4 * Kernel internal timers
6 * Copyright (C) 1991, 1992 Linus Torvalds
8 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
10 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
11 * "A Kernel Model for Precision Timekeeping" by Dave Mills
12 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
13 * serialize accesses to xtime/lost_ticks).
14 * Copyright (C) 1998 Andrea Arcangeli
15 * 1999-03-10 Improved NTP compatibility by Ulrich Windl
16 * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love
17 * 2000-10-05 Implemented scalable SMP per-CPU timer handling.
18 * Copyright (C) 2000, 2001, 2002 Ingo Molnar
19 * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
22 #include <linux/kernel_stat.h>
23 #include <linux/export.h>
24 #include <linux/interrupt.h>
25 #include <linux/percpu.h>
26 #include <linux/init.h>
28 #include <linux/swap.h>
29 #include <linux/pid_namespace.h>
30 #include <linux/notifier.h>
31 #include <linux/thread_info.h>
32 #include <linux/time.h>
33 #include <linux/jiffies.h>
34 #include <linux/posix-timers.h>
35 #include <linux/cpu.h>
36 #include <linux/syscalls.h>
37 #include <linux/delay.h>
38 #include <linux/tick.h>
39 #include <linux/kallsyms.h>
40 #include <linux/irq_work.h>
41 #include <linux/sched.h>
42 #include <linux/sched/sysctl.h>
43 #include <linux/slab.h>
44 #include <linux/compat.h>
46 #include <asm/uaccess.h>
47 #include <asm/unistd.h>
48 #include <asm/div64.h>
49 #include <asm/timex.h>
52 #include "tick-internal.h"
54 #define CREATE_TRACE_POINTS
55 #include <trace/events/timer.h>
57 __visible u64 jiffies_64 __cacheline_aligned_in_smp
= INITIAL_JIFFIES
;
59 EXPORT_SYMBOL(jiffies_64
);
62 * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
63 * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
64 * level has a different granularity.
66 * The level granularity is: LVL_CLK_DIV ^ lvl
67 * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level)
69 * The array level of a newly armed timer depends on the relative expiry
70 * time. The farther the expiry time is away the higher the array level and
71 * therefor the granularity becomes.
73 * Contrary to the original timer wheel implementation, which aims for 'exact'
74 * expiry of the timers, this implementation removes the need for recascading
75 * the timers into the lower array levels. The previous 'classic' timer wheel
76 * implementation of the kernel already violated the 'exact' expiry by adding
77 * slack to the expiry time to provide batched expiration. The granularity
78 * levels provide implicit batching.
80 * This is an optimization of the original timer wheel implementation for the
81 * majority of the timer wheel use cases: timeouts. The vast majority of
82 * timeout timers (networking, disk I/O ...) are canceled before expiry. If
83 * the timeout expires it indicates that normal operation is disturbed, so it
84 * does not matter much whether the timeout comes with a slight delay.
86 * The only exception to this are networking timers with a small expiry
87 * time. They rely on the granularity. Those fit into the first wheel level,
88 * which has HZ granularity.
90 * We don't have cascading anymore. timers with a expiry time above the
91 * capacity of the last wheel level are force expired at the maximum timeout
92 * value of the last wheel level. From data sampling we know that the maximum
93 * value observed is 5 days (network connection tracking), so this should not
96 * The currently chosen array constants values are a good compromise between
97 * array size and granularity.
99 * This results in the following granularity and range levels:
102 * Level Offset Granularity Range
103 * 0 0 1 ms 0 ms - 63 ms
104 * 1 64 8 ms 64 ms - 511 ms
105 * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s)
106 * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s)
107 * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m)
108 * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m)
109 * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h)
110 * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d)
111 * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d)
114 * Level Offset Granularity Range
115 * 0 0 3 ms 0 ms - 210 ms
116 * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s)
117 * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s)
118 * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m)
119 * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m)
120 * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h)
121 * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h)
122 * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d)
123 * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
126 * Level Offset Granularity Range
127 * 0 0 4 ms 0 ms - 255 ms
128 * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s)
129 * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s)
130 * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m)
131 * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m)
132 * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h)
133 * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h)
134 * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d)
135 * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
138 * Level Offset Granularity Range
139 * 0 0 10 ms 0 ms - 630 ms
140 * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s)
141 * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s)
142 * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m)
143 * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m)
144 * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h)
145 * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d)
146 * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
149 /* Clock divisor for the next level */
150 #define LVL_CLK_SHIFT 3
151 #define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT)
152 #define LVL_CLK_MASK (LVL_CLK_DIV - 1)
153 #define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT)
154 #define LVL_GRAN(n) (1UL << LVL_SHIFT(n))
157 * The time start value for each level to select the bucket at enqueue
160 #define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
162 /* Size of each clock level */
164 #define LVL_SIZE (1UL << LVL_BITS)
165 #define LVL_MASK (LVL_SIZE - 1)
166 #define LVL_OFFS(n) ((n) * LVL_SIZE)
175 /* The cutoff (max. capacity of the wheel) */
176 #define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH))
177 #define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
180 * The resulting wheel size. If NOHZ is configured we allocate two
181 * wheels so we have a separate storage for the deferrable timers.
183 #define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH)
185 #ifdef CONFIG_NO_HZ_COMMON
197 struct timer_list
*running_timer
;
200 bool migration_enabled
;
202 DECLARE_BITMAP(pending_map
, WHEEL_SIZE
);
203 struct hlist_head vectors
[WHEEL_SIZE
];
204 } ____cacheline_aligned
;
206 static DEFINE_PER_CPU(struct timer_base
, timer_bases
[NR_BASES
]);
208 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
209 unsigned int sysctl_timer_migration
= 1;
211 void timers_update_migration(bool update_nohz
)
213 bool on
= sysctl_timer_migration
&& tick_nohz_active
;
216 /* Avoid the loop, if nothing to update */
217 if (this_cpu_read(timer_bases
[BASE_STD
].migration_enabled
) == on
)
220 for_each_possible_cpu(cpu
) {
221 per_cpu(timer_bases
[BASE_STD
].migration_enabled
, cpu
) = on
;
222 per_cpu(timer_bases
[BASE_DEF
].migration_enabled
, cpu
) = on
;
223 per_cpu(hrtimer_bases
.migration_enabled
, cpu
) = on
;
226 per_cpu(timer_bases
[BASE_STD
].nohz_active
, cpu
) = true;
227 per_cpu(timer_bases
[BASE_DEF
].nohz_active
, cpu
) = true;
228 per_cpu(hrtimer_bases
.nohz_active
, cpu
) = true;
232 int timer_migration_handler(struct ctl_table
*table
, int write
,
233 void __user
*buffer
, size_t *lenp
,
236 static DEFINE_MUTEX(mutex
);
240 ret
= proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
242 timers_update_migration(false);
243 mutex_unlock(&mutex
);
248 static unsigned long round_jiffies_common(unsigned long j
, int cpu
,
252 unsigned long original
= j
;
255 * We don't want all cpus firing their timers at once hitting the
256 * same lock or cachelines, so we skew each extra cpu with an extra
257 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
259 * The skew is done by adding 3*cpunr, then round, then subtract this
260 * extra offset again.
