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 unsigned int tbase_get_irqsafe(struct tvec_base
*base
)
100 return ((unsigned int)(unsigned long)base
& TIMER_IRQSAFE
);
103 static inline struct tvec_base
*tbase_get_base(struct tvec_base
*base
)
105 return ((struct tvec_base
*)((unsigned long)base
& ~TIMER_FLAG_MASK
));
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 do_init_timer(struct timer_list
*timer
, unsigned int flags
,
568 const char *name
, struct lock_class_key
*key
);
570 void init_timer_on_stack_key(struct timer_list
*timer
, unsigned int flags
,
571 const char *name
, struct lock_class_key
*key
)
573 debug_object_init_on_stack(timer
, &timer_debug_descr
);
574 do_init_timer(timer
, flags
, name
, key
);
576 EXPORT_SYMBOL_GPL(init_timer_on_stack_key
);
578 void destroy_timer_on_stack(struct timer_list
*timer
)
580 debug_object_free(timer
, &timer_debug_descr
);
582 EXPORT_SYMBOL_GPL(destroy_timer_on_stack
);
585 static inline void debug_timer_init(struct timer_list
*timer
) { }
586 static inline void debug_timer_activate(struct timer_list
*timer
) { }
587 static inline void debug_timer_deactivate(struct timer_list
*timer
) { }
588 static inline void debug_timer_assert_init(struct timer_list
*timer
) { }
591 static inline void debug_init(struct timer_list
*timer
)
593 debug_timer_init(timer
);
594 trace_timer_init(timer
);
598 debug_activate(struct timer_list
*timer
, unsigned long expires
)
600 debug_timer_activate(timer
);
601 trace_timer_start(timer
, expires
);
604 static inline void debug_deactivate(struct timer_list
*timer
)
606 debug_timer_deactivate(timer
);
607 trace_timer_cancel(timer
);
610 static inline void debug_assert_init(struct timer_list
*timer
)
612 debug_timer_assert_init(timer
);
615 static void do_init_timer(struct timer_list
*timer
, unsigned int flags
,
616 const char *name
, struct lock_class_key
*key
)
618 struct tvec_base
*base
= __raw_get_cpu_var(tvec_bases
);
620 timer
->entry
.next
= NULL
;
621 timer
->base
= (void *)((unsigned long)base
| flags
);
623 #ifdef CONFIG_TIMER_STATS
624 timer
->start_site
= NULL
;
625 timer
->start_pid
= -1;
626 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
628 lockdep_init_map(&timer
->lockdep_map
, name
, key
, 0);
632 * init_timer_key - initialize a timer
633 * @timer: the timer to be initialized
634 * @flags: timer flags
635 * @name: name of the timer
636 * @key: lockdep class key of the fake lock used for tracking timer
637 * sync lock dependencies
639 * init_timer_key() must be done to a timer prior calling *any* of the
640 * other timer functions.
642 void init_timer_key(struct timer_list
*timer
, unsigned int flags
,
643 const char *name
, struct lock_class_key
*key
)
646 do_init_timer(timer
, flags
, name
, key
);
648 EXPORT_SYMBOL(init_timer_key
);
650 static inline void detach_timer(struct timer_list
*timer
, bool clear_pending
)
652 struct list_head
*entry
= &timer
->entry
;
654 debug_deactivate(timer
);
656 __list_del(entry
->prev
, entry
->next
);
659 entry
->prev
= LIST_POISON2
;
663 detach_expired_timer(struct timer_list
*timer
, struct tvec_base
*base
)
665 detach_timer(timer
, true);
666 if (!tbase_get_deferrable(timer
->base
))
667 base
->active_timers
--;
670 static int detach_if_pending(struct timer_list
*timer
, struct tvec_base
*base
,
673 if (!timer_pending(timer
))
676 detach_timer(timer
, clear_pending
);
677 if (!tbase_get_deferrable(timer
->base
)) {
678 base
->active_timers
--;
679 if (timer
->expires
== base
->next_timer
)
680 base
->next_timer
= base
->timer_jiffies
;
686 * We are using hashed locking: holding per_cpu(tvec_bases).lock
687 * means that all timers which are tied to this base via timer->base are
688 * locked, and the base itself is locked too.
690 * So __run_timers/migrate_timers can safely modify all timers which could
691 * be found on ->tvX lists.
