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/module.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/perf_counter.h>
42 #include <asm/uaccess.h>
43 #include <asm/unistd.h>
44 #include <asm/div64.h>
45 #include <asm/timex.h>
48 u64 jiffies_64 __cacheline_aligned_in_smp
= INITIAL_JIFFIES
;
50 EXPORT_SYMBOL(jiffies_64
);
53 * per-CPU timer vector definitions:
55 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
56 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
57 #define TVN_SIZE (1 << TVN_BITS)
58 #define TVR_SIZE (1 << TVR_BITS)
59 #define TVN_MASK (TVN_SIZE - 1)
60 #define TVR_MASK (TVR_SIZE - 1)
63 struct list_head vec
[TVN_SIZE
];
67 struct list_head vec
[TVR_SIZE
];
72 struct timer_list
*running_timer
;
73 unsigned long timer_jiffies
;
79 } ____cacheline_aligned
;
81 struct tvec_base boot_tvec_bases
;
82 EXPORT_SYMBOL(boot_tvec_bases
);
83 static DEFINE_PER_CPU(struct tvec_base
*, tvec_bases
) = &boot_tvec_bases
;
86 * Note that all tvec_bases are 2 byte aligned and lower bit of
87 * base in timer_list is guaranteed to be zero. Use the LSB for
88 * the new flag to indicate whether the timer is deferrable
90 #define TBASE_DEFERRABLE_FLAG (0x1)
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
& TBASE_DEFERRABLE_FLAG
);
98 static inline struct tvec_base
*tbase_get_base(struct tvec_base
*base
)
100 return ((struct tvec_base
*)((unsigned long)base
& ~TBASE_DEFERRABLE_FLAG
));
103 static inline void timer_set_deferrable(struct timer_list
*timer
)
105 timer
->base
= ((struct tvec_base
*)((unsigned long)(timer
->base
) |
106 TBASE_DEFERRABLE_FLAG
));
110 timer_set_base(struct timer_list
*timer
, struct tvec_base
*new_base
)
112 timer
->base
= (struct tvec_base
*)((unsigned long)(new_base
) |
113 tbase_get_deferrable(timer
->base
));
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 static inline void set_running_timer(struct tvec_base
*base
,
318 struct timer_list
*timer
)
321 base
->running_timer
= timer
;
325 static void internal_add_timer(struct tvec_base
*base
, struct timer_list
*timer
)
327 unsigned long expires
= timer
->expires
;
328 unsigned long idx
= expires
- base
->timer_jiffies
;
329 struct list_head
*vec
;
331 if (idx
< TVR_SIZE
) {
332 int i
= expires
& TVR_MASK
;
333 vec
= base
->tv1
.vec
+ i
;
334 } else if (idx
< 1 << (TVR_BITS
+ TVN_BITS
)) {
335 int i
= (expires
>> TVR_BITS
) & TVN_MASK
;
336 vec
= base
->tv2
.vec
+ i
;
337 } else if (idx
< 1 << (TVR_BITS
+ 2 * TVN_BITS
)) {
338 int i
= (expires
>> (TVR_BITS
+ TVN_BITS
)) & TVN_MASK
;
339 vec
= base
->tv3
.vec
+ i
;
340 } else if (idx
< 1 << (TVR_BITS
+ 3 * TVN_BITS
)) {
341 int i
= (expires
>> (TVR_BITS
+ 2 * TVN_BITS
)) & TVN_MASK
;
342 vec
= base
->tv4
.vec
+ i
;
343 } else if ((signed long) idx
< 0) {
345 * Can happen if you add a timer with expires == jiffies,
346 * or you set a timer to go off in the past
348 vec
= base
->tv1
.vec
+ (base
->timer_jiffies
& TVR_MASK
);
351 /* If the timeout is larger than 0xffffffff on 64-bit
352 * architectures then we use the maximum timeout:
354 if (idx
> 0xffffffffUL
) {
356 expires
= idx
+ base
->timer_jiffies
;
358 i
= (expires
>> (TVR_BITS
+ 3 * TVN_BITS
)) & TVN_MASK
;
359 vec
= base
->tv5
.vec
+ i
;
364 list_add_tail(&timer
->entry
, vec
);
367 #ifdef CONFIG_TIMER_STATS
368 void __timer_stats_timer_set_start_info(struct timer_list
*timer
, void *addr
)
370 if (timer
->start_site
)
373 timer
->start_site
= addr
;
374 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
375 timer
->start_pid
= current
->pid
;
378 static void timer_stats_account_timer(struct timer_list
*timer
)
380 unsigned int flag
= 0;
382 if (unlikely(tbase_get_deferrable(timer
->base
)))
383 flag
|= TIMER_STATS_FLAG_DEFERRABLE
;
385 timer_stats_update_stats(timer
, timer
->start_pid
, timer
->start_site
,
386 timer
->function
, timer
->start_comm
, flag
);
390 static void timer_stats_account_timer(struct timer_list
*timer
) {}
393 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
395 static struct debug_obj_descr timer_debug_descr
;
398 * fixup_init is called when:
399 * - an active object is initialized
401 static int timer_fixup_init(void *addr
, enum debug_obj_state state
)
403 struct timer_list
*timer
= addr
;
406 case ODEBUG_STATE_ACTIVE
:
407 del_timer_sync(timer
);
408 debug_object_init(timer
, &timer_debug_descr
);
416 * fixup_activate is called when:
417 * - an active object is activated
418 * - an unknown object is activated (might be a statically initialized object)
420 static int timer_fixup_activate(void *addr
, enum debug_obj_state state
)
422 struct timer_list
*timer
= addr
;
426 case ODEBUG_STATE_NOTAVAILABLE
:
428 * This is not really a fixup. The timer was
429 * statically initialized. We just make sure that it
430 * is tracked in the object tracker.
432 if (timer
->entry
.next
== NULL
&&
433 timer
->entry
.prev
== TIMER_ENTRY_STATIC
) {
434 debug_object_init(timer
, &timer_debug_descr
);
435 debug_object_activate(timer
, &timer_debug_descr
);
442 case ODEBUG_STATE_ACTIVE
:
451 * fixup_free is called when:
452 * - an active object is freed
454 static int timer_fixup_free(void *addr
, enum debug_obj_state state
)
456 struct timer_list
*timer
= addr
;
459 case ODEBUG_STATE_ACTIVE
:
460 del_timer_sync(timer
);
461 debug_object_free(timer
, &timer_debug_descr
);
468 static struct debug_obj_descr timer_debug_descr
= {
469 .name
= "timer_list",
470 .fixup_init
= timer_fixup_init
,
471 .fixup_activate
= timer_fixup_activate
,
472 .fixup_free
= timer_fixup_free
,
475 static inline void debug_timer_init(struct timer_list
*timer
)
477 debug_object_init(timer
, &timer_debug_descr
);
480 static inline void debug_timer_activate(struct timer_list
*timer
)
482 debug_object_activate(timer
, &timer_debug_descr
);
485 static inline void debug_timer_deactivate(struct timer_list
*timer
)
487 debug_object_deactivate(timer
, &timer_debug_descr
);
490 static inline void debug_timer_free(struct timer_list
*timer
)
492 debug_object_free(timer
, &timer_debug_descr
);
495 static void __init_timer(struct timer_list
*timer
,
497 struct lock_class_key
*key
);
499 void init_timer_on_stack_key(struct timer_list
*timer
,
501 struct lock_class_key
*key
)
503 debug_object_init_on_stack(timer
, &timer_debug_descr
);
504 __init_timer(timer
, name
, key
);
506 EXPORT_SYMBOL_GPL(init_timer_on_stack_key
);
508 void destroy_timer_on_stack(struct timer_list
*timer
)
510 debug_object_free(timer
, &timer_debug_descr
);
512 EXPORT_SYMBOL_GPL(destroy_timer_on_stack
);
515 static inline void debug_timer_init(struct timer_list
*timer
) { }
516 static inline void debug_timer_activate(struct timer_list
*timer
) { }
517 static inline void debug_timer_deactivate(struct timer_list
*timer
) { }
520 static void __init_timer(struct timer_list
*timer
,
522 struct lock_class_key
*key
)
524 timer
->entry
.next
= NULL
;
525 timer
->base
= __raw_get_cpu_var(tvec_bases
);
526 #ifdef CONFIG_TIMER_STATS
527 timer
->start_site
= NULL
;
528 timer
->start_pid
= -1;
529 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
531 lockdep_init_map(&timer
->lockdep_map
, name
, key
, 0);
535 * init_timer - initialize a timer.
536 * @timer: the timer to be initialized
538 * init_timer() must be done to a timer prior calling *any* of the
539 * other timer functions.
541 void init_timer_key(struct timer_list
*timer
,
543 struct lock_class_key
*key
)
545 debug_timer_init(timer
);
546 __init_timer(timer
, name
, key
);
548 EXPORT_SYMBOL(init_timer_key
);
550 void init_timer_deferrable_key(struct timer_list
*timer
,
552 struct lock_class_key
*key
)
554 init_timer_key(timer
, name
, key
);
555 timer_set_deferrable(timer
);
557 EXPORT_SYMBOL(init_timer_deferrable_key
);
559 static inline void detach_timer(struct timer_list
*timer
,
562 struct list_head
*entry
= &timer
->entry
;
564 debug_timer_deactivate(timer
);
566 __list_del(entry
->prev
, entry
->next
);
569 entry
->prev
= LIST_POISON2
;
573 * We are using hashed locking: holding per_cpu(tvec_bases).lock
574 * means that all timers which are tied to this base via timer->base are
575 * locked, and the base itself is locked too.
577 * So __run_timers/migrate_timers can safely modify all timers which could
578 * be found on ->tvX lists.
580 * When the timer's base is locked, and the timer removed from list, it is
581 * possible to set timer->base = NULL and drop the lock: the timer remains
584 static struct tvec_base
*lock_timer_base(struct timer_list
*timer
,
585 unsigned long *flags
)
586 __acquires(timer
->base
->lock
)
588 struct tvec_base
*base
;
591 struct tvec_base
*prelock_base
= timer
->base
;
592 base
= tbase_get_base(prelock_base
);
593 if (likely(base
!= NULL
)) {
594 spin_lock_irqsave(&base
->lock
, *flags
);
595 if (likely(prelock_base
== timer
->base
))
597 /* The timer has migrated to another CPU */
598 spin_unlock_irqrestore(&base
->lock
, *flags
);
605 __mod_timer(struct timer_list
*timer
, unsigned long expires
, bool pending_only
)
607 struct tvec_base
*base
, *new_base
;
613 timer_stats_timer_set_start_info(timer
);
614 BUG_ON(!timer
->function
);
616 base
= lock_timer_base(timer
, &flags
);
618 if (timer_pending(timer
)) {
619 detach_timer(timer
, 0);
626 debug_timer_activate(timer
);
628 new_base
= __get_cpu_var(tvec_bases
);
630 if (base
!= new_base
) {
632 * We are trying to schedule the timer on the local CPU.
633 * However we can't change timer's base while it is running,
634 * otherwise del_timer_sync() can't detect that the timer's
635 * handler yet has not finished. This also guarantees that
636 * the timer is serialized wrt itself.
638 if (likely(base
->running_timer
!= timer
)) {
639 /* See the comment in lock_timer_base() */
640 timer_set_base(timer
, NULL
);
641 spin_unlock(&base
->lock
);
643 spin_lock(&base
->lock
);
644 timer_set_base(timer
, base
);
648 timer
->expires
= expires
;
649 internal_add_timer(base
, timer
);
652 spin_unlock_irqrestore(&base
->lock
, flags
);
658 * mod_timer_pending - modify a pending timer's timeout
659 * @timer: the pending timer to be modified
660 * @expires: new timeout in jiffies
662 * mod_timer_pending() is the same for pending timers as mod_timer(),
663 * but will not re-activate and modify already deleted timers.
665 * It is useful for unserialized use of timers.
667 int mod_timer_pending(struct timer_list
*timer
, unsigned long expires
)
669 return __mod_timer(timer
, expires
, true);
671 EXPORT_SYMBOL(mod_timer_pending
);
674 * mod_timer - modify a timer's timeout
675 * @timer: the timer to be modified
676 * @expires: new timeout in jiffies
678 * mod_timer() is a more efficient way to update the expire field of an
679 * active timer (if the timer is inactive it will be activated)
681 * mod_timer(timer, expires) is equivalent to:
683 * del_timer(timer); timer->expires = expires; add_timer(timer);
685 * Note that if there are multiple unserialized concurrent users of the
686 * same timer, then mod_timer() is the only safe way to modify the timeout,
687 * since add_timer() cannot modify an already running timer.
689 * The function returns whether it has modified a pending timer or not.
690 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
691 * active timer returns 1.)
693 int mod_timer(struct timer_list
*timer
, unsigned long expires
)
696 * This is a common optimization triggered by the
697 * networking code - if the timer is re-modified
698 * to be the same thing then just return:
700 if (timer
->expires
== expires
&& timer_pending(timer
))
703 return __mod_timer(timer
, expires
, false);
705 EXPORT_SYMBOL(mod_timer
);
708 * add_timer - start a timer
709 * @timer: the timer to be added
711 * The kernel will do a ->function(->data) callback from the
712 * timer interrupt at the ->expires point in the future. The
713 * current time is 'jiffies'.
715 * The timer's ->expires, ->function (and if the handler uses it, ->data)
716 * fields must be set prior calling this function.
718 * Timers with an ->expires field in the past will be executed in the next
721 void add_timer(struct timer_list
*timer
)
723 BUG_ON(timer_pending(timer
));
724 mod_timer(timer
, timer
->expires
);
726 EXPORT_SYMBOL(add_timer
);
729 * add_timer_on - start a timer on a particular CPU
730 * @timer: the timer to be added
731 * @cpu: the CPU to start it on
733 * This is not very scalable on SMP. Double adds are not possible.
735 void add_timer_on(struct timer_list
*timer
, int cpu
)
737 struct tvec_base
*base
= per_cpu(tvec_bases
, cpu
);
740 timer_stats_timer_set_start_info(timer
);
741 BUG_ON(timer_pending(timer
) || !timer
->function
);
742 spin_lock_irqsave(&base
->lock
, flags
);
743 timer_set_base(timer
, base
);
744 debug_timer_activate(timer
);
745 internal_add_timer(base
, timer
);
747 * Check whether the other CPU is idle and needs to be
748 * triggered to reevaluate the timer wheel when nohz is
749 * active. We are protected against the other CPU fiddling
750 * with the timer by holding the timer base lock. This also
751 * makes sure that a CPU on the way to idle can not evaluate
754 wake_up_idle_cpu(cpu
);
755 spin_unlock_irqrestore(&base
->lock
, flags
);
759 * del_timer - deactive a timer.
760 * @timer: the timer to be deactivated
762 * del_timer() deactivates a timer - this works on both active and inactive
765 * The function returns whether it has deactivated a pending timer or not.
766 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
767 * active timer returns 1.)
769 int del_timer(struct timer_list
*timer
)
771 struct tvec_base
*base
;
775 timer_stats_timer_clear_start_info(timer
);
776 if (timer_pending(timer
)) {
777 base
= lock_timer_base(timer
, &flags
);
778 if (timer_pending(timer
)) {
779 detach_timer(timer
, 1);
782 spin_unlock_irqrestore(&base
->lock
, flags
);
787 EXPORT_SYMBOL(del_timer
);
791 * try_to_del_timer_sync - Try to deactivate a timer
792 * @timer: timer do del
794 * This function tries to deactivate a timer. Upon successful (ret >= 0)
795 * exit the timer is not queued and the handler is not running on any CPU.
797 * It must not be called from interrupt contexts.
799 int try_to_del_timer_sync(struct timer_list
*timer
)
801 struct tvec_base
*base
;
805 base
= lock_timer_base(timer
, &flags
);
807 if (base
->running_timer
== timer
)
811 if (timer_pending(timer
)) {
812 detach_timer(timer
, 1);
816 spin_unlock_irqrestore(&base
->lock
, flags
);
820 EXPORT_SYMBOL(try_to_del_timer_sync
);
823 * del_timer_sync - deactivate a timer and wait for the handler to finish.
824 * @timer: the timer to be deactivated
826 * This function only differs from del_timer() on SMP: besides deactivating
827 * the timer it also makes sure the handler has finished executing on other
830 * Synchronization rules: Callers must prevent restarting of the timer,
831 * otherwise this function is meaningless. It must not be called from
832 * interrupt contexts. The caller must not hold locks which would prevent
833 * completion of the timer's handler. The timer's handler must not call
834 * add_timer_on(). Upon exit the timer is not queued and the handler is
835 * not running on any CPU.
837 * The function returns whether it has deactivated a pending timer or not.
839 int del_timer_sync(struct timer_list
*timer
)
841 #ifdef CONFIG_LOCKDEP
844 local_irq_save(flags
);
845 lock_map_acquire(&timer
->lockdep_map
);
846 lock_map_release(&timer
->lockdep_map
);
847 local_irq_restore(flags
);
851 int ret
= try_to_del_timer_sync(timer
);
857 EXPORT_SYMBOL(del_timer_sync
);
860 static int cascade(struct tvec_base
*base
, struct tvec
*tv
, int index
)
862 /* cascade all the timers from tv up one level */
863 struct timer_list
*timer
, *tmp
;
864 struct list_head tv_list
;
866 list_replace_init(tv
->vec
+ index
, &tv_list
);
869 * We are removing _all_ timers from the list, so we
870 * don't have to detach them individually.
872 list_for_each_entry_safe(timer
, tmp
, &tv_list
, entry
) {
873 BUG_ON(tbase_get_base(timer
->base
) != base
);
874 internal_add_timer(base
, timer
);
880 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
883 * __run_timers - run all expired timers (if any) on this CPU.
884 * @base: the timer vector to be processed.
886 * This function cascades all vectors and executes all expired timer
889 static inline void __run_timers(struct tvec_base
*base
)
891 struct timer_list
*timer
;
893 spin_lock_irq(&base
->lock
);
894 while (time_after_eq(jiffies
, base
->timer_jiffies
)) {
895 struct list_head work_list
;
896 struct list_head
*head
= &work_list
;
897 int index
= base
->timer_jiffies
& TVR_MASK
;
903 (!cascade(base
, &base
->tv2
, INDEX(0))) &&
904 (!cascade(base
, &base
->tv3
, INDEX(1))) &&
905 !cascade(base
, &base
->tv4
, INDEX(2)))
906 cascade(base
, &base
->tv5
, INDEX(3));
907 ++base
->timer_jiffies
;
908 list_replace_init(base
->tv1
.vec
+ index
, &work_list
);
909 while (!list_empty(head
)) {
910 void (*fn
)(unsigned long);
913 timer
= list_first_entry(head
, struct timer_list
,entry
);
914 fn
= timer
->function
;
917 timer_stats_account_timer(timer
);
919 set_running_timer(base
, timer
);
920 detach_timer(timer
, 1);
922 spin_unlock_irq(&base
->lock
);
924 int preempt_count
= preempt_count();
926 #ifdef CONFIG_LOCKDEP
928 * It is permissible to free the timer from
929 * inside the function that is called from
930 * it, this we need to take into account for
931 * lockdep too. To avoid bogus "held lock
932 * freed" warnings as well as problems when
933 * looking into timer->lockdep_map, make a
934 * copy and use that here.
936 struct lockdep_map lockdep_map
=
940 * Couple the lock chain with the lock chain at
941 * del_timer_sync() by acquiring the lock_map
942 * around the fn() call here and in
945 lock_map_acquire(&lockdep_map
);
949 lock_map_release(&lockdep_map
);
951 if (preempt_count
!= preempt_count()) {
952 printk(KERN_ERR
"huh, entered %p "
953 "with preempt_count %08x, exited"
960 spin_lock_irq(&base
->lock
);
963 set_running_timer(base
, NULL
);
964 spin_unlock_irq(&base
->lock
);
969 * Find out when the next timer event is due to happen. This
970 * is used on S/390 to stop all activity when a cpus is idle.
971 * This functions needs to be called disabled.
973 static unsigned long __next_timer_interrupt(struct tvec_base
*base
)
975 unsigned long timer_jiffies
= base
->timer_jiffies
;
976 unsigned long expires
= timer_jiffies
+ NEXT_TIMER_MAX_DELTA
;
977 int index
, slot
, array
, found
= 0;
978 struct timer_list
*nte
;
979 struct tvec
*varray
[4];
981 /* Look for timer events in tv1. */
982 index
= slot
= timer_jiffies
& TVR_MASK
;
984 list_for_each_entry(nte
, base
->tv1
.vec
+ slot
, entry
) {
985 if (tbase_get_deferrable(nte
->base
))
989 expires
= nte
->expires
;
990 /* Look at the cascade bucket(s)? */
991 if (!index
|| slot
< index
)
995 slot
= (slot
+ 1) & TVR_MASK
;
996 } while (slot
!= index
);
999 /* Calculate the next cascade event */
1001 timer_jiffies
+= TVR_SIZE
- index
;
1002 timer_jiffies
>>= TVR_BITS
;
1004 /* Check tv2-tv5. */
1005 varray
[0] = &base
->tv2
;
1006 varray
[1] = &base
->tv3
;
1007 varray
[2] = &base
->tv4
;
1008 varray
[3] = &base
->tv5
;
1010 for (array
= 0; array
< 4; array
++) {
1011 struct tvec
*varp
= varray
[array
];
1013 index
= slot
= timer_jiffies
& TVN_MASK
;
1015 list_for_each_entry(nte
, varp
->vec
+ slot
, entry
) {
1017 if (time_before(nte
->expires
, expires
))
1018 expires
= nte
->expires
;
1021 * Do we still search for the first timer or are
1022 * we looking up the cascade buckets ?
1025 /* Look at the cascade bucket(s)? */
1026 if (!index
|| slot
< index
)
1030 slot
= (slot
+ 1) & TVN_MASK
;
1031 } while (slot
!= index
);
1034 timer_jiffies
+= TVN_SIZE
- index
;
1035 timer_jiffies
>>= TVN_BITS
;
1041 * Check, if the next hrtimer event is before the next timer wheel
1044 static unsigned long cmp_next_hrtimer_event(unsigned long now
,
1045 unsigned long expires
)
1047 ktime_t hr_delta
= hrtimer_get_next_event();
1048 struct timespec tsdelta
;
1049 unsigned long delta
;
1051 if (hr_delta
.tv64
== KTIME_MAX
)
1055 * Expired timer available, let it expire in the next tick
1057 if (hr_delta
.tv64
<= 0)
1060 tsdelta
= ktime_to_timespec(hr_delta
);
1061 delta
= timespec_to_jiffies(&tsdelta
);
1064 * Limit the delta to the max value, which is checked in
1065 * tick_nohz_stop_sched_tick():
1067 if (delta
> NEXT_TIMER_MAX_DELTA
)
1068 delta
= NEXT_TIMER_MAX_DELTA
;
1071 * Take rounding errors in to account and make sure, that it
1072 * expires in the next tick. Otherwise we go into an endless
1073 * ping pong due to tick_nohz_stop_sched_tick() retriggering
1079 if (time_before(now
, expires
))
1085 * get_next_timer_interrupt - return the jiffy of the next pending timer
1086 * @now: current time (in jiffies)
1088 unsigned long get_next_timer_interrupt(unsigned long now
)
1090 struct tvec_base
*base
= __get_cpu_var(tvec_bases
);
1091 unsigned long expires
;
1093 spin_lock(&base
->lock
);
1094 expires
= __next_timer_interrupt(base
);
1095 spin_unlock(&base
->lock
);
1097 if (time_before_eq(expires
, now
))
1100 return cmp_next_hrtimer_event(now
, expires
);
1105 * Called from the timer interrupt handler to charge one tick to the current
1106 * process. user_tick is 1 if the tick is user time, 0 for system.
1108 void update_process_times(int user_tick
)
1110 struct task_struct
*p
= current
;
1111 int cpu
= smp_processor_id();
1113 /* Note: this timer irq context must be accounted for as well. */
1114 account_process_tick(p
, user_tick
);
1116 if (rcu_pending(cpu
))
1117 rcu_check_callbacks(cpu
, user_tick
);
1120 run_posix_cpu_timers(p
);
1124 * Nr of active tasks - counted in fixed-point numbers
1126 static unsigned long count_active_tasks(void)
1128 return nr_active() * FIXED_1
;
1132 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1133 * imply that avenrun[] is the standard name for this kind of thing.
1134 * Nothing else seems to be standardized: the fractional size etc
1135 * all seem to differ on different machines.
1137 * Requires xtime_lock to access.
1139 unsigned long avenrun
[3];
1141 EXPORT_SYMBOL(avenrun
);
1144 * calc_load - given tick count, update the avenrun load estimates.
1145 * This is called while holding a write_lock on xtime_lock.
1147 static inline void calc_load(unsigned long ticks
)
1149 unsigned long active_tasks
; /* fixed-point */
1150 static int count
= LOAD_FREQ
;
1153 if (unlikely(count
< 0)) {
1154 active_tasks
= count_active_tasks();
1156 CALC_LOAD(avenrun
[0], EXP_1
, active_tasks
);
1157 CALC_LOAD(avenrun
[1], EXP_5
, active_tasks
);
1158 CALC_LOAD(avenrun
[2], EXP_15
, active_tasks
);
1160 } while (count
< 0);
1165 * This function runs timers and the timer-tq in bottom half context.
1167 static void run_timer_softirq(struct softirq_action
*h
)
1169 struct tvec_base
*base
= __get_cpu_var(tvec_bases
);
1171 perf_counter_do_pending();
1173 hrtimer_run_pending();
1175 if (time_after_eq(jiffies
, base
->timer_jiffies
))
1180 * Called by the local, per-CPU timer interrupt on SMP.
1182 void run_local_timers(void)
1184 hrtimer_run_queues();
1185 raise_softirq(TIMER_SOFTIRQ
);
1190 * Called by the timer interrupt. xtime_lock must already be taken
1193 static inline void update_times(unsigned long ticks
)
1200 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1201 * without sampling the sequence number in xtime_lock.
1202 * jiffies is defined in the linker script...
1205 void do_timer(unsigned long ticks
)
1207 jiffies_64
+= ticks
;
1208 update_times(ticks
);
1211 #ifdef __ARCH_WANT_SYS_ALARM
1214 * For backwards compatibility? This can be done in libc so Alpha
1215 * and all newer ports shouldn't need it.
1217 SYSCALL_DEFINE1(alarm
, unsigned int, seconds
)
1219 return alarm_setitimer(seconds
);
1227 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1228 * should be moved into arch/i386 instead?
1232 * sys_getpid - return the thread group id of the current process
1234 * Note, despite the name, this returns the tgid not the pid. The tgid and
1235 * the pid are identical unless CLONE_THREAD was specified on clone() in
1236 * which case the tgid is the same in all threads of the same group.
1238 * This is SMP safe as current->tgid does not change.
1240 SYSCALL_DEFINE0(getpid
)
1242 return task_tgid_vnr(current
);
1246 * Accessing ->real_parent is not SMP-safe, it could
1247 * change from under us. However, we can use a stale
1248 * value of ->real_parent under rcu_read_lock(), see
1249 * release_task()->call_rcu(delayed_put_task_struct).
1251 SYSCALL_DEFINE0(getppid
)
1256 pid
= task_tgid_vnr(current
->real_parent
);
1262 SYSCALL_DEFINE0(getuid
)
1264 /* Only we change this so SMP safe */
1265 return current_uid();
1268 SYSCALL_DEFINE0(geteuid
)
1270 /* Only we change this so SMP safe */
1271 return current_euid();
1274 SYSCALL_DEFINE0(getgid
)
1276 /* Only we change this so SMP safe */
1277 return current_gid();
1280 SYSCALL_DEFINE0(getegid
)
1282 /* Only we change this so SMP safe */
1283 return current_egid();
1288 static void process_timeout(unsigned long __data
)
1290 wake_up_process((struct task_struct
*)__data
);
1294 * schedule_timeout - sleep until timeout
1295 * @timeout: timeout value in jiffies
1297 * Make the current task sleep until @timeout jiffies have
1298 * elapsed. The routine will return immediately unless
1299 * the current task state has been set (see set_current_state()).
1301 * You can set the task state as follows -
1303 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1304 * pass before the routine returns. The routine will return 0
1306 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1307 * delivered to the current task. In this case the remaining time
1308 * in jiffies will be returned, or 0 if the timer expired in time
1310 * The current task state is guaranteed to be TASK_RUNNING when this
1313 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1314 * the CPU away without a bound on the timeout. In this case the return
1315 * value will be %MAX_SCHEDULE_TIMEOUT.
1317 * In all cases the return value is guaranteed to be non-negative.
1319 signed long __sched
schedule_timeout(signed long timeout
)
1321 struct timer_list timer
;
1322 unsigned long expire
;
1326 case MAX_SCHEDULE_TIMEOUT
:
1328 * These two special cases are useful to be comfortable
1329 * in the caller. Nothing more. We could take
1330 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1331 * but I' d like to return a valid offset (>=0) to allow
1332 * the caller to do everything it want with the retval.
1338 * Another bit of PARANOID. Note that the retval will be
1339 * 0 since no piece of kernel is supposed to do a check
1340 * for a negative retval of schedule_timeout() (since it
1341 * should never happens anyway). You just have the printk()
1342 * that will tell you if something is gone wrong and where.
1345 printk(KERN_ERR
"schedule_timeout: wrong timeout "
1346 "value %lx\n", timeout
);
1348 current
->state
= TASK_RUNNING
;
1353 expire
= timeout
+ jiffies
;
1355 setup_timer_on_stack(&timer
, process_timeout
, (unsigned long)current
);
1356 __mod_timer(&timer
, expire
, false);
1358 del_singleshot_timer_sync(&timer
);
1360 /* Remove the timer from the object tracker */
1361 destroy_timer_on_stack(&timer
);
1363 timeout
= expire
- jiffies
;
1366 return timeout
< 0 ? 0 : timeout
;
1368 EXPORT_SYMBOL(schedule_timeout
);
1371 * We can use __set_current_state() here because schedule_timeout() calls
1372 * schedule() unconditionally.
1374 signed long __sched
schedule_timeout_interruptible(signed long timeout
)
1376 __set_current_state(TASK_INTERRUPTIBLE
);
1377 return schedule_timeout(timeout
);
1379 EXPORT_SYMBOL(schedule_timeout_interruptible
);
1381 signed long __sched
schedule_timeout_killable(signed long timeout
)
1383 __set_current_state(TASK_KILLABLE
);
1384 return schedule_timeout(timeout
);
1386 EXPORT_SYMBOL(schedule_timeout_killable
);
1388 signed long __sched
schedule_timeout_uninterruptible(signed long timeout
)
1390 __set_current_state(TASK_UNINTERRUPTIBLE
);
1391 return schedule_timeout(timeout
);
1393 EXPORT_SYMBOL(schedule_timeout_uninterruptible
);
1395 /* Thread ID - the internal kernel "pid" */
1396 SYSCALL_DEFINE0(gettid
)
1398 return task_pid_vnr(current
);
1402 * do_sysinfo - fill in sysinfo struct
1403 * @info: pointer to buffer to fill
1405 int do_sysinfo(struct sysinfo
*info
)
1407 unsigned long mem_total
, sav_total
;
1408 unsigned int mem_unit
, bitcount
;
1411 memset(info
, 0, sizeof(struct sysinfo
));
1415 seq
= read_seqbegin(&xtime_lock
);
1418 * This is annoying. The below is the same thing
1419 * posix_get_clock_monotonic() does, but it wants to
1420 * take the lock which we want to cover the loads stuff
1424 getnstimeofday(&tp
);
1425 tp
.tv_sec
+= wall_to_monotonic
.tv_sec
;
1426 tp
.tv_nsec
+= wall_to_monotonic
.tv_nsec
;
1427 monotonic_to_bootbased(&tp
);
1428 if (tp
.tv_nsec
- NSEC_PER_SEC
>= 0) {
1429 tp
.tv_nsec
= tp
.tv_nsec
- NSEC_PER_SEC
;
1432 info
->uptime
= tp
.tv_sec
+ (tp
.tv_nsec
? 1 : 0);
1434 info
->loads
[0] = avenrun
[0] << (SI_LOAD_SHIFT
- FSHIFT
);
1435 info
->loads
[1] = avenrun
[1] << (SI_LOAD_SHIFT
- FSHIFT
);
1436 info
->loads
[2] = avenrun
[2] << (SI_LOAD_SHIFT
- FSHIFT
);
1438 info
->procs
= nr_threads
;
1439 } while (read_seqretry(&xtime_lock
, seq
));
1445 * If the sum of all the available memory (i.e. ram + swap)
1446 * is less than can be stored in a 32 bit unsigned long then
1447 * we can be binary compatible with 2.2.x kernels. If not,
1448 * well, in that case 2.2.x was broken anyways...
1450 * -Erik Andersen <andersee@debian.org>
1453 mem_total
= info
->totalram
+ info
->totalswap
;
1454 if (mem_total
< info
->totalram
|| mem_total
< info
->totalswap
)
1457 mem_unit
= info
->mem_unit
;
1458 while (mem_unit
> 1) {
1461 sav_total
= mem_total
;
1463 if (mem_total
< sav_total
)
1468 * If mem_total did not overflow, multiply all memory values by
1469 * info->mem_unit and set it to 1. This leaves things compatible
1470 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1475 info
->totalram
<<= bitcount
;
1476 info
->freeram
<<= bitcount
;
1477 info
->sharedram
<<= bitcount
;
1478 info
->bufferram
<<= bitcount
;
1479 info
->totalswap
<<= bitcount
;
1480 info
->freeswap
<<= bitcount
;
1481 info
->totalhigh
<<= bitcount
;
1482 info
->freehigh
<<= bitcount
;
1488 SYSCALL_DEFINE1(sysinfo
, struct sysinfo __user
*, info
)
1494 if (copy_to_user(info
, &val
, sizeof(struct sysinfo
)))
1500 static int __cpuinit
init_timers_cpu(int cpu
)
1503 struct tvec_base
*base
;
1504 static char __cpuinitdata tvec_base_done
[NR_CPUS
];
1506 if (!tvec_base_done
[cpu
]) {
1507 static char boot_done
;
1511 * The APs use this path later in boot
1513 base
= kmalloc_node(sizeof(*base
),
1514 GFP_KERNEL
| __GFP_ZERO
,
1519 /* Make sure that tvec_base is 2 byte aligned */
1520 if (tbase_get_deferrable(base
)) {
1525 per_cpu(tvec_bases
, cpu
) = base
;
1528 * This is for the boot CPU - we use compile-time
1529 * static initialisation because per-cpu memory isn't
1530 * ready yet and because the memory allocators are not
1531 * initialised either.
1534 base
= &boot_tvec_bases
;
1536 tvec_base_done
[cpu
] = 1;
1538 base
= per_cpu(tvec_bases
, cpu
);
1541 spin_lock_init(&base
->lock
);
1543 for (j
= 0; j
< TVN_SIZE
; j
++) {
1544 INIT_LIST_HEAD(base
->tv5
.vec
+ j
);
1545 INIT_LIST_HEAD(base
->tv4
.vec
+ j
);
1546 INIT_LIST_HEAD(base
->tv3
.vec
+ j
);
1547 INIT_LIST_HEAD(base
->tv2
.vec
+ j
);
1549 for (j
= 0; j
< TVR_SIZE
; j
++)
1550 INIT_LIST_HEAD(base
->tv1
.vec
+ j
);
1552 base
->timer_jiffies
= jiffies
;
1556 #ifdef CONFIG_HOTPLUG_CPU
1557 static void migrate_timer_list(struct tvec_base
*new_base
, struct list_head
*head
)
1559 struct timer_list
*timer
;
1561 while (!list_empty(head
)) {
1562 timer
= list_first_entry(head
, struct timer_list
, entry
);
1563 detach_timer(timer
, 0);
1564 timer_set_base(timer
, new_base
);
1565 internal_add_timer(new_base
, timer
);
1569 static void __cpuinit
migrate_timers(int cpu
)
1571 struct tvec_base
*old_base
;
1572 struct tvec_base
*new_base
;
1575 BUG_ON(cpu_online(cpu
));
1576 old_base
= per_cpu(tvec_bases
, cpu
);
1577 new_base
= get_cpu_var(tvec_bases
);
1579 * The caller is globally serialized and nobody else
1580 * takes two locks at once, deadlock is not possible.
1582 spin_lock_irq(&new_base
->lock
);
1583 spin_lock_nested(&old_base
->lock
, SINGLE_DEPTH_NESTING
);
1585 BUG_ON(old_base
->running_timer
);
1587 for (i
= 0; i
< TVR_SIZE
; i
++)
1588 migrate_timer_list(new_base
, old_base
->tv1
.vec
+ i
);
1589 for (i
= 0; i
< TVN_SIZE
; i
++) {
1590 migrate_timer_list(new_base
, old_base
->tv2
.vec
+ i
);
1591 migrate_timer_list(new_base
, old_base
->tv3
.vec
+ i
);
1592 migrate_timer_list(new_base
, old_base
->tv4
.vec
+ i
);
1593 migrate_timer_list(new_base
, old_base
->tv5
.vec
+ i
);
1596 spin_unlock(&old_base
->lock
);
1597 spin_unlock_irq(&new_base
->lock
);
1598 put_cpu_var(tvec_bases
);
1600 #endif /* CONFIG_HOTPLUG_CPU */
1602 static int __cpuinit
timer_cpu_notify(struct notifier_block
*self
,
1603 unsigned long action
, void *hcpu
)
1605 long cpu
= (long)hcpu
;
1607 case CPU_UP_PREPARE
:
1608 case CPU_UP_PREPARE_FROZEN
:
1609 if (init_timers_cpu(cpu
) < 0)
1612 #ifdef CONFIG_HOTPLUG_CPU
1614 case CPU_DEAD_FROZEN
:
1615 migrate_timers(cpu
);
1624 static struct notifier_block __cpuinitdata timers_nb
= {
1625 .notifier_call
= timer_cpu_notify
,
1629 void __init
init_timers(void)
1631 int err
= timer_cpu_notify(&timers_nb
, (unsigned long)CPU_UP_PREPARE
,
1632 (void *)(long)smp_processor_id());
1636 BUG_ON(err
== NOTIFY_BAD
);
1637 register_cpu_notifier(&timers_nb
);
1638 open_softirq(TIMER_SOFTIRQ
, run_timer_softirq
);
1642 * msleep - sleep safely even with waitqueue interruptions
1643 * @msecs: Time in milliseconds to sleep for
1645 void msleep(unsigned int msecs
)
1647 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1650 timeout
= schedule_timeout_uninterruptible(timeout
);
1653 EXPORT_SYMBOL(msleep
);
1656 * msleep_interruptible - sleep waiting for signals
1657 * @msecs: Time in milliseconds to sleep for
1659 unsigned long msleep_interruptible(unsigned int msecs
)
1661 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1663 while (timeout
&& !signal_pending(current
))
1664 timeout
= schedule_timeout_interruptible(timeout
);
1665 return jiffies_to_msecs(timeout
);
1668 EXPORT_SYMBOL(msleep_interruptible
);