4 * Kernel internal timers, kernel timekeeping, 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/notifier.h>
30 #include <linux/thread_info.h>
31 #include <linux/time.h>
32 #include <linux/jiffies.h>
33 #include <linux/posix-timers.h>
34 #include <linux/cpu.h>
35 #include <linux/syscalls.h>
36 #include <linux/delay.h>
37 #include <linux/tick.h>
38 #include <linux/kallsyms.h>
40 #include <asm/uaccess.h>
41 #include <asm/unistd.h>
42 #include <asm/div64.h>
43 #include <asm/timex.h>
46 u64 jiffies_64 __cacheline_aligned_in_smp
= INITIAL_JIFFIES
;
48 EXPORT_SYMBOL(jiffies_64
);
51 * per-CPU timer vector definitions:
53 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
54 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
55 #define TVN_SIZE (1 << TVN_BITS)
56 #define TVR_SIZE (1 << TVR_BITS)
57 #define TVN_MASK (TVN_SIZE - 1)
58 #define TVR_MASK (TVR_SIZE - 1)
60 typedef struct tvec_s
{
61 struct list_head vec
[TVN_SIZE
];
64 typedef struct tvec_root_s
{
65 struct list_head vec
[TVR_SIZE
];
68 struct tvec_t_base_s
{
70 struct timer_list
*running_timer
;
71 unsigned long timer_jiffies
;
77 } ____cacheline_aligned_in_smp
;
79 typedef struct tvec_t_base_s tvec_base_t
;
81 tvec_base_t boot_tvec_bases
;
82 EXPORT_SYMBOL(boot_tvec_bases
);
83 static DEFINE_PER_CPU(tvec_base_t
*, tvec_bases
) = &boot_tvec_bases
;
86 * __round_jiffies - function to round jiffies to a full second
87 * @j: the time in (absolute) jiffies that should be rounded
88 * @cpu: the processor number on which the timeout will happen
90 * __round_jiffies() rounds an absolute time in the future (in jiffies)
91 * up or down to (approximately) full seconds. This is useful for timers
92 * for which the exact time they fire does not matter too much, as long as
93 * they fire approximately every X seconds.
95 * By rounding these timers to whole seconds, all such timers will fire
96 * at the same time, rather than at various times spread out. The goal
97 * of this is to have the CPU wake up less, which saves power.
99 * The exact rounding is skewed for each processor to avoid all
100 * processors firing at the exact same time, which could lead
101 * to lock contention or spurious cache line bouncing.
103 * The return value is the rounded version of the @j parameter.
105 unsigned long __round_jiffies(unsigned long j
, int cpu
)
108 unsigned long original
= j
;
111 * We don't want all cpus firing their timers at once hitting the
112 * same lock or cachelines, so we skew each extra cpu with an extra
113 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
115 * The skew is done by adding 3*cpunr, then round, then subtract this
116 * extra offset again.
123 * If the target jiffie is just after a whole second (which can happen
124 * due to delays of the timer irq, long irq off times etc etc) then
125 * we should round down to the whole second, not up. Use 1/4th second
126 * as cutoff for this rounding as an extreme upper bound for this.
128 if (rem
< HZ
/4) /* round down */
133 /* now that we have rounded, subtract the extra skew again */
136 if (j
<= jiffies
) /* rounding ate our timeout entirely; */
140 EXPORT_SYMBOL_GPL(__round_jiffies
);
143 * __round_jiffies_relative - function to round jiffies to a full second
144 * @j: the time in (relative) jiffies that should be rounded
145 * @cpu: the processor number on which the timeout will happen
147 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
148 * up or down to (approximately) full seconds. This is useful for timers
149 * for which the exact time they fire does not matter too much, as long as
150 * they fire approximately every X seconds.
152 * By rounding these timers to whole seconds, all such timers will fire
153 * at the same time, rather than at various times spread out. The goal
154 * of this is to have the CPU wake up less, which saves power.
156 * The exact rounding is skewed for each processor to avoid all
157 * processors firing at the exact same time, which could lead
158 * to lock contention or spurious cache line bouncing.
160 * The return value is the rounded version of the @j parameter.
162 unsigned long __round_jiffies_relative(unsigned long j
, int cpu
)
165 * In theory the following code can skip a jiffy in case jiffies
166 * increments right between the addition and the later subtraction.
167 * However since the entire point of this function is to use approximate
168 * timeouts, it's entirely ok to not handle that.
170 return __round_jiffies(j
+ jiffies
, cpu
) - jiffies
;
172 EXPORT_SYMBOL_GPL(__round_jiffies_relative
);
175 * round_jiffies - function to round jiffies to a full second
176 * @j: the time in (absolute) jiffies that should be rounded
178 * round_jiffies() rounds an absolute time in the future (in jiffies)
179 * up or down to (approximately) full seconds. This is useful for timers
180 * for which the exact time they fire does not matter too much, as long as
181 * they fire approximately every X seconds.
183 * By rounding these timers to whole seconds, all such timers will fire
184 * at the same time, rather than at various times spread out. The goal
185 * of this is to have the CPU wake up less, which saves power.
187 * The return value is the rounded version of the @j parameter.
189 unsigned long round_jiffies(unsigned long j
)
191 return __round_jiffies(j
, raw_smp_processor_id());
193 EXPORT_SYMBOL_GPL(round_jiffies
);
196 * round_jiffies_relative - function to round jiffies to a full second
197 * @j: the time in (relative) jiffies that should be rounded
199 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
200 * up or down to (approximately) full seconds. This is useful for timers
201 * for which the exact time they fire does not matter too much, as long as
202 * they fire approximately every X seconds.
204 * By rounding these timers to whole seconds, all such timers will fire
205 * at the same time, rather than at various times spread out. The goal
206 * of this is to have the CPU wake up less, which saves power.
208 * The return value is the rounded version of the @j parameter.
210 unsigned long round_jiffies_relative(unsigned long j
)
212 return __round_jiffies_relative(j
, raw_smp_processor_id());
214 EXPORT_SYMBOL_GPL(round_jiffies_relative
);
217 static inline void set_running_timer(tvec_base_t
*base
,
218 struct timer_list
*timer
)
221 base
->running_timer
= timer
;
225 static void internal_add_timer(tvec_base_t
*base
, struct timer_list
*timer
)
227 unsigned long expires
= timer
->expires
;
228 unsigned long idx
= expires
- base
->timer_jiffies
;
229 struct list_head
*vec
;
231 if (idx
< TVR_SIZE
) {
232 int i
= expires
& TVR_MASK
;
233 vec
= base
->tv1
.vec
+ i
;
234 } else if (idx
< 1 << (TVR_BITS
+ TVN_BITS
)) {
235 int i
= (expires
>> TVR_BITS
) & TVN_MASK
;
236 vec
= base
->tv2
.vec
+ i
;
237 } else if (idx
< 1 << (TVR_BITS
+ 2 * TVN_BITS
)) {
238 int i
= (expires
>> (TVR_BITS
+ TVN_BITS
)) & TVN_MASK
;
239 vec
= base
->tv3
.vec
+ i
;
240 } else if (idx
< 1 << (TVR_BITS
+ 3 * TVN_BITS
)) {
241 int i
= (expires
>> (TVR_BITS
+ 2 * TVN_BITS
)) & TVN_MASK
;
242 vec
= base
->tv4
.vec
+ i
;
243 } else if ((signed long) idx
< 0) {
245 * Can happen if you add a timer with expires == jiffies,
246 * or you set a timer to go off in the past
248 vec
= base
->tv1
.vec
+ (base
->timer_jiffies
& TVR_MASK
);
251 /* If the timeout is larger than 0xffffffff on 64-bit
252 * architectures then we use the maximum timeout:
254 if (idx
> 0xffffffffUL
) {
256 expires
= idx
+ base
->timer_jiffies
;
258 i
= (expires
>> (TVR_BITS
+ 3 * TVN_BITS
)) & TVN_MASK
;
259 vec
= base
->tv5
.vec
+ i
;
264 list_add_tail(&timer
->entry
, vec
);
267 #ifdef CONFIG_TIMER_STATS
268 void __timer_stats_timer_set_start_info(struct timer_list
*timer
, void *addr
)
270 if (timer
->start_site
)
273 timer
->start_site
= addr
;
274 memcpy(timer
->start_comm
, current
->comm
, TASK_COMM_LEN
);
275 timer
->start_pid
= current
->pid
;
280 * init_timer - initialize a timer.
281 * @timer: the timer to be initialized
283 * init_timer() must be done to a timer prior calling *any* of the
284 * other timer functions.
286 void fastcall
init_timer(struct timer_list
*timer
)
288 timer
->entry
.next
= NULL
;
289 timer
->base
= __raw_get_cpu_var(tvec_bases
);
290 #ifdef CONFIG_TIMER_STATS
291 timer
->start_site
= NULL
;
292 timer
->start_pid
= -1;
293 memset(timer
->start_comm
, 0, TASK_COMM_LEN
);
296 EXPORT_SYMBOL(init_timer
);
298 static inline void detach_timer(struct timer_list
*timer
,
301 struct list_head
*entry
= &timer
->entry
;
303 __list_del(entry
->prev
, entry
->next
);
306 entry
->prev
= LIST_POISON2
;
310 * We are using hashed locking: holding per_cpu(tvec_bases).lock
311 * means that all timers which are tied to this base via timer->base are
312 * locked, and the base itself is locked too.
314 * So __run_timers/migrate_timers can safely modify all timers which could
315 * be found on ->tvX lists.
317 * When the timer's base is locked, and the timer removed from list, it is
318 * possible to set timer->base = NULL and drop the lock: the timer remains
321 static tvec_base_t
*lock_timer_base(struct timer_list
*timer
,
322 unsigned long *flags
)
323 __acquires(timer
->base
->lock
)
329 if (likely(base
!= NULL
)) {
330 spin_lock_irqsave(&base
->lock
, *flags
);
331 if (likely(base
== timer
->base
))
333 /* The timer has migrated to another CPU */
334 spin_unlock_irqrestore(&base
->lock
, *flags
);
340 int __mod_timer(struct timer_list
*timer
, unsigned long expires
)
342 tvec_base_t
*base
, *new_base
;
346 timer_stats_timer_set_start_info(timer
);
347 BUG_ON(!timer
->function
);
349 base
= lock_timer_base(timer
, &flags
);
351 if (timer_pending(timer
)) {
352 detach_timer(timer
, 0);
356 new_base
= __get_cpu_var(tvec_bases
);
358 if (base
!= new_base
) {
360 * We are trying to schedule the timer on the local CPU.
361 * However we can't change timer's base while it is running,
362 * otherwise del_timer_sync() can't detect that the timer's
363 * handler yet has not finished. This also guarantees that
364 * the timer is serialized wrt itself.
366 if (likely(base
->running_timer
!= timer
)) {
367 /* See the comment in lock_timer_base() */
369 spin_unlock(&base
->lock
);
371 spin_lock(&base
->lock
);
376 timer
->expires
= expires
;
377 internal_add_timer(base
, timer
);
378 spin_unlock_irqrestore(&base
->lock
, flags
);
383 EXPORT_SYMBOL(__mod_timer
);
386 * add_timer_on - start a timer on a particular CPU
387 * @timer: the timer to be added
388 * @cpu: the CPU to start it on
390 * This is not very scalable on SMP. Double adds are not possible.
392 void add_timer_on(struct timer_list
*timer
, int cpu
)
394 tvec_base_t
*base
= per_cpu(tvec_bases
, cpu
);
397 timer_stats_timer_set_start_info(timer
);
398 BUG_ON(timer_pending(timer
) || !timer
->function
);
399 spin_lock_irqsave(&base
->lock
, flags
);
401 internal_add_timer(base
, timer
);
402 spin_unlock_irqrestore(&base
->lock
, flags
);
407 * mod_timer - modify a timer's timeout
408 * @timer: the timer to be modified
409 * @expires: new timeout in jiffies
411 * mod_timer() is a more efficient way to update the expire field of an
412 * active timer (if the timer is inactive it will be activated)
414 * mod_timer(timer, expires) is equivalent to:
416 * del_timer(timer); timer->expires = expires; add_timer(timer);
418 * Note that if there are multiple unserialized concurrent users of the
419 * same timer, then mod_timer() is the only safe way to modify the timeout,
420 * since add_timer() cannot modify an already running timer.
422 * The function returns whether it has modified a pending timer or not.
423 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
424 * active timer returns 1.)
426 int mod_timer(struct timer_list
*timer
, unsigned long expires
)
428 BUG_ON(!timer
->function
);
430 timer_stats_timer_set_start_info(timer
);
432 * This is a common optimization triggered by the
433 * networking code - if the timer is re-modified
434 * to be the same thing then just return:
436 if (timer
->expires
== expires
&& timer_pending(timer
))
439 return __mod_timer(timer
, expires
);
442 EXPORT_SYMBOL(mod_timer
);
445 * del_timer - deactive a timer.
446 * @timer: the timer to be deactivated
448 * del_timer() deactivates a timer - this works on both active and inactive
451 * The function returns whether it has deactivated a pending timer or not.
452 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
453 * active timer returns 1.)
455 int del_timer(struct timer_list
*timer
)
461 timer_stats_timer_clear_start_info(timer
);
462 if (timer_pending(timer
)) {
463 base
= lock_timer_base(timer
, &flags
);
464 if (timer_pending(timer
)) {
465 detach_timer(timer
, 1);
468 spin_unlock_irqrestore(&base
->lock
, flags
);
474 EXPORT_SYMBOL(del_timer
);
478 * try_to_del_timer_sync - Try to deactivate a timer
479 * @timer: timer do del
481 * This function tries to deactivate a timer. Upon successful (ret >= 0)
482 * exit the timer is not queued and the handler is not running on any CPU.
484 * It must not be called from interrupt contexts.
486 int try_to_del_timer_sync(struct timer_list
*timer
)
492 base
= lock_timer_base(timer
, &flags
);
494 if (base
->running_timer
== timer
)
498 if (timer_pending(timer
)) {
499 detach_timer(timer
, 1);
503 spin_unlock_irqrestore(&base
->lock
, flags
);
509 * del_timer_sync - deactivate a timer and wait for the handler to finish.
510 * @timer: the timer to be deactivated
512 * This function only differs from del_timer() on SMP: besides deactivating
513 * the timer it also makes sure the handler has finished executing on other
516 * Synchronization rules: Callers must prevent restarting of the timer,
517 * otherwise this function is meaningless. It must not be called from
518 * interrupt contexts. The caller must not hold locks which would prevent
519 * completion of the timer's handler. The timer's handler must not call
520 * add_timer_on(). Upon exit the timer is not queued and the handler is
521 * not running on any CPU.
523 * The function returns whether it has deactivated a pending timer or not.
525 int del_timer_sync(struct timer_list
*timer
)
528 int ret
= try_to_del_timer_sync(timer
);
535 EXPORT_SYMBOL(del_timer_sync
);
538 static int cascade(tvec_base_t
*base
, tvec_t
*tv
, int index
)
540 /* cascade all the timers from tv up one level */
541 struct timer_list
*timer
, *tmp
;
542 struct list_head tv_list
;
544 list_replace_init(tv
->vec
+ index
, &tv_list
);
547 * We are removing _all_ timers from the list, so we
548 * don't have to detach them individually.
550 list_for_each_entry_safe(timer
, tmp
, &tv_list
, entry
) {
551 BUG_ON(timer
->base
!= base
);
552 internal_add_timer(base
, timer
);
558 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
561 * __run_timers - run all expired timers (if any) on this CPU.
562 * @base: the timer vector to be processed.
564 * This function cascades all vectors and executes all expired timer
567 static inline void __run_timers(tvec_base_t
*base
)
569 struct timer_list
*timer
;
571 spin_lock_irq(&base
->lock
);
572 while (time_after_eq(jiffies
, base
->timer_jiffies
)) {
573 struct list_head work_list
;
574 struct list_head
*head
= &work_list
;
575 int index
= base
->timer_jiffies
& TVR_MASK
;
581 (!cascade(base
, &base
->tv2
, INDEX(0))) &&
582 (!cascade(base
, &base
->tv3
, INDEX(1))) &&
583 !cascade(base
, &base
->tv4
, INDEX(2)))
584 cascade(base
, &base
->tv5
, INDEX(3));
585 ++base
->timer_jiffies
;
586 list_replace_init(base
->tv1
.vec
+ index
, &work_list
);
587 while (!list_empty(head
)) {
588 void (*fn
)(unsigned long);
591 timer
= list_entry(head
->next
,struct timer_list
,entry
);
592 fn
= timer
->function
;
595 timer_stats_account_timer(timer
);
597 set_running_timer(base
, timer
);
598 detach_timer(timer
, 1);
599 spin_unlock_irq(&base
->lock
);
601 int preempt_count
= preempt_count();
603 if (preempt_count
!= preempt_count()) {
604 printk(KERN_WARNING
"huh, entered %p "
605 "with preempt_count %08x, exited"
612 spin_lock_irq(&base
->lock
);
615 set_running_timer(base
, NULL
);
616 spin_unlock_irq(&base
->lock
);
619 #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
621 * Find out when the next timer event is due to happen. This
622 * is used on S/390 to stop all activity when a cpus is idle.
623 * This functions needs to be called disabled.
625 static unsigned long __next_timer_interrupt(tvec_base_t
*base
)
627 unsigned long timer_jiffies
= base
->timer_jiffies
;
628 unsigned long expires
= timer_jiffies
+ (LONG_MAX
>> 1);
629 int index
, slot
, array
, found
= 0;
630 struct timer_list
*nte
;
633 /* Look for timer events in tv1. */
634 index
= slot
= timer_jiffies
& TVR_MASK
;
636 list_for_each_entry(nte
, base
->tv1
.vec
+ slot
, entry
) {
638 expires
= nte
->expires
;
639 /* Look at the cascade bucket(s)? */
640 if (!index
|| slot
< index
)
644 slot
= (slot
+ 1) & TVR_MASK
;
645 } while (slot
!= index
);
648 /* Calculate the next cascade event */
650 timer_jiffies
+= TVR_SIZE
- index
;
651 timer_jiffies
>>= TVR_BITS
;
654 varray
[0] = &base
->tv2
;
655 varray
[1] = &base
->tv3
;
656 varray
[2] = &base
->tv4
;
657 varray
[3] = &base
->tv5
;
659 for (array
= 0; array
< 4; array
++) {
660 tvec_t
*varp
= varray
[array
];
662 index
= slot
= timer_jiffies
& TVN_MASK
;
664 list_for_each_entry(nte
, varp
->vec
+ slot
, entry
) {
666 if (time_before(nte
->expires
, expires
))
667 expires
= nte
->expires
;
670 * Do we still search for the first timer or are
671 * we looking up the cascade buckets ?
674 /* Look at the cascade bucket(s)? */
675 if (!index
|| slot
< index
)
679 slot
= (slot
+ 1) & TVN_MASK
;
680 } while (slot
!= index
);
683 timer_jiffies
+= TVN_SIZE
- index
;
684 timer_jiffies
>>= TVN_BITS
;
690 * Check, if the next hrtimer event is before the next timer wheel
693 static unsigned long cmp_next_hrtimer_event(unsigned long now
,
694 unsigned long expires
)
696 ktime_t hr_delta
= hrtimer_get_next_event();
697 struct timespec tsdelta
;
700 if (hr_delta
.tv64
== KTIME_MAX
)
704 * Expired timer available, let it expire in the next tick
706 if (hr_delta
.tv64
<= 0)
709 tsdelta
= ktime_to_timespec(hr_delta
);
710 delta
= timespec_to_jiffies(&tsdelta
);
712 * Take rounding errors in to account and make sure, that it
713 * expires in the next tick. Otherwise we go into an endless
714 * ping pong due to tick_nohz_stop_sched_tick() retriggering
720 if (time_before(now
, expires
))
726 * next_timer_interrupt - return the jiffy of the next pending timer
727 * @now: current time (in jiffies)
729 unsigned long get_next_timer_interrupt(unsigned long now
)
731 tvec_base_t
*base
= __get_cpu_var(tvec_bases
);
732 unsigned long expires
;
734 spin_lock(&base
->lock
);
735 expires
= __next_timer_interrupt(base
);
736 spin_unlock(&base
->lock
);
738 if (time_before_eq(expires
, now
))
741 return cmp_next_hrtimer_event(now
, expires
);
744 #ifdef CONFIG_NO_IDLE_HZ
745 unsigned long next_timer_interrupt(void)
747 return get_next_timer_interrupt(jiffies
);
753 /******************************************************************/
757 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
758 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
759 * at zero at system boot time, so wall_to_monotonic will be negative,
760 * however, we will ALWAYS keep the tv_nsec part positive so we can use
761 * the usual normalization.
763 struct timespec xtime
__attribute__ ((aligned (16)));
764 struct timespec wall_to_monotonic
__attribute__ ((aligned (16)));
766 EXPORT_SYMBOL(xtime
);
769 /* XXX - all of this timekeeping code should be later moved to time.c */
770 #include <linux/clocksource.h>
771 static struct clocksource
*clock
; /* pointer to current clocksource */
773 #ifdef CONFIG_GENERIC_TIME
775 * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook
777 * private function, must hold xtime_lock lock when being
778 * called. Returns the number of nanoseconds since the
779 * last call to update_wall_time() (adjusted by NTP scaling)
781 static inline s64
__get_nsec_offset(void)
783 cycle_t cycle_now
, cycle_delta
;
786 /* read clocksource: */
787 cycle_now
= clocksource_read(clock
);
789 /* calculate the delta since the last update_wall_time: */
790 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
792 /* convert to nanoseconds: */
793 ns_offset
= cyc2ns(clock
, cycle_delta
);
799 * __get_realtime_clock_ts - Returns the time of day in a timespec
800 * @ts: pointer to the timespec to be set
802 * Returns the time of day in a timespec. Used by
803 * do_gettimeofday() and get_realtime_clock_ts().
805 static inline void __get_realtime_clock_ts(struct timespec
*ts
)
811 seq
= read_seqbegin(&xtime_lock
);
814 nsecs
= __get_nsec_offset();
816 } while (read_seqretry(&xtime_lock
, seq
));
818 timespec_add_ns(ts
, nsecs
);
822 * getnstimeofday - Returns the time of day in a timespec
823 * @ts: pointer to the timespec to be set
825 * Returns the time of day in a timespec.
827 void getnstimeofday(struct timespec
*ts
)
829 __get_realtime_clock_ts(ts
);
832 EXPORT_SYMBOL(getnstimeofday
);
835 * do_gettimeofday - Returns the time of day in a timeval
836 * @tv: pointer to the timeval to be set
838 * NOTE: Users should be converted to using get_realtime_clock_ts()
840 void do_gettimeofday(struct timeval
*tv
)
844 __get_realtime_clock_ts(&now
);
845 tv
->tv_sec
= now
.tv_sec
;
846 tv
->tv_usec
= now
.tv_nsec
/1000;
849 EXPORT_SYMBOL(do_gettimeofday
);
851 * do_settimeofday - Sets the time of day
852 * @tv: pointer to the timespec variable containing the new time
854 * Sets the time of day to the new time and update NTP and notify hrtimers
856 int do_settimeofday(struct timespec
*tv
)
859 time_t wtm_sec
, sec
= tv
->tv_sec
;
860 long wtm_nsec
, nsec
= tv
->tv_nsec
;
862 if ((unsigned long)tv
->tv_nsec
>= NSEC_PER_SEC
)
865 write_seqlock_irqsave(&xtime_lock
, flags
);
867 nsec
-= __get_nsec_offset();
869 wtm_sec
= wall_to_monotonic
.tv_sec
+ (xtime
.tv_sec
- sec
);
870 wtm_nsec
= wall_to_monotonic
.tv_nsec
+ (xtime
.tv_nsec
- nsec
);
872 set_normalized_timespec(&xtime
, sec
, nsec
);
873 set_normalized_timespec(&wall_to_monotonic
, wtm_sec
, wtm_nsec
);
878 update_vsyscall(&xtime
, clock
);
880 write_sequnlock_irqrestore(&xtime_lock
, flags
);
882 /* signal hrtimers about time change */
888 EXPORT_SYMBOL(do_settimeofday
);
891 * change_clocksource - Swaps clocksources if a new one is available
893 * Accumulates current time interval and initializes new clocksource
895 static void change_clocksource(void)
897 struct clocksource
*new;
901 new = clocksource_get_next();
906 now
= clocksource_read(new);
907 nsec
= __get_nsec_offset();
908 timespec_add_ns(&xtime
, nsec
);
911 clock
->cycle_last
= now
;
914 clock
->xtime_nsec
= 0;
915 clocksource_calculate_interval(clock
, NTP_INTERVAL_LENGTH
);
919 printk(KERN_INFO
"Time: %s clocksource has been installed.\n",
923 static inline void change_clocksource(void) { }
927 * timekeeping_is_continuous - check to see if timekeeping is free running
929 int timekeeping_is_continuous(void)
935 seq
= read_seqbegin(&xtime_lock
);
937 ret
= clock
->flags
& CLOCK_SOURCE_VALID_FOR_HRES
;
939 } while (read_seqretry(&xtime_lock
, seq
));
945 * read_persistent_clock - Return time in seconds from the persistent clock.
947 * Weak dummy function for arches that do not yet support it.
948 * Returns seconds from epoch using the battery backed persistent clock.
949 * Returns zero if unsupported.
951 * XXX - Do be sure to remove it once all arches implement it.
953 unsigned long __attribute__((weak
)) read_persistent_clock(void)
959 * timekeeping_init - Initializes the clocksource and common timekeeping values
961 void __init
timekeeping_init(void)
964 unsigned long sec
= read_persistent_clock();
966 write_seqlock_irqsave(&xtime_lock
, flags
);
970 clock
= clocksource_get_next();
971 clocksource_calculate_interval(clock
, NTP_INTERVAL_LENGTH
);
972 clock
->cycle_last
= clocksource_read(clock
);
976 set_normalized_timespec(&wall_to_monotonic
,
977 -xtime
.tv_sec
, -xtime
.tv_nsec
);
979 write_sequnlock_irqrestore(&xtime_lock
, flags
);
982 /* flag for if timekeeping is suspended */
983 static int timekeeping_suspended
;
984 /* time in seconds when suspend began */
985 static unsigned long timekeeping_suspend_time
;
988 * timekeeping_resume - Resumes the generic timekeeping subsystem.
991 * This is for the generic clocksource timekeeping.
992 * xtime/wall_to_monotonic/jiffies/etc are
993 * still managed by arch specific suspend/resume code.
995 static int timekeeping_resume(struct sys_device
*dev
)
998 unsigned long now
= read_persistent_clock();
1000 write_seqlock_irqsave(&xtime_lock
, flags
);
1002 if (now
&& (now
> timekeeping_suspend_time
)) {
1003 unsigned long sleep_length
= now
- timekeeping_suspend_time
;
1005 xtime
.tv_sec
+= sleep_length
;
1006 wall_to_monotonic
.tv_sec
-= sleep_length
;
1008 /* re-base the last cycle value */
1009 clock
->cycle_last
= clocksource_read(clock
);
1011 timekeeping_suspended
= 0;
1012 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1014 touch_softlockup_watchdog();
1016 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME
, NULL
);
1018 /* Resume hrtimers */
1019 hres_timers_resume();
1024 static int timekeeping_suspend(struct sys_device
*dev
, pm_message_t state
)
1026 unsigned long flags
;
1028 write_seqlock_irqsave(&xtime_lock
, flags
);
1029 timekeeping_suspended
= 1;
1030 timekeeping_suspend_time
= read_persistent_clock();
1031 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1033 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND
, NULL
);
1038 /* sysfs resume/suspend bits for timekeeping */
1039 static struct sysdev_class timekeeping_sysclass
= {
1040 .resume
= timekeeping_resume
,
1041 .suspend
= timekeeping_suspend
,
1042 set_kset_name("timekeeping"),
1045 static struct sys_device device_timer
= {
1047 .cls
= &timekeeping_sysclass
,
1050 static int __init
timekeeping_init_device(void)
1052 int error
= sysdev_class_register(&timekeeping_sysclass
);
1054 error
= sysdev_register(&device_timer
);
1058 device_initcall(timekeeping_init_device
);
1061 * If the error is already larger, we look ahead even further
1062 * to compensate for late or lost adjustments.
1064 static __always_inline
int clocksource_bigadjust(s64 error
, s64
*interval
,
1068 u32 look_ahead
, adj
;
1072 * Use the current error value to determine how much to look ahead.
1073 * The larger the error the slower we adjust for it to avoid problems
1074 * with losing too many ticks, otherwise we would overadjust and
1075 * produce an even larger error. The smaller the adjustment the
1076 * faster we try to adjust for it, as lost ticks can do less harm
1077 * here. This is tuned so that an error of about 1 msec is adusted
1078 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1080 error2
= clock
->error
>> (TICK_LENGTH_SHIFT
+ 22 - 2 * SHIFT_HZ
);
1081 error2
= abs(error2
);
1082 for (look_ahead
= 0; error2
> 0; look_ahead
++)
1086 * Now calculate the error in (1 << look_ahead) ticks, but first
1087 * remove the single look ahead already included in the error.
1089 tick_error
= current_tick_length() >>
1090 (TICK_LENGTH_SHIFT
- clock
->shift
+ 1);
1091 tick_error
-= clock
->xtime_interval
>> 1;
1092 error
= ((error
- tick_error
) >> look_ahead
) + tick_error
;
1094 /* Finally calculate the adjustment shift value. */
1099 *interval
= -*interval
;
1103 for (adj
= 0; error
> i
; adj
++)
1112 * Adjust the multiplier to reduce the error value,
1113 * this is optimized for the most common adjustments of -1,0,1,
1114 * for other values we can do a bit more work.
1116 static void clocksource_adjust(struct clocksource
*clock
, s64 offset
)
1118 s64 error
, interval
= clock
->cycle_interval
;
1121 error
= clock
->error
>> (TICK_LENGTH_SHIFT
- clock
->shift
- 1);
1122 if (error
> interval
) {
1124 if (likely(error
<= interval
))
1127 adj
= clocksource_bigadjust(error
, &interval
, &offset
);
1128 } else if (error
< -interval
) {
1130 if (likely(error
>= -interval
)) {
1132 interval
= -interval
;
1135 adj
= clocksource_bigadjust(error
, &interval
, &offset
);
1140 clock
->xtime_interval
+= interval
;
1141 clock
->xtime_nsec
-= offset
;
1142 clock
->error
-= (interval
- offset
) <<
1143 (TICK_LENGTH_SHIFT
- clock
->shift
);
1147 * update_wall_time - Uses the current clocksource to increment the wall time
1149 * Called from the timer interrupt, must hold a write on xtime_lock.
1151 static void update_wall_time(void)
1155 /* Make sure we're fully resumed: */
1156 if (unlikely(timekeeping_suspended
))
1159 #ifdef CONFIG_GENERIC_TIME
1160 offset
= (clocksource_read(clock
) - clock
->cycle_last
) & clock
->mask
;
1162 offset
= clock
->cycle_interval
;
1164 clock
->xtime_nsec
+= (s64
)xtime
.tv_nsec
<< clock
->shift
;
1166 /* normally this loop will run just once, however in the
1167 * case of lost or late ticks, it will accumulate correctly.
1169 while (offset
>= clock
->cycle_interval
) {
1170 /* accumulate one interval */
1171 clock
->xtime_nsec
+= clock
->xtime_interval
;
1172 clock
->cycle_last
+= clock
->cycle_interval
;
1173 offset
-= clock
->cycle_interval
;
1175 if (clock
->xtime_nsec
>= (u64
)NSEC_PER_SEC
<< clock
->shift
) {
1176 clock
->xtime_nsec
-= (u64
)NSEC_PER_SEC
<< clock
->shift
;
1181 /* interpolator bits */
1182 time_interpolator_update(clock
->xtime_interval
1185 /* accumulate error between NTP and clock interval */
1186 clock
->error
+= current_tick_length();
1187 clock
->error
-= clock
->xtime_interval
<< (TICK_LENGTH_SHIFT
- clock
->shift
);
1190 /* correct the clock when NTP error is too big */
1191 clocksource_adjust(clock
, offset
);
1193 /* store full nanoseconds into xtime */
1194 xtime
.tv_nsec
= (s64
)clock
->xtime_nsec
>> clock
->shift
;
1195 clock
->xtime_nsec
-= (s64
)xtime
.tv_nsec
<< clock
->shift
;
1197 /* check to see if there is a new clocksource to use */
1198 change_clocksource();
1199 update_vsyscall(&xtime
, clock
);
1203 * Called from the timer interrupt handler to charge one tick to the current
1204 * process. user_tick is 1 if the tick is user time, 0 for system.
1206 void update_process_times(int user_tick
)
1208 struct task_struct
*p
= current
;
1209 int cpu
= smp_processor_id();
1211 /* Note: this timer irq context must be accounted for as well. */
1213 account_user_time(p
, jiffies_to_cputime(1));
1215 account_system_time(p
, HARDIRQ_OFFSET
, jiffies_to_cputime(1));
1217 if (rcu_pending(cpu
))
1218 rcu_check_callbacks(cpu
, user_tick
);
1220 run_posix_cpu_timers(p
);
1224 * Nr of active tasks - counted in fixed-point numbers
1226 static unsigned long count_active_tasks(void)
1228 return nr_active() * FIXED_1
;
1232 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1233 * imply that avenrun[] is the standard name for this kind of thing.
1234 * Nothing else seems to be standardized: the fractional size etc
1235 * all seem to differ on different machines.
1237 * Requires xtime_lock to access.
1239 unsigned long avenrun
[3];
1241 EXPORT_SYMBOL(avenrun
);
1244 * calc_load - given tick count, update the avenrun load estimates.
1245 * This is called while holding a write_lock on xtime_lock.
1247 static inline void calc_load(unsigned long ticks
)
1249 unsigned long active_tasks
; /* fixed-point */
1250 static int count
= LOAD_FREQ
;
1253 if (unlikely(count
< 0)) {
1254 active_tasks
= count_active_tasks();
1256 CALC_LOAD(avenrun
[0], EXP_1
, active_tasks
);
1257 CALC_LOAD(avenrun
[1], EXP_5
, active_tasks
);
1258 CALC_LOAD(avenrun
[2], EXP_15
, active_tasks
);
1260 } while (count
< 0);
1265 * This read-write spinlock protects us from races in SMP while
1266 * playing with xtime and avenrun.
1268 __attribute__((weak
)) __cacheline_aligned_in_smp
DEFINE_SEQLOCK(xtime_lock
);
1270 EXPORT_SYMBOL(xtime_lock
);
1273 * This function runs timers and the timer-tq in bottom half context.
1275 static void run_timer_softirq(struct softirq_action
*h
)
1277 tvec_base_t
*base
= __get_cpu_var(tvec_bases
);
1279 hrtimer_run_queues();
1281 if (time_after_eq(jiffies
, base
->timer_jiffies
))
1286 * Called by the local, per-CPU timer interrupt on SMP.
1288 void run_local_timers(void)
1290 raise_softirq(TIMER_SOFTIRQ
);
1295 * Called by the timer interrupt. xtime_lock must already be taken
1298 static inline void update_times(unsigned long ticks
)
1305 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1306 * without sampling the sequence number in xtime_lock.
1307 * jiffies is defined in the linker script...
1310 void do_timer(unsigned long ticks
)
1312 jiffies_64
+= ticks
;
1313 update_times(ticks
);
1316 #ifdef __ARCH_WANT_SYS_ALARM
1319 * For backwards compatibility? This can be done in libc so Alpha
1320 * and all newer ports shouldn't need it.
1322 asmlinkage
unsigned long sys_alarm(unsigned int seconds
)
1324 return alarm_setitimer(seconds
);
1332 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1333 * should be moved into arch/i386 instead?
1337 * sys_getpid - return the thread group id of the current process
1339 * Note, despite the name, this returns the tgid not the pid. The tgid and
1340 * the pid are identical unless CLONE_THREAD was specified on clone() in
1341 * which case the tgid is the same in all threads of the same group.
1343 * This is SMP safe as current->tgid does not change.
1345 asmlinkage
long sys_getpid(void)
1347 return current
->tgid
;
1351 * Accessing ->real_parent is not SMP-safe, it could
1352 * change from under us. However, we can use a stale
1353 * value of ->real_parent under rcu_read_lock(), see
1354 * release_task()->call_rcu(delayed_put_task_struct).
1356 asmlinkage
long sys_getppid(void)
1361 pid
= rcu_dereference(current
->real_parent
)->tgid
;
1367 asmlinkage
long sys_getuid(void)
1369 /* Only we change this so SMP safe */
1370 return current
->uid
;
1373 asmlinkage
long sys_geteuid(void)
1375 /* Only we change this so SMP safe */
1376 return current
->euid
;
1379 asmlinkage
long sys_getgid(void)
1381 /* Only we change this so SMP safe */
1382 return current
->gid
;
1385 asmlinkage
long sys_getegid(void)
1387 /* Only we change this so SMP safe */
1388 return current
->egid
;
1393 static void process_timeout(unsigned long __data
)
1395 wake_up_process((struct task_struct
*)__data
);
1399 * schedule_timeout - sleep until timeout
1400 * @timeout: timeout value in jiffies
1402 * Make the current task sleep until @timeout jiffies have
1403 * elapsed. The routine will return immediately unless
1404 * the current task state has been set (see set_current_state()).
1406 * You can set the task state as follows -
1408 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1409 * pass before the routine returns. The routine will return 0
1411 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1412 * delivered to the current task. In this case the remaining time
1413 * in jiffies will be returned, or 0 if the timer expired in time
1415 * The current task state is guaranteed to be TASK_RUNNING when this
1418 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1419 * the CPU away without a bound on the timeout. In this case the return
1420 * value will be %MAX_SCHEDULE_TIMEOUT.
1422 * In all cases the return value is guaranteed to be non-negative.
1424 fastcall
signed long __sched
schedule_timeout(signed long timeout
)
1426 struct timer_list timer
;
1427 unsigned long expire
;
1431 case MAX_SCHEDULE_TIMEOUT
:
1433 * These two special cases are useful to be comfortable
1434 * in the caller. Nothing more. We could take
1435 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1436 * but I' d like to return a valid offset (>=0) to allow
1437 * the caller to do everything it want with the retval.
1443 * Another bit of PARANOID. Note that the retval will be
1444 * 0 since no piece of kernel is supposed to do a check
1445 * for a negative retval of schedule_timeout() (since it
1446 * should never happens anyway). You just have the printk()
1447 * that will tell you if something is gone wrong and where.
1450 printk(KERN_ERR
"schedule_timeout: wrong timeout "
1451 "value %lx\n", timeout
);
1453 current
->state
= TASK_RUNNING
;
1458 expire
= timeout
+ jiffies
;
1460 setup_timer(&timer
, process_timeout
, (unsigned long)current
);
1461 __mod_timer(&timer
, expire
);
1463 del_singleshot_timer_sync(&timer
);
1465 timeout
= expire
- jiffies
;
1468 return timeout
< 0 ? 0 : timeout
;
1470 EXPORT_SYMBOL(schedule_timeout
);
1473 * We can use __set_current_state() here because schedule_timeout() calls
1474 * schedule() unconditionally.
1476 signed long __sched
schedule_timeout_interruptible(signed long timeout
)
1478 __set_current_state(TASK_INTERRUPTIBLE
);
1479 return schedule_timeout(timeout
);
1481 EXPORT_SYMBOL(schedule_timeout_interruptible
);
1483 signed long __sched
schedule_timeout_uninterruptible(signed long timeout
)
1485 __set_current_state(TASK_UNINTERRUPTIBLE
);
1486 return schedule_timeout(timeout
);
1488 EXPORT_SYMBOL(schedule_timeout_uninterruptible
);
1490 /* Thread ID - the internal kernel "pid" */
1491 asmlinkage
long sys_gettid(void)
1493 return current
->pid
;
1497 * do_sysinfo - fill in sysinfo struct
1498 * @info: pointer to buffer to fill
1500 int do_sysinfo(struct sysinfo
*info
)
1502 unsigned long mem_total
, sav_total
;
1503 unsigned int mem_unit
, bitcount
;
1506 memset(info
, 0, sizeof(struct sysinfo
));
1510 seq
= read_seqbegin(&xtime_lock
);
1513 * This is annoying. The below is the same thing
1514 * posix_get_clock_monotonic() does, but it wants to
1515 * take the lock which we want to cover the loads stuff
1519 getnstimeofday(&tp
);
1520 tp
.tv_sec
+= wall_to_monotonic
.tv_sec
;
1521 tp
.tv_nsec
+= wall_to_monotonic
.tv_nsec
;
1522 if (tp
.tv_nsec
- NSEC_PER_SEC
>= 0) {
1523 tp
.tv_nsec
= tp
.tv_nsec
- NSEC_PER_SEC
;
1526 info
->uptime
= tp
.tv_sec
+ (tp
.tv_nsec
? 1 : 0);
1528 info
->loads
[0] = avenrun
[0] << (SI_LOAD_SHIFT
- FSHIFT
);
1529 info
->loads
[1] = avenrun
[1] << (SI_LOAD_SHIFT
- FSHIFT
);
1530 info
->loads
[2] = avenrun
[2] << (SI_LOAD_SHIFT
- FSHIFT
);
1532 info
->procs
= nr_threads
;
1533 } while (read_seqretry(&xtime_lock
, seq
));
1539 * If the sum of all the available memory (i.e. ram + swap)
1540 * is less than can be stored in a 32 bit unsigned long then
1541 * we can be binary compatible with 2.2.x kernels. If not,
1542 * well, in that case 2.2.x was broken anyways...
1544 * -Erik Andersen <andersee@debian.org>
1547 mem_total
= info
->totalram
+ info
->totalswap
;
1548 if (mem_total
< info
->totalram
|| mem_total
< info
->totalswap
)
1551 mem_unit
= info
->mem_unit
;
1552 while (mem_unit
> 1) {
1555 sav_total
= mem_total
;
1557 if (mem_total
< sav_total
)
1562 * If mem_total did not overflow, multiply all memory values by
1563 * info->mem_unit and set it to 1. This leaves things compatible
1564 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1569 info
->totalram
<<= bitcount
;
1570 info
->freeram
<<= bitcount
;
1571 info
->sharedram
<<= bitcount
;
1572 info
->bufferram
<<= bitcount
;
1573 info
->totalswap
<<= bitcount
;
1574 info
->freeswap
<<= bitcount
;
1575 info
->totalhigh
<<= bitcount
;
1576 info
->freehigh
<<= bitcount
;
1582 asmlinkage
long sys_sysinfo(struct sysinfo __user
*info
)
1588 if (copy_to_user(info
, &val
, sizeof(struct sysinfo
)))
1595 * lockdep: we want to track each per-CPU base as a separate lock-class,
1596 * but timer-bases are kmalloc()-ed, so we need to attach separate
1599 static struct lock_class_key base_lock_keys
[NR_CPUS
];
1601 static int __devinit
init_timers_cpu(int cpu
)
1605 static char __devinitdata tvec_base_done
[NR_CPUS
];
1607 if (!tvec_base_done
[cpu
]) {
1608 static char boot_done
;
1612 * The APs use this path later in boot
1614 base
= kmalloc_node(sizeof(*base
), GFP_KERNEL
,
1618 memset(base
, 0, sizeof(*base
));
1619 per_cpu(tvec_bases
, cpu
) = base
;
1622 * This is for the boot CPU - we use compile-time
1623 * static initialisation because per-cpu memory isn't
1624 * ready yet and because the memory allocators are not
1625 * initialised either.
1628 base
= &boot_tvec_bases
;
1630 tvec_base_done
[cpu
] = 1;
1632 base
= per_cpu(tvec_bases
, cpu
);
1635 spin_lock_init(&base
->lock
);
1636 lockdep_set_class(&base
->lock
, base_lock_keys
+ cpu
);
1638 for (j
= 0; j
< TVN_SIZE
; j
++) {
1639 INIT_LIST_HEAD(base
->tv5
.vec
+ j
);
1640 INIT_LIST_HEAD(base
->tv4
.vec
+ j
);
1641 INIT_LIST_HEAD(base
->tv3
.vec
+ j
);
1642 INIT_LIST_HEAD(base
->tv2
.vec
+ j
);
1644 for (j
= 0; j
< TVR_SIZE
; j
++)
1645 INIT_LIST_HEAD(base
->tv1
.vec
+ j
);
1647 base
->timer_jiffies
= jiffies
;
1651 #ifdef CONFIG_HOTPLUG_CPU
1652 static void migrate_timer_list(tvec_base_t
*new_base
, struct list_head
*head
)
1654 struct timer_list
*timer
;
1656 while (!list_empty(head
)) {
1657 timer
= list_entry(head
->next
, struct timer_list
, entry
);
1658 detach_timer(timer
, 0);
1659 timer
->base
= new_base
;
1660 internal_add_timer(new_base
, timer
);
1664 static void __devinit
migrate_timers(int cpu
)
1666 tvec_base_t
*old_base
;
1667 tvec_base_t
*new_base
;
1670 BUG_ON(cpu_online(cpu
));
1671 old_base
= per_cpu(tvec_bases
, cpu
);
1672 new_base
= get_cpu_var(tvec_bases
);
1674 local_irq_disable();
1675 double_spin_lock(&new_base
->lock
, &old_base
->lock
,
1676 smp_processor_id() < cpu
);
1678 BUG_ON(old_base
->running_timer
);
1680 for (i
= 0; i
< TVR_SIZE
; i
++)
1681 migrate_timer_list(new_base
, old_base
->tv1
.vec
+ i
);
1682 for (i
= 0; i
< TVN_SIZE
; i
++) {
1683 migrate_timer_list(new_base
, old_base
->tv2
.vec
+ i
);
1684 migrate_timer_list(new_base
, old_base
->tv3
.vec
+ i
);
1685 migrate_timer_list(new_base
, old_base
->tv4
.vec
+ i
);
1686 migrate_timer_list(new_base
, old_base
->tv5
.vec
+ i
);
1689 double_spin_unlock(&new_base
->lock
, &old_base
->lock
,
1690 smp_processor_id() < cpu
);
1692 put_cpu_var(tvec_bases
);
1694 #endif /* CONFIG_HOTPLUG_CPU */
1696 static int __cpuinit
timer_cpu_notify(struct notifier_block
*self
,
1697 unsigned long action
, void *hcpu
)
1699 long cpu
= (long)hcpu
;
1701 case CPU_UP_PREPARE
:
1702 if (init_timers_cpu(cpu
) < 0)
1705 #ifdef CONFIG_HOTPLUG_CPU
1707 migrate_timers(cpu
);
1716 static struct notifier_block __cpuinitdata timers_nb
= {
1717 .notifier_call
= timer_cpu_notify
,
1721 void __init
init_timers(void)
1723 int err
= timer_cpu_notify(&timers_nb
, (unsigned long)CPU_UP_PREPARE
,
1724 (void *)(long)smp_processor_id());
1728 BUG_ON(err
== NOTIFY_BAD
);
1729 register_cpu_notifier(&timers_nb
);
1730 open_softirq(TIMER_SOFTIRQ
, run_timer_softirq
, NULL
);
1733 #ifdef CONFIG_TIME_INTERPOLATION
1735 struct time_interpolator
*time_interpolator __read_mostly
;
1736 static struct time_interpolator
*time_interpolator_list __read_mostly
;
1737 static DEFINE_SPINLOCK(time_interpolator_lock
);
1739 static inline cycles_t
time_interpolator_get_cycles(unsigned int src
)
1741 unsigned long (*x
)(void);
1745 case TIME_SOURCE_FUNCTION
:
1746 x
= time_interpolator
->addr
;
1749 case TIME_SOURCE_MMIO64
:
1750 return readq_relaxed((void __iomem
*)time_interpolator
->addr
);
1752 case TIME_SOURCE_MMIO32
:
1753 return readl_relaxed((void __iomem
*)time_interpolator
->addr
);
1755 default: return get_cycles();
1759 static inline u64
time_interpolator_get_counter(int writelock
)
1761 unsigned int src
= time_interpolator
->source
;
1763 if (time_interpolator
->jitter
)
1769 lcycle
= time_interpolator
->last_cycle
;
1770 now
= time_interpolator_get_cycles(src
);
1771 if (lcycle
&& time_after(lcycle
, now
))
1774 /* When holding the xtime write lock, there's no need
1775 * to add the overhead of the cmpxchg. Readers are
1776 * force to retry until the write lock is released.
1779 time_interpolator
->last_cycle
= now
;
1782 /* Keep track of the last timer value returned. The use of cmpxchg here
1783 * will cause contention in an SMP environment.
1785 } while (unlikely(cmpxchg(&time_interpolator
->last_cycle
, lcycle
, now
) != lcycle
));
1789 return time_interpolator_get_cycles(src
);
1792 void time_interpolator_reset(void)
1794 time_interpolator
->offset
= 0;
1795 time_interpolator
->last_counter
= time_interpolator_get_counter(1);
1798 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1800 unsigned long time_interpolator_get_offset(void)
1802 /* If we do not have a time interpolator set up then just return zero */
1803 if (!time_interpolator
)
1806 return time_interpolator
->offset
+
1807 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator
);
1810 #define INTERPOLATOR_ADJUST 65536
1811 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1813 void time_interpolator_update(long delta_nsec
)
1816 unsigned long offset
;
1818 /* If there is no time interpolator set up then do nothing */
1819 if (!time_interpolator
)
1823 * The interpolator compensates for late ticks by accumulating the late
1824 * time in time_interpolator->offset. A tick earlier than expected will
1825 * lead to a reset of the offset and a corresponding jump of the clock
1826 * forward. Again this only works if the interpolator clock is running
1827 * slightly slower than the regular clock and the tuning logic insures
1831 counter
= time_interpolator_get_counter(1);
1832 offset
= time_interpolator
->offset
+
1833 GET_TI_NSECS(counter
, time_interpolator
);
1835 if (delta_nsec
< 0 || (unsigned long) delta_nsec
< offset
)
1836 time_interpolator
->offset
= offset
- delta_nsec
;
1838 time_interpolator
->skips
++;
1839 time_interpolator
->ns_skipped
+= delta_nsec
- offset
;
1840 time_interpolator
->offset
= 0;
1842 time_interpolator
->last_counter
= counter
;
1844 /* Tuning logic for time interpolator invoked every minute or so.
1845 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1846 * Increase interpolator clock speed if we skip too much time.
1848 if (jiffies
% INTERPOLATOR_ADJUST
== 0)
1850 if (time_interpolator
->skips
== 0 && time_interpolator
->offset
> tick_nsec
)
1851 time_interpolator
->nsec_per_cyc
--;
1852 if (time_interpolator
->ns_skipped
> INTERPOLATOR_MAX_SKIP
&& time_interpolator
->offset
== 0)
1853 time_interpolator
->nsec_per_cyc
++;
1854 time_interpolator
->skips
= 0;
1855 time_interpolator
->ns_skipped
= 0;
1860 is_better_time_interpolator(struct time_interpolator
*new)
1862 if (!time_interpolator
)
1864 return new->frequency
> 2*time_interpolator
->frequency
||
1865 (unsigned long)new->drift
< (unsigned long)time_interpolator
->drift
;
1869 register_time_interpolator(struct time_interpolator
*ti
)
1871 unsigned long flags
;
1874 BUG_ON(ti
->frequency
== 0 || ti
->mask
== 0);
1876 ti
->nsec_per_cyc
= ((u64
)NSEC_PER_SEC
<< ti
->shift
) / ti
->frequency
;
1877 spin_lock(&time_interpolator_lock
);
1878 write_seqlock_irqsave(&xtime_lock
, flags
);
1879 if (is_better_time_interpolator(ti
)) {
1880 time_interpolator
= ti
;
1881 time_interpolator_reset();
1883 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1885 ti
->next
= time_interpolator_list
;
1886 time_interpolator_list
= ti
;
1887 spin_unlock(&time_interpolator_lock
);
1891 unregister_time_interpolator(struct time_interpolator
*ti
)
1893 struct time_interpolator
*curr
, **prev
;
1894 unsigned long flags
;
1896 spin_lock(&time_interpolator_lock
);
1897 prev
= &time_interpolator_list
;
1898 for (curr
= *prev
; curr
; curr
= curr
->next
) {
1906 write_seqlock_irqsave(&xtime_lock
, flags
);
1907 if (ti
== time_interpolator
) {
1908 /* we lost the best time-interpolator: */
1909 time_interpolator
= NULL
;
1910 /* find the next-best interpolator */
1911 for (curr
= time_interpolator_list
; curr
; curr
= curr
->next
)
1912 if (is_better_time_interpolator(curr
))
1913 time_interpolator
= curr
;
1914 time_interpolator_reset();
1916 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1917 spin_unlock(&time_interpolator_lock
);
1919 #endif /* CONFIG_TIME_INTERPOLATION */
1922 * msleep - sleep safely even with waitqueue interruptions
1923 * @msecs: Time in milliseconds to sleep for
1925 void msleep(unsigned int msecs
)
1927 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1930 timeout
= schedule_timeout_uninterruptible(timeout
);
1933 EXPORT_SYMBOL(msleep
);
1936 * msleep_interruptible - sleep waiting for signals
1937 * @msecs: Time in milliseconds to sleep for
1939 unsigned long msleep_interruptible(unsigned int msecs
)
1941 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1943 while (timeout
&& !signal_pending(current
))
1944 timeout
= schedule_timeout_interruptible(timeout
);
1945 return jiffies_to_msecs(timeout
);
1948 EXPORT_SYMBOL(msleep_interruptible
);