693 * When the timer's base is locked, and the timer removed from list, it is
694 * possible to set timer->base = NULL and drop the lock: the timer remains
697 static struct tvec_base
*lock_timer_base(struct timer_list
*timer
,
698 unsigned long *flags
)
699 __acquires(timer
->base
->lock
)
701 struct tvec_base
*base
;
704 struct tvec_base
*prelock_base
= timer
->base
;
705 base
= tbase_get_base(prelock_base
);
706 if (likely(base
!= NULL
)) {
707 spin_lock_irqsave(&base
->lock
, *flags
);
708 if (likely(prelock_base
== timer
->base
))
710 /* The timer has migrated to another CPU */
711 spin_unlock_irqrestore(&base
->lock
, *flags
);
718 __mod_timer(struct timer_list
*timer
, unsigned long expires
,
719 bool pending_only
, int pinned
)
721 struct tvec_base
*base
, *new_base
;
725 timer_stats_timer_set_start_info(timer
);
726 BUG_ON(!timer
->function
);
728 base
= lock_timer_base(timer
, &flags
);
730 ret
= detach_if_pending(timer
, base
, false);
731 if (!ret
&& pending_only
)
734 debug_activate(timer
, expires
);
736 cpu
= smp_processor_id();
738 #if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
739 if (!pinned
&& get_sysctl_timer_migration() && idle_cpu(cpu
))
740 cpu
= get_nohz_timer_target();
742 new_base
= per_cpu(tvec_bases
, cpu
);
744 if (base
!= new_base
) {
746 * We are trying to schedule the timer on the local CPU.
747 * However we can't change timer's base while it is running,
748 * otherwise del_timer_sync() can't detect that the timer's
749 * handler yet has not finished. This also guarantees that
750 * the timer is serialized wrt itself.
752 if (likely(base
->running_timer
!= timer
)) {
753 /* See the comment in lock_timer_base() */
754 timer_set_base(timer
, NULL
);
755 spin_unlock(&base
->lock
);
757 spin_lock(&base
->lock
);
758 timer_set_base(timer
, base
);
762 timer
->expires
= expires
;
763 internal_add_timer(base
, timer
);
766 spin_unlock_irqrestore(&base
->lock
, flags
);
772 * mod_timer_pending - modify a pending timer's timeout
773 * @timer: the pending timer to be modified
774 * @expires: new timeout in jiffies
776 * mod_timer_pending() is the same for pending timers as mod_timer(),
777 * but will not re-activate and modify already deleted timers.
779 * It is useful for unserialized use of timers.
781 int mod_timer_pending(struct timer_list
*timer
, unsigned long expires
)
783 return __mod_timer(timer
, expires
, true, TIMER_NOT_PINNED
);
785 EXPORT_SYMBOL(mod_timer_pending
);
788 * Decide where to put the timer while taking the slack into account
791 * 1) calculate the maximum (absolute) time
792 * 2) calculate the highest bit where the expires and new max are different
793 * 3) use this bit to make a mask
794 * 4) use the bitmask to round down the maximum time, so that all last
798 unsigned long apply_slack(struct timer_list
*timer
, unsigned long expires
)
800 unsigned long expires_limit
, mask
;
803 if (timer
->slack
>= 0) {
804 expires_limit
= expires
+ timer
->slack
;
806 long delta
= expires
- jiffies
;
811 expires_limit
= expires
+ delta
/ 256;
813 mask
= expires
^ expires_limit
;
817 bit
= find_last_bit(&mask
, BITS_PER_LONG
);
819 mask
= (1 << bit
) - 1;
821 expires_limit
= expires_limit
& ~(mask
);
823 return expires_limit
;
827 * mod_timer - modify a timer's timeout
828 * @timer: the timer to be modified
829 * @expires: new timeout in jiffies
831 * mod_timer() is a more efficient way to update the expire field of an
832 * active timer (if the timer is inactive it will be activated)
834 * mod_timer(timer, expires) is equivalent to:
836 * del_timer(timer); timer->expires = expires; add_timer(timer);
838 * Note that if there are multiple unserialized concurrent users of the
839 * same timer, then mod_timer() is the only safe way to modify the timeout,
840 * since add_timer() cannot modify an already running timer.
842 * The function returns whether it has modified a pending timer or not.
843 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
844 * active timer returns 1.)
846 int mod_timer(struct timer_list
*timer
, unsigned long expires
)
848 expires
= apply_slack(timer
, expires
);
851 * This is a common optimization triggered by the
852 * networking code - if the timer is re-modified
853 * to be the same thing then just return:
855 if (timer_pending(timer
) && timer
->expires
== expires
)
858 return __mod_timer(timer
, expires
, false, TIMER_NOT_PINNED
);
860 EXPORT_SYMBOL(mod_timer
);
863 * mod_timer_pinned - modify a timer's timeout
864 * @timer: the timer to be modified
865 * @expires: new timeout in jiffies
867 * mod_timer_pinned() is a way to update the expire field of an
868 * active timer (if the timer is inactive it will be activated)
869 * and to ensure that the timer is scheduled on the current CPU.
