4 * Kernel internal timers, basic process system calls
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/slab.h>
44 #include <asm/uaccess.h>
45 #include <asm/unistd.h>
46 #include <asm/div64.h>
47 #include <asm/timex.h>
50 #define CREATE_TRACE_POINTS
51 #include <trace/events/timer.h>
53 u64 jiffies_64 __cacheline_aligned_in_smp
= INITIAL_JIFFIES
;
55 EXPORT_SYMBOL(jiffies_64
);
58 * per-CPU timer vector definitions:
60 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
61 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
62 #define TVN_SIZE (1 << TVN_BITS)
63 #define TVR_SIZE (1 << TVR_BITS)
64 #define TVN_MASK (TVN_SIZE - 1)
65 #define TVR_MASK (TVR_SIZE - 1)
68 struct list_head vec
[TVN_SIZE
];
72 struct list_head vec
[TVR_SIZE
];
77 struct timer_list
*running_timer
;
78 unsigned long timer_jiffies
;
79 unsigned long next_timer
;
80 unsigned long active_timers
;
86 } ____cacheline_aligned
;
88 struct tvec_base boot_tvec_bases
;
89 EXPORT_SYMBOL(boot_tvec_bases
);
90 static DEFINE_PER_CPU(struct tvec_base
*, tvec_bases
) = &boot_tvec_bases
;
92 /* Functions below help us manage 'deferrable' flag */
93 static inline unsigned int tbase_get_deferrable(struct tvec_base
*base
)
95 return ((unsigned int)(unsigned long)base
& TIMER_DEFERRABLE
);
98 static inline struct tvec_base
*tbase_get_base(struct tvec_base
*base
)
100 return ((struct tvec_base
*)((unsigned long)base
& ~TIMER_FLAG_MASK
));
103 static inline void timer_set_deferrable(struct timer_list
*timer
)
105 timer
->base
= TBASE_MAKE_DEFERRED(timer
->base
);
109 timer_set_base(struct timer_list
*timer
, struct tvec_base
*new_base
)
111 unsigned long flags
= (unsigned long)timer
->base
& TIMER_FLAG_MASK
;
113 timer
->base
= (struct tvec_base
*)((unsigned long)(new_base
) | flags
);
116 static unsigned long round_jiffies_common(unsigned long j
, int cpu
,
120 unsigned long original
= j
;
123 * We don't want all cpus firing their timers at once hitting the
124 * same lock or cachelines, so we skew each extra cpu with an extra
125 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
127 * The skew is done by adding 3*cpunr, then round, then subtract this
128 * extra offset again.
135 * If the target jiffie is just after a whole second (which can happen
136 * due to delays of the timer irq, long irq off times etc etc) then
137 * we should round down to the whole second, not up. Use 1/4th second
138 * as cutoff for this rounding as an extreme upper bound for this.
139 * But never round down if @force_up is set.
141 if (rem
< HZ
/4 && !force_up
) /* round down */
146 /* now that we have rounded, subtract the extra skew again */
149 if (j
<= jiffies
) /* rounding ate our timeout entirely; */
155 * __round_jiffies - function to round jiffies to a full second
156 * @j: the time in (absolute) jiffies that should be rounded
157 * @cpu: the processor number on which the timeout will happen
159 * __round_jiffies() rounds an absolute time in the future (in jiffies)
160 * up or down to (approximately) full seconds. This is useful for timers
161 * for which the exact time they fire does not matter too much, as long as
162 * they fire approximately every X seconds.
164 * By rounding these timers to whole seconds, all such timers will fire
165 * at the same time, rather than at various times spread out. The goal
166 * of this is to have the CPU wake up less, which saves power.
168 * The exact rounding is skewed for each processor to avoid all
169 * processors firing at the exact same time, which could lead
170 * to lock contention or spurious cache line bouncing.
172 * The return value is the rounded version of the @j parameter.
174 unsigned long __round_jiffies(unsigned long j
, int cpu
)
176 return round_jiffies_common(j
, cpu
, false);
178 EXPORT_SYMBOL_GPL(__round_jiffies
);
181 * __round_jiffies_relative - function to round jiffies to a full second
182 * @j: the time in (relative) jiffies that should be rounded
183 * @cpu: the processor number on which the timeout will happen
185 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
186 * up or down to (approximately) full seconds. This is useful for timers
187 * for which the exact time they fire does not matter too much, as long as
188 * they fire approximately every X seconds.
190 * By rounding these timers to whole seconds, all such timers will fire
191 * at the same time, rather than at various times spread out. The goal
192 * of this is to have the CPU wake up less, which saves power.
194 * The exact rounding is skewed for each processor to avoid all
195 * processors firing at the exact same time, which could lead
196 * to lock contention or spurious cache line bouncing.
198 * The return value is the rounded version of the @j parameter.
200 unsigned long __round_jiffies_relative(unsigned long j
, int cpu
)
202 unsigned long j0
= jiffies
;
204 /* Use j0 because jiffies might change while we run */
205 return round_jiffies_common(j
+ j0
, cpu
, false) - j0
;
207 EXPORT_SYMBOL_GPL(__round_jiffies_relative
);
210 * round_jiffies - function to round jiffies to a full second
211 * @j: the time in (absolute) jiffies that should be rounded
213 * round_jiffies() rounds an absolute time in the future (in jiffies)
214 * up or down to (approximately) full seconds. This is useful for timers
215 * for which the exact time they fire does not matter too much, as long as
216 * they fire approximately every X seconds.
218 * By rounding these timers to whole seconds, all such timers will fire
219 * at the same time, rather than at various times spread out. The goal
220 * of this is to have the CPU wake up less, which saves power.
222 * The return value is the rounded version of the @j parameter.
224 unsigned long round_jiffies(unsigned long j
)
226 return round_jiffies_common(j
, raw_smp_processor_id(), false);
228 EXPORT_SYMBOL_GPL(round_jiffies
);
231 * round_jiffies_relative - function to round jiffies to a full second
232 * @j: the time in (relative) jiffies that should be rounded
234 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
235 * up or down to (approximately) full seconds. This is useful for timers
236 * for which the exact time they fire does not matter too much, as long as
237 * they fire approximately every X seconds.
239 * By rounding these timers to whole seconds, all such timers will fire
240 * at the same time, rather than at various times spread out. The goal
241 * of this is to have the CPU wake up less, which saves power.
243 * The return value is the rounded version of the @j parameter.
