1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
4 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
5 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
7 * High-resolution kernel timers
9 * In contrast to the low-resolution timeout API, aka timer wheel,
10 * hrtimers provide finer resolution and accuracy depending on system
11 * configuration and capabilities.
13 * Started by: Thomas Gleixner and Ingo Molnar
16 * Based on the original timer wheel code
18 * Help, testing, suggestions, bugfixes, improvements were
21 * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
25 #include <linux/cpu.h>
26 #include <linux/export.h>
27 #include <linux/percpu.h>
28 #include <linux/hrtimer.h>
29 #include <linux/notifier.h>
30 #include <linux/syscalls.h>
31 #include <linux/interrupt.h>
32 #include <linux/tick.h>
33 #include <linux/err.h>
34 #include <linux/debugobjects.h>
35 #include <linux/sched/signal.h>
36 #include <linux/sched/sysctl.h>
37 #include <linux/sched/rt.h>
38 #include <linux/sched/deadline.h>
39 #include <linux/sched/nohz.h>
40 #include <linux/sched/debug.h>
41 #include <linux/timer.h>
42 #include <linux/freezer.h>
43 #include <linux/compat.h>
45 #include <linux/uaccess.h>
47 #include <trace/events/timer.h>
49 #include "tick-internal.h"
52 * Masks for selecting the soft and hard context timers from
55 #define MASK_SHIFT (HRTIMER_BASE_MONOTONIC_SOFT)
56 #define HRTIMER_ACTIVE_HARD ((1U << MASK_SHIFT) - 1)
57 #define HRTIMER_ACTIVE_SOFT (HRTIMER_ACTIVE_HARD << MASK_SHIFT)
58 #define HRTIMER_ACTIVE_ALL (HRTIMER_ACTIVE_SOFT | HRTIMER_ACTIVE_HARD)
63 * There are more clockids than hrtimer bases. Thus, we index
64 * into the timer bases by the hrtimer_base_type enum. When trying
65 * to reach a base using a clockid, hrtimer_clockid_to_base()
66 * is used to convert from clockid to the proper hrtimer_base_type.
68 DEFINE_PER_CPU(struct hrtimer_cpu_base
, hrtimer_bases
) =
70 .lock
= __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases
.lock
),
74 .index
= HRTIMER_BASE_MONOTONIC
,
75 .clockid
= CLOCK_MONOTONIC
,
76 .get_time
= &ktime_get
,
79 .index
= HRTIMER_BASE_REALTIME
,
80 .clockid
= CLOCK_REALTIME
,
81 .get_time
= &ktime_get_real
,
84 .index
= HRTIMER_BASE_BOOTTIME
,
85 .clockid
= CLOCK_BOOTTIME
,
86 .get_time
= &ktime_get_boottime
,
89 .index
= HRTIMER_BASE_TAI
,
91 .get_time
= &ktime_get_clocktai
,
94 .index
= HRTIMER_BASE_MONOTONIC_SOFT
,
95 .clockid
= CLOCK_MONOTONIC
,
96 .get_time
= &ktime_get
,
99 .index
= HRTIMER_BASE_REALTIME_SOFT
,
100 .clockid
= CLOCK_REALTIME
,
101 .get_time
= &ktime_get_real
,
104 .index
= HRTIMER_BASE_BOOTTIME_SOFT
,
105 .clockid
= CLOCK_BOOTTIME
,
106 .get_time
= &ktime_get_boottime
,
109 .index
= HRTIMER_BASE_TAI_SOFT
,
110 .clockid
= CLOCK_TAI
,
111 .get_time
= &ktime_get_clocktai
,
116 static const int hrtimer_clock_to_base_table
[MAX_CLOCKS
] = {
117 /* Make sure we catch unsupported clockids */
118 [0 ... MAX_CLOCKS
- 1] = HRTIMER_MAX_CLOCK_BASES
,
120 [CLOCK_REALTIME
] = HRTIMER_BASE_REALTIME
,
121 [CLOCK_MONOTONIC
] = HRTIMER_BASE_MONOTONIC
,
122 [CLOCK_BOOTTIME
] = HRTIMER_BASE_BOOTTIME
,
123 [CLOCK_TAI
] = HRTIMER_BASE_TAI
,
127 * Functions and macros which are different for UP/SMP systems are kept in a
133 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
134 * such that hrtimer_callback_running() can unconditionally dereference
135 * timer->base->cpu_base
137 static struct hrtimer_cpu_base migration_cpu_base
= {
138 .clock_base
= { { .cpu_base
= &migration_cpu_base
, }, },
141 #define migration_base migration_cpu_base.clock_base[0]
144 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
145 * means that all timers which are tied to this base via timer->base are
146 * locked, and the base itself is locked too.
148 * So __run_timers/migrate_timers can safely modify all timers which could
149 * be found on the lists/queues.
151 * When the timer's base is locked, and the timer removed from list, it is
152 * possible to set timer->base = &migration_base and drop the lock: the timer
156 struct hrtimer_clock_base
*lock_hrtimer_base(const struct hrtimer
*timer
,
157 unsigned long *flags
)
159 struct hrtimer_clock_base
*base
;
163 if (likely(base
!= &migration_base
)) {
164 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
165 if (likely(base
== timer
->base
))
167 /* The timer has migrated to another CPU: */
168 raw_spin_unlock_irqrestore(&base
->cpu_base
->lock
, *flags
);
175 * We do not migrate the timer when it is expiring before the next
176 * event on the target cpu. When high resolution is enabled, we cannot
177 * reprogram the target cpu hardware and we would cause it to fire
178 * late. To keep it simple, we handle the high resolution enabled and
179 * disabled case similar.
181 * Called with cpu_base->lock of target cpu held.
184 hrtimer_check_target(struct hrtimer
*timer
, struct hrtimer_clock_base
*new_base
)
188 expires
= ktime_sub(hrtimer_get_expires(timer
), new_base
->offset
);
189 return expires
< new_base
->cpu_base
->expires_next
;
193 struct hrtimer_cpu_base
*get_target_base(struct hrtimer_cpu_base
*base
,
196 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
197 if (static_branch_likely(&timers_migration_enabled
) && !pinned
)
198 return &per_cpu(hrtimer_bases
, get_nohz_timer_target());
204 * We switch the timer base to a power-optimized selected CPU target,
206 * - NO_HZ_COMMON is enabled
207 * - timer migration is enabled
208 * - the timer callback is not running
209 * - the timer is not the first expiring timer on the new target
211 * If one of the above requirements is not fulfilled we move the timer
212 * to the current CPU or leave it on the previously assigned CPU if
213 * the timer callback is currently running.
215 static inline struct hrtimer_clock_base
*
216 switch_hrtimer_base(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
,
219 struct hrtimer_cpu_base
*new_cpu_base
, *this_cpu_base
;
220 struct hrtimer_clock_base
*new_base
;
221 int basenum
= base
->index
;
223 this_cpu_base
= this_cpu_ptr(&hrtimer_bases
);
224 new_cpu_base
= get_target_base(this_cpu_base
, pinned
);
226 new_base
= &new_cpu_base
->clock_base
[basenum
];
228 if (base
!= new_base
) {
230 * We are trying to move timer to new_base.
231 * However we can't change timer's base while it is running,
232 * so we keep it on the same CPU. No hassle vs. reprogramming
233 * the event source in the high resolution case. The softirq
234 * code will take care of this when the timer function has
235 * completed. There is no conflict as we hold the lock until
236 * the timer is enqueued.