871 * Note that this does not prevent the timer from being migrated
872 * when the current CPU goes offline. If this is a problem for
873 * you, use CPU-hotplug notifiers to handle it correctly, for
874 * example, cancelling the timer when the corresponding CPU goes
877 * mod_timer_pinned(timer, expires) is equivalent to:
879 * del_timer(timer); timer->expires = expires; add_timer(timer);
881 int mod_timer_pinned(struct timer_list
*timer
, unsigned long expires
)
883 if (timer
->expires
== expires
&& timer_pending(timer
))
886 return __mod_timer(timer
, expires
, false, TIMER_PINNED
);
888 EXPORT_SYMBOL(mod_timer_pinned
);
891 * add_timer - start a timer
892 * @timer: the timer to be added
894 * The kernel will do a ->function(->data) callback from the
895 * timer interrupt at the ->expires point in the future. The
896 * current time is 'jiffies'.
898 * The timer's ->expires, ->function (and if the handler uses it, ->data)
899 * fields must be set prior calling this function.
901 * Timers with an ->expires field in the past will be executed in the next
904 void add_timer(struct timer_list
*timer
)
906 BUG_ON(timer_pending(timer
));
907 mod_timer(timer
, timer
->expires
);
909 EXPORT_SYMBOL(add_timer
);
912 * add_timer_on - start a timer on a particular CPU
913 * @timer: the timer to be added
914 * @cpu: the CPU to start it on
916 * This is not very scalable on SMP. Double adds are not possible.
918 void add_timer_on(struct timer_list
*timer
, int cpu
)
920 struct tvec_base
*base
= per_cpu(tvec_bases
, cpu
);
923 timer_stats_timer_set_start_info(timer
);
924 BUG_ON(timer_pending(timer
) || !timer
->function
);
925 spin_lock_irqsave(&base
->lock
, flags
);
926 timer_set_base(timer
, base
);
927 debug_activate(timer
, timer
->expires
);
928 internal_add_timer(base
, timer
);
930 * Check whether the other CPU is idle and needs to be
931 * triggered to reevaluate the timer wheel when nohz is
932 * active. We are protected against the other CPU fiddling
933 * with the timer by holding the timer base lock. This also
934 * makes sure that a CPU on the way to idle can not evaluate
937 wake_up_idle_cpu(cpu
);
938 spin_unlock_irqrestore(&base
->lock
, flags
);
940 EXPORT_SYMBOL_GPL(add_timer_on
);
943 * del_timer - deactive a timer.
944 * @timer: the timer to be deactivated
946 * del_timer() deactivates a timer - this works on both active and inactive
949 * The function returns whether it has deactivated a pending timer or not.
950 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
951 * active timer returns 1.)
953 int del_timer(struct timer_list
*timer
)
955 struct tvec_base
*base
;
959 debug_assert_init(timer
);
961 timer_stats_timer_clear_start_info(timer
);
962 if (timer_pending(timer
)) {
963 base
= lock_timer_base(timer
, &flags
);
964 ret
= detach_if_pending(timer
, base
, true);
965 spin_unlock_irqrestore(&base
->lock
, flags
);
970 EXPORT_SYMBOL(del_timer
);
973 * try_to_del_timer_sync - Try to deactivate a timer
974 * @timer: timer do del
976 * This function tries to deactivate a timer. Upon successful (ret >= 0)
977 * exit the timer is not queued and the handler is not running on any CPU.
979 int try_to_del_timer_sync(struct timer_list
*timer
)
981 struct tvec_base
*base
;
985 debug_assert_init(timer
);
987 base
= lock_timer_base(timer
, &flags
);
989 if (base
->running_timer
!= timer
) {
990 timer_stats_timer_clear_start_info(timer
);
991 ret
= detach_if_pending(timer
, base
, true);
993 spin_unlock_irqrestore(&base
->lock
, flags
);
997 EXPORT_SYMBOL(try_to_del_timer_sync
);
1001 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1002 * @timer: the timer to be deactivated
1004 * This function only differs from del_timer() on SMP: besides deactivating
1005 * the timer it also makes sure the handler has finished executing on other
1008 * Synchronization rules: Callers must prevent restarting of the timer,
1009 * otherwise this function is meaningless. It must not be called from
1010 * interrupt contexts unless the timer is an irqsafe one. The caller must
1011 * not hold locks which would prevent completion of the timer's
1012 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1013 * timer is not queued and the handler is not running on any CPU.
1015 * Note: For !irqsafe timers, you must not hold locks that are held in
1016 * interrupt context while calling this function. Even if the lock has
1017 * nothing to do with the timer in question. Here's why:
1023 * base->running_timer = mytimer;
1024 * spin_lock_irq(somelock);
1026 * spin_lock(somelock);
1027 * del_timer_sync(mytimer);
1028 * while (base->running_timer == mytimer);
1030 * Now del_timer_sync() will never return and never release somelock.
1031 * The interrupt on the other CPU is waiting to grab somelock but
1032 * it has interrupted the softirq that CPU0 is waiting to finish.