245 unsigned long round_jiffies_relative(unsigned long j
)
247 return __round_jiffies_relative(j
, raw_smp_processor_id());
249 EXPORT_SYMBOL_GPL(round_jiffies_relative
);
252 * __round_jiffies_up - function to round jiffies up to a full second
253 * @j: the time in (absolute) jiffies that should be rounded
254 * @cpu: the processor number on which the timeout will happen
256 * This is the same as __round_jiffies() except that it will never
257 * round down. This is useful for timeouts for which the exact time
258 * of firing does not matter too much, as long as they don't fire too
261 unsigned long __round_jiffies_up(unsigned long j
, int cpu
)
263 return round_jiffies_common(j
, cpu
, true);
265 EXPORT_SYMBOL_GPL(__round_jiffies_up
);
268 * __round_jiffies_up_relative - function to round jiffies up to a full second
269 * @j: the time in (relative) jiffies that should be rounded
270 * @cpu: the processor number on which the timeout will happen
272 * This is the same as __round_jiffies_relative() except that it will never
273 * round down. This is useful for timeouts for which the exact time
274 * of firing does not matter too much, as long as they don't fire too
277 unsigned long __round_jiffies_up_relative(unsigned long j
, int cpu
)
279 unsigned long j0
= jiffies
;
281 /* Use j0 because jiffies might change while we run */
282 return round_jiffies_common(j
+ j0
, cpu
, true) - j0
;
284 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative
);
287 * round_jiffies_up - function to round jiffies up to a full second
288 * @j: the time in (absolute) jiffies that should be rounded
290 * This is the same as round_jiffies() except that it will never
291 * round down. This is useful for timeouts for which the exact time
292 * of firing does not matter too much, as long as they don't fire too
295 unsigned long round_jiffies_up(unsigned long j
)
297 return round_jiffies_common(j
, raw_smp_processor_id(), true);
299 EXPORT_SYMBOL_GPL(round_jiffies_up
);
302 * round_jiffies_up_relative - function to round jiffies up to a full second
303 * @j: the time in (relative) jiffies that should be rounded
305 * This is the same as round_jiffies_relative() except that it will never
306 * round down. This is useful for timeouts for which the exact time
307 * of firing does not matter too much, as long as they don't fire too
310 unsigned long round_jiffies_up_relative(unsigned long j
)
312 return __round_jiffies_up_relative(j
, raw_smp_processor_id());
314 EXPORT_SYMBOL_GPL(round_jiffies_up_relative
);
317 * set_timer_slack - set the allowed slack for a timer
318 * @timer: the timer to be modified
319 * @slack_hz: the amount of time (in jiffies) allowed for rounding
321 * Set the amount of time, in jiffies, that a certain timer has
322 * in terms of slack. By setting this value, the timer subsystem
323 * will schedule the actual timer somewhere between
324 * the time mod_timer() asks for, and that time plus the slack.
326 * By setting the slack to -1, a percentage of the delay is used
329 void set_timer_slack(struct timer_list
*timer
, int slack_hz
)
331 timer
->slack
= slack_hz
;
333 EXPORT_SYMBOL_GPL(set_timer_slack
);
336 __internal_add_timer(struct tvec_base
*base
, struct timer_list
*timer
)
338 unsigned long expires
= timer
->expires
;
339 unsigned long idx
= expires
- base
->timer_jiffies
;
340 struct list_head
*vec
;
342 if (idx
< TVR_SIZE
) {
343 int i
= expires
& TVR_MASK
;
344 vec
= base
->tv1
.vec
+ i
;
345 } else if (idx
< 1 << (TVR_BITS
+ TVN_BITS
)) {
346 int i
= (expires
>> TVR_BITS
) & TVN_MASK
;
347 vec
= base
->tv2
.vec
+ i
;
348 } else if (idx
< 1 << (TVR_BITS
+ 2 * TVN_BITS
)) {
349 int i
= (expires
>> (TVR_BITS
+ TVN_BITS
)) & TVN_MASK
;
350 vec
= base
->tv3
.vec
+ i
;
351 } else if (idx
< 1 << (TVR_BITS
+ 3 * TVN_BITS
)) {
352 int i
= (expires
>> (TVR_BITS
+ 2 * TVN_BITS
)) & TVN_MASK
;
353 vec
= base
->tv4
.vec
+ i
;
354 } else if ((signed long) idx
< 0) {
356 * Can happen if you add a timer with expires == jiffies,
357 * or you set a timer to go off in the past
359 vec
= base
->tv1
.vec
+ (base
->timer_jiffies
& TVR_MASK
);
362 /* If the timeout is larger than 0xffffffff on 64-bit
363 * architectures then we use the maximum timeout:
365 if (idx
> 0xffffffffUL
) {
367 expires
= idx
+ base
->timer_jiffies
;
369 i
= (expires
>> (TVR_BITS
+ 3 * TVN_BITS
)) & TVN_MASK
;
370 vec
= base
->tv5
.vec
+ i
;
375 list_add_tail(&timer
->entry
, vec
);
378 static void internal_add_timer(struct tvec_base
*base
, struct timer_list
*timer
)
380 __internal_add_timer(base
, timer
);
382 * Update base->active_timers and base->next_timer
384 if (!tbase_get_deferrable(timer
->base
)) {
385 if (time_before(timer
->expires
, base
->next_timer
))
386 base
->next_timer
= timer
->expires
;
387 base
->active_timers
++;
391 #ifdef CONFIG_TIMER_STATS
392 void __timer_stats_timer_set_start_info(struct timer_list
*timer
, void *addr
)
394 if (timer
->start_site
)
397 timer
->start_site
= addr
;
398 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
399 timer
->start_pid
= current
->pid
;
402 static void timer_stats_account_timer(struct timer_list
*timer
)
404 unsigned int flag
= 0;
406 if (likely(!timer
->start_site
))
408 if (unlikely(tbase_get_deferrable(timer
->base
)))
409 flag
|= TIMER_STATS_FLAG_DEFERRABLE
;
411 timer_stats_update_stats(timer
, timer
->start_pid
, timer
->start_site
,
412 timer
->function
, timer
->start_comm
, flag
);
416 static void timer_stats_account_timer(struct timer_list
*timer
) {}
419 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
421 static struct debug_obj_descr timer_debug_descr
;
423 static void *timer_debug_hint(void *addr
)
425 return ((struct timer_list
*) addr
)->function
;
429 * fixup_init is called when:
430 * - an active object is initialized
432 static int timer_fixup_init(void *addr
, enum debug_obj_state state
)
434 struct timer_list
*timer
= addr
;
437 case ODEBUG_STATE_ACTIVE
:
438 del_timer_sync(timer
);
439 debug_object_init(timer
, &timer_debug_descr
);
446 /* Stub timer callback for improperly used timers. */
447 static void stub_timer(unsigned long data
)
453 * fixup_activate is called when:
454 * - an active object is activated
455 * - an unknown object is activated (might be a statically initialized object)
457 static int timer_fixup_activate(void *addr
, enum debug_obj_state state
)
459 struct timer_list
*timer
= addr
;
463 case ODEBUG_STATE_NOTAVAILABLE
:
465 * This is not really a fixup. The timer was
466 * statically initialized. We just make sure that it
467 * is tracked in the object tracker.