238 if (unlikely(hrtimer_callback_running(timer
)))
241 /* See the comment in lock_hrtimer_base() */
242 timer
->base
= &migration_base
;
243 raw_spin_unlock(&base
->cpu_base
->lock
);
244 raw_spin_lock(&new_base
->cpu_base
->lock
);
246 if (new_cpu_base
!= this_cpu_base
&&
247 hrtimer_check_target(timer
, new_base
)) {
248 raw_spin_unlock(&new_base
->cpu_base
->lock
);
249 raw_spin_lock(&base
->cpu_base
->lock
);
250 new_cpu_base
= this_cpu_base
;
254 timer
->base
= new_base
;
256 if (new_cpu_base
!= this_cpu_base
&&
257 hrtimer_check_target(timer
, new_base
)) {
258 new_cpu_base
= this_cpu_base
;
265 #else /* CONFIG_SMP */
267 static inline struct hrtimer_clock_base
*
268 lock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
270 struct hrtimer_clock_base
*base
= timer
->base
;
272 raw_spin_lock_irqsave(&base
->cpu_base
->lock
, *flags
);
277 # define switch_hrtimer_base(t, b, p) (b)
279 #endif /* !CONFIG_SMP */
282 * Functions for the union type storage format of ktime_t which are
283 * too large for inlining:
285 #if BITS_PER_LONG < 64
287 * Divide a ktime value by a nanosecond value
289 s64
__ktime_divns(const ktime_t kt
, s64 div
)
295 dclc
= ktime_to_ns(kt
);
296 tmp
= dclc
< 0 ? -dclc
: dclc
;
298 /* Make sure the divisor is less than 2^32: */
304 do_div(tmp
, (unsigned long) div
);
305 return dclc
< 0 ? -tmp
: tmp
;
307 EXPORT_SYMBOL_GPL(__ktime_divns
);
308 #endif /* BITS_PER_LONG >= 64 */
311 * Add two ktime values and do a safety check for overflow:
313 ktime_t
ktime_add_safe(const ktime_t lhs
, const ktime_t rhs
)
315 ktime_t res
= ktime_add_unsafe(lhs
, rhs
);
318 * We use KTIME_SEC_MAX here, the maximum timeout which we can
319 * return to user space in a timespec:
321 if (res
< 0 || res
< lhs
|| res
< rhs
)
322 res
= ktime_set(KTIME_SEC_MAX
, 0);
327 EXPORT_SYMBOL_GPL(ktime_add_safe
);
329 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
331 static struct debug_obj_descr hrtimer_debug_descr
;
333 static void *hrtimer_debug_hint(void *addr
)
335 return ((struct hrtimer
*) addr
)->function
;
339 * fixup_init is called when:
340 * - an active object is initialized
342 static bool hrtimer_fixup_init(void *addr
, enum debug_obj_state state
)
344 struct hrtimer
*timer
= addr
;
347 case ODEBUG_STATE_ACTIVE
:
348 hrtimer_cancel(timer
);
349 debug_object_init(timer
, &hrtimer_debug_descr
);
357 * fixup_activate is called when:
358 * - an active object is activated
359 * - an unknown non-static object is activated
361 static bool hrtimer_fixup_activate(void *addr
, enum debug_obj_state state
)
364 case ODEBUG_STATE_ACTIVE
:
373 * fixup_free is called when:
374 * - an active object is freed
376 static bool hrtimer_fixup_free(void *addr
, enum debug_obj_state state
)
378 struct hrtimer
*timer
= addr
;
381 case ODEBUG_STATE_ACTIVE
:
382 hrtimer_cancel(timer
);
383 debug_object_free(timer
, &hrtimer_debug_descr
);
390 static struct debug_obj_descr hrtimer_debug_descr
= {
392 .debug_hint
= hrtimer_debug_hint
,
393 .fixup_init
= hrtimer_fixup_init
,
394 .fixup_activate
= hrtimer_fixup_activate
,
395 .fixup_free
= hrtimer_fixup_free
,
398 static inline void debug_hrtimer_init(struct hrtimer
*timer
)
400 debug_object_init(timer
, &hrtimer_debug_descr
);
403 static inline void debug_hrtimer_activate(struct hrtimer
*timer
,
404 enum hrtimer_mode mode
)
406 debug_object_activate(timer
, &hrtimer_debug_descr
);
409 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
)
411 debug_object_deactivate(timer
, &hrtimer_debug_descr
);
414 static inline void debug_hrtimer_free(struct hrtimer
*timer
)
416 debug_object_free(timer
, &hrtimer_debug_descr
);
419 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
420 enum hrtimer_mode mode
);
422 void hrtimer_init_on_stack(struct hrtimer
*timer
, clockid_t clock_id
,
423 enum hrtimer_mode mode
)
425 debug_object_init_on_stack(timer
, &hrtimer_debug_descr
);
426 __hrtimer_init(timer
, clock_id
, mode
);
428 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack
);
430 void destroy_hrtimer_on_stack(struct hrtimer
*timer
)
432 debug_object_free(timer
, &hrtimer_debug_descr
);
434 EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack
);
438 static inline void debug_hrtimer_init(struct hrtimer
*timer
) { }
439 static inline void debug_hrtimer_activate(struct hrtimer
*timer
,
440 enum hrtimer_mode mode
) { }
441 static inline void debug_hrtimer_deactivate(struct hrtimer
*timer
) { }
445 debug_init(struct hrtimer
*timer
, clockid_t clockid
,
446 enum hrtimer_mode mode
)
448 debug_hrtimer_init(timer
);
449 trace_hrtimer_init(timer
, clockid
, mode
);
452 static inline void debug_activate(struct hrtimer
*timer
,
453 enum hrtimer_mode mode
)
455 debug_hrtimer_activate(timer
, mode
);
456 trace_hrtimer_start(timer
, mode
);
459 static inline void debug_deactivate(struct hrtimer
*timer
)
461 debug_hrtimer_deactivate(timer
);
462 trace_hrtimer_cancel(timer
);
465 static struct hrtimer_clock_base
*
466 __next_base(struct hrtimer_cpu_base
*cpu_base
, unsigned int *active
)
473 idx
= __ffs(*active
);
474 *active
&= ~(1U << idx
);
476 return &cpu_base
->clock_base
[idx
];
479 #define for_each_active_base(base, cpu_base, active) \
480 while ((base = __next_base((cpu_base), &(active))))
482 static ktime_t
__hrtimer_next_event_base(struct hrtimer_cpu_base
*cpu_base
,
483 const struct hrtimer
*exclude
,
485 ktime_t expires_next
)
487 struct hrtimer_clock_base
*base
;
490 for_each_active_base(base
, cpu_base
, active
) {
491 struct timerqueue_node
*next
;
492 struct hrtimer
*timer
;
494 next
= timerqueue_getnext(&base
->active
);
495 timer
= container_of(next
, struct hrtimer
, node
);
496 if (timer
== exclude
) {
497 /* Get to the next timer in the queue. */
498 next
= timerqueue_iterate_next(next
);
502 timer
= container_of(next
, struct hrtimer
, node
);
504 expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
505 if (expires
< expires_next
) {
506 expires_next
= expires
;
508 /* Skip cpu_base update if a timer is being excluded. */
513 cpu_base
->softirq_next_timer
= timer
;
515 cpu_base
->next_timer
= timer
;
519 * clock_was_set() might have changed base->offset of any of
520 * the clock bases so the result might be negative. Fix it up
521 * to prevent a false positive in clockevents_program_event().
523 if (expires_next
< 0)
529 * Recomputes cpu_base::*next_timer and returns the earliest expires_next but
530 * does not set cpu_base::*expires_next, that is done by hrtimer_reprogram.
532 * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases,
533 * those timers will get run whenever the softirq gets handled, at the end of
534 * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases.
536 * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases.
537 * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual
538 * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD.
540 * @active_mask must be one of:
541 * - HRTIMER_ACTIVE_ALL,
542 * - HRTIMER_ACTIVE_SOFT, or
543 * - HRTIMER_ACTIVE_HARD.