267 * If the target jiffie is just after a whole second (which can happen
268 * due to delays of the timer irq, long irq off times etc etc) then
269 * we should round down to the whole second, not up. Use 1/4th second
270 * as cutoff for this rounding as an extreme upper bound for this.
271 * But never round down if @force_up is set.
273 if (rem
< HZ
/4 && !force_up
) /* round down */
278 /* now that we have rounded, subtract the extra skew again */
282 * Make sure j is still in the future. Otherwise return the
285 return time_is_after_jiffies(j
) ? j
: original
;
289 * __round_jiffies - function to round jiffies to a full second
290 * @j: the time in (absolute) jiffies that should be rounded
291 * @cpu: the processor number on which the timeout will happen
293 * __round_jiffies() rounds an absolute time in the future (in jiffies)
294 * up or down to (approximately) full seconds. This is useful for timers
295 * for which the exact time they fire does not matter too much, as long as
296 * they fire approximately every X seconds.
298 * By rounding these timers to whole seconds, all such timers will fire
299 * at the same time, rather than at various times spread out. The goal
300 * of this is to have the CPU wake up less, which saves power.
302 * The exact rounding is skewed for each processor to avoid all
303 * processors firing at the exact same time, which could lead
304 * to lock contention or spurious cache line bouncing.
306 * The return value is the rounded version of the @j parameter.
308 unsigned long __round_jiffies(unsigned long j
, int cpu
)
310 return round_jiffies_common(j
, cpu
, false);
312 EXPORT_SYMBOL_GPL(__round_jiffies
);
315 * __round_jiffies_relative - function to round jiffies to a full second
316 * @j: the time in (relative) jiffies that should be rounded
317 * @cpu: the processor number on which the timeout will happen
319 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
320 * up or down to (approximately) full seconds. This is useful for timers
321 * for which the exact time they fire does not matter too much, as long as
322 * they fire approximately every X seconds.
324 * By rounding these timers to whole seconds, all such timers will fire
325 * at the same time, rather than at various times spread out. The goal
326 * of this is to have the CPU wake up less, which saves power.
328 * The exact rounding is skewed for each processor to avoid all
329 * processors firing at the exact same time, which could lead
330 * to lock contention or spurious cache line bouncing.
332 * The return value is the rounded version of the @j parameter.
334 unsigned long __round_jiffies_relative(unsigned long j
, int cpu
)
336 unsigned long j0
= jiffies
;
338 /* Use j0 because jiffies might change while we run */
339 return round_jiffies_common(j
+ j0
, cpu
, false) - j0
;
341 EXPORT_SYMBOL_GPL(__round_jiffies_relative
);
344 * round_jiffies - function to round jiffies to a full second
345 * @j: the time in (absolute) jiffies that should be rounded
347 * round_jiffies() rounds an absolute time in the future (in jiffies)
348 * up or down to (approximately) full seconds. This is useful for timers
349 * for which the exact time they fire does not matter too much, as long as
350 * they fire approximately every X seconds.
352 * By rounding these timers to whole seconds, all such timers will fire
353 * at the same time, rather than at various times spread out. The goal
354 * of this is to have the CPU wake up less, which saves power.
356 * The return value is the rounded version of the @j parameter.
358 unsigned long round_jiffies(unsigned long j
)
360 return round_jiffies_common(j
, raw_smp_processor_id(), false);
362 EXPORT_SYMBOL_GPL(round_jiffies
);
365 * round_jiffies_relative - function to round jiffies to a full second
366 * @j: the time in (relative) jiffies that should be rounded
368 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
369 * up or down to (approximately) full seconds. This is useful for timers
370 * for which the exact time they fire does not matter too much, as long as
371 * they fire approximately every X seconds.
373 * By rounding these timers to whole seconds, all such timers will fire
374 * at the same time, rather than at various times spread out. The goal
375 * of this is to have the CPU wake up less, which saves power.
377 * The return value is the rounded version of the @j parameter.
