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>
38 #include <asm/uaccess.h>
39 #include <asm/unistd.h>
40 #include <asm/div64.h>
41 #include <asm/timex.h>
44 u64 jiffies_64 __cacheline_aligned_in_smp
= INITIAL_JIFFIES
;
46 EXPORT_SYMBOL(jiffies_64
);
49 * per-CPU timer vector definitions:
51 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
52 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
53 #define TVN_SIZE (1 << TVN_BITS)
54 #define TVR_SIZE (1 << TVR_BITS)
55 #define TVN_MASK (TVN_SIZE - 1)
56 #define TVR_MASK (TVR_SIZE - 1)
58 typedef struct tvec_s
{
59 struct list_head vec
[TVN_SIZE
];
62 typedef struct tvec_root_s
{
63 struct list_head vec
[TVR_SIZE
];
66 struct tvec_t_base_s
{
68 struct timer_list
*running_timer
;
69 unsigned long timer_jiffies
;
75 } ____cacheline_aligned_in_smp
;
77 typedef struct tvec_t_base_s tvec_base_t
;
79 tvec_base_t boot_tvec_bases
;
80 EXPORT_SYMBOL(boot_tvec_bases
);
81 static DEFINE_PER_CPU(tvec_base_t
*, tvec_bases
) = &boot_tvec_bases
;
84 * __round_jiffies - function to round jiffies to a full second
85 * @j: the time in (absolute) jiffies that should be rounded
86 * @cpu: the processor number on which the timeout will happen
88 * __round_jiffies() rounds an absolute time in the future (in jiffies)
89 * up or down to (approximately) full seconds. This is useful for timers
90 * for which the exact time they fire does not matter too much, as long as
91 * they fire approximately every X seconds.
93 * By rounding these timers to whole seconds, all such timers will fire
94 * at the same time, rather than at various times spread out. The goal
95 * of this is to have the CPU wake up less, which saves power.
97 * The exact rounding is skewed for each processor to avoid all
98 * processors firing at the exact same time, which could lead
99 * to lock contention or spurious cache line bouncing.
101 * The return value is the rounded version of the @j parameter.
103 unsigned long __round_jiffies(unsigned long j
, int cpu
)
106 unsigned long original
= j
;
109 * We don't want all cpus firing their timers at once hitting the
110 * same lock or cachelines, so we skew each extra cpu with an extra
111 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
113 * The skew is done by adding 3*cpunr, then round, then subtract this
114 * extra offset again.
121 * If the target jiffie is just after a whole second (which can happen
122 * due to delays of the timer irq, long irq off times etc etc) then
123 * we should round down to the whole second, not up. Use 1/4th second
124 * as cutoff for this rounding as an extreme upper bound for this.
126 if (rem
< HZ
/4) /* round down */
131 /* now that we have rounded, subtract the extra skew again */
134 if (j
<= jiffies
) /* rounding ate our timeout entirely; */
138 EXPORT_SYMBOL_GPL(__round_jiffies
);
141 * __round_jiffies_relative - function to round jiffies to a full second
142 * @j: the time in (relative) jiffies that should be rounded
143 * @cpu: the processor number on which the timeout will happen
145 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
146 * up or down to (approximately) full seconds. This is useful for timers
147 * for which the exact time they fire does not matter too much, as long as
148 * they fire approximately every X seconds.
150 * By rounding these timers to whole seconds, all such timers will fire
151 * at the same time, rather than at various times spread out. The goal
152 * of this is to have the CPU wake up less, which saves power.
154 * The exact rounding is skewed for each processor to avoid all
155 * processors firing at the exact same time, which could lead
156 * to lock contention or spurious cache line bouncing.
158 * The return value is the rounded version of the @j parameter.
160 unsigned long __round_jiffies_relative(unsigned long j
, int cpu
)
163 * In theory the following code can skip a jiffy in case jiffies
164 * increments right between the addition and the later subtraction.
165 * However since the entire point of this function is to use approximate
166 * timeouts, it's entirely ok to not handle that.
168 return __round_jiffies(j
+ jiffies
, cpu
) - jiffies
;
170 EXPORT_SYMBOL_GPL(__round_jiffies_relative
);
173 * round_jiffies - function to round jiffies to a full second
174 * @j: the time in (absolute) jiffies that should be rounded
176 * round_jiffies() rounds an absolute time in the future (in jiffies)
177 * up or down to (approximately) full seconds. This is useful for timers
178 * for which the exact time they fire does not matter too much, as long as
179 * they fire approximately every X seconds.
181 * By rounding these timers to whole seconds, all such timers will fire
182 * at the same time, rather than at various times spread out. The goal
183 * of this is to have the CPU wake up less, which saves power.
185 * The return value is the rounded version of the @j parameter.
187 unsigned long round_jiffies(unsigned long j
)
189 return __round_jiffies(j
, raw_smp_processor_id());
191 EXPORT_SYMBOL_GPL(round_jiffies
);
194 * round_jiffies_relative - function to round jiffies to a full second
195 * @j: the time in (relative) jiffies that should be rounded
197 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
198 * up or down to (approximately) full seconds. This is useful for timers
199 * for which the exact time they fire does not matter too much, as long as
200 * they fire approximately every X seconds.
202 * By rounding these timers to whole seconds, all such timers will fire
203 * at the same time, rather than at various times spread out. The goal
204 * of this is to have the CPU wake up less, which saves power.
206 * The return value is the rounded version of the @j parameter.
208 unsigned long round_jiffies_relative(unsigned long j
)
210 return __round_jiffies_relative(j
, raw_smp_processor_id());
212 EXPORT_SYMBOL_GPL(round_jiffies_relative
);
215 static inline void set_running_timer(tvec_base_t
*base
,
216 struct timer_list
*timer
)
219 base
->running_timer
= timer
;
223 static void internal_add_timer(tvec_base_t
*base
, struct timer_list
*timer
)
225 unsigned long expires
= timer
->expires
;
226 unsigned long idx
= expires
- base
->timer_jiffies
;
227 struct list_head
*vec
;
229 if (idx
< TVR_SIZE
) {
230 int i
= expires
& TVR_MASK
;
231 vec
= base
->tv1
.vec
+ i
;
232 } else if (idx
< 1 << (TVR_BITS
+ TVN_BITS
)) {
233 int i
= (expires
>> TVR_BITS
) & TVN_MASK
;
234 vec
= base
->tv2
.vec
+ i
;
235 } else if (idx
< 1 << (TVR_BITS
+ 2 * TVN_BITS
)) {
236 int i
= (expires
>> (TVR_BITS
+ TVN_BITS
)) & TVN_MASK
;
237 vec
= base
->tv3
.vec
+ i
;
238 } else if (idx
< 1 << (TVR_BITS
+ 3 * TVN_BITS
)) {
239 int i
= (expires
>> (TVR_BITS
+ 2 * TVN_BITS
)) & TVN_MASK
;
240 vec
= base
->tv4
.vec
+ i
;
241 } else if ((signed long) idx
< 0) {
243 * Can happen if you add a timer with expires == jiffies,
244 * or you set a timer to go off in the past
246 vec
= base
->tv1
.vec
+ (base
->timer_jiffies
& TVR_MASK
);
249 /* If the timeout is larger than 0xffffffff on 64-bit
250 * architectures then we use the maximum timeout:
252 if (idx
> 0xffffffffUL
) {
254 expires
= idx
+ base
->timer_jiffies
;
256 i
= (expires
>> (TVR_BITS
+ 3 * TVN_BITS
)) & TVN_MASK
;
257 vec
= base
->tv5
.vec
+ i
;
262 list_add_tail(&timer
->entry
, vec
);
266 * init_timer - initialize a timer.
267 * @timer: the timer to be initialized
269 * init_timer() must be done to a timer prior calling *any* of the
270 * other timer functions.
272 void fastcall
init_timer(struct timer_list
*timer
)
274 timer
->entry
.next
= NULL
;
275 timer
->base
= __raw_get_cpu_var(tvec_bases
);
277 EXPORT_SYMBOL(init_timer
);
279 static inline void detach_timer(struct timer_list
*timer
,
282 struct list_head
*entry
= &timer
->entry
;
284 __list_del(entry
->prev
, entry
->next
);
287 entry
->prev
= LIST_POISON2
;
291 * We are using hashed locking: holding per_cpu(tvec_bases).lock
292 * means that all timers which are tied to this base via timer->base are
293 * locked, and the base itself is locked too.
295 * So __run_timers/migrate_timers can safely modify all timers which could
296 * be found on ->tvX lists.
298 * When the timer's base is locked, and the timer removed from list, it is
299 * possible to set timer->base = NULL and drop the lock: the timer remains
302 static tvec_base_t
*lock_timer_base(struct timer_list
*timer
,
303 unsigned long *flags
)
304 __acquires(timer
->base
->lock
)
310 if (likely(base
!= NULL
)) {
311 spin_lock_irqsave(&base
->lock
, *flags
);
312 if (likely(base
== timer
->base
))
314 /* The timer has migrated to another CPU */
315 spin_unlock_irqrestore(&base
->lock
, *flags
);
321 int __mod_timer(struct timer_list
*timer
, unsigned long expires
)
323 tvec_base_t
*base
, *new_base
;
327 BUG_ON(!timer
->function
);
329 base
= lock_timer_base(timer
, &flags
);
331 if (timer_pending(timer
)) {
332 detach_timer(timer
, 0);
336 new_base
= __get_cpu_var(tvec_bases
);
338 if (base
!= new_base
) {
340 * We are trying to schedule the timer on the local CPU.
341 * However we can't change timer's base while it is running,
342 * otherwise del_timer_sync() can't detect that the timer's
343 * handler yet has not finished. This also guarantees that
344 * the timer is serialized wrt itself.
346 if (likely(base
->running_timer
!= timer
)) {
347 /* See the comment in lock_timer_base() */
349 spin_unlock(&base
->lock
);
351 spin_lock(&base
->lock
);
356 timer
->expires
= expires
;
357 internal_add_timer(base
, timer
);
358 spin_unlock_irqrestore(&base
->lock
, flags
);
363 EXPORT_SYMBOL(__mod_timer
);
366 * add_timer_on - start a timer on a particular CPU
367 * @timer: the timer to be added
368 * @cpu: the CPU to start it on
370 * This is not very scalable on SMP. Double adds are not possible.
372 void add_timer_on(struct timer_list
*timer
, int cpu
)
374 tvec_base_t
*base
= per_cpu(tvec_bases
, cpu
);
377 BUG_ON(timer_pending(timer
) || !timer
->function
);
378 spin_lock_irqsave(&base
->lock
, flags
);
380 internal_add_timer(base
, timer
);
381 spin_unlock_irqrestore(&base
->lock
, flags
);
386 * mod_timer - modify a timer's timeout
387 * @timer: the timer to be modified
388 * @expires: new timeout in jiffies
390 * mod_timer() is a more efficient way to update the expire field of an
391 * active timer (if the timer is inactive it will be activated)
393 * mod_timer(timer, expires) is equivalent to:
395 * del_timer(timer); timer->expires = expires; add_timer(timer);
397 * Note that if there are multiple unserialized concurrent users of the
398 * same timer, then mod_timer() is the only safe way to modify the timeout,
399 * since add_timer() cannot modify an already running timer.
401 * The function returns whether it has modified a pending timer or not.
402 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
403 * active timer returns 1.)
405 int mod_timer(struct timer_list
*timer
, unsigned long expires
)
407 BUG_ON(!timer
->function
);
410 * This is a common optimization triggered by the
411 * networking code - if the timer is re-modified
412 * to be the same thing then just return:
414 if (timer
->expires
== expires
&& timer_pending(timer
))
417 return __mod_timer(timer
, expires
);
420 EXPORT_SYMBOL(mod_timer
);
423 * del_timer - deactive a timer.
424 * @timer: the timer to be deactivated
426 * del_timer() deactivates a timer - this works on both active and inactive
429 * The function returns whether it has deactivated a pending timer or not.
430 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
431 * active timer returns 1.)
433 int del_timer(struct timer_list
*timer
)
439 if (timer_pending(timer
)) {
440 base
= lock_timer_base(timer
, &flags
);
441 if (timer_pending(timer
)) {
442 detach_timer(timer
, 1);
445 spin_unlock_irqrestore(&base
->lock
, flags
);
451 EXPORT_SYMBOL(del_timer
);
455 * try_to_del_timer_sync - Try to deactivate a timer
456 * @timer: timer do del
458 * This function tries to deactivate a timer. Upon successful (ret >= 0)
459 * exit the timer is not queued and the handler is not running on any CPU.
461 * It must not be called from interrupt contexts.
463 int try_to_del_timer_sync(struct timer_list
*timer
)
469 base
= lock_timer_base(timer
, &flags
);
471 if (base
->running_timer
== timer
)
475 if (timer_pending(timer
)) {
476 detach_timer(timer
, 1);
480 spin_unlock_irqrestore(&base
->lock
, flags
);
486 * del_timer_sync - deactivate a timer and wait for the handler to finish.
487 * @timer: the timer to be deactivated
489 * This function only differs from del_timer() on SMP: besides deactivating
490 * the timer it also makes sure the handler has finished executing on other
493 * Synchronization rules: Callers must prevent restarting of the timer,
494 * otherwise this function is meaningless. It must not be called from
495 * interrupt contexts. The caller must not hold locks which would prevent
496 * completion of the timer's handler. The timer's handler must not call
497 * add_timer_on(). Upon exit the timer is not queued and the handler is
498 * not running on any CPU.
500 * The function returns whether it has deactivated a pending timer or not.
502 int del_timer_sync(struct timer_list
*timer
)
505 int ret
= try_to_del_timer_sync(timer
);
512 EXPORT_SYMBOL(del_timer_sync
);
515 static int cascade(tvec_base_t
*base
, tvec_t
*tv
, int index
)
517 /* cascade all the timers from tv up one level */
518 struct timer_list
*timer
, *tmp
;
519 struct list_head tv_list
;
521 list_replace_init(tv
->vec
+ index
, &tv_list
);
524 * We are removing _all_ timers from the list, so we
525 * don't have to detach them individually.
527 list_for_each_entry_safe(timer
, tmp
, &tv_list
, entry
) {
528 BUG_ON(timer
->base
!= base
);
529 internal_add_timer(base
, timer
);
535 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
538 * __run_timers - run all expired timers (if any) on this CPU.
539 * @base: the timer vector to be processed.
541 * This function cascades all vectors and executes all expired timer
544 static inline void __run_timers(tvec_base_t
*base
)
546 struct timer_list
*timer
;
548 spin_lock_irq(&base
->lock
);
549 while (time_after_eq(jiffies
, base
->timer_jiffies
)) {
550 struct list_head work_list
;
551 struct list_head
*head
= &work_list
;
552 int index
= base
->timer_jiffies
& TVR_MASK
;
558 (!cascade(base
, &base
->tv2
, INDEX(0))) &&
559 (!cascade(base
, &base
->tv3
, INDEX(1))) &&
560 !cascade(base
, &base
->tv4
, INDEX(2)))
561 cascade(base
, &base
->tv5
, INDEX(3));
562 ++base
->timer_jiffies
;
563 list_replace_init(base
->tv1
.vec
+ index
, &work_list
);
564 while (!list_empty(head
)) {
565 void (*fn
)(unsigned long);
568 timer
= list_entry(head
->next
,struct timer_list
,entry
);
569 fn
= timer
->function
;
572 set_running_timer(base
, timer
);
573 detach_timer(timer
, 1);
574 spin_unlock_irq(&base
->lock
);
576 int preempt_count
= preempt_count();
578 if (preempt_count
!= preempt_count()) {
579 printk(KERN_WARNING
"huh, entered %p "
580 "with preempt_count %08x, exited"
587 spin_lock_irq(&base
->lock
);
590 set_running_timer(base
, NULL
);
591 spin_unlock_irq(&base
->lock
);
594 #ifdef CONFIG_NO_IDLE_HZ
596 * Find out when the next timer event is due to happen. This
597 * is used on S/390 to stop all activity when a cpus is idle.
598 * This functions needs to be called disabled.
600 static unsigned long __next_timer_interrupt(tvec_base_t