546 __hrtimer_get_next_event(struct hrtimer_cpu_base
*cpu_base
, unsigned int active_mask
)
549 struct hrtimer
*next_timer
= NULL
;
550 ktime_t expires_next
= KTIME_MAX
;
552 if (!cpu_base
->softirq_activated
&& (active_mask
& HRTIMER_ACTIVE_SOFT
)) {
553 active
= cpu_base
->active_bases
& HRTIMER_ACTIVE_SOFT
;
554 cpu_base
->softirq_next_timer
= NULL
;
555 expires_next
= __hrtimer_next_event_base(cpu_base
, NULL
,
558 next_timer
= cpu_base
->softirq_next_timer
;
561 if (active_mask
& HRTIMER_ACTIVE_HARD
) {
562 active
= cpu_base
->active_bases
& HRTIMER_ACTIVE_HARD
;
563 cpu_base
->next_timer
= next_timer
;
564 expires_next
= __hrtimer_next_event_base(cpu_base
, NULL
, active
,
571 static inline ktime_t
hrtimer_update_base(struct hrtimer_cpu_base
*base
)
573 ktime_t
*offs_real
= &base
->clock_base
[HRTIMER_BASE_REALTIME
].offset
;
574 ktime_t
*offs_boot
= &base
->clock_base
[HRTIMER_BASE_BOOTTIME
].offset
;
575 ktime_t
*offs_tai
= &base
->clock_base
[HRTIMER_BASE_TAI
].offset
;
577 ktime_t now
= ktime_get_update_offsets_now(&base
->clock_was_set_seq
,
578 offs_real
, offs_boot
, offs_tai
);
580 base
->clock_base
[HRTIMER_BASE_REALTIME_SOFT
].offset
= *offs_real
;
581 base
->clock_base
[HRTIMER_BASE_BOOTTIME_SOFT
].offset
= *offs_boot
;
582 base
->clock_base
[HRTIMER_BASE_TAI_SOFT
].offset
= *offs_tai
;
588 * Is the high resolution mode active ?
590 static inline int __hrtimer_hres_active(struct hrtimer_cpu_base
*cpu_base
)
592 return IS_ENABLED(CONFIG_HIGH_RES_TIMERS
) ?
593 cpu_base
->hres_active
: 0;
596 static inline int hrtimer_hres_active(void)
598 return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases
));
602 * Reprogram the event source with checking both queues for the
604 * Called with interrupts disabled and base->lock held
607 hrtimer_force_reprogram(struct hrtimer_cpu_base
*cpu_base
, int skip_equal
)
609 ktime_t expires_next
;
612 * Find the current next expiration time.
614 expires_next
= __hrtimer_get_next_event(cpu_base
, HRTIMER_ACTIVE_ALL
);
616 if (cpu_base
->next_timer
&& cpu_base
->next_timer
->is_soft
) {
618 * When the softirq is activated, hrtimer has to be
619 * programmed with the first hard hrtimer because soft
620 * timer interrupt could occur too late.
622 if (cpu_base
->softirq_activated
)
623 expires_next
= __hrtimer_get_next_event(cpu_base
,
624 HRTIMER_ACTIVE_HARD
);
626 cpu_base
->softirq_expires_next
= expires_next
;
629 if (skip_equal
&& expires_next
== cpu_base
->expires_next
)
632 cpu_base
->expires_next
= expires_next
;
635 * If hres is not active, hardware does not have to be
638 * If a hang was detected in the last timer interrupt then we
639 * leave the hang delay active in the hardware. We want the
640 * system to make progress. That also prevents the following
642 * T1 expires 50ms from now
643 * T2 expires 5s from now
645 * T1 is removed, so this code is called and would reprogram
646 * the hardware to 5s from now. Any hrtimer_start after that
647 * will not reprogram the hardware due to hang_detected being
648 * set. So we'd effectivly block all timers until the T2 event
651 if (!__hrtimer_hres_active(cpu_base
) || cpu_base
->hang_detected
)
654 tick_program_event(cpu_base
->expires_next
, 1);
657 /* High resolution timer related functions */
658 #ifdef CONFIG_HIGH_RES_TIMERS
661 * High resolution timer enabled ?
663 static bool hrtimer_hres_enabled __read_mostly
= true;
664 unsigned int hrtimer_resolution __read_mostly
= LOW_RES_NSEC
;
665 EXPORT_SYMBOL_GPL(hrtimer_resolution
);
668 * Enable / Disable high resolution mode
670 static int __init
setup_hrtimer_hres(char *str
)
672 return (kstrtobool(str
, &hrtimer_hres_enabled
) == 0);
675 __setup("highres=", setup_hrtimer_hres
);
678 * hrtimer_high_res_enabled - query, if the highres mode is enabled
680 static inline int hrtimer_is_hres_enabled(void)
682 return hrtimer_hres_enabled
;
686 * Retrigger next event is called after clock was set
688 * Called with interrupts disabled via on_each_cpu()
690 static void retrigger_next_event(void *arg
)
692 struct hrtimer_cpu_base
*base
= this_cpu_ptr(&hrtimer_bases
);
694 if (!__hrtimer_hres_active(base
))
697 raw_spin_lock(&base
->lock
);
698 hrtimer_update_base(base
);
699 hrtimer_force_reprogram(base
, 0);
700 raw_spin_unlock(&base
->lock
);
704 * Switch to high resolution mode
706 static void hrtimer_switch_to_hres(void)
708 struct hrtimer_cpu_base
*base
= this_cpu_ptr(&hrtimer_bases
);
710 if (tick_init_highres()) {
711 pr_warn("Could not switch to high resolution mode on CPU %u\n",
715 base
->hres_active
= 1;
716 hrtimer_resolution
= HIGH_RES_NSEC
;
718 tick_setup_sched_timer();
719 /* "Retrigger" the interrupt to get things going */
720 retrigger_next_event(NULL
);
723 static void clock_was_set_work(struct work_struct
*work
)
728 static DECLARE_WORK(hrtimer_work
, clock_was_set_work
);
731 * Called from timekeeping and resume code to reprogram the hrtimer
732 * interrupt device on all cpus.
734 void clock_was_set_delayed(void)
736 schedule_work(&hrtimer_work
);
741 static inline int hrtimer_is_hres_enabled(void) { return 0; }
742 static inline void hrtimer_switch_to_hres(void) { }
743 static inline void retrigger_next_event(void *arg
) { }
745 #endif /* CONFIG_HIGH_RES_TIMERS */
748 * When a timer is enqueued and expires earlier than the already enqueued
749 * timers, we have to check, whether it expires earlier than the timer for
750 * which the clock event device was armed.
752 * Called with interrupts disabled and base->cpu_base.lock held
754 static void hrtimer_reprogram(struct hrtimer
*timer
, bool reprogram
)
756 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
757 struct hrtimer_clock_base
*base
= timer
->base
;
758 ktime_t expires
= ktime_sub(hrtimer_get_expires(timer
), base
->offset
);
760 WARN_ON_ONCE(hrtimer_get_expires_tv64(timer
) < 0);
763 * CLOCK_REALTIME timer might be requested with an absolute
764 * expiry time which is less than base->offset. Set it to 0.
769 if (timer
->is_soft
) {
771 * soft hrtimer could be started on a remote CPU. In this
772 * case softirq_expires_next needs to be updated on the
773 * remote CPU. The soft hrtimer will not expire before the
774 * first hard hrtimer on the remote CPU -
775 * hrtimer_check_target() prevents this case.