469 if (timer
->entry
.next
== NULL
&&
470 timer
->entry
.prev
== TIMER_ENTRY_STATIC
) {
471 debug_object_init(timer
, &timer_debug_descr
);
472 debug_object_activate(timer
, &timer_debug_descr
);
475 setup_timer(timer
, stub_timer
, 0);
480 case ODEBUG_STATE_ACTIVE
:
489 * fixup_free is called when:
490 * - an active object is freed
492 static int timer_fixup_free(void *addr
, enum debug_obj_state state
)
494 struct timer_list
*timer
= addr
;
497 case ODEBUG_STATE_ACTIVE
:
498 del_timer_sync(timer
);
499 debug_object_free(timer
, &timer_debug_descr
);
507 * fixup_assert_init is called when:
508 * - an untracked/uninit-ed object is found
510 static int timer_fixup_assert_init(void *addr
, enum debug_obj_state state
)
512 struct timer_list
*timer
= addr
;
515 case ODEBUG_STATE_NOTAVAILABLE
:
516 if (timer
->entry
.prev
== TIMER_ENTRY_STATIC
) {
518 * This is not really a fixup. The timer was
519 * statically initialized. We just make sure that it
520 * is tracked in the object tracker.
522 debug_object_init(timer
, &timer_debug_descr
);
525 setup_timer(timer
, stub_timer
, 0);
533 static struct debug_obj_descr timer_debug_descr
= {
534 .name
= "timer_list",
535 .debug_hint
= timer_debug_hint
,
536 .fixup_init
= timer_fixup_init
,
537 .fixup_activate
= timer_fixup_activate
,
538 .fixup_free
= timer_fixup_free
,
539 .fixup_assert_init
= timer_fixup_assert_init
,
542 static inline void debug_timer_init(struct timer_list
*timer
)
544 debug_object_init(timer
, &timer_debug_descr
);
547 static inline void debug_timer_activate(struct timer_list
*timer
)
549 debug_object_activate(timer
, &timer_debug_descr
);
552 static inline void debug_timer_deactivate(struct timer_list
*timer
)
554 debug_object_deactivate(timer
, &timer_debug_descr
);
557 static inline void debug_timer_free(struct timer_list
*timer
)
559 debug_object_free(timer
, &timer_debug_descr
);
562 static inline void debug_timer_assert_init(struct timer_list
*timer
)
564 debug_object_assert_init(timer
, &timer_debug_descr
);
567 static void __init_timer(struct timer_list
*timer
,
569 struct lock_class_key
*key
);
571 void init_timer_on_stack_key(struct timer_list
*timer
,
573 struct lock_class_key
*key
)
575 debug_object_init_on_stack(timer
, &timer_debug_descr
);
576 __init_timer(timer
, name
, key
);
578 EXPORT_SYMBOL_GPL(init_timer_on_stack_key
);
580 void destroy_timer_on_stack(struct timer_list
*timer
)
582 debug_object_free(timer
, &timer_debug_descr
);
584 EXPORT_SYMBOL_GPL(destroy_timer_on_stack
);
587 static inline void debug_timer_init(struct timer_list
*timer
) { }
588 static inline void debug_timer_activate(struct timer_list
*timer
) { }
589 static inline void debug_timer_deactivate(struct timer_list
*timer
) { }
590 static inline void debug_timer_assert_init(struct timer_list
*timer
) { }
593 static inline void debug_init(struct timer_list
*timer
)
595 debug_timer_init(timer
);
596 trace_timer_init(timer
);
600 debug_activate(struct timer_list
*timer
, unsigned long expires
)
602 debug_timer_activate(timer
);
603 trace_timer_start(timer
, expires
);
606 static inline void debug_deactivate(struct timer_list
*timer
)
608 debug_timer_deactivate(timer
);
609 trace_timer_cancel(timer
);
612 static inline void debug_assert_init(struct timer_list
*timer
)
614 debug_timer_assert_init(timer
);
617 static void __init_timer(struct timer_list
*timer
,
619 struct lock_class_key
*key
)
621 timer
->entry
.next
= NULL
;
622 timer
->base
= __raw_get_cpu_var(tvec_bases
);
624 #ifdef CONFIG_TIMER_STATS
625 timer
->start_site
= NULL
;
626 timer
->start_pid
= -1;
627 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
629 lockdep_init_map(&timer
->lockdep_map
, name
, key
, 0);
632 void setup_deferrable_timer_on_stack_key(struct timer_list
*timer
,
634 struct lock_class_key
*key
,
635 void (*function
)(unsigned long),
638 timer
->function
= function
;
640 init_timer_on_stack_key(timer
, name
, key
);
641 timer_set_deferrable(timer
);
643 EXPORT_SYMBOL_GPL(setup_deferrable_timer_on_stack_key
);
646 * init_timer_key - initialize a timer
647 * @timer: the timer to be initialized
648 * @name: name of the timer
649 * @key: lockdep class key of the fake lock used for tracking timer
650 * sync lock dependencies
652 * init_timer_key() must be done to a timer prior calling *any* of the
653 * other timer functions.
655 void init_timer_key(struct timer_list
*timer
,
657 struct lock_class_key
*key
)
660 __init_timer(timer
, name
, key
);
662 EXPORT_SYMBOL(init_timer_key
);
664 void init_timer_deferrable_key(struct timer_list
*timer
,
666 struct lock_class_key
*key
)
668 init_timer_key(timer
, name
, key
);
669 timer_set_deferrable(timer
);
671 EXPORT_SYMBOL(init_timer_deferrable_key
);
673 static inline void detach_timer(struct timer_list
*timer
, bool clear_pending
)
675 struct list_head
*entry
= &timer
->entry
;
677 debug_deactivate(timer
);
679 __list_del(entry
->prev
, entry
->next
);
682 entry
->prev
= LIST_POISON2
;
686 detach_expired_timer(struct timer_list
*timer
, struct tvec_base
*base
)
688 detach_timer(timer
, true);
689 if (!tbase_get_deferrable(timer
->base
))
690 base
->active_timers
--;
693 static int detach_if_pending(struct timer_list
*timer
, struct tvec_base
*base
,
696 if (!timer_pending(timer
))
699 detach_timer(timer
, clear_pending
);
700 if (!tbase_get_deferrable(timer
->base
)) {
701 base
->active_timers
--;
702 if (timer
->expires
== base
->next_timer
)
703 base
->next_timer
= base
->timer_jiffies
;
709 * We are using hashed locking: holding per_cpu(tvec_bases).lock
710 * means that all timers which are tied to this base via timer->base are
711 * locked, and the base itself is locked too.