379 unsigned long round_jiffies_relative(unsigned long j
)
381 return __round_jiffies_relative(j
, raw_smp_processor_id());
383 EXPORT_SYMBOL_GPL(round_jiffies_relative
);
386 * __round_jiffies_up - function to round jiffies up to a full second
387 * @j: the time in (absolute) jiffies that should be rounded
388 * @cpu: the processor number on which the timeout will happen
390 * This is the same as __round_jiffies() except that it will never
391 * round down. This is useful for timeouts for which the exact time
392 * of firing does not matter too much, as long as they don't fire too
395 unsigned long __round_jiffies_up(unsigned long j
, int cpu
)
397 return round_jiffies_common(j
, cpu
, true);
399 EXPORT_SYMBOL_GPL(__round_jiffies_up
);
402 * __round_jiffies_up_relative - function to round jiffies up to a full second
403 * @j: the time in (relative) jiffies that should be rounded
404 * @cpu: the processor number on which the timeout will happen
406 * This is the same as __round_jiffies_relative() except that it will never
407 * round down. This is useful for timeouts for which the exact time
408 * of firing does not matter too much, as long as they don't fire too
411 unsigned long __round_jiffies_up_relative(unsigned long j
, int cpu
)
413 unsigned long j0
= jiffies
;
415 /* Use j0 because jiffies might change while we run */
416 return round_jiffies_common(j
+ j0
, cpu
, true) - j0
;
418 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative
);
421 * round_jiffies_up - function to round jiffies up to a full second
422 * @j: the time in (absolute) jiffies that should be rounded
424 * This is the same as round_jiffies() except that it will never
425 * round down. This is useful for timeouts for which the exact time
426 * of firing does not matter too much, as long as they don't fire too
429 unsigned long round_jiffies_up(unsigned long j
)
431 return round_jiffies_common(j
, raw_smp_processor_id(), true);
433 EXPORT_SYMBOL_GPL(round_jiffies_up
);
436 * round_jiffies_up_relative - function to round jiffies up to a full second
437 * @j: the time in (relative) jiffies that should be rounded
439 * This is the same as round_jiffies_relative() except that it will never
440 * round down. This is useful for timeouts for which the exact time
441 * of firing does not matter too much, as long as they don't fire too
444 unsigned long round_jiffies_up_relative(unsigned long j
)
446 return __round_jiffies_up_relative(j
, raw_smp_processor_id());
448 EXPORT_SYMBOL_GPL(round_jiffies_up_relative
);
451 static inline unsigned int timer_get_idx(struct timer_list
*timer
)
453 return (timer
->flags
& TIMER_ARRAYMASK
) >> TIMER_ARRAYSHIFT
;
456 static inline void timer_set_idx(struct timer_list
*timer
, unsigned int idx
)
458 timer
->flags
= (timer
->flags
& ~TIMER_ARRAYMASK
) |
459 idx
<< TIMER_ARRAYSHIFT
;
463 * Helper function to calculate the array index for a given expiry
466 static inline unsigned calc_index(unsigned expires
, unsigned lvl
)
468 expires
= (expires
+ LVL_GRAN(lvl
)) >> LVL_SHIFT(lvl
);
469 return LVL_OFFS(lvl
) + (expires
& LVL_MASK
);
473 __internal_add_timer(struct timer_base
*base
, struct timer_list
*timer
)
475 unsigned long expires
= timer
->expires
;
476 unsigned long delta
= expires
- base
->clk
;
477 struct hlist_head
*vec
;
480 if (delta
< LVL_START(1)) {
481 idx
= calc_index(expires
, 0);
482 } else if (delta
< LVL_START(2)) {
483 idx
= calc_index(expires
, 1);
484 } else if (delta
< LVL_START(3)) {
485 idx
= calc_index(expires
, 2);
486 } else if (delta
< LVL_START(4)) {
487 idx
= calc_index(expires
, 3);
488 } else if (delta
< LVL_START(5)) {
489 idx
= calc_index(expires
, 4);
490 } else if (delta
< LVL_START(6)) {
491 idx
= calc_index(expires
, 5);
492 } else if (delta
< LVL_START(7)) {
493 idx
= calc_index(expires
, 6);
494 } else if (LVL_DEPTH
> 8 && delta
< LVL_START(8)) {
495 idx
= calc_index(expires
, 7);
496 } else if ((long) delta
< 0) {
497 idx
= base
->clk
& LVL_MASK
;
500 * Force expire obscene large timeouts to expire at the
501 * capacity limit of the wheel.
503 if (expires
>= WHEEL_TIMEOUT_CUTOFF
)
504 expires
= WHEEL_TIMEOUT_MAX
;
506 idx
= calc_index(expires
, LVL_DEPTH
- 1);
509 * Enqueue the timer into the array bucket, mark it pending in
510 * the bitmap and store the index in the timer flags.
512 vec
= base
->vectors
+ idx
;
513 hlist_add_head(&timer
->entry
, vec
);
514 __set_bit(idx
, base
->pending_map
);
515 timer_set_idx(timer
, idx
);
518 static void internal_add_timer(struct timer_base
*base
, struct timer_list
*timer
)
520 __internal_add_timer(base
, timer
);
523 * Check whether the other CPU is in dynticks mode and needs
524 * to be triggered to reevaluate the timer wheel. We are
525 * protected against the other CPU fiddling with the timer by
526 * holding the timer base lock. This also makes sure that a
527 * CPU on the way to stop its tick can not evaluate the timer
530 * Spare the IPI for deferrable timers on idle targets though.
531 * The next busy ticks will take care of it. Except full dynticks
532 * require special care against races with idle_cpu(), lets deal
535 if (IS_ENABLED(CONFIG_NO_HZ_COMMON
) && base
->nohz_active
) {
536 if (!(timer
->flags
& TIMER_DEFERRABLE
) ||
537 tick_nohz_full_cpu(base
->cpu
))
538 wake_up_nohz_cpu(base
->cpu
);
542 #ifdef CONFIG_TIMER_STATS
543 void __timer_stats_timer_set_start_info(struct timer_list
*timer
, void *addr
)
545 if (timer
->start_site
)
548 timer
->start_site
= addr
;
549 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
550 timer
->start_pid
= current
->pid
;
553 static void timer_stats_account_timer(struct timer_list
*timer
)
558 * start_site can be concurrently reset by
559 * timer_stats_timer_clear_start_info()
561 site
= READ_ONCE(timer
->start_site
);
565 timer_stats_update_stats(timer
, timer
->start_pid
, site
,
566 timer
->function
, timer
->start_comm
,
571 static void timer_stats_account_timer(struct timer_list
*timer
) {}
574 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
576 static struct debug_obj_descr timer_debug_descr
;
578 static void *timer_debug_hint(void *addr
)
580 return ((struct timer_list
*) addr
)->function
;
583 static bool timer_is_static_object(void *addr
)
585 struct timer_list
*timer
= addr
;
587 return (timer
->entry
.pprev
== NULL
&&
588 timer
->entry
.next
== TIMER_ENTRY_STATIC
);
592 * fixup_init is called when:
593 * - an active object is initialized
595 static bool timer_fixup_init(void *addr
, enum debug_obj_state state
)
597 struct timer_list
*timer
= addr
;
600 case ODEBUG_STATE_ACTIVE
:
601 del_timer_sync(timer
);
602 debug_object_init(timer
, &timer_debug_descr
);
609 /* Stub timer callback for improperly used timers. */
610 static void stub_timer(unsigned long data
)
616 * fixup_activate is called when:
617 * - an active object is activated
618 * - an unknown non-static object is activated
620 static bool timer_fixup_activate(void *addr
, enum debug_obj_state state
)
622 struct timer_list
*timer
= addr
;
625 case ODEBUG_STATE_NOTAVAILABLE
:
626 setup_timer(timer
, stub_timer
, 0);
629 case ODEBUG_STATE_ACTIVE
:
638 * fixup_free is called when:
639 * - an active object is freed
641 static bool timer_fixup_free(void *addr
, enum debug_obj_state state
)
643 struct timer_list
*timer
= addr
;
646 case ODEBUG_STATE_ACTIVE
:
647 del_timer_sync(timer
);
648 debug_object_free(timer
, &timer_debug_descr
);
656 * fixup_assert_init is called when:
657 * - an untracked/uninit-ed object is found
659 static bool timer_fixup_assert_init(void *addr
, enum debug_obj_state state
)
661 struct timer_list
*timer
= addr
;
664 case ODEBUG_STATE_NOTAVAILABLE
:
665 setup_timer(timer
, stub_timer
, 0);
672 static struct debug_obj_descr timer_debug_descr
= {
673 .name
= "timer_list",
674 .debug_hint
= timer_debug_hint
,
675 .is_static_object
= timer_is_static_object
,
676 .fixup_init
= timer_fixup_init
,
677 .fixup_activate
= timer_fixup_activate
,
678 .fixup_free
= timer_fixup_free
,
679 .fixup_assert_init
= timer_fixup_assert_init
,
682 static inline void debug_timer_init(struct timer_list
*timer
)
684 debug_object_init(timer
, &timer_debug_descr
);
687 static inline void debug_timer_activate(struct timer_list
*timer
)
689 debug_object_activate(timer
, &timer_debug_descr
);
692 static inline void debug_timer_deactivate(struct timer_list
*timer
)
694 debug_object_deactivate(timer
, &timer_debug_descr
);
697 static inline void debug_timer_free(struct timer_list
*timer
)
699 debug_object_free(timer
, &timer_debug_descr
);
702 static inline void debug_timer_assert_init(struct timer_list
*timer
)
704 debug_object_assert_init(timer
, &timer_debug_descr
);
707 static void do_init_timer(struct timer_list
*timer
, unsigned int flags
,
708 const char *name
, struct lock_class_key
*key
);
710 void init_timer_on_stack_key(struct timer_list
*timer
, unsigned int flags
,
711 const char *name
, struct lock_class_key
*key
)
713 debug_object_init_on_stack(timer
, &timer_debug_descr
);
714 do_init_timer(timer
, flags
, name
, key
);
716 EXPORT_SYMBOL_GPL(init_timer_on_stack_key
);
718 void destroy_timer_on_stack(struct timer_list
*timer
)
720 debug_object_free(timer
, &timer_debug_descr
);
722 EXPORT_SYMBOL_GPL(destroy_timer_on_stack
);
725 static inline void debug_timer_init(struct timer_list
*timer
) { }
726 static inline void debug_timer_activate(struct timer_list
*timer
) { }
727 static inline void debug_timer_deactivate(struct timer_list
*timer
) { }
728 static inline void debug_timer_assert_init(struct timer_list
*timer
) { }
731 static inline void debug_init(struct timer_list
*timer
)
733 debug_timer_init(timer
);
734 trace_timer_init(timer
);
738 debug_activate(struct timer_list
*timer
, unsigned long expires
)
740 debug_timer_activate(timer
);
741 trace_timer_start(timer
, expires
, timer
->flags
);
744 static inline void debug_deactivate(struct timer_list
*timer
)
746 debug_timer_deactivate(timer
);
747 trace_timer_cancel(timer
);
750 static inline void debug_assert_init(struct timer_list
*timer
)
752 debug_timer_assert_init(timer
);
755 static void do_init_timer(struct timer_list
*timer
, unsigned int flags
,
756 const char *name
, struct lock_class_key
*key
)
758 timer
->entry
.pprev
= NULL
;
759 timer
->flags
= flags
| raw_smp_processor_id();
760 #ifdef CONFIG_TIMER_STATS
761 timer
->start_site
= NULL
;
762 timer
->start_pid
= -1;
763 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
765 lockdep_init_map(&timer
->lockdep_map
, name
, key
, 0);
769 * init_timer_key - initialize a timer
770 * @timer: the timer to be initialized
771 * @flags: timer flags
772 * @name: name of the timer
773 * @key: lockdep class key of the fake lock used for tracking timer
774 * sync lock dependencies
776 * init_timer_key() must be done to a timer prior calling *any* of the
777 * other timer functions.
779 void init_timer_key(struct timer_list
*timer
, unsigned int flags
,
780 const char *name
, struct lock_class_key
*key
)
783 do_init_timer(timer
, flags
, name
, key
);
785 EXPORT_SYMBOL(init_timer_key
);
787 static inline void detach_timer(struct timer_list
*timer
, bool clear_pending
)
789 struct hlist_node
*entry
= &timer
->entry
;
791 debug_deactivate(timer
);
796 entry
->next
= LIST_POISON2
;
799 static int detach_if_pending(struct timer_list
*timer
, struct timer_base
*base
,
802 unsigned idx
= timer_get_idx(timer
);
804 if (!timer_pending(timer
))
807 if (hlist_is_singular_node(&timer
->entry
, base
->vectors
+ idx
))
808 __clear_bit(idx
, base
->pending_map
);
810 detach_timer(timer
, clear_pending
);
814 static inline struct timer_base
*get_timer_cpu_base(u32 tflags
, u32 cpu
)
816 struct timer_base
*base
= per_cpu_ptr(&timer_bases
[BASE_STD
], cpu
);
819 * If the timer is deferrable and nohz is active then we need to use
820 * the deferrable base.
822 if (IS_ENABLED(CONFIG_NO_HZ_COMMON
) && base
->nohz_active
&&
823 (tflags
& TIMER_DEFERRABLE
))
824 base
= per_cpu_ptr(&timer_bases
[BASE_DEF
], cpu
);
828 static inline struct timer_base
*get_timer_this_cpu_base(u32 tflags
)
830 struct timer_base
*base
= this_cpu_ptr(&timer_bases
[BASE_STD
]);
833 * If the timer is deferrable and nohz is active then we need to use
834 * the deferrable base.
836 if (IS_ENABLED(CONFIG_NO_HZ_COMMON
) && base
->nohz_active
&&
837 (tflags
& TIMER_DEFERRABLE
))
838 base
= this_cpu_ptr(&timer_bases
[BASE_DEF
]);
842 static inline struct timer_base
*get_timer_base(u32 tflags
)
844 return get_timer_cpu_base(tflags
, tflags
& TIMER_CPUMASK
);
847 static inline struct timer_base
*get_target_base(struct timer_base
*base
,
850 #if defined(CONFIG_NO_HZ_COMMON) && defined(CONFIG_SMP)
851 if ((tflags
& TIMER_PINNED
) || !base
->migration_enabled
)
852 return get_timer_this_cpu_base(tflags
);
853 return get_timer_cpu_base(tflags
, get_nohz_timer_target());
855 return get_timer_this_cpu_base(tflags
);
860 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
861 * that all timers which are tied to this base are locked, and the base itself
864 * So __run_timers/migrate_timers can safely modify all timers which could
865 * be found in the base->vectors array.
867 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
868 * to wait until the migration is done.
870 static struct timer_base
*lock_timer_base(struct timer_list
*timer
,
871 unsigned long *flags
)
872 __acquires(timer
->base
->lock
)
875 struct timer_base
*base
;
876 u32 tf
= timer
->flags
;
878 if (!(tf
& TIMER_MIGRATING
)) {
879 base
= get_timer_base(tf
);
880 spin_lock_irqsave(&base
->lock
, *flags
);
881 if (timer
->flags
== tf
)
883 spin_unlock_irqrestore(&base
->lock
, *flags
);
890 __mod_timer(struct timer_list
*timer
, unsigned long expires
, bool pending_only
)
892 struct timer_base
*base
, *new_base
;
897 * TODO: Calculate the array bucket of the timer right here w/o
898 * holding the base lock. This allows to check not only
899 * timer->expires == expires below, but also whether the timer
900 * ends up in the same bucket. If we really need to requeue
901 * the timer then we check whether base->clk have
902 * advanced between here and locking the timer base. If
903 * jiffies advanced we have to recalc the array bucket with the
908 * This is a common optimization triggered by the
909 * networking code - if the timer is re-modified
910 * to be the same thing then just return:
912 if (timer_pending(timer
)) {
913 if (timer
->expires
== expires
)
917 timer_stats_timer_set_start_info(timer
);
918 BUG_ON(!timer
->function
);
920 base
= lock_timer_base(timer
, &flags
);
922 ret
= detach_if_pending(timer
, base
, false);
923 if (!ret
&& pending_only
)
926 debug_activate(timer
, expires
);
928 new_base
= get_target_base(base
, timer
->flags
);
930 if (base
!= new_base
) {
932 * We are trying to schedule the timer on the new base.
933 * However we can't change timer's base while it is running,
934 * otherwise del_timer_sync() can't detect that the timer's
935 * handler yet has not finished. This also guarantees that the
936 * timer is serialized wrt itself.
938 if (likely(base
->running_timer
!= timer
)) {
939 /* See the comment in lock_timer_base() */
940 timer
->flags
|= TIMER_MIGRATING
;
942 spin_unlock(&base
->lock
);
944 spin_lock(&base
->lock
);
945 WRITE_ONCE(timer
->flags
,
946 (timer
->flags
& ~TIMER_BASEMASK
) | base
->cpu
);
950 timer
->expires
= expires
;
951 internal_add_timer(base
, timer
);
954 spin_unlock_irqrestore(&base
->lock
, flags
);
960 * mod_timer_pending - modify a pending timer's timeout
961 * @timer: the pending timer to be modified
962 * @expires: new timeout in jiffies
964 * mod_timer_pending() is the same for pending timers as mod_timer(),
965 * but will not re-activate and modify already deleted timers.
967 * It is useful for unserialized use of timers.
969 int mod_timer_pending(struct timer_list
*timer
, unsigned long expires
)
971 return __mod_timer(timer
, expires
, true);
973 EXPORT_SYMBOL(mod_timer_pending
);
976 * mod_timer - modify a timer's timeout
977 * @timer: the timer to be modified
978 * @expires: new timeout in jiffies
980 * mod_timer() is a more efficient way to update the expire field of an
981 * active timer (if the timer is inactive it will be activated)
983 * mod_timer(timer, expires) is equivalent to:
985 * del_timer(timer); timer->expires = expires; add_timer(timer);
987 * Note that if there are multiple unserialized concurrent users of the
988 * same timer, then mod_timer() is the only safe way to modify the timeout,
989 * since add_timer() cannot modify an already running timer.
991 * The function returns whether it has modified a pending timer or not.
992 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
993 * active timer returns 1.)
995 int mod_timer(struct timer_list
*timer
, unsigned long expires
)
997 return __mod_timer(timer
, expires
, false);
999 EXPORT_SYMBOL(mod_timer
);
1002 * add_timer - start a timer
1003 * @timer: the timer to be added
1005 * The kernel will do a ->function(->data) callback from the
1006 * timer interrupt at the ->expires point in the future. The
1007 * current time is 'jiffies'.
1009 * The timer's ->expires, ->function (and if the handler uses it, ->data)
1010 * fields must be set prior calling this function.
1012 * Timers with an ->expires field in the past will be executed in the next
1015 void add_timer(struct timer_list
*timer
)
1017 BUG_ON(timer_pending(timer
));
1018 mod_timer(timer
, timer
->expires
);
1020 EXPORT_SYMBOL(add_timer
);
1023 * add_timer_on - start a timer on a particular CPU
1024 * @timer: the timer to be added
1025 * @cpu: the CPU to start it on
1027 * This is not very scalable on SMP. Double adds are not possible.
1029 void add_timer_on(struct timer_list
*timer
, int cpu
)
1031 struct timer_base
*new_base
, *base
;
1032 unsigned long flags
;
1034 timer_stats_timer_set_start_info(timer
);
1035 BUG_ON(timer_pending(timer
) || !timer
->function
);
1037 new_base
= get_timer_cpu_base(timer
->flags
, cpu
);
1040 * If @timer was on a different CPU, it should be migrated with the
1041 * old base locked to prevent other operations proceeding with the
1042 * wrong base locked. See lock_timer_base().
1044 base
= lock_timer_base(timer
, &flags
);
1045 if (base
!= new_base
) {
1046 timer
->flags
|= TIMER_MIGRATING
;
1048 spin_unlock(&base
->lock
);
1050 spin_lock(&base
->lock
);
1051 WRITE_ONCE(timer
->flags
,
1052 (timer
->flags
& ~TIMER_BASEMASK
) | cpu
);
1055 debug_activate(timer
, timer
->expires
);
1056 internal_add_timer(base
, timer
);
1057 spin_unlock_irqrestore(&base
->lock
, flags
);
1059 EXPORT_SYMBOL_GPL(add_timer_on
);
1062 * del_timer - deactive a timer.
1063 * @timer: the timer to be deactivated
1065 * del_timer() deactivates a timer - this works on both active and inactive
1068 * The function returns whether it has deactivated a pending timer or not.
1069 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1070 * active timer returns 1.)
1072 int del_timer(struct timer_list
*timer
)
1074 struct timer_base
*base
;
1075 unsigned long flags
;
1078 debug_assert_init(timer
);
1080 timer_stats_timer_clear_start_info(timer
);
1081 if (timer_pending(timer
)) {
1082 base
= lock_timer_base(timer
, &flags
);
1083 ret
= detach_if_pending(timer
, base
, true);
1084 spin_unlock_irqrestore(&base
->lock
, flags
);
1089 EXPORT_SYMBOL(del_timer
);
1092 * try_to_del_timer_sync - Try to deactivate a timer
1093 * @timer: timer do del
1095 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1096 * exit the timer is not queued and the handler is not running on any CPU.
1098 int try_to_del_timer_sync(struct timer_list
*timer
)
1100 struct timer_base
*base
;
1101 unsigned long flags
;
1104 debug_assert_init(timer
);
1106 base
= lock_timer_base(timer
, &flags
);
1108 if (base
->running_timer
!= timer
) {
1109 timer_stats_timer_clear_start_info(timer
);
1110 ret
= detach_if_pending(timer
, base
, true);
1112 spin_unlock_irqrestore(&base
->lock
, flags
);
1116 EXPORT_SYMBOL(try_to_del_timer_sync
);
1120 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1121 * @timer: the timer to be deactivated
1123 * This function only differs from del_timer() on SMP: besides deactivating
1124 * the timer it also makes sure the handler has finished executing on other
1127 * Synchronization rules: Callers must prevent restarting of the timer,
1128 * otherwise this function is meaningless. It must not be called from
1129 * interrupt contexts unless the timer is an irqsafe one. The caller must
1130 * not hold locks which would prevent completion of the timer's
1131 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1132 * timer is not queued and the handler is not running on any CPU.
1134 * Note: For !irqsafe timers, you must not hold locks that are held in
1135 * interrupt context while calling this function. Even if the lock has
1136 * nothing to do with the timer in question. Here's why:
1142 * base->running_timer = mytimer;
1143 * spin_lock_irq(somelock);
1145 * spin_lock(somelock);
1146 * del_timer_sync(mytimer);
1147 * while (base->running_timer == mytimer);
1149 * Now del_timer_sync() will never return and never release somelock.
1150 * The interrupt on the other CPU is waiting to grab somelock but
1151 * it has interrupted the softirq that CPU0 is waiting to finish.
1153 * The function returns whether it has deactivated a pending timer or not.
1155 int del_timer_sync(struct timer_list
*timer
)
1157 #ifdef CONFIG_LOCKDEP
1158 unsigned long flags
;
1161 * If lockdep gives a backtrace here, please reference
1162 * the synchronization rules above.
1164 local_irq_save(flags
);
1165 lock_map_acquire(&timer
->lockdep_map
);
1166 lock_map_release(&timer
->lockdep_map
);
1167 local_irq_restore(flags
);
1170 * don't use it in hardirq context, because it
1171 * could lead to deadlock.
1173 WARN_ON(in_irq() && !(timer
->flags
& TIMER_IRQSAFE
));
1175 int ret
= try_to_del_timer_sync(timer
);
1181 EXPORT_SYMBOL(del_timer_sync
);
1184 static void call_timer_fn(struct timer_list
*timer
, void (*fn
)(unsigned long),
1187 int count
= preempt_count();
1189 #ifdef CONFIG_LOCKDEP
1191 * It is permissible to free the timer from inside the
1192 * function that is called from it, this we need to take into
1193 * account for lockdep too. To avoid bogus "held lock freed"
1194 * warnings as well as problems when looking into
1195 * timer->lockdep_map, make a copy and use that here.
1197 struct lockdep_map lockdep_map
;
1199 lockdep_copy_map(&lockdep_map
, &timer
->lockdep_map
);
1202 * Couple the lock chain with the lock chain at
1203 * del_timer_sync() by acquiring the lock_map around the fn()
1204 * call here and in del_timer_sync().
1206 lock_map_acquire(&lockdep_map
);
1208 trace_timer_expire_entry(timer
);
1210 trace_timer_expire_exit(timer
);
1212 lock_map_release(&lockdep_map
);
1214 if (count
!= preempt_count()) {
1215 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1216 fn
, count
, preempt_count());
1218 * Restore the preempt count. That gives us a decent
1219 * chance to survive and extract information. If the
1220 * callback kept a lock held, bad luck, but not worse
1221 * than the BUG() we had.
1223 preempt_count_set(count
);
1227 static void expire_timers(struct timer_base
*base
, struct hlist_head
*head
)
1229 while (!hlist_empty(head
)) {
1230 struct timer_list
*timer
;
1231 void (*fn
)(unsigned long);
1234 timer
= hlist_entry(head
->first
, struct timer_list
, entry
);
1235 timer_stats_account_timer(timer
);
1237 base
->running_timer
= timer
;
1238 detach_timer(timer
, true);
1240 fn
= timer
->function
;
1243 if (timer
->flags
& TIMER_IRQSAFE
) {
1244 spin_unlock(&base
->lock
);
1245 call_timer_fn(timer
, fn
, data
);
1246 spin_lock(&base
->lock
);
1248 spin_unlock_irq(&base
->lock
);
1249 call_timer_fn(timer
, fn
, data
);
1250 spin_lock_irq(&base
->lock
);
1255 static int collect_expired_timers(struct timer_base
*base
,
1256 struct hlist_head
*heads
)
1258 unsigned long clk
= base
->clk
;
1259 struct hlist_head
*vec
;
1263 for (i
= 0; i
< LVL_DEPTH
; i
++) {
1264 idx
= (clk
& LVL_MASK
) + i
* LVL_SIZE
;
1266 if (__test_and_clear_bit(idx
, base
->pending_map
)) {
1267 vec
= base
->vectors
+ idx
;
1268 hlist_move_list(vec
, heads
++);
1271 /* Is it time to look at the next level? */
1272 if (clk
& LVL_CLK_MASK
)
1274 /* Shift clock for the next level granularity */
1275 clk
>>= LVL_CLK_SHIFT
;
1280 #ifdef CONFIG_NO_HZ_COMMON
1282 * Find the next pending bucket of a level. Search from @offset + @clk upwards
1283 * and if nothing there, search from start of the level (@offset) up to
1286 static int next_pending_bucket(struct timer_base
*base
, unsigned offset
,
1289 unsigned pos
, start
= offset
+ clk
;
1290 unsigned end
= offset
+ LVL_SIZE
;
1292 pos
= find_next_bit(base
->pending_map
, end
, start
);
1296 pos
= find_next_bit(base
->pending_map
, start
, offset
);
1297 return pos
< start
? pos
+ LVL_SIZE
- start
: -1;
1301 * Search the first expiring timer in the various clock levels.
1303 static unsigned long __next_timer_interrupt(struct timer_base
*base
)
1305 unsigned long clk
, next
, adj
;
1306 unsigned lvl
, offset
= 0;
1308 spin_lock(&base
->lock
);
1309 next
= base
->clk
+ NEXT_TIMER_MAX_DELTA
;
1311 for (lvl
= 0; lvl
< LVL_DEPTH
; lvl
++, offset
+= LVL_SIZE
) {
1312 int pos
= next_pending_bucket(base
, offset
, clk
& LVL_MASK
);
1315 unsigned long tmp
= clk
+ (unsigned long) pos
;
1317 tmp
<<= LVL_SHIFT(lvl
);
1318 if (time_before(tmp
, next
))
1322 * Clock for the next level. If the current level clock lower
1323 * bits are zero, we look at the next level as is. If not we
1324 * need to advance it by one because that's going to be the
1325 * next expiring bucket in that level. base->clk is the next
1326 * expiring jiffie. So in case of:
1328 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1331 * we have to look at all levels @index 0. With
1333 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1336 * LVL0 has the next expiring bucket @index 2. The upper
1337 * levels have the next expiring bucket @index 1.
1339 * In case that the propagation wraps the next level the same
1342 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1345 * So after looking at LVL0 we get:
1347 * LVL5 LVL4 LVL3 LVL2 LVL1
1350 * So no propagation from LVL1 to LVL2 because that happened
1351 * with the add already, but then we need to propagate further
1352 * from LVL2 to LVL3.
1354 * So the simple check whether the lower bits of the current
1355 * level are 0 or not is sufficient for all cases.
1357 adj
= clk
& LVL_CLK_MASK
? 1 : 0;
1358 clk
>>= LVL_CLK_SHIFT
;
1361 spin_unlock(&base
->lock
);
1366 * Check, if the next hrtimer event is before the next timer wheel
1369 static u64
cmp_next_hrtimer_event(u64 basem
, u64 expires
)
1371 u64 nextevt
= hrtimer_get_next_event();
1374 * If high resolution timers are enabled
1375 * hrtimer_get_next_event() returns KTIME_MAX.
1377 if (expires
<= nextevt
)
1381 * If the next timer is already expired, return the tick base
1382 * time so the tick is fired immediately.
1384 if (nextevt
<= basem
)
1388 * Round up to the next jiffie. High resolution timers are
1389 * off, so the hrtimers are expired in the tick and we need to
1390 * make sure that this tick really expires the timer to avoid
1391 * a ping pong of the nohz stop code.
1393 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1395 return DIV_ROUND_UP_ULL(nextevt
, TICK_NSEC
) * TICK_NSEC
;
1399 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1400 * @basej: base time jiffies
1401 * @basem: base time clock monotonic
1403 * Returns the tick aligned clock monotonic time of the next pending
1404 * timer or KTIME_MAX if no timer is pending.
1406 u64
get_next_timer_interrupt(unsigned long basej
, u64 basem
)
1408 struct timer_base
*base
= this_cpu_ptr(&timer_bases
[BASE_STD
]);
1409 u64 expires
= KTIME_MAX
;
1410 unsigned long nextevt
;
1413 * Pretend that there is no timer pending if the cpu is offline.
1414 * Possible pending timers will be migrated later to an active cpu.
1416 if (cpu_is_offline(smp_processor_id()))
1419 nextevt
= __next_timer_interrupt(base
);
1420 if (time_before_eq(nextevt
, basej
))
1423 expires
= basem
+ (nextevt
- basej
) * TICK_NSEC
;
1425 return cmp_next_hrtimer_event(basem
, expires
);
1430 * Called from the timer interrupt handler to charge one tick to the current
1431 * process. user_tick is 1 if the tick is user time, 0 for system.
1433 void update_process_times(int user_tick
)
1435 struct task_struct
*p
= current
;
1437 /* Note: this timer irq context must be accounted for as well. */
1438 account_process_tick(p
, user_tick
);
1440 rcu_check_callbacks(user_tick
);
1441 #ifdef CONFIG_IRQ_WORK
1446 run_posix_cpu_timers(p
);
1450 * __run_timers - run all expired timers (if any) on this CPU.
1451 * @base: the timer vector to be processed.
1453 static inline void __run_timers(struct timer_base
*base
)
1455 struct hlist_head heads
[LVL_DEPTH
];
1458 if (!time_after_eq(jiffies
, base
->clk
))
1461 spin_lock_irq(&base
->lock
);
1463 while (time_after_eq(jiffies
, base
->clk
)) {
1465 levels
= collect_expired_timers(base
, heads
);
1469 expire_timers(base
, heads
+ levels
);
1471 base
->running_timer
= NULL
;
1472 spin_unlock_irq(&base
->lock
);
1476 * This function runs timers and the timer-tq in bottom half context.
1478 static void run_timer_softirq(struct softirq_action
*h
)
1480 struct timer_base
*base
= this_cpu_ptr(&timer_bases
[BASE_STD
]);
1483 if (IS_ENABLED(CONFIG_NO_HZ_COMMON
) && base
->nohz_active
)
1484 __run_timers(this_cpu_ptr(&timer_bases
[BASE_DEF
]));
1488 * Called by the local, per-CPU timer interrupt on SMP.
1490 void run_local_timers(void)
1492 hrtimer_run_queues();
1493 raise_softirq(TIMER_SOFTIRQ
);
1496 #ifdef __ARCH_WANT_SYS_ALARM
1499 * For backwards compatibility? This can be done in libc so Alpha
1500 * and all newer ports shouldn't need it.
1502 SYSCALL_DEFINE1(alarm
, unsigned int, seconds
)
1504 return alarm_setitimer(seconds
);
1509 static void process_timeout(unsigned long __data
)
1511 wake_up_process((struct task_struct
*)__data
);
1515 * schedule_timeout - sleep until timeout
1516 * @timeout: timeout value in jiffies
1518 * Make the current task sleep until @timeout jiffies have
1519 * elapsed. The routine will return immediately unless
1520 * the current task state has been set (see set_current_state()).
1522 * You can set the task state as follows -
1524 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1525 * pass before the routine returns. The routine will return 0
1527 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1528 * delivered to the current task. In this case the remaining time
1529 * in jiffies will be returned, or 0 if the timer expired in time
1531 * The current task state is guaranteed to be TASK_RUNNING when this
1534 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1535 * the CPU away without a bound on the timeout. In this case the return
1536 * value will be %MAX_SCHEDULE_TIMEOUT.
1538 * In all cases the return value is guaranteed to be non-negative.
1540 signed long __sched
schedule_timeout(signed long timeout
)
1542 struct timer_list timer
;
1543 unsigned long expire
;
1547 case MAX_SCHEDULE_TIMEOUT
:
1549 * These two special cases are useful to be comfortable
1550 * in the caller. Nothing more. We could take
1551 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1552 * but I' d like to return a valid offset (>=0) to allow
1553 * the caller to do everything it want with the retval.
1559 * Another bit of PARANOID. Note that the retval will be
1560 * 0 since no piece of kernel is supposed to do a check
1561 * for a negative retval of schedule_timeout() (since it
1562 * should never happens anyway). You just have the printk()
1563 * that will tell you if something is gone wrong and where.
1566 printk(KERN_ERR
"schedule_timeout: wrong timeout "
1567 "value %lx\n", timeout
);
1569 current
->state
= TASK_RUNNING
;
1574 expire
= timeout
+ jiffies
;
1576 setup_timer_on_stack(&timer
, process_timeout
, (unsigned long)current
);
1577 __mod_timer(&timer
, expire
, false);
1579 del_singleshot_timer_sync(&timer
);
1581 /* Remove the timer from the object tracker */
1582 destroy_timer_on_stack(&timer
);
1584 timeout
= expire
- jiffies
;
1587 return timeout
< 0 ? 0 : timeout
;
1589 EXPORT_SYMBOL(schedule_timeout
);
1592 * We can use __set_current_state() here because schedule_timeout() calls
1593 * schedule() unconditionally.
1595 signed long __sched
schedule_timeout_interruptible(signed long timeout
)
1597 __set_current_state(TASK_INTERRUPTIBLE
);
1598 return schedule_timeout(timeout
);
1600 EXPORT_SYMBOL(schedule_timeout_interruptible
);
1602 signed long __sched
schedule_timeout_killable(signed long timeout
)
1604 __set_current_state(TASK_KILLABLE
);
1605 return schedule_timeout(timeout
);
1607 EXPORT_SYMBOL(schedule_timeout_killable
);
1609 signed long __sched
schedule_timeout_uninterruptible(signed long timeout
)
1611 __set_current_state(TASK_UNINTERRUPTIBLE
);
1612 return schedule_timeout(timeout
);
1614 EXPORT_SYMBOL(schedule_timeout_uninterruptible
);
1617 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1620 signed long __sched
schedule_timeout_idle(signed long timeout
)
1622 __set_current_state(TASK_IDLE
);
1623 return schedule_timeout(timeout
);
1625 EXPORT_SYMBOL(schedule_timeout_idle
);
1627 #ifdef CONFIG_HOTPLUG_CPU
1628 static void migrate_timer_list(struct timer_base
*new_base
, struct hlist_head
*head
)
1630 struct timer_list
*timer
;
1631 int cpu
= new_base
->cpu
;
1633 while (!hlist_empty(head
)) {
1634 timer
= hlist_entry(head
->first
, struct timer_list
, entry
);
1635 detach_timer(timer
, false);
1636 timer
->flags
= (timer
->flags
& ~TIMER_BASEMASK
) | cpu
;
1637 internal_add_timer(new_base
, timer
);
1641 static void migrate_timers(int cpu
)
1643 struct timer_base
*old_base
;
1644 struct timer_base
*new_base
;
1647 BUG_ON(cpu_online(cpu
));
1649 for (b
= 0; b
< NR_BASES
; b
++) {
1650 old_base
= per_cpu_ptr(&timer_bases
[b
], cpu
);
1651 new_base
= get_cpu_ptr(&timer_bases
[b
]);
1653 * The caller is globally serialized and nobody else
1654 * takes two locks at once, deadlock is not possible.
1656 spin_lock_irq(&new_base
->lock
);
1657 spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1659 BUG_ON(old_base
->running_timer
);
1661 for (i
= 0; i
< WHEEL_SIZE
; i
++)
1662 migrate_timer_list(new_base
, old_base
->vectors
+ i
);
1664 spin_unlock(&old_base
->lock
);
1665 spin_unlock_irq(&new_base
->lock
);
1666 put_cpu_ptr(&timer_bases
);
1670 static int timer_cpu_notify(struct notifier_block
*self
,
1671 unsigned long action
, void *hcpu
)
1675 case CPU_DEAD_FROZEN
:
1676 migrate_timers((long)hcpu
);
1685 static inline void timer_register_cpu_notifier(void)
1687 cpu_notifier(timer_cpu_notify
, 0);
1690 static inline void timer_register_cpu_notifier(void) { }
1691 #endif /* CONFIG_HOTPLUG_CPU */
1693 static void __init
init_timer_cpu(int cpu
)
1695 struct timer_base
*base
;
1698 for (i
= 0; i
< NR_BASES
; i
++) {
1699 base
= per_cpu_ptr(&timer_bases
[i
], cpu
);
1701 spin_lock_init(&base
->lock
);
1702 base
->clk
= jiffies
;
1706 static void __init
init_timer_cpus(void)
1710 for_each_possible_cpu(cpu
)
1711 init_timer_cpu(cpu
);
1714 void __init
init_timers(void)
1718 timer_register_cpu_notifier();
1719 open_softirq(TIMER_SOFTIRQ
, run_timer_softirq
);
1723 * msleep - sleep safely even with waitqueue interruptions
1724 * @msecs: Time in milliseconds to sleep for
1726 void msleep(unsigned int msecs
)
1728 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1731 timeout
= schedule_timeout_uninterruptible(timeout
);
1734 EXPORT_SYMBOL(msleep
);
1737 * msleep_interruptible - sleep waiting for signals
1738 * @msecs: Time in milliseconds to sleep for
1740 unsigned long msleep_interruptible(unsigned int msecs
)
1742 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1744 while (timeout
&& !signal_pending(current
))
1745 timeout
= schedule_timeout_interruptible(timeout
);
1746 return jiffies_to_msecs(timeout
);
1749 EXPORT_SYMBOL(msleep_interruptible
);
1751 static void __sched
do_usleep_range(unsigned long min
, unsigned long max
)
1756 kmin
= ktime_set(0, min
* NSEC_PER_USEC
);
1757 delta
= (u64
)(max
- min
) * NSEC_PER_USEC
;
1758 schedule_hrtimeout_range(&kmin
, delta
, HRTIMER_MODE_REL
);
1762 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1763 * @min: Minimum time in usecs to sleep
1764 * @max: Maximum time in usecs to sleep
1766 void __sched
usleep_range(unsigned long min
, unsigned long max
)
1768 __set_current_state(TASK_UNINTERRUPTIBLE
);
1769 do_usleep_range(min
, max
);
1771 EXPORT_SYMBOL(usleep_range
);