777 struct hrtimer_cpu_base
*timer_cpu_base
= base
->cpu_base
;
779 if (timer_cpu_base
->softirq_activated
)
782 if (!ktime_before(expires
, timer_cpu_base
->softirq_expires_next
))
785 timer_cpu_base
->softirq_next_timer
= timer
;
786 timer_cpu_base
->softirq_expires_next
= expires
;
788 if (!ktime_before(expires
, timer_cpu_base
->expires_next
) ||
794 * If the timer is not on the current cpu, we cannot reprogram
795 * the other cpus clock event device.
797 if (base
->cpu_base
!= cpu_base
)
801 * If the hrtimer interrupt is running, then it will
802 * reevaluate the clock bases and reprogram the clock event
803 * device. The callbacks are always executed in hard interrupt
804 * context so we don't need an extra check for a running
807 if (cpu_base
->in_hrtirq
)
810 if (expires
>= cpu_base
->expires_next
)
813 /* Update the pointer to the next expiring timer */
814 cpu_base
->next_timer
= timer
;
815 cpu_base
->expires_next
= expires
;
818 * If hres is not active, hardware does not have to be
821 * If a hang was detected in the last timer interrupt then we
822 * do not schedule a timer which is earlier than the expiry
823 * which we enforced in the hang detection. We want the system
826 if (!__hrtimer_hres_active(cpu_base
) || cpu_base
->hang_detected
)
830 * Program the timer hardware. We enforce the expiry for
831 * events which are already in the past.
833 tick_program_event(expires
, 1);
837 * Clock realtime was set
839 * Change the offset of the realtime clock vs. the monotonic
842 * We might have to reprogram the high resolution timer interrupt. On
843 * SMP we call the architecture specific code to retrigger _all_ high
844 * resolution timer interrupts. On UP we just disable interrupts and
845 * call the high resolution interrupt code.
847 void clock_was_set(void)
849 #ifdef CONFIG_HIGH_RES_TIMERS
850 /* Retrigger the CPU local events everywhere */
851 on_each_cpu(retrigger_next_event
, NULL
, 1);
853 timerfd_clock_was_set();
857 * During resume we might have to reprogram the high resolution timer
858 * interrupt on all online CPUs. However, all other CPUs will be
859 * stopped with IRQs interrupts disabled so the clock_was_set() call
862 void hrtimers_resume(void)
864 lockdep_assert_irqs_disabled();
865 /* Retrigger on the local CPU */
866 retrigger_next_event(NULL
);
867 /* And schedule a retrigger for all others */
868 clock_was_set_delayed();
872 * Counterpart to lock_hrtimer_base above:
875 void unlock_hrtimer_base(const struct hrtimer
*timer
, unsigned long *flags
)
877 raw_spin_unlock_irqrestore(&timer
->base
->cpu_base
->lock
, *flags
);
881 * hrtimer_forward - forward the timer expiry
882 * @timer: hrtimer to forward
883 * @now: forward past this time
884 * @interval: the interval to forward
886 * Forward the timer expiry so it will expire in the future.
887 * Returns the number of overruns.
889 * Can be safely called from the callback function of @timer. If
890 * called from other contexts @timer must neither be enqueued nor
891 * running the callback and the caller needs to take care of
894 * Note: This only updates the timer expiry value and does not requeue
897 u64
hrtimer_forward(struct hrtimer
*timer
, ktime_t now
, ktime_t interval
)
902 delta
= ktime_sub(now
, hrtimer_get_expires(timer
));
907 if (WARN_ON(timer
->state
& HRTIMER_STATE_ENQUEUED
))
910 if (interval
< hrtimer_resolution
)
911 interval
= hrtimer_resolution
;
913 if (unlikely(delta
>= interval
)) {
914 s64 incr
= ktime_to_ns(interval
);
916 orun
= ktime_divns(delta
, incr
);
917 hrtimer_add_expires_ns(timer
, incr
* orun
);
918 if (hrtimer_get_expires_tv64(timer
) > now
)
921 * This (and the ktime_add() below) is the
922 * correction for exact:
926 hrtimer_add_expires(timer
, interval
);
930 EXPORT_SYMBOL_GPL(hrtimer_forward
);
933 * enqueue_hrtimer - internal function to (re)start a timer
935 * The timer is inserted in expiry order. Insertion into the
936 * red black tree is O(log(n)). Must hold the base lock.
938 * Returns 1 when the new timer is the leftmost timer in the tree.
940 static int enqueue_hrtimer(struct hrtimer
*timer
,
941 struct hrtimer_clock_base
*base
,
942 enum hrtimer_mode mode
)
944 debug_activate(timer
, mode
);
946 base
->cpu_base
->active_bases
|= 1 << base
->index
;
948 /* Pairs with the lockless read in hrtimer_is_queued() */
949 WRITE_ONCE(timer
->state
, HRTIMER_STATE_ENQUEUED
);
951 return timerqueue_add(&base
->active
, &timer
->node
);
955 * __remove_hrtimer - internal function to remove a timer
957 * Caller must hold the base lock.
959 * High resolution timer mode reprograms the clock event device when the
960 * timer is the one which expires next. The caller can disable this by setting
961 * reprogram to zero. This is useful, when the context does a reprogramming
962 * anyway (e.g. timer interrupt)
964 static void __remove_hrtimer(struct hrtimer
*timer
,
965 struct hrtimer_clock_base
*base
,
966 u8 newstate
, int reprogram
)
968 struct hrtimer_cpu_base
*cpu_base
= base
->cpu_base
;
969 u8 state
= timer
->state
;
971 /* Pairs with the lockless read in hrtimer_is_queued() */
972 WRITE_ONCE(timer
->state
, newstate
);
973 if (!(state
& HRTIMER_STATE_ENQUEUED
))
976 if (!timerqueue_del(&base
->active
, &timer
->node
))
977 cpu_base
->active_bases
&= ~(1 << base
->index
);
980 * Note: If reprogram is false we do not update
981 * cpu_base->next_timer. This happens when we remove the first
982 * timer on a remote cpu. No harm as we never dereference
983 * cpu_base->next_timer. So the worst thing what can happen is
984 * an superflous call to hrtimer_force_reprogram() on the
985 * remote cpu later on if the same timer gets enqueued again.
987 if (reprogram
&& timer
== cpu_base
->next_timer
)
988 hrtimer_force_reprogram(cpu_base
, 1);
992 * remove hrtimer, called with base lock held
995 remove_hrtimer(struct hrtimer
*timer
, struct hrtimer_clock_base
*base
, bool restart
)
997 u8 state
= timer
->state
;
999 if (state
& HRTIMER_STATE_ENQUEUED
) {
1003 * Remove the timer and force reprogramming when high
1004 * resolution mode is active and the timer is on the current
1005 * CPU. If we remove a timer on another CPU, reprogramming is
1006 * skipped. The interrupt event on this CPU is fired and
1007 * reprogramming happens in the interrupt handler. This is a
1008 * rare case and less expensive than a smp call.
1010 debug_deactivate(timer
);
1011 reprogram
= base
->cpu_base
== this_cpu_ptr(&hrtimer_bases
);
1014 state
= HRTIMER_STATE_INACTIVE
;
1016 __remove_hrtimer(timer
, base
, state
, reprogram
);
1022 static inline ktime_t
hrtimer_update_lowres(struct hrtimer
*timer
, ktime_t tim
,
1023 const enum hrtimer_mode mode
)
1025 #ifdef CONFIG_TIME_LOW_RES
1027 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1028 * granular time values. For relative timers we add hrtimer_resolution
1029 * (i.e. one jiffie) to prevent short timeouts.
1031 timer
->is_rel
= mode
& HRTIMER_MODE_REL
;
1033 tim
= ktime_add_safe(tim
, hrtimer_resolution
);
1039 hrtimer_update_softirq_timer(struct hrtimer_cpu_base
*cpu_base
, bool reprogram
)
1044 * Find the next SOFT expiration.
1046 expires
= __hrtimer_get_next_event(cpu_base
, HRTIMER_ACTIVE_SOFT
);
1049 * reprogramming needs to be triggered, even if the next soft
1050 * hrtimer expires at the same time than the next hard
1051 * hrtimer. cpu_base->softirq_expires_next needs to be updated!
1053 if (expires
== KTIME_MAX
)
1057 * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event()
1058 * cpu_base->*expires_next is only set by hrtimer_reprogram()
1060 hrtimer_reprogram(cpu_base
->softirq_next_timer
, reprogram
);
1063 static int __hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
1064 u64 delta_ns
, const enum hrtimer_mode mode
,
1065 struct hrtimer_clock_base
*base
)
1067 struct hrtimer_clock_base
*new_base
;
1069 /* Remove an active timer from the queue: */
1070 remove_hrtimer(timer
, base
, true);
1072 if (mode
& HRTIMER_MODE_REL
)
1073 tim
= ktime_add_safe(tim
, base
->get_time());
1075 tim
= hrtimer_update_lowres(timer
, tim
, mode
);
1077 hrtimer_set_expires_range_ns(timer
, tim
, delta_ns
);
1079 /* Switch the timer base, if necessary: */
1080 new_base
= switch_hrtimer_base(timer
, base
, mode
& HRTIMER_MODE_PINNED
);
1082 return enqueue_hrtimer(timer
, new_base
, mode
);
1086 * hrtimer_start_range_ns - (re)start an hrtimer
1087 * @timer: the timer to be added
1089 * @delta_ns: "slack" range for the timer
1090 * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or
1091 * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);
1092 * softirq based mode is considered for debug purpose only!
1094 void hrtimer_start_range_ns(struct hrtimer
*timer
, ktime_t tim
,
1095 u64 delta_ns
, const enum hrtimer_mode mode
)
1097 struct hrtimer_clock_base
*base
;
1098 unsigned long flags
;
1101 * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
1104 WARN_ON_ONCE(!(mode
& HRTIMER_MODE_SOFT
) ^ !timer
->is_soft
);
1106 base
= lock_hrtimer_base(timer
, &flags
);
1108 if (__hrtimer_start_range_ns(timer
, tim
, delta_ns
, mode
, base
))
1109 hrtimer_reprogram(timer
, true);
1111 unlock_hrtimer_base(timer
, &flags
);
1113 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns
);
1116 * hrtimer_try_to_cancel - try to deactivate a timer
1117 * @timer: hrtimer to stop
1121 * * 0 when the timer was not active
1122 * * 1 when the timer was active
1123 * * -1 when the timer is currently executing the callback function and
1126 int hrtimer_try_to_cancel(struct hrtimer
*timer
)
1128 struct hrtimer_clock_base
*base
;
1129 unsigned long flags
;
1133 * Check lockless first. If the timer is not active (neither
1134 * enqueued nor running the callback, nothing to do here. The
1135 * base lock does not serialize against a concurrent enqueue,
1136 * so we can avoid taking it.
1138 if (!hrtimer_active(timer
))
1141 base
= lock_hrtimer_base(timer
, &flags
);
1143 if (!hrtimer_callback_running(timer
))
1144 ret
= remove_hrtimer(timer
, base
, false);
1146 unlock_hrtimer_base(timer
, &flags
);
1151 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel
);
1154 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1155 * @timer: the timer to be cancelled
1158 * 0 when the timer was not active
1159 * 1 when the timer was active
1161 int hrtimer_cancel(struct hrtimer
*timer
)
1164 int ret
= hrtimer_try_to_cancel(timer
);
1171 EXPORT_SYMBOL_GPL(hrtimer_cancel
);
1174 * hrtimer_get_remaining - get remaining time for the timer
1175 * @timer: the timer to read
1176 * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y
1178 ktime_t
__hrtimer_get_remaining(const struct hrtimer
*timer
, bool adjust
)
1180 unsigned long flags
;
1183 lock_hrtimer_base(timer
, &flags
);
1184 if (IS_ENABLED(CONFIG_TIME_LOW_RES
) && adjust
)
1185 rem
= hrtimer_expires_remaining_adjusted(timer
);
1187 rem
= hrtimer_expires_remaining(timer
);
1188 unlock_hrtimer_base(timer
, &flags
);
1192 EXPORT_SYMBOL_GPL(__hrtimer_get_remaining
);
1194 #ifdef CONFIG_NO_HZ_COMMON
1196 * hrtimer_get_next_event - get the time until next expiry event
1198 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1200 u64
hrtimer_get_next_event(void)
1202 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1203 u64 expires
= KTIME_MAX
;
1204 unsigned long flags
;
1206 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1208 if (!__hrtimer_hres_active(cpu_base
))
1209 expires
= __hrtimer_get_next_event(cpu_base
, HRTIMER_ACTIVE_ALL
);
1211 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1217 * hrtimer_next_event_without - time until next expiry event w/o one timer
1218 * @exclude: timer to exclude
1220 * Returns the next expiry time over all timers except for the @exclude one or
1221 * KTIME_MAX if none of them is pending.
1223 u64
hrtimer_next_event_without(const struct hrtimer
*exclude
)
1225 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1226 u64 expires
= KTIME_MAX
;
1227 unsigned long flags
;
1229 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1231 if (__hrtimer_hres_active(cpu_base
)) {
1232 unsigned int active
;
1234 if (!cpu_base
->softirq_activated
) {
1235 active
= cpu_base
->active_bases
& HRTIMER_ACTIVE_SOFT
;
1236 expires
= __hrtimer_next_event_base(cpu_base
, exclude
,
1239 active
= cpu_base
->active_bases
& HRTIMER_ACTIVE_HARD
;
1240 expires
= __hrtimer_next_event_base(cpu_base
, exclude
, active
,
1244 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1250 static inline int hrtimer_clockid_to_base(clockid_t clock_id
)
1252 if (likely(clock_id
< MAX_CLOCKS
)) {
1253 int base
= hrtimer_clock_to_base_table
[clock_id
];
1255 if (likely(base
!= HRTIMER_MAX_CLOCK_BASES
))
1258 WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id
);
1259 return HRTIMER_BASE_MONOTONIC
;
1262 static void __hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1263 enum hrtimer_mode mode
)
1265 bool softtimer
= !!(mode
& HRTIMER_MODE_SOFT
);
1266 int base
= softtimer
? HRTIMER_MAX_CLOCK_BASES
/ 2 : 0;
1267 struct hrtimer_cpu_base
*cpu_base
;
1269 memset(timer
, 0, sizeof(struct hrtimer
));
1271 cpu_base
= raw_cpu_ptr(&hrtimer_bases
);
1274 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1275 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1276 * ensure POSIX compliance.
1278 if (clock_id
== CLOCK_REALTIME
&& mode
& HRTIMER_MODE_REL
)
1279 clock_id
= CLOCK_MONOTONIC
;
1281 base
+= hrtimer_clockid_to_base(clock_id
);
1282 timer
->is_soft
= softtimer
;
1283 timer
->base
= &cpu_base
->clock_base
[base
];
1284 timerqueue_init(&timer
->node
);
1288 * hrtimer_init - initialize a timer to the given clock
1289 * @timer: the timer to be initialized
1290 * @clock_id: the clock to be used
1291 * @mode: The modes which are relevant for intitialization:
1292 * HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT,
1293 * HRTIMER_MODE_REL_SOFT
1295 * The PINNED variants of the above can be handed in,
1296 * but the PINNED bit is ignored as pinning happens
1297 * when the hrtimer is started
1299 void hrtimer_init(struct hrtimer
*timer
, clockid_t clock_id
,
1300 enum hrtimer_mode mode
)
1302 debug_init(timer
, clock_id
, mode
);
1303 __hrtimer_init(timer
, clock_id
, mode
);
1305 EXPORT_SYMBOL_GPL(hrtimer_init
);
1308 * A timer is active, when it is enqueued into the rbtree or the
1309 * callback function is running or it's in the state of being migrated
1312 * It is important for this function to not return a false negative.
1314 bool hrtimer_active(const struct hrtimer
*timer
)
1316 struct hrtimer_clock_base
*base
;
1320 base
= READ_ONCE(timer
->base
);
1321 seq
= raw_read_seqcount_begin(&base
->seq
);
1323 if (timer
->state
!= HRTIMER_STATE_INACTIVE
||
1324 base
->running
== timer
)
1327 } while (read_seqcount_retry(&base
->seq
, seq
) ||
1328 base
!= READ_ONCE(timer
->base
));
1332 EXPORT_SYMBOL_GPL(hrtimer_active
);
1335 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1336 * distinct sections:
1338 * - queued: the timer is queued
1339 * - callback: the timer is being ran
1340 * - post: the timer is inactive or (re)queued
1342 * On the read side we ensure we observe timer->state and cpu_base->running
1343 * from the same section, if anything changed while we looked at it, we retry.
1344 * This includes timer->base changing because sequence numbers alone are
1345 * insufficient for that.
1347 * The sequence numbers are required because otherwise we could still observe
1348 * a false negative if the read side got smeared over multiple consequtive
1349 * __run_hrtimer() invocations.
1352 static void __run_hrtimer(struct hrtimer_cpu_base
*cpu_base
,
1353 struct hrtimer_clock_base
*base
,
1354 struct hrtimer
*timer
, ktime_t
*now
,
1355 unsigned long flags
)
1357 enum hrtimer_restart (*fn
)(struct hrtimer
*);
1360 lockdep_assert_held(&cpu_base
->lock
);
1362 debug_deactivate(timer
);
1363 base
->running
= timer
;
1366 * Separate the ->running assignment from the ->state assignment.
1368 * As with a regular write barrier, this ensures the read side in
1369 * hrtimer_active() cannot observe base->running == NULL &&
1370 * timer->state == INACTIVE.
1372 raw_write_seqcount_barrier(&base
->seq
);
1374 __remove_hrtimer(timer
, base
, HRTIMER_STATE_INACTIVE
, 0);
1375 fn
= timer
->function
;
1378 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1379 * timer is restarted with a period then it becomes an absolute
1380 * timer. If its not restarted it does not matter.
1382 if (IS_ENABLED(CONFIG_TIME_LOW_RES
))
1383 timer
->is_rel
= false;
1386 * The timer is marked as running in the CPU base, so it is
1387 * protected against migration to a different CPU even if the lock
1390 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1391 trace_hrtimer_expire_entry(timer
, now
);
1392 restart
= fn(timer
);
1393 trace_hrtimer_expire_exit(timer
);
1394 raw_spin_lock_irq(&cpu_base
->lock
);
1397 * Note: We clear the running state after enqueue_hrtimer and
1398 * we do not reprogram the event hardware. Happens either in
1399 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1401 * Note: Because we dropped the cpu_base->lock above,
1402 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1405 if (restart
!= HRTIMER_NORESTART
&&
1406 !(timer
->state
& HRTIMER_STATE_ENQUEUED
))
1407 enqueue_hrtimer(timer
, base
, HRTIMER_MODE_ABS
);
1410 * Separate the ->running assignment from the ->state assignment.
1412 * As with a regular write barrier, this ensures the read side in
1413 * hrtimer_active() cannot observe base->running.timer == NULL &&
1414 * timer->state == INACTIVE.
1416 raw_write_seqcount_barrier(&base
->seq
);
1418 WARN_ON_ONCE(base
->running
!= timer
);
1419 base
->running
= NULL
;
1422 static void __hrtimer_run_queues(struct hrtimer_cpu_base
*cpu_base
, ktime_t now
,
1423 unsigned long flags
, unsigned int active_mask
)
1425 struct hrtimer_clock_base
*base
;
1426 unsigned int active
= cpu_base
->active_bases
& active_mask
;
1428 for_each_active_base(base
, cpu_base
, active
) {
1429 struct timerqueue_node
*node
;
1432 basenow
= ktime_add(now
, base
->offset
);
1434 while ((node
= timerqueue_getnext(&base
->active
))) {
1435 struct hrtimer
*timer
;
1437 timer
= container_of(node
, struct hrtimer
, node
);
1440 * The immediate goal for using the softexpires is
1441 * minimizing wakeups, not running timers at the
1442 * earliest interrupt after their soft expiration.
1443 * This allows us to avoid using a Priority Search
1444 * Tree, which can answer a stabbing querry for
1445 * overlapping intervals and instead use the simple
1446 * BST we already have.
1447 * We don't add extra wakeups by delaying timers that
1448 * are right-of a not yet expired timer, because that
1449 * timer will have to trigger a wakeup anyway.
1451 if (basenow
< hrtimer_get_softexpires_tv64(timer
))
1454 __run_hrtimer(cpu_base
, base
, timer
, &basenow
, flags
);
1459 static __latent_entropy
void hrtimer_run_softirq(struct softirq_action
*h
)
1461 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1462 unsigned long flags
;
1465 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1467 now
= hrtimer_update_base(cpu_base
);
1468 __hrtimer_run_queues(cpu_base
, now
, flags
, HRTIMER_ACTIVE_SOFT
);
1470 cpu_base
->softirq_activated
= 0;
1471 hrtimer_update_softirq_timer(cpu_base
, true);
1473 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1476 #ifdef CONFIG_HIGH_RES_TIMERS
1479 * High resolution timer interrupt
1480 * Called with interrupts disabled
1482 void hrtimer_interrupt(struct clock_event_device
*dev
)
1484 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1485 ktime_t expires_next
, now
, entry_time
, delta
;
1486 unsigned long flags
;
1489 BUG_ON(!cpu_base
->hres_active
);
1490 cpu_base
->nr_events
++;
1491 dev
->next_event
= KTIME_MAX
;
1493 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1494 entry_time
= now
= hrtimer_update_base(cpu_base
);
1496 cpu_base
->in_hrtirq
= 1;
1498 * We set expires_next to KTIME_MAX here with cpu_base->lock
1499 * held to prevent that a timer is enqueued in our queue via
1500 * the migration code. This does not affect enqueueing of
1501 * timers which run their callback and need to be requeued on
1504 cpu_base
->expires_next
= KTIME_MAX
;
1506 if (!ktime_before(now
, cpu_base
->softirq_expires_next
)) {
1507 cpu_base
->softirq_expires_next
= KTIME_MAX
;
1508 cpu_base
->softirq_activated
= 1;
1509 raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
1512 __hrtimer_run_queues(cpu_base
, now
, flags
, HRTIMER_ACTIVE_HARD
);
1514 /* Reevaluate the clock bases for the next expiry */
1515 expires_next
= __hrtimer_get_next_event(cpu_base
, HRTIMER_ACTIVE_ALL
);
1517 * Store the new expiry value so the migration code can verify
1520 cpu_base
->expires_next
= expires_next
;
1521 cpu_base
->in_hrtirq
= 0;
1522 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1524 /* Reprogramming necessary ? */
1525 if (!tick_program_event(expires_next
, 0)) {
1526 cpu_base
->hang_detected
= 0;
1531 * The next timer was already expired due to:
1533 * - long lasting callbacks
1534 * - being scheduled away when running in a VM
1536 * We need to prevent that we loop forever in the hrtimer
1537 * interrupt routine. We give it 3 attempts to avoid
1538 * overreacting on some spurious event.
1540 * Acquire base lock for updating the offsets and retrieving
1543 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1544 now
= hrtimer_update_base(cpu_base
);
1545 cpu_base
->nr_retries
++;
1549 * Give the system a chance to do something else than looping
1550 * here. We stored the entry time, so we know exactly how long
1551 * we spent here. We schedule the next event this amount of
1554 cpu_base
->nr_hangs
++;
1555 cpu_base
->hang_detected
= 1;
1556 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1558 delta
= ktime_sub(now
, entry_time
);
1559 if ((unsigned int)delta
> cpu_base
->max_hang_time
)
1560 cpu_base
->max_hang_time
= (unsigned int) delta
;
1562 * Limit it to a sensible value as we enforce a longer
1563 * delay. Give the CPU at least 100ms to catch up.
1565 if (delta
> 100 * NSEC_PER_MSEC
)
1566 expires_next
= ktime_add_ns(now
, 100 * NSEC_PER_MSEC
);
1568 expires_next
= ktime_add(now
, delta
);
1569 tick_program_event(expires_next
, 1);
1570 pr_warn_once("hrtimer: interrupt took %llu ns\n", ktime_to_ns(delta
));
1573 /* called with interrupts disabled */
1574 static inline void __hrtimer_peek_ahead_timers(void)
1576 struct tick_device
*td
;
1578 if (!hrtimer_hres_active())
1581 td
= this_cpu_ptr(&tick_cpu_device
);
1582 if (td
&& td
->evtdev
)
1583 hrtimer_interrupt(td
->evtdev
);
1586 #else /* CONFIG_HIGH_RES_TIMERS */
1588 static inline void __hrtimer_peek_ahead_timers(void) { }
1590 #endif /* !CONFIG_HIGH_RES_TIMERS */
1593 * Called from run_local_timers in hardirq context every jiffy
1595 void hrtimer_run_queues(void)
1597 struct hrtimer_cpu_base
*cpu_base
= this_cpu_ptr(&hrtimer_bases
);
1598 unsigned long flags
;
1601 if (__hrtimer_hres_active(cpu_base
))
1605 * This _is_ ugly: We have to check periodically, whether we
1606 * can switch to highres and / or nohz mode. The clocksource
1607 * switch happens with xtime_lock held. Notification from
1608 * there only sets the check bit in the tick_oneshot code,
1609 * otherwise we might deadlock vs. xtime_lock.
1611 if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1612 hrtimer_switch_to_hres();
1616 raw_spin_lock_irqsave(&cpu_base
->lock
, flags
);
1617 now
= hrtimer_update_base(cpu_base
);
1619 if (!ktime_before(now
, cpu_base
->softirq_expires_next
)) {
1620 cpu_base
->softirq_expires_next
= KTIME_MAX
;
1621 cpu_base
->softirq_activated
= 1;
1622 raise_softirq_irqoff(HRTIMER_SOFTIRQ
);
1625 __hrtimer_run_queues(cpu_base
, now
, flags
, HRTIMER_ACTIVE_HARD
);
1626 raw_spin_unlock_irqrestore(&cpu_base
->lock
, flags
);
1630 * Sleep related functions:
1632 static enum hrtimer_restart
hrtimer_wakeup(struct hrtimer
*timer
)
1634 struct hrtimer_sleeper
*t
=
1635 container_of(timer
, struct hrtimer_sleeper
, timer
);
1636 struct task_struct
*task
= t
->task
;
1640 wake_up_process(task
);
1642 return HRTIMER_NORESTART
;
1645 void hrtimer_init_sleeper(struct hrtimer_sleeper
*sl
, struct task_struct
*task
)
1647 sl
->timer
.function
= hrtimer_wakeup
;
1650 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper
);
1652 int nanosleep_copyout(struct restart_block
*restart
, struct timespec64
*ts
)
1654 switch(restart
->nanosleep
.type
) {
1655 #ifdef CONFIG_COMPAT_32BIT_TIME
1657 if (put_old_timespec32(ts
, restart
->nanosleep
.compat_rmtp
))
1662 if (put_timespec64(ts
, restart
->nanosleep
.rmtp
))
1668 return -ERESTART_RESTARTBLOCK
;
1671 static int __sched
do_nanosleep(struct hrtimer_sleeper
*t
, enum hrtimer_mode mode
)
1673 struct restart_block
*restart
;
1675 hrtimer_init_sleeper(t
, current
);
1678 set_current_state(TASK_INTERRUPTIBLE
);
1679 hrtimer_start_expires(&t
->timer
, mode
);
1681 if (likely(t
->task
))
1682 freezable_schedule();
1684 hrtimer_cancel(&t
->timer
);
1685 mode
= HRTIMER_MODE_ABS
;
1687 } while (t
->task
&& !signal_pending(current
));
1689 __set_current_state(TASK_RUNNING
);
1694 restart
= ¤t
->restart_block
;
1695 if (restart
->nanosleep
.type
!= TT_NONE
) {
1696 ktime_t rem
= hrtimer_expires_remaining(&t
->timer
);
1697 struct timespec64 rmt
;
1701 rmt
= ktime_to_timespec64(rem
);
1703 return nanosleep_copyout(restart
, &rmt
);
1705 return -ERESTART_RESTARTBLOCK
;
1708 static long __sched
hrtimer_nanosleep_restart(struct restart_block
*restart
)
1710 struct hrtimer_sleeper t
;
1713 hrtimer_init_on_stack(&t
.timer
, restart
->nanosleep
.clockid
,
1715 hrtimer_set_expires_tv64(&t
.timer
, restart
->nanosleep
.expires
);
1717 ret
= do_nanosleep(&t
, HRTIMER_MODE_ABS
);
1718 destroy_hrtimer_on_stack(&t
.timer
);
1722 long hrtimer_nanosleep(const struct timespec64
*rqtp
,
1723 const enum hrtimer_mode mode
, const clockid_t clockid
)
1725 struct restart_block
*restart
;
1726 struct hrtimer_sleeper t
;
1730 slack
= current
->timer_slack_ns
;
1731 if (dl_task(current
) || rt_task(current
))
1734 hrtimer_init_on_stack(&t
.timer
, clockid
, mode
);
1735 hrtimer_set_expires_range_ns(&t
.timer
, timespec64_to_ktime(*rqtp
), slack
);
1736 ret
= do_nanosleep(&t
, mode
);
1737 if (ret
!= -ERESTART_RESTARTBLOCK
)
1740 /* Absolute timers do not update the rmtp value and restart: */
1741 if (mode
== HRTIMER_MODE_ABS
) {
1742 ret
= -ERESTARTNOHAND
;
1746 restart
= ¤t
->restart_block
;
1747 restart
->fn
= hrtimer_nanosleep_restart
;
1748 restart
->nanosleep
.clockid
= t
.timer
.base
->clockid
;
1749 restart
->nanosleep
.expires
= hrtimer_get_expires_tv64(&t
.timer
);
1751 destroy_hrtimer_on_stack(&t
.timer
);
1755 #if !defined(CONFIG_64BIT_TIME) || defined(CONFIG_64BIT)
1757 SYSCALL_DEFINE2(nanosleep
, struct __kernel_timespec __user
*, rqtp
,
1758 struct __kernel_timespec __user
*, rmtp
)
1760 struct timespec64 tu
;
1762 if (get_timespec64(&tu
, rqtp
))
1765 if (!timespec64_valid(&tu
))
1768 current
->restart_block
.nanosleep
.type
= rmtp
? TT_NATIVE
: TT_NONE
;
1769 current
->restart_block
.nanosleep
.rmtp
= rmtp
;
1770 return hrtimer_nanosleep(&tu
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1775 #ifdef CONFIG_COMPAT_32BIT_TIME
1777 SYSCALL_DEFINE2(nanosleep_time32
, struct old_timespec32 __user
*, rqtp
,
1778 struct old_timespec32 __user
*, rmtp
)
1780 struct timespec64 tu
;
1782 if (get_old_timespec32(&tu
, rqtp
))
1785 if (!timespec64_valid(&tu
))
1788 current
->restart_block
.nanosleep
.type
= rmtp
? TT_COMPAT
: TT_NONE
;
1789 current
->restart_block
.nanosleep
.compat_rmtp
= rmtp
;
1790 return hrtimer_nanosleep(&tu
, HRTIMER_MODE_REL
, CLOCK_MONOTONIC
);
1795 * Functions related to boot-time initialization:
1797 int hrtimers_prepare_cpu(unsigned int cpu
)
1799 struct hrtimer_cpu_base
*cpu_base
= &per_cpu(hrtimer_bases
, cpu
);
1802 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1803 cpu_base
->clock_base
[i
].cpu_base
= cpu_base
;
1804 timerqueue_init_head(&cpu_base
->clock_base
[i
].active
);
1807 cpu_base
->cpu
= cpu
;
1808 cpu_base
->active_bases
= 0;
1809 cpu_base
->hres_active
= 0;
1810 cpu_base
->hang_detected
= 0;
1811 cpu_base
->next_timer
= NULL
;
1812 cpu_base
->softirq_next_timer
= NULL
;
1813 cpu_base
->expires_next
= KTIME_MAX
;
1814 cpu_base
->softirq_expires_next
= KTIME_MAX
;
1818 #ifdef CONFIG_HOTPLUG_CPU
1820 static void migrate_hrtimer_list(struct hrtimer_clock_base
*old_base
,
1821 struct hrtimer_clock_base
*new_base
)
1823 struct hrtimer
*timer
;
1824 struct timerqueue_node
*node
;
1826 while ((node
= timerqueue_getnext(&old_base
->active
))) {
1827 timer
= container_of(node
, struct hrtimer
, node
);
1828 BUG_ON(hrtimer_callback_running(timer
));
1829 debug_deactivate(timer
);
1832 * Mark it as ENQUEUED not INACTIVE otherwise the
1833 * timer could be seen as !active and just vanish away
1834 * under us on another CPU
1836 __remove_hrtimer(timer
, old_base
, HRTIMER_STATE_ENQUEUED
, 0);
1837 timer
->base
= new_base
;
1839 * Enqueue the timers on the new cpu. This does not
1840 * reprogram the event device in case the timer
1841 * expires before the earliest on this CPU, but we run
1842 * hrtimer_interrupt after we migrated everything to
1843 * sort out already expired timers and reprogram the
1846 enqueue_hrtimer(timer
, new_base
, HRTIMER_MODE_ABS
);
1850 int hrtimers_dead_cpu(unsigned int scpu
)
1852 struct hrtimer_cpu_base
*old_base
, *new_base
;
1855 BUG_ON(cpu_online(scpu
));
1856 tick_cancel_sched_timer(scpu
);
1859 * this BH disable ensures that raise_softirq_irqoff() does
1860 * not wakeup ksoftirqd (and acquire the pi-lock) while
1861 * holding the cpu_base lock
1864 local_irq_disable();
1865 old_base
= &per_cpu(hrtimer_bases
, scpu
);
1866 new_base
= this_cpu_ptr(&hrtimer_bases
);
1868 * The caller is globally serialized and nobody else
1869 * takes two locks at once, deadlock is not possible.
1871 raw_spin_lock(&new_base
->lock
);
1872 raw_spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1874 for (i
= 0; i
< HRTIMER_MAX_CLOCK_BASES
; i
++) {
1875 migrate_hrtimer_list(&old_base
->clock_base
[i
],
1876 &new_base
->clock_base
[i
]);
1880 * The migration might have changed the first expiring softirq
1881 * timer on this CPU. Update it.
1883 hrtimer_update_softirq_timer(new_base
, false);
1885 raw_spin_unlock(&old_base
->lock
);
1886 raw_spin_unlock(&new_base
->lock
);
1888 /* Check, if we got expired work to do */
1889 __hrtimer_peek_ahead_timers();
1895 #endif /* CONFIG_HOTPLUG_CPU */
1897 void __init
hrtimers_init(void)
1899 hrtimers_prepare_cpu(smp_processor_id());
1900 open_softirq(HRTIMER_SOFTIRQ
, hrtimer_run_softirq
);
1904 * schedule_hrtimeout_range_clock - sleep until timeout
1905 * @expires: timeout value (ktime_t)
1906 * @delta: slack in expires timeout (ktime_t)
1908 * @clock_id: timer clock to be used
1911 schedule_hrtimeout_range_clock(ktime_t
*expires
, u64 delta
,
1912 const enum hrtimer_mode mode
, clockid_t clock_id
)
1914 struct hrtimer_sleeper t
;
1917 * Optimize when a zero timeout value is given. It does not
1918 * matter whether this is an absolute or a relative time.
1920 if (expires
&& *expires
== 0) {
1921 __set_current_state(TASK_RUNNING
);
1926 * A NULL parameter means "infinite"
1933 hrtimer_init_on_stack(&t
.timer
, clock_id
, mode
);
1934 hrtimer_set_expires_range_ns(&t
.timer
, *expires
, delta
);
1936 hrtimer_init_sleeper(&t
, current
);
1938 hrtimer_start_expires(&t
.timer
, mode
);
1943 hrtimer_cancel(&t
.timer
);
1944 destroy_hrtimer_on_stack(&t
.timer
);
1946 __set_current_state(TASK_RUNNING
);
1948 return !t
.task
? 0 : -EINTR
;
1952 * schedule_hrtimeout_range - sleep until timeout
1953 * @expires: timeout value (ktime_t)
1954 * @delta: slack in expires timeout (ktime_t)
1957 * Make the current task sleep until the given expiry time has
1958 * elapsed. The routine will return immediately unless
1959 * the current task state has been set (see set_current_state()).
1961 * The @delta argument gives the kernel the freedom to schedule the
1962 * actual wakeup to a time that is both power and performance friendly.
1963 * The kernel give the normal best effort behavior for "@expires+@delta",
1964 * but may decide to fire the timer earlier, but no earlier than @expires.
1966 * You can set the task state as follows -
1968 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1969 * pass before the routine returns unless the current task is explicitly
1970 * woken up, (e.g. by wake_up_process()).
1972 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1973 * delivered to the current task or the current task is explicitly woken
1976 * The current task state is guaranteed to be TASK_RUNNING when this
1979 * Returns 0 when the timer has expired. If the task was woken before the
1980 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
1981 * by an explicit wakeup, it returns -EINTR.
1983 int __sched
schedule_hrtimeout_range(ktime_t
*expires
, u64 delta
,
1984 const enum hrtimer_mode mode
)
1986 return schedule_hrtimeout_range_clock(expires
, delta
, mode
,
1989 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range
);
1992 * schedule_hrtimeout - sleep until timeout
1993 * @expires: timeout value (ktime_t)
1996 * Make the current task sleep until the given expiry time has
1997 * elapsed. The routine will return immediately unless
1998 * the current task state has been set (see set_current_state()).
2000 * You can set the task state as follows -
2002 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2003 * pass before the routine returns unless the current task is explicitly
2004 * woken up, (e.g. by wake_up_process()).
2006 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2007 * delivered to the current task or the current task is explicitly woken
2010 * The current task state is guaranteed to be TASK_RUNNING when this
2013 * Returns 0 when the timer has expired. If the task was woken before the
2014 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2015 * by an explicit wakeup, it returns -EINTR.
2017 int __sched
schedule_hrtimeout(ktime_t
*expires
,
2018 const enum hrtimer_mode mode
)
2020 return schedule_hrtimeout_range(expires
, 0, mode
);
2022 EXPORT_SYMBOL_GPL(schedule_hrtimeout
);