713 * So __run_timers/migrate_timers can safely modify all timers which could
714 * be found on ->tvX lists.
716 * When the timer's base is locked, and the timer removed from list, it is
717 * possible to set timer->base = NULL and drop the lock: the timer remains
720 static struct tvec_base
*lock_timer_base(struct timer_list
*timer
,
721 unsigned long *flags
)
722 __acquires(timer
->base
->lock
)
724 struct tvec_base
*base
;
727 struct tvec_base
*prelock_base
= timer
->base
;
728 base
= tbase_get_base(prelock_base
);
729 if (likely(base
!= NULL
)) {
730 spin_lock_irqsave(&base
->lock
, *flags
);
731 if (likely(prelock_base
== timer
->base
))
733 /* The timer has migrated to another CPU */
734 spin_unlock_irqrestore(&base
->lock
, *flags
);
741 __mod_timer(struct timer_list
*timer
, unsigned long expires
,
742 bool pending_only
, int pinned
)
744 struct tvec_base
*base
, *new_base
;
748 timer_stats_timer_set_start_info(timer
);
749 BUG_ON(!timer
->function
);
751 base
= lock_timer_base(timer
, &flags
);
753 ret
= detach_if_pending(timer
, base
, false);
754 if (!ret
&& pending_only
)
757 debug_activate(timer
, expires
);
759 cpu
= smp_processor_id();
761 #if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
762 if (!pinned
&& get_sysctl_timer_migration() && idle_cpu(cpu
))
763 cpu
= get_nohz_timer_target();
765 new_base
= per_cpu(tvec_bases
, cpu
);
767 if (base
!= new_base
) {
769 * We are trying to schedule the timer on the local CPU.
770 * However we can't change timer's base while it is running,
771 * otherwise del_timer_sync() can't detect that the timer's
772 * handler yet has not finished. This also guarantees that
773 * the timer is serialized wrt itself.
775 if (likely(base
->running_timer
!= timer
)) {
776 /* See the comment in lock_timer_base() */
777 timer_set_base(timer
, NULL
);
778 spin_unlock(&base
->lock
);
780 spin_lock(&base
->lock
);
781 timer_set_base(timer
, base
);
785 timer
->expires
= expires
;
786 internal_add_timer(base
, timer
);
789 spin_unlock_irqrestore(&base
->lock
, flags
);
795 * mod_timer_pending - modify a pending timer's timeout
796 * @timer: the pending timer to be modified
797 * @expires: new timeout in jiffies
799 * mod_timer_pending() is the same for pending timers as mod_timer(),
800 * but will not re-activate and modify already deleted timers.
802 * It is useful for unserialized use of timers.
804 int mod_timer_pending(struct timer_list
*timer
, unsigned long expires
)
806 return __mod_timer(timer
, expires
, true, TIMER_NOT_PINNED
);
808 EXPORT_SYMBOL(mod_timer_pending
);
811 * Decide where to put the timer while taking the slack into account
814 * 1) calculate the maximum (absolute) time
815 * 2) calculate the highest bit where the expires and new max are different
816 * 3) use this bit to make a mask
817 * 4) use the bitmask to round down the maximum time, so that all last
821 unsigned long apply_slack(struct timer_list
*timer
, unsigned long expires
)
823 unsigned long expires_limit
, mask
;
826 if (timer
->slack
>= 0) {
827 expires_limit
= expires
+ timer
->slack
;
829 long delta
= expires
- jiffies
;
834 expires_limit
= expires
+ delta
/ 256;
836 mask
= expires
^ expires_limit
;
840 bit
= find_last_bit(&mask
, BITS_PER_LONG
);
842 mask
= (1 << bit
) - 1;
844 expires_limit
= expires_limit
& ~(mask
);
846 return expires_limit
;
850 * mod_timer - modify a timer's timeout
851 * @timer: the timer to be modified
852 * @expires: new timeout in jiffies
854 * mod_timer() is a more efficient way to update the expire field of an
855 * active timer (if the timer is inactive it will be activated)
857 * mod_timer(timer, expires) is equivalent to:
859 * del_timer(timer); timer->expires = expires; add_timer(timer);
861 * Note that if there are multiple unserialized concurrent users of the
862 * same timer, then mod_timer() is the only safe way to modify the timeout,
863 * since add_timer() cannot modify an already running timer.
865 * The function returns whether it has modified a pending timer or not.
866 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
867 * active timer returns 1.)
869 int mod_timer(struct timer_list
*timer
, unsigned long expires
)
871 expires
= apply_slack(timer
, expires
);
874 * This is a common optimization triggered by the
875 * networking code - if the timer is re-modified
876 * to be the same thing then just return:
878 if (timer_pending(timer
) && timer
->expires
== expires
)
881 return __mod_timer(timer
, expires
, false, TIMER_NOT_PINNED
);
883 EXPORT_SYMBOL(mod_timer
);
886 * mod_timer_pinned - modify a timer's timeout
887 * @timer: the timer to be modified
888 * @expires: new timeout in jiffies
890 * mod_timer_pinned() is a way to update the expire field of an
891 * active timer (if the timer is inactive it will be activated)
892 * and to ensure that the timer is scheduled on the current CPU.
894 * Note that this does not prevent the timer from being migrated
895 * when the current CPU goes offline. If this is a problem for
896 * you, use CPU-hotplug notifiers to handle it correctly, for
897 * example, cancelling the timer when the corresponding CPU goes
900 * mod_timer_pinned(timer, expires) is equivalent to:
902 * del_timer(timer); timer->expires = expires; add_timer(timer);
904 int mod_timer_pinned(struct timer_list
*timer
, unsigned long expires
)
906 if (timer
->expires
== expires
&& timer_pending(timer
))
909 return __mod_timer(timer
, expires
, false, TIMER_PINNED
);
911 EXPORT_SYMBOL(mod_timer_pinned
);
914 * add_timer - start a timer
915 * @timer: the timer to be added
917 * The kernel will do a ->function(->data) callback from the
918 * timer interrupt at the ->expires point in the future. The
919 * current time is 'jiffies'.
921 * The timer's ->expires, ->function (and if the handler uses it, ->data)
922 * fields must be set prior calling this function.
924 * Timers with an ->expires field in the past will be executed in the next
927 void add_timer(struct timer_list
*timer
)
929 BUG_ON(timer_pending(timer
));
930 mod_timer(timer
, timer
->expires
);
932 EXPORT_SYMBOL(add_timer
);
935 * add_timer_on - start a timer on a particular CPU
936 * @timer: the timer to be added
937 * @cpu: the CPU to start it on
939 * This is not very scalable on SMP. Double adds are not possible.