1034 * The function returns whether it has deactivated a pending timer or not.
1036 int del_timer_sync(struct timer_list
*timer
)
1038 #ifdef CONFIG_LOCKDEP
1039 unsigned long flags
;
1042 * If lockdep gives a backtrace here, please reference
1043 * the synchronization rules above.
1045 local_irq_save(flags
);
1046 lock_map_acquire(&timer
->lockdep_map
);
1047 lock_map_release(&timer
->lockdep_map
);
1048 local_irq_restore(flags
);
1051 * don't use it in hardirq context, because it
1052 * could lead to deadlock.
1054 WARN_ON(in_irq() && !tbase_get_irqsafe(timer
->base
));
1056 int ret
= try_to_del_timer_sync(timer
);
1062 EXPORT_SYMBOL(del_timer_sync
);
1065 static int cascade(struct tvec_base
*base
, struct tvec
*tv
, int index
)
1067 /* cascade all the timers from tv up one level */
1068 struct timer_list
*timer
, *tmp
;
1069 struct list_head tv_list
;
1071 list_replace_init(tv
->vec
+ index
, &tv_list
);
1074 * We are removing _all_ timers from the list, so we
1075 * don't have to detach them individually.
1077 list_for_each_entry_safe(timer
, tmp
, &tv_list
, entry
) {
1078 BUG_ON(tbase_get_base(timer
->base
) != base
);
1079 /* No accounting, while moving them */
1080 __internal_add_timer(base
, timer
);
1086 static void call_timer_fn(struct timer_list
*timer
, void (*fn
)(unsigned long),
1089 int preempt_count
= preempt_count();
1091 #ifdef CONFIG_LOCKDEP
1093 * It is permissible to free the timer from inside the
1094 * function that is called from it, this we need to take into
1095 * account for lockdep too. To avoid bogus "held lock freed"
1096 * warnings as well as problems when looking into
1097 * timer->lockdep_map, make a copy and use that here.
1099 struct lockdep_map lockdep_map
;
1101 lockdep_copy_map(&lockdep_map
, &timer
->lockdep_map
);
1104 * Couple the lock chain with the lock chain at
1105 * del_timer_sync() by acquiring the lock_map around the fn()
1106 * call here and in del_timer_sync().
1108 lock_map_acquire(&lockdep_map
);
1110 trace_timer_expire_entry(timer
);
1112 trace_timer_expire_exit(timer
);
1114 lock_map_release(&lockdep_map
);
1116 if (preempt_count
!= preempt_count()) {
1117 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1118 fn
, preempt_count
, preempt_count());
1120 * Restore the preempt count. That gives us a decent
1121 * chance to survive and extract information. If the
1122 * callback kept a lock held, bad luck, but not worse
1123 * than the BUG() we had.
1125 preempt_count() = preempt_count
;
1129 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1132 * __run_timers - run all expired timers (if any) on this CPU.
1133 * @base: the timer vector to be processed.
1135 * This function cascades all vectors and executes all expired timer
1138 static inline void __run_timers(struct tvec_base
*base
)
1140 struct timer_list
*timer
;
1142 spin_lock_irq(&base
->lock
);
1143 while (time_after_eq(jiffies
, base
->timer_jiffies
)) {
1144 struct list_head work_list
;
1145 struct list_head
*head
= &work_list
;
1146 int index
= base
->timer_jiffies
& TVR_MASK
;
1152 (!cascade(base
, &base
->tv2
, INDEX(0))) &&
1153 (!cascade(base
, &base
->tv3
, INDEX(1))) &&
1154 !cascade(base
, &base
->tv4
, INDEX(2)))
1155 cascade(base
, &base
->tv5
, INDEX(3));
1156 ++base
->timer_jiffies
;
1157 list_replace_init(base
->tv1
.vec
+ index
, &work_list
);
1158 while (!list_empty(head
)) {
1159 void (*fn
)(unsigned long);
1163 timer
= list_first_entry(head
, struct timer_list
,entry
);
1164 fn
= timer
->function
;
1166 irqsafe
= tbase_get_irqsafe(timer
->base
);
1168 timer_stats_account_timer(timer
);
1170 base
->running_timer
= timer
;
1171 detach_expired_timer(timer
, base
);
1174 spin_unlock(&base
->lock
);
1175 call_timer_fn(timer
, fn
, data
);
1176 spin_lock(&base
->lock
);
1178 spin_unlock_irq(&base
->lock
);
1179 call_timer_fn(timer
, fn
, data
);
1180 spin_lock_irq(&base
->lock
);
1184 base
->running_timer
= NULL
;
1185 spin_unlock_irq(&base
->lock
);
1190 * Find out when the next timer event is due to happen. This
1191 * is used on S/390 to stop all activity when a CPU is idle.
1192 * This function needs to be called with interrupts disabled.
1194 static unsigned long __next_timer_interrupt(struct tvec_base
*base
)
1196 unsigned long timer_jiffies
= base
->timer_jiffies
;
1197 unsigned long expires
= timer_jiffies
+ NEXT_TIMER_MAX_DELTA
;
1198 int index
, slot
, array
, found
= 0;
1199 struct timer_list
*nte
;
1200 struct tvec
*varray
[4];
1202 /* Look for timer events in tv1. */
1203 index
= slot
= timer_jiffies
& TVR_MASK
;
1205 list_for_each_entry(nte
, base
->tv1
.vec
+ slot
, entry
) {
1206 if (tbase_get_deferrable(nte
->base
))
1210 expires
= nte
->expires
;
1211 /* Look at the cascade bucket(s)? */
1212 if (!index
|| slot
< index
)
1216 slot
= (slot
+ 1) & TVR_MASK
;
1217 } while (slot
!= index
);
1220 /* Calculate the next cascade event */
1222 timer_jiffies
+= TVR_SIZE
- index
;
1223 timer_jiffies
>>= TVR_BITS
;
1225 /* Check tv2-tv5. */
1226 varray
[0] = &base
->tv2
;
1227 varray
[1] = &base
->tv3
;
1228 varray
[2] = &base
->tv4
;
1229 varray
[3] = &base
->tv5
;
1231 for (array
= 0; array
< 4; array
++) {
1232 struct tvec
*varp
= varray
[array
];
1234 index
= slot
= timer_jiffies
& TVN_MASK
;
1236 list_for_each_entry(nte
, varp
->vec
+ slot
, entry
) {
1237 if (tbase_get_deferrable(nte
->base
))
1241 if (time_before(nte
->expires
, expires
))
1242 expires
= nte
->expires
;
1245 * Do we still search for the first timer or are
1246 * we looking up the cascade buckets ?
1249 /* Look at the cascade bucket(s)? */
1250 if (!index
|| slot
< index
)
1254 slot
= (slot
+ 1) & TVN_MASK
;
1255 } while (slot
!= index
);
1258 timer_jiffies
+= TVN_SIZE
- index
;
1259 timer_jiffies
>>= TVN_BITS
;
1265 * Check, if the next hrtimer event is before the next timer wheel
1268 static unsigned long cmp_next_hrtimer_event(unsigned long now
,
1269 unsigned long expires
)
1271 ktime_t hr_delta
= hrtimer_get_next_event();
1272 struct timespec tsdelta
;
1273 unsigned long delta
;
1275 if (hr_delta
.tv64
== KTIME_MAX
)
1279 * Expired timer available, let it expire in the next tick
1281 if (hr_delta
.tv64
<= 0)
1284 tsdelta
= ktime_to_timespec(hr_delta
);
1285 delta
= timespec_to_jiffies(&tsdelta
);
1288 * Limit the delta to the max value, which is checked in
1289 * tick_nohz_stop_sched_tick():
1291 if (delta
> NEXT_TIMER_MAX_DELTA
)
1292 delta
= NEXT_TIMER_MAX_DELTA
;
1295 * Take rounding errors in to account and make sure, that it
1296 * expires in the next tick. Otherwise we go into an endless
1297 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1303 if (time_before(now
, expires
))
1309 * get_next_timer_interrupt - return the jiffy of the next pending timer
1310 * @now: current time (in jiffies)
1312 unsigned long get_next_timer_interrupt(unsigned long now
)
1314 struct tvec_base
*base
= __this_cpu_read(tvec_bases
);
1315 unsigned long expires
= now
+ NEXT_TIMER_MAX_DELTA
;
1318 * Pretend that there is no timer pending if the cpu is offline.
1319 * Possible pending timers will be migrated later to an active cpu.
1321 if (cpu_is_offline(smp_processor_id()))
1324 spin_lock(&base
->lock
);
1325 if (base
->active_timers
) {
1326 if (time_before_eq(base
->next_timer
, base
->timer_jiffies
))
1327 base
->next_timer
= __next_timer_interrupt(base
);
1328 expires
= base
->next_timer
;
1330 spin_unlock(&base
->lock
);
1332 if (time_before_eq(expires
, now
))
1335 return cmp_next_hrtimer_event(now
, expires
);
1340 * Called from the timer interrupt handler to charge one tick to the current
1341 * process. user_tick is 1 if the tick is user time, 0 for system.
1343 void update_process_times(int user_tick
)
1345 struct task_struct
*p
= current
;
1346 int cpu
= smp_processor_id();
1348 /* Note: this timer irq context must be accounted for as well. */
1349 account_process_tick(p
, user_tick
);
1351 rcu_check_callbacks(cpu
, user_tick
);
1353 #ifdef CONFIG_IRQ_WORK
1358 run_posix_cpu_timers(p
);
1362 * This function runs timers and the timer-tq in bottom half context.
1364 static void run_timer_softirq(struct softirq_action
*h
)
1366 struct tvec_base
*base
= __this_cpu_read(tvec_bases
);
1368 hrtimer_run_pending();
1370 if (time_after_eq(jiffies
, base
->timer_jiffies
))
1375 * Called by the local, per-CPU timer interrupt on SMP.
1377 void run_local_timers(void)
1379 hrtimer_run_queues();
1380 raise_softirq(TIMER_SOFTIRQ
);
1383 #ifdef __ARCH_WANT_SYS_ALARM
1386 * For backwards compatibility? This can be done in libc so Alpha
1387 * and all newer ports shouldn't need it.
1389 SYSCALL_DEFINE1(alarm
, unsigned int, seconds
)
1391 return alarm_setitimer(seconds
);
1399 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1400 * should be moved into arch/i386 instead?
1404 * sys_getpid - return the thread group id of the current process
1406 * Note, despite the name, this returns the tgid not the pid. The tgid and
1407 * the pid are identical unless CLONE_THREAD was specified on clone() in
1408 * which case the tgid is the same in all threads of the same group.
1410 * This is SMP safe as current->tgid does not change.
1412 SYSCALL_DEFINE0(getpid
)
1414 return task_tgid_vnr(current
);
1418 * Accessing ->real_parent is not SMP-safe, it could
1419 * change from under us. However, we can use a stale
1420 * value of ->real_parent under rcu_read_lock(), see
1421 * release_task()->call_rcu(delayed_put_task_struct).
1423 SYSCALL_DEFINE0(getppid
)
1428 pid
= task_tgid_vnr(rcu_dereference(current
->real_parent
));
1434 SYSCALL_DEFINE0(getuid
)
1436 /* Only we change this so SMP safe */
1437 return from_kuid_munged(current_user_ns(), current_uid());
1440 SYSCALL_DEFINE0(geteuid
)
1442 /* Only we change this so SMP safe */
1443 return from_kuid_munged(current_user_ns(), current_euid());
1446 SYSCALL_DEFINE0(getgid
)
1448 /* Only we change this so SMP safe */
1449 return from_kgid_munged(current_user_ns(), current_gid());
1452 SYSCALL_DEFINE0(getegid
)
1454 /* Only we change this so SMP safe */
1455 return from_kgid_munged(current_user_ns(), current_egid());
1460 static void process_timeout(unsigned long __data
)
1462 wake_up_process((struct task_struct
*)__data
);
1466 * schedule_timeout - sleep until timeout
1467 * @timeout: timeout value in jiffies
1469 * Make the current task sleep until @timeout jiffies have
1470 * elapsed. The routine will return immediately unless
1471 * the current task state has been set (see set_current_state()).
1473 * You can set the task state as follows -
1475 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1476 * pass before the routine returns. The routine will return 0
1478 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1479 * delivered to the current task. In this case the remaining time
1480 * in jiffies will be returned, or 0 if the timer expired in time
1482 * The current task state is guaranteed to be TASK_RUNNING when this
1485 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1486 * the CPU away without a bound on the timeout. In this case the return
1487 * value will be %MAX_SCHEDULE_TIMEOUT.
1489 * In all cases the return value is guaranteed to be non-negative.
1491 signed long __sched
schedule_timeout(signed long timeout
)
1493 struct timer_list timer
;
1494 unsigned long expire
;
1498 case MAX_SCHEDULE_TIMEOUT
:
1500 * These two special cases are useful to be comfortable
1501 * in the caller. Nothing more. We could take
1502 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1503 * but I' d like to return a valid offset (>=0) to allow
1504 * the caller to do everything it want with the retval.
1510 * Another bit of PARANOID. Note that the retval will be
1511 * 0 since no piece of kernel is supposed to do a check
1512 * for a negative retval of schedule_timeout() (since it
1513 * should never happens anyway). You just have the printk()
1514 * that will tell you if something is gone wrong and where.
1517 printk(KERN_ERR
"schedule_timeout: wrong timeout "
1518 "value %lx\n", timeout
);
1520 current
->state
= TASK_RUNNING
;
1525 expire
= timeout
+ jiffies
;
1527 setup_timer_on_stack(&timer
, process_timeout
, (unsigned long)current
);
1528 __mod_timer(&timer
, expire
, false, TIMER_NOT_PINNED
);
1530 del_singleshot_timer_sync(&timer
);
1532 /* Remove the timer from the object tracker */
1533 destroy_timer_on_stack(&timer
);
1535 timeout
= expire
- jiffies
;
1538 return timeout
< 0 ? 0 : timeout
;
1540 EXPORT_SYMBOL(schedule_timeout
);
1543 * We can use __set_current_state() here because schedule_timeout() calls
1544 * schedule() unconditionally.
1546 signed long __sched
schedule_timeout_interruptible(signed long timeout
)
1548 __set_current_state(TASK_INTERRUPTIBLE
);
1549 return schedule_timeout(timeout
);
1551 EXPORT_SYMBOL(schedule_timeout_interruptible
);
1553 signed long __sched
schedule_timeout_killable(signed long timeout
)
1555 __set_current_state(TASK_KILLABLE
);
1556 return schedule_timeout(timeout
);
1558 EXPORT_SYMBOL(schedule_timeout_killable
);
1560 signed long __sched
schedule_timeout_uninterruptible(signed long timeout
)
1562 __set_current_state(TASK_UNINTERRUPTIBLE
);
1563 return schedule_timeout(timeout
);
1565 EXPORT_SYMBOL(schedule_timeout_uninterruptible
);
1567 /* Thread ID - the internal kernel "pid" */
1568 SYSCALL_DEFINE0(gettid
)
1570 return task_pid_vnr(current
);
1574 * do_sysinfo - fill in sysinfo struct
1575 * @info: pointer to buffer to fill
1577 int do_sysinfo(struct sysinfo
*info
)
1579 unsigned long mem_total
, sav_total
;
1580 unsigned int mem_unit
, bitcount
;
1583 memset(info
, 0, sizeof(struct sysinfo
));
1586 monotonic_to_bootbased(&tp
);
1587 info
->uptime
= tp
.tv_sec
+ (tp
.tv_nsec
? 1 : 0);
1589 get_avenrun(info
->loads
, 0, SI_LOAD_SHIFT
- FSHIFT
);
1591 info
->procs
= nr_threads
;
1597 * If the sum of all the available memory (i.e. ram + swap)
1598 * is less than can be stored in a 32 bit unsigned long then
1599 * we can be binary compatible with 2.2.x kernels. If not,
1600 * well, in that case 2.2.x was broken anyways...
1602 * -Erik Andersen <andersee@debian.org>
1605 mem_total
= info
->totalram
+ info
->totalswap
;
1606 if (mem_total
< info
->totalram
|| mem_total
< info
->totalswap
)
1609 mem_unit
= info
->mem_unit
;
1610 while (mem_unit
> 1) {
1613 sav_total
= mem_total
;
1615 if (mem_total
< sav_total
)
1620 * If mem_total did not overflow, multiply all memory values by
1621 * info->mem_unit and set it to 1. This leaves things compatible
1622 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1627 info
->totalram
<<= bitcount
;
1628 info
->freeram
<<= bitcount
;
1629 info
->sharedram
<<= bitcount
;
1630 info
->bufferram
<<= bitcount
;
1631 info
->totalswap
<<= bitcount
;
1632 info
->freeswap
<<= bitcount
;
1633 info
->totalhigh
<<= bitcount
;
1634 info
->freehigh
<<= bitcount
;
1640 SYSCALL_DEFINE1(sysinfo
, struct sysinfo __user
*, info
)
1646 if (copy_to_user(info
, &val
, sizeof(struct sysinfo
)))
1652 static int __cpuinit
init_timers_cpu(int cpu
)
1655 struct tvec_base
*base
;
1656 static char __cpuinitdata tvec_base_done
[NR_CPUS
];
1658 if (!tvec_base_done
[cpu
]) {
1659 static char boot_done
;
1663 * The APs use this path later in boot
1665 base
= kmalloc_node(sizeof(*base
),
1666 GFP_KERNEL
| __GFP_ZERO
,
1671 /* Make sure that tvec_base is 2 byte aligned */
1672 if (tbase_get_deferrable(base
)) {
1677 per_cpu(tvec_bases
, cpu
) = base
;
1680 * This is for the boot CPU - we use compile-time
1681 * static initialisation because per-cpu memory isn't
1682 * ready yet and because the memory allocators are not
1683 * initialised either.
1686 base
= &boot_tvec_bases
;
1688 tvec_base_done
[cpu
] = 1;
1690 base
= per_cpu(tvec_bases
, cpu
);
1693 spin_lock_init(&base
->lock
);
1695 for (j
= 0; j
< TVN_SIZE
; j
++) {
1696 INIT_LIST_HEAD(base
->tv5
.vec
+ j
);
1697 INIT_LIST_HEAD(base
->tv4
.vec
+ j
);
1698 INIT_LIST_HEAD(base
->tv3
.vec
+ j
);
1699 INIT_LIST_HEAD(base
->tv2
.vec
+ j
);
1701 for (j
= 0; j
< TVR_SIZE
; j
++)
1702 INIT_LIST_HEAD(base
->tv1
.vec
+ j
);
1704 base
->timer_jiffies
= jiffies
;
1705 base
->next_timer
= base
->timer_jiffies
;
1706 base
->active_timers
= 0;
1710 #ifdef CONFIG_HOTPLUG_CPU
1711 static void migrate_timer_list(struct tvec_base
*new_base
, struct list_head
*head
)
1713 struct timer_list
*timer
;
1715 while (!list_empty(head
)) {
1716 timer
= list_first_entry(head
, struct timer_list
, entry
);
1717 /* We ignore the accounting on the dying cpu */
1718 detach_timer(timer
, false);
1719 timer_set_base(timer
, new_base
);
1720 internal_add_timer(new_base
, timer
);
1724 static void __cpuinit
migrate_timers(int cpu
)
1726 struct tvec_base
*old_base
;
1727 struct tvec_base
*new_base
;
1730 BUG_ON(cpu_online(cpu
));
1731 old_base
= per_cpu(tvec_bases
, cpu
);
1732 new_base
= get_cpu_var(tvec_bases
);
1734 * The caller is globally serialized and nobody else
1735 * takes two locks at once, deadlock is not possible.
1737 spin_lock_irq(&new_base
->lock
);
1738 spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1740 BUG_ON(old_base
->running_timer
);
1742 for (i
= 0; i
< TVR_SIZE
; i
++)
1743 migrate_timer_list(new_base
, old_base
->tv1
.vec
+ i
);
1744 for (i
= 0; i
< TVN_SIZE
; i
++) {
1745 migrate_timer_list(new_base
, old_base
->tv2
.vec
+ i
);
1746 migrate_timer_list(new_base
, old_base
->tv3
.vec
+ i
);
1747 migrate_timer_list(new_base
, old_base
->tv4
.vec
+ i
);
1748 migrate_timer_list(new_base
, old_base
->tv5
.vec
+ i
);
1751 spin_unlock(&old_base
->lock
);
1752 spin_unlock_irq(&new_base
->lock
);
1753 put_cpu_var(tvec_bases
);
1755 #endif /* CONFIG_HOTPLUG_CPU */
1757 static int __cpuinit
timer_cpu_notify(struct notifier_block
*self
,
1758 unsigned long action
, void *hcpu
)
1760 long cpu
= (long)hcpu
;
1764 case CPU_UP_PREPARE
:
1765 case CPU_UP_PREPARE_FROZEN
:
1766 err
= init_timers_cpu(cpu
);
1768 return notifier_from_errno(err
);
1770 #ifdef CONFIG_HOTPLUG_CPU
1772 case CPU_DEAD_FROZEN
:
1773 migrate_timers(cpu
);
1782 static struct notifier_block __cpuinitdata timers_nb
= {
1783 .notifier_call
= timer_cpu_notify
,
1787 void __init
init_timers(void)
1791 /* ensure there are enough low bits for flags in timer->base pointer */
1792 BUILD_BUG_ON(__alignof__(struct tvec_base
) & TIMER_FLAG_MASK
);
1794 err
= timer_cpu_notify(&timers_nb
, (unsigned long)CPU_UP_PREPARE
,
1795 (void *)(long)smp_processor_id());
1798 BUG_ON(err
!= NOTIFY_OK
);
1799 register_cpu_notifier(&timers_nb
);
1800 open_softirq(TIMER_SOFTIRQ
, run_timer_softirq
);
1804 * msleep - sleep safely even with waitqueue interruptions
1805 * @msecs: Time in milliseconds to sleep for
1807 void msleep(unsigned int msecs
)
1809 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1812 timeout
= schedule_timeout_uninterruptible(timeout
);
1815 EXPORT_SYMBOL(msleep
);
1818 * msleep_interruptible - sleep waiting for signals
1819 * @msecs: Time in milliseconds to sleep for
1821 unsigned long msleep_interruptible(unsigned int msecs
)
1823 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1825 while (timeout
&& !signal_pending(current
))
1826 timeout
= schedule_timeout_interruptible(timeout
);
1827 return jiffies_to_msecs(timeout
);
1830 EXPORT_SYMBOL(msleep_interruptible
);
1832 static int __sched
do_usleep_range(unsigned long min
, unsigned long max
)
1835 unsigned long delta
;
1837 kmin
= ktime_set(0, min
* NSEC_PER_USEC
);
1838 delta
= (max
- min
) * NSEC_PER_USEC
;
1839 return schedule_hrtimeout_range(&kmin
, delta
, HRTIMER_MODE_REL
);
1843 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1844 * @min: Minimum time in usecs to sleep
1845 * @max: Maximum time in usecs to sleep
1847 void usleep_range(unsigned long min
, unsigned long max
)
1849 __set_current_state(TASK_UNINTERRUPTIBLE
);
1850 do_usleep_range(min
, max
);
1852 EXPORT_SYMBOL(usleep_range
);