = restart->nanosleep.rmtp;
1743 ret = update_rmtp(&t.timer, rmtp);
1748 /* The other values in restart are already filled in */
1749 ret = -ERESTART_RESTARTBLOCK;
1751 destroy_hrtimer_on_stack(&t.timer);
1756 __hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1757 const enum hrtimer_mode mode, const clockid_t clockid,
1758 unsigned long state)
1760 struct restart_block *restart;
1761 struct hrtimer_sleeper t;
1763 unsigned long slack;
1765 slack = current->timer_slack_ns;
1766 if (dl_task(current) || rt_task(current))
1769 hrtimer_init_on_stack(&t.timer, clockid, mode);
1770 hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1771 if (do_nanosleep(&t, mode, state))
1774 /* Absolute timers do not update the rmtp value and restart: */
1775 if (mode == HRTIMER_MODE_ABS) {
1776 ret = -ERESTARTNOHAND;
1781 ret = update_rmtp(&t.timer, rmtp);
1786 restart = ¤t->restart_block;
1787 restart->fn = hrtimer_nanosleep_restart;
1788 restart->nanosleep.clockid = t.timer.base->clockid;
1789 restart->nanosleep.rmtp = rmtp;
1790 restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1792 ret = -ERESTART_RESTARTBLOCK;
1794 destroy_hrtimer_on_stack(&t.timer);
1798 long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1799 const enum hrtimer_mode mode, const clockid_t clockid)
1801 return __hrtimer_nanosleep(rqtp, rmtp, mode, clockid, TASK_INTERRUPTIBLE);
1804 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1805 struct timespec __user *, rmtp)
1809 if (copy_from_user(&tu, rqtp, sizeof(tu)))
1812 if (!timespec_valid(&tu))
1815 return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1818 #ifdef CONFIG_PREEMPT_RT_FULL
1820 * Sleep for 1 ms in hope whoever holds what we want will let it go.
1822 void cpu_chill(void)
1824 struct timespec tu = {
1825 .tv_nsec = NSEC_PER_MSEC,
1827 unsigned int freeze_flag = current->flags & PF_NOFREEZE;
1829 current->flags |= PF_NOFREEZE;
1830 __hrtimer_nanosleep(&tu, NULL, HRTIMER_MODE_REL, CLOCK_MONOTONIC,
1831 TASK_UNINTERRUPTIBLE);
1833 current->flags &= ~PF_NOFREEZE;
1835 EXPORT_SYMBOL(cpu_chill);
1839 * Functions related to boot-time initialization:
1841 static void init_hrtimers_cpu(int cpu)
1843 struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1846 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1847 cpu_base->clock_base[i].cpu_base = cpu_base;
1848 timerqueue_init_head(&cpu_base->clock_base[i].active);
1849 INIT_LIST_HEAD(&cpu_base->clock_base[i].expired);
1852 cpu_base->cpu = cpu;
1853 hrtimer_init_hres(cpu_base);
1854 #ifdef CONFIG_PREEMPT_RT_BASE
1855 init_waitqueue_head(&cpu_base->wait);
1859 #ifdef CONFIG_HOTPLUG_CPU
1861 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1862 struct hrtimer_clock_base *new_base)
1864 struct hrtimer *timer;
1865 struct timerqueue_node *node;
1867 while ((node = timerqueue_getnext(&old_base->active))) {
1868 timer = container_of(node, struct hrtimer, node);
1869 BUG_ON(hrtimer_callback_running(timer));
1870 debug_deactivate(timer);
1873 * Mark it as ENQUEUED not INACTIVE otherwise the
1874 * timer could be seen as !active and just vanish away
1875 * under us on another CPU
1877 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
1878 timer->base = new_base;
1880 * Enqueue the timers on the new cpu. This does not
1881 * reprogram the event device in case the timer
1882 * expires before the earliest on this CPU, but we run
1883 * hrtimer_interrupt after we migrated everything to
1884 * sort out already expired timers and reprogram the
1887 enqueue_hrtimer(timer, new_base);
1891 static void migrate_hrtimers(int scpu)
1893 struct hrtimer_cpu_base *old_base, *new_base;
1896 BUG_ON(cpu_online(scpu));
1897 tick_cancel_sched_timer(scpu);
1899 local_irq_disable();
1900 old_base = &per_cpu(hrtimer_bases, scpu);
1901 new_base = this_cpu_ptr(&hrtimer_bases);
1903 * The caller is globally serialized and nobody else
1904 * takes two locks at once, deadlock is not possible.
1906 raw_spin_lock(&new_base->lock);
1907 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1909 for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1910 migrate_hrtimer_list(&old_base->clock_base[i],
1911 &new_base->clock_base[i]);
1914 raw_spin_unlock(&old_base->lock);
1915 raw_spin_unlock(&new_base->lock);
1917 /* Check, if we got expired work to do */
1918 __hrtimer_peek_ahead_timers();
1922 #endif /* CONFIG_HOTPLUG_CPU */
1924 static int hrtimer_cpu_notify(struct notifier_block *self,
1925 unsigned long action, void *hcpu)
1927 int scpu = (long)hcpu;
1931 case CPU_UP_PREPARE:
1932 case CPU_UP_PREPARE_FROZEN:
1933 init_hrtimers_cpu(scpu);
1936 #ifdef CONFIG_HOTPLUG_CPU
1938 case CPU_DEAD_FROZEN:
1939 migrate_hrtimers(scpu);
1950 static struct notifier_block hrtimers_nb = {
1951 .notifier_call = hrtimer_cpu_notify,
1954 #ifdef CONFIG_PREEMPT_RT_BASE
1955 static void run_hrtimer_softirq(struct softirq_action *h)
1957 hrtimer_rt_run_pending();
1961 void __init hrtimers_init(void)
1963 hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1964 (void *)(long)smp_processor_id());
1965 register_cpu_notifier(&hrtimers_nb);
1966 #ifdef CONFIG_PREEMPT_RT_BASE
1967 open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1972 * schedule_hrtimeout_range_clock - sleep until timeout
1973 * @expires: timeout value (ktime_t)
1974 * @delta: slack in expires timeout (ktime_t)
1975 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1976 * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1979 schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1980 const enum hrtimer_mode mode, int clock)
1982 struct hrtimer_sleeper t;
1985 * Optimize when a zero timeout value is given. It does not
1986 * matter whether this is an absolute or a relative time.
1988 if (expires && !expires->tv64) {
1989 __set_current_state(TASK_RUNNING);
1994 * A NULL parameter means "infinite"
2001 hrtimer_init_on_stack(&t.timer, clock, mode);
2002 hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
2004 hrtimer_init_sleeper(&t, current);
2006 hrtimer_start_expires(&t.timer, mode);
2011 hrtimer_cancel(&t.timer);
2012 destroy_hrtimer_on_stack(&t.timer);
2014 __set_current_state(TASK_RUNNING);
2016 return !t.task ? 0 : -EINTR;
2020 * schedule_hrtimeout_range - sleep until timeout
2021 * @expires: timeout value (ktime_t)
2022 * @delta: slack in expires timeout (ktime_t)
2023 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
2025 * Make the current task sleep until the given expiry time has
2026 * elapsed. The routine will return immediately unless
2027 * the current task state has been set (see set_current_state()).
2029 * The @delta argument gives the kernel the freedom to schedule the
2030 * actual wakeup to a time that is both power and performance friendly.
2031 * The kernel give the normal best effort behavior for "@expires+@delta",
2032 * but may decide to fire the timer earlier, but no earlier than @expires.
2034 * You can set the task state as follows -
2036 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2037 * pass before the routine returns.
2039 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2040 * delivered to the current task.
2042 * The current task state is guaranteed to be TASK_RUNNING when this
2045 * Returns 0 when the timer has expired otherwise -EINTR
2047 int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
2048 const enum hrtimer_mode mode)
2050 return schedule_hrtimeout_range_clock(expires, delta, mode,
2053 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
2056 * schedule_hrtimeout - sleep until timeout
2057 * @expires: timeout value (ktime_t)
2058 * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
2060 * Make the current task sleep until the given expiry time has
2061 * elapsed. The routine will return immediately unless
2062 * the current task state has been set (see set_current_state()).
2064 * You can set the task state as follows -
2066 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2067 * pass before the routine returns.
2069 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2070 * delivered to the current task.
2072 * The current task state is guaranteed to be TASK_RUNNING when this
2075 * Returns 0 when the timer has expired otherwise -EINTR
2077 int __sched schedule_hrtimeout(ktime_t *expires,
2078 const enum hrtimer_mode mode)
2080 return schedule_hrtimeout_range(expires, 0, mode);
2082 EXPORT_SYMBOL_GPL(schedule_hrtimeout);