941 void add_timer_on(struct timer_list
*timer
, int cpu
)
943 struct tvec_base
*base
= per_cpu(tvec_bases
, cpu
);
946 timer_stats_timer_set_start_info(timer
);
947 BUG_ON(timer_pending(timer
) || !timer
->function
);
948 spin_lock_irqsave(&base
->lock
, flags
);
949 timer_set_base(timer
, base
);
950 debug_activate(timer
, timer
->expires
);
951 internal_add_timer(base
, timer
);
953 * Check whether the other CPU is idle and needs to be
954 * triggered to reevaluate the timer wheel when nohz is
955 * active. We are protected against the other CPU fiddling
956 * with the timer by holding the timer base lock. This also
957 * makes sure that a CPU on the way to idle can not evaluate
960 wake_up_idle_cpu(cpu
);
961 spin_unlock_irqrestore(&base
->lock
, flags
);
963 EXPORT_SYMBOL_GPL(add_timer_on
);
966 * del_timer - deactive a timer.
967 * @timer: the timer to be deactivated
969 * del_timer() deactivates a timer - this works on both active and inactive
972 * The function returns whether it has deactivated a pending timer or not.
973 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
974 * active timer returns 1.)
976 int del_timer(struct timer_list
*timer
)
978 struct tvec_base
*base
;
982 debug_assert_init(timer
);
984 timer_stats_timer_clear_start_info(timer
);
985 if (timer_pending(timer
)) {
986 base
= lock_timer_base(timer
, &flags
);
987 ret
= detach_if_pending(timer
, base
, true);
988 spin_unlock_irqrestore(&base
->lock
, flags
);
993 EXPORT_SYMBOL(del_timer
);
996 * try_to_del_timer_sync - Try to deactivate a timer
997 * @timer: timer do del
999 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1000 * exit the timer is not queued and the handler is not running on any CPU.
1002 int try_to_del_timer_sync(struct timer_list
*timer
)
1004 struct tvec_base
*base
;
1005 unsigned long flags
;
1008 debug_assert_init(timer
);
1010 base
= lock_timer_base(timer
, &flags
);
1012 if (base
->running_timer
!= timer
) {
1013 timer_stats_timer_clear_start_info(timer
);
1014 ret
= detach_if_pending(timer
, base
, true);
1016 spin_unlock_irqrestore(&base
->lock
, flags
);
1020 EXPORT_SYMBOL(try_to_del_timer_sync
);
1024 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1025 * @timer: the timer to be deactivated
1027 * This function only differs from del_timer() on SMP: besides deactivating
1028 * the timer it also makes sure the handler has finished executing on other
1031 * Synchronization rules: Callers must prevent restarting of the timer,
1032 * otherwise this function is meaningless. It must not be called from
1033 * interrupt contexts. The caller must not hold locks which would prevent
1034 * completion of the timer's handler. The timer's handler must not call
1035 * add_timer_on(). Upon exit the timer is not queued and the handler is
1036 * not running on any CPU.
1038 * Note: You must not hold locks that are held in interrupt context
1039 * while calling this function. Even if the lock has nothing to do
1040 * with the timer in question. Here's why:
1046 * base->running_timer = mytimer;
1047 * spin_lock_irq(somelock);
1049 * spin_lock(somelock);
1050 * del_timer_sync(mytimer);
1051 * while (base->running_timer == mytimer);
1053 * Now del_timer_sync() will never return and never release somelock.
1054 * The interrupt on the other CPU is waiting to grab somelock but
1055 * it has interrupted the softirq that CPU0 is waiting to finish.
1057 * The function returns whether it has deactivated a pending timer or not.
1059 int del_timer_sync(struct timer_list
*timer
)
1061 #ifdef CONFIG_LOCKDEP
1062 unsigned long flags
;
1065 * If lockdep gives a backtrace here, please reference
1066 * the synchronization rules above.
1068 local_irq_save(flags
);
1069 lock_map_acquire(&timer
->lockdep_map
);
1070 lock_map_release(&timer
->lockdep_map
);
1071 local_irq_restore(flags
);
1074 * don't use it in hardirq context, because it
1075 * could lead to deadlock.
1079 int ret
= try_to_del_timer_sync(timer
);
1085 EXPORT_SYMBOL(del_timer_sync
);
1088 static int cascade(struct tvec_base
*base
, struct tvec
*tv
, int index
)
1090 /* cascade all the timers from tv up one level */
1091 struct timer_list
*timer
, *tmp
;
1092 struct list_head tv_list
;
1094 list_replace_init(tv
->vec
+ index
, &tv_list
);
1097 * We are removing _all_ timers from the list, so we
1098 * don't have to detach them individually.
1100 list_for_each_entry_safe(timer
, tmp
, &tv_list
, entry
) {
1101 BUG_ON(tbase_get_base(timer
->base
) != base
);
1102 /* No accounting, while moving them */
1103 __internal_add_timer(base
, timer
);
1109 static void call_timer_fn(struct timer_list
*timer
, void (*fn
)(unsigned long),
1112 int preempt_count
= preempt_count();
1114 #ifdef CONFIG_LOCKDEP
1116 * It is permissible to free the timer from inside the
1117 * function that is called from it, this we need to take into
1118 * account for lockdep too. To avoid bogus "held lock freed"
1119 * warnings as well as problems when looking into
1120 * timer->lockdep_map, make a copy and use that here.
1122 struct lockdep_map lockdep_map
;
1124 lockdep_copy_map(&lockdep_map
, &timer
->lockdep_map
);
1127 * Couple the lock chain with the lock chain at
1128 * del_timer_sync() by acquiring the lock_map around the fn()
1129 * call here and in del_timer_sync().
1131 lock_map_acquire(&lockdep_map
);
1133 trace_timer_expire_entry(timer
);
1135 trace_timer_expire_exit(timer
);
1137 lock_map_release(&lockdep_map
);
1139 if (preempt_count
!= preempt_count()) {
1140 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1141 fn
, preempt_count
, preempt_count());
1143 * Restore the preempt count. That gives us a decent
1144 * chance to survive and extract information. If the
1145 * callback kept a lock held, bad luck, but not worse
1146 * than the BUG() we had.
1148 preempt_count() = preempt_count
;
1152 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1155 * __run_timers - run all expired timers (if any) on this CPU.
1156 * @base: the timer vector to be processed.
1158 * This function cascades all vectors and executes all expired timer
1161 static inline void __run_timers(struct tvec_base
*base
)
1163 struct timer_list
*timer
;
1165 spin_lock_irq(&base
->lock
);
1166 while (time_after_eq(jiffies
, base
->timer_jiffies
)) {
1167 struct list_head work_list
;
1168 struct list_head
*head
= &work_list
;
1169 int index
= base
->timer_jiffies
& TVR_MASK
;
1175 (!cascade(base
, &base
->tv2
, INDEX(0))) &&
1176 (!cascade(base
, &base
->tv3
, INDEX(1))) &&
1177 !cascade(base
, &base
->tv4
, INDEX(2)))
1178 cascade(base
, &base
->tv5
, INDEX(3));
1179 ++base
->timer_jiffies
;
1180 list_replace_init(base
->tv1
.vec
+ index
, &work_list
);
1181 while (!list_empty(head
)) {
1182 void (*fn
)(unsigned long);
1185 timer
= list_first_entry(head
, struct timer_list
,entry
);
1186 fn
= timer
->function
;
1189 timer_stats_account_timer(timer
);
1191 base
->running_timer
= timer
;
1192 detach_expired_timer(timer
, base
);
1194 spin_unlock_irq(&base
->lock
);
1195 call_timer_fn(timer
, fn
, data
);
1196 spin_lock_irq(&base
->lock
);
1199 base
->running_timer
= NULL
;
1200 spin_unlock_irq(&base
->lock
);
1205 * Find out when the next timer event is due to happen. This
1206 * is used on S/390 to stop all activity when a CPU is idle.
1207 * This function needs to be called with interrupts disabled.
1209 static unsigned long __next_timer_interrupt(struct tvec_base
*base
)
1211 unsigned long timer_jiffies
= base
->timer_jiffies
;
1212 unsigned long expires
= timer_jiffies
+ NEXT_TIMER_MAX_DELTA
;
1213 int index
, slot
, array
, found
= 0;
1214 struct timer_list
*nte
;
1215 struct tvec
*varray
[4];
1217 /* Look for timer events in tv1. */
1218 index
= slot
= timer_jiffies
& TVR_MASK
;
1220 list_for_each_entry(nte
, base
->tv1
.vec
+ slot
, entry
) {
1221 if (tbase_get_deferrable(nte
->base
))
1225 expires
= nte
->expires
;
1226 /* Look at the cascade bucket(s)? */
1227 if (!index
|| slot
< index
)
1231 slot
= (slot
+ 1) & TVR_MASK
;
1232 } while (slot
!= index
);
1235 /* Calculate the next cascade event */
1237 timer_jiffies
+= TVR_SIZE
- index
;
1238 timer_jiffies
>>= TVR_BITS
;
1240 /* Check tv2-tv5. */
1241 varray
[0] = &base
->tv2
;
1242 varray
[1] = &base
->tv3
;
1243 varray
[2] = &base
->tv4
;
1244 varray
[3] = &base
->tv5
;
1246 for (array
= 0; array
< 4; array
++) {
1247 struct tvec
*varp
= varray
[array
];
1249 index
= slot
= timer_jiffies
& TVN_MASK
;
1251 list_for_each_entry(nte
, varp
->vec
+ slot
, entry
) {
1252 if (tbase_get_deferrable(nte
->base
))
1256 if (time_before(nte
->expires
, expires
))
1257 expires
= nte
->expires
;
1260 * Do we still search for the first timer or are
1261 * we looking up the cascade buckets ?
1264 /* Look at the cascade bucket(s)? */
1265 if (!index
|| slot
< index
)
1269 slot
= (slot
+ 1) & TVN_MASK
;
1270 } while (slot
!= index
);
1273 timer_jiffies
+= TVN_SIZE
- index
;
1274 timer_jiffies
>>= TVN_BITS
;
1280 * Check, if the next hrtimer event is before the next timer wheel
1283 static unsigned long cmp_next_hrtimer_event(unsigned long now
,
1284 unsigned long expires
)
1286 ktime_t hr_delta
= hrtimer_get_next_event();
1287 struct timespec tsdelta
;
1288 unsigned long delta
;
1290 if (hr_delta
.tv64
== KTIME_MAX
)
1294 * Expired timer available, let it expire in the next tick
1296 if (hr_delta
.tv64
<= 0)
1299 tsdelta
= ktime_to_timespec(hr_delta
);
1300 delta
= timespec_to_jiffies(&tsdelta
);
1303 * Limit the delta to the max value, which is checked in
1304 * tick_nohz_stop_sched_tick():
1306 if (delta
> NEXT_TIMER_MAX_DELTA
)
1307 delta
= NEXT_TIMER_MAX_DELTA
;
1310 * Take rounding errors in to account and make sure, that it
1311 * expires in the next tick. Otherwise we go into an endless
1312 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1318 if (time_before(now
, expires
))
1324 * get_next_timer_interrupt - return the jiffy of the next pending timer
1325 * @now: current time (in jiffies)
1327 unsigned long get_next_timer_interrupt(unsigned long now
)
1329 struct tvec_base
*base
= __this_cpu_read(tvec_bases
);
1330 unsigned long expires
= now
+ NEXT_TIMER_MAX_DELTA
;
1333 * Pretend that there is no timer pending if the cpu is offline.
1334 * Possible pending timers will be migrated later to an active cpu.
1336 if (cpu_is_offline(smp_processor_id()))
1339 spin_lock(&base
->lock
);
1340 if (base
->active_timers
) {
1341 if (time_before_eq(base
->next_timer
, base
->timer_jiffies
))
1342 base
->next_timer
= __next_timer_interrupt(base
);
1343 expires
= base
->next_timer
;
1345 spin_unlock(&base
->lock
);
1347 if (time_before_eq(expires
, now
))
1350 return cmp_next_hrtimer_event(now
, expires
);
1355 * Called from the timer interrupt handler to charge one tick to the current
1356 * process. user_tick is 1 if the tick is user time, 0 for system.
1358 void update_process_times(int user_tick
)
1360 struct task_struct
*p
= current
;
1361 int cpu
= smp_processor_id();
1363 /* Note: this timer irq context must be accounted for as well. */
1364 account_process_tick(p
, user_tick
);
1366 rcu_check_callbacks(cpu
, user_tick
);
1368 #ifdef CONFIG_IRQ_WORK
1373 run_posix_cpu_timers(p
);
1377 * This function runs timers and the timer-tq in bottom half context.
1379 static void run_timer_softirq(struct softirq_action
*h
)
1381 struct tvec_base
*base
= __this_cpu_read(tvec_bases
);
1383 hrtimer_run_pending();
1385 if (time_after_eq(jiffies
, base
->timer_jiffies
))
1390 * Called by the local, per-CPU timer interrupt on SMP.
1392 void run_local_timers(void)
1394 hrtimer_run_queues();
1395 raise_softirq(TIMER_SOFTIRQ
);
1398 #ifdef __ARCH_WANT_SYS_ALARM
1401 * For backwards compatibility? This can be done in libc so Alpha
1402 * and all newer ports shouldn't need it.
1404 SYSCALL_DEFINE1(alarm
, unsigned int, seconds
)
1406 return alarm_setitimer(seconds
);
1414 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1415 * should be moved into arch/i386 instead?
1419 * sys_getpid - return the thread group id of the current process
1421 * Note, despite the name, this returns the tgid not the pid. The tgid and
1422 * the pid are identical unless CLONE_THREAD was specified on clone() in
1423 * which case the tgid is the same in all threads of the same group.
1425 * This is SMP safe as current->tgid does not change.
1427 SYSCALL_DEFINE0(getpid
)
1429 return task_tgid_vnr(current
);
1433 * Accessing ->real_parent is not SMP-safe, it could
1434 * change from under us. However, we can use a stale
1435 * value of ->real_parent under rcu_read_lock(), see
1436 * release_task()->call_rcu(delayed_put_task_struct).
1438 SYSCALL_DEFINE0(getppid
)
1443 pid
= task_tgid_vnr(rcu_dereference(current
->real_parent
));
1449 SYSCALL_DEFINE0(getuid
)
1451 /* Only we change this so SMP safe */
1452 return from_kuid_munged(current_user_ns(), current_uid());
1455 SYSCALL_DEFINE0(geteuid
)
1457 /* Only we change this so SMP safe */
1458 return from_kuid_munged(current_user_ns(), current_euid());
1461 SYSCALL_DEFINE0(getgid
)
1463 /* Only we change this so SMP safe */
1464 return from_kgid_munged(current_user_ns(), current_gid());
1467 SYSCALL_DEFINE0(getegid
)
1469 /* Only we change this so SMP safe */
1470 return from_kgid_munged(current_user_ns(), current_egid());
1475 static void process_timeout(unsigned long __data
)
1477 wake_up_process((struct task_struct
*)__data
);
1481 * schedule_timeout - sleep until timeout
1482 * @timeout: timeout value in jiffies
1484 * Make the current task sleep until @timeout jiffies have
1485 * elapsed. The routine will return immediately unless
1486 * the current task state has been set (see set_current_state()).
1488 * You can set the task state as follows -
1490 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1491 * pass before the routine returns. The routine will return 0
1493 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1494 * delivered to the current task. In this case the remaining time
1495 * in jiffies will be returned, or 0 if the timer expired in time
1497 * The current task state is guaranteed to be TASK_RUNNING when this
1500 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1501 * the CPU away without a bound on the timeout. In this case the return
1502 * value will be %MAX_SCHEDULE_TIMEOUT.
1504 * In all cases the return value is guaranteed to be non-negative.
1506 signed long __sched
schedule_timeout(signed long timeout
)
1508 struct timer_list timer
;
1509 unsigned long expire
;
1513 case MAX_SCHEDULE_TIMEOUT
:
1515 * These two special cases are useful to be comfortable
1516 * in the caller. Nothing more. We could take
1517 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1518 * but I' d like to return a valid offset (>=0) to allow
1519 * the caller to do everything it want with the retval.
1525 * Another bit of PARANOID. Note that the retval will be
1526 * 0 since no piece of kernel is supposed to do a check
1527 * for a negative retval of schedule_timeout() (since it
1528 * should never happens anyway). You just have the printk()
1529 * that will tell you if something is gone wrong and where.
1532 printk(KERN_ERR
"schedule_timeout: wrong timeout "
1533 "value %lx\n", timeout
);
1535 current
->state
= TASK_RUNNING
;
1540 expire
= timeout
+ jiffies
;
1542 setup_timer_on_stack(&timer
, process_timeout
, (unsigned long)current
);
1543 __mod_timer(&timer
, expire
, false, TIMER_NOT_PINNED
);
1545 del_singleshot_timer_sync(&timer
);
1547 /* Remove the timer from the object tracker */
1548 destroy_timer_on_stack(&timer
);
1550 timeout
= expire
- jiffies
;
1553 return timeout
< 0 ? 0 : timeout
;
1555 EXPORT_SYMBOL(schedule_timeout
);
1558 * We can use __set_current_state() here because schedule_timeout() calls
1559 * schedule() unconditionally.
1561 signed long __sched
schedule_timeout_interruptible(signed long timeout
)
1563 __set_current_state(TASK_INTERRUPTIBLE
);
1564 return schedule_timeout(timeout
);
1566 EXPORT_SYMBOL(schedule_timeout_interruptible
);
1568 signed long __sched
schedule_timeout_killable(signed long timeout
)
1570 __set_current_state(TASK_KILLABLE
);
1571 return schedule_timeout(timeout
);
1573 EXPORT_SYMBOL(schedule_timeout_killable
);
1575 signed long __sched
schedule_timeout_uninterruptible(signed long timeout
)
1577 __set_current_state(TASK_UNINTERRUPTIBLE
);
1578 return schedule_timeout(timeout
);
1580 EXPORT_SYMBOL(schedule_timeout_uninterruptible
);
1582 /* Thread ID - the internal kernel "pid" */
1583 SYSCALL_DEFINE0(gettid
)
1585 return task_pid_vnr(current
);
1589 * do_sysinfo - fill in sysinfo struct
1590 * @info: pointer to buffer to fill
1592 int do_sysinfo(struct sysinfo
*info
)
1594 unsigned long mem_total
, sav_total
;
1595 unsigned int mem_unit
, bitcount
;
1598 memset(info
, 0, sizeof(struct sysinfo
));
1601 monotonic_to_bootbased(&tp
);
1602 info
->uptime
= tp
.tv_sec
+ (tp
.tv_nsec
? 1 : 0);
1604 get_avenrun(info
->loads
, 0, SI_LOAD_SHIFT
- FSHIFT
);
1606 info
->procs
= nr_threads
;
1612 * If the sum of all the available memory (i.e. ram + swap)
1613 * is less than can be stored in a 32 bit unsigned long then
1614 * we can be binary compatible with 2.2.x kernels. If not,
1615 * well, in that case 2.2.x was broken anyways...
1617 * -Erik Andersen <andersee@debian.org>
1620 mem_total
= info
->totalram
+ info
->totalswap
;
1621 if (mem_total
< info
->totalram
|| mem_total
< info
->totalswap
)
1624 mem_unit
= info
->mem_unit
;
1625 while (mem_unit
> 1) {
1628 sav_total
= mem_total
;
1630 if (mem_total
< sav_total
)
1635 * If mem_total did not overflow, multiply all memory values by
1636 * info->mem_unit and set it to 1. This leaves things compatible
1637 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1642 info
->totalram
<<= bitcount
;
1643 info
->freeram
<<= bitcount
;
1644 info
->sharedram
<<= bitcount
;
1645 info
->bufferram
<<= bitcount
;
1646 info
->totalswap
<<= bitcount
;
1647 info
->freeswap
<<= bitcount
;
1648 info
->totalhigh
<<= bitcount
;
1649 info
->freehigh
<<= bitcount
;
1655 SYSCALL_DEFINE1(sysinfo
, struct sysinfo __user
*, info
)
1661 if (copy_to_user(info
, &val
, sizeof(struct sysinfo
)))
1667 static int __cpuinit
init_timers_cpu(int cpu
)
1670 struct tvec_base
*base
;
1671 static char __cpuinitdata tvec_base_done
[NR_CPUS
];
1673 if (!tvec_base_done
[cpu
]) {
1674 static char boot_done
;
1678 * The APs use this path later in boot
1680 base
= kmalloc_node(sizeof(*base
),
1681 GFP_KERNEL
| __GFP_ZERO
,
1686 /* Make sure that tvec_base is 2 byte aligned */
1687 if (tbase_get_deferrable(base
)) {
1692 per_cpu(tvec_bases
, cpu
) = base
;
1695 * This is for the boot CPU - we use compile-time
1696 * static initialisation because per-cpu memory isn't
1697 * ready yet and because the memory allocators are not
1698 * initialised either.
1701 base
= &boot_tvec_bases
;
1703 tvec_base_done
[cpu
] = 1;
1705 base
= per_cpu(tvec_bases
, cpu
);
1708 spin_lock_init(&base
->lock
);
1710 for (j
= 0; j
< TVN_SIZE
; j
++) {
1711 INIT_LIST_HEAD(base
->tv5
.vec
+ j
);
1712 INIT_LIST_HEAD(base
->tv4
.vec
+ j
);
1713 INIT_LIST_HEAD(base
->tv3
.vec
+ j
);
1714 INIT_LIST_HEAD(base
->tv2
.vec
+ j
);
1716 for (j
= 0; j
< TVR_SIZE
; j
++)
1717 INIT_LIST_HEAD(base
->tv1
.vec
+ j
);
1719 base
->timer_jiffies
= jiffies
;
1720 base
->next_timer
= base
->timer_jiffies
;
1721 base
->active_timers
= 0;
1725 #ifdef CONFIG_HOTPLUG_CPU
1726 static void migrate_timer_list(struct tvec_base
*new_base
, struct list_head
*head
)
1728 struct timer_list
*timer
;
1730 while (!list_empty(head
)) {
1731 timer
= list_first_entry(head
, struct timer_list
, entry
);
1732 /* We ignore the accounting on the dying cpu */
1733 detach_timer(timer
, false);
1734 timer_set_base(timer
, new_base
);
1735 internal_add_timer(new_base
, timer
);
1739 static void __cpuinit
migrate_timers(int cpu
)
1741 struct tvec_base
*old_base
;
1742 struct tvec_base
*new_base
;
1745 BUG_ON(cpu_online(cpu
));
1746 old_base
= per_cpu(tvec_bases
, cpu
);
1747 new_base
= get_cpu_var(tvec_bases
);
1749 * The caller is globally serialized and nobody else
1750 * takes two locks at once, deadlock is not possible.
1752 spin_lock_irq(&new_base
->lock
);
1753 spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1755 BUG_ON(old_base
->running_timer
);
1757 for (i
= 0; i
< TVR_SIZE
; i
++)
1758 migrate_timer_list(new_base
, old_base
->tv1
.vec
+ i
);
1759 for (i
= 0; i
< TVN_SIZE
; i
++) {
1760 migrate_timer_list(new_base
, old_base
->tv2
.vec
+ i
);
1761 migrate_timer_list(new_base
, old_base
->tv3
.vec
+ i
);
1762 migrate_timer_list(new_base
, old_base
->tv4
.vec
+ i
);
1763 migrate_timer_list(new_base
, old_base
->tv5
.vec
+ i
);
1766 spin_unlock(&old_base
->lock
);
1767 spin_unlock_irq(&new_base
->lock
);
1768 put_cpu_var(tvec_bases
);
1770 #endif /* CONFIG_HOTPLUG_CPU */
1772 static int __cpuinit
timer_cpu_notify(struct notifier_block
*self
,
1773 unsigned long action
, void *hcpu
)
1775 long cpu
= (long)hcpu
;
1779 case CPU_UP_PREPARE
:
1780 case CPU_UP_PREPARE_FROZEN
:
1781 err
= init_timers_cpu(cpu
);
1783 return notifier_from_errno(err
);
1785 #ifdef CONFIG_HOTPLUG_CPU
1787 case CPU_DEAD_FROZEN
:
1788 migrate_timers(cpu
);
1797 static struct notifier_block __cpuinitdata timers_nb
= {
1798 .notifier_call
= timer_cpu_notify
,
1802 void __init
init_timers(void)
1806 /* ensure there are enough low bits for flags in timer->base pointer */
1807 BUILD_BUG_ON(__alignof__(struct tvec_base
) & TIMER_FLAG_MASK
);
1809 err
= timer_cpu_notify(&timers_nb
, (unsigned long)CPU_UP_PREPARE
,
1810 (void *)(long)smp_processor_id());
1813 BUG_ON(err
!= NOTIFY_OK
);
1814 register_cpu_notifier(&timers_nb
);
1815 open_softirq(TIMER_SOFTIRQ
, run_timer_softirq
);
1819 * msleep - sleep safely even with waitqueue interruptions
1820 * @msecs: Time in milliseconds to sleep for
1822 void msleep(unsigned int msecs
)
1824 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1827 timeout
= schedule_timeout_uninterruptible(timeout
);
1830 EXPORT_SYMBOL(msleep
);
1833 * msleep_interruptible - sleep waiting for signals
1834 * @msecs: Time in milliseconds to sleep for
1836 unsigned long msleep_interruptible(unsigned int msecs
)
1838 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1840 while (timeout
&& !signal_pending(current
))
1841 timeout
= schedule_timeout_interruptible(timeout
);
1842 return jiffies_to_msecs(timeout
);
1845 EXPORT_SYMBOL(msleep_interruptible
);
1847 static int __sched
do_usleep_range(unsigned long min
, unsigned long max
)
1850 unsigned long delta
;
1852 kmin
= ktime_set(0, min
* NSEC_PER_USEC
);
1853 delta
= (max
- min
) * NSEC_PER_USEC
;
1854 return schedule_hrtimeout_range(&kmin
, delta
, HRTIMER_MODE_REL
);
1858 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1859 * @min: Minimum time in usecs to sleep
1860 * @max: Maximum time in usecs to sleep
1862 void usleep_range(unsigned long min
, unsigned long max
)
1864 __set_current_state(TASK_UNINTERRUPTIBLE
);
1865 do_usleep_range(min
, max
);
1867 EXPORT_SYMBOL(usleep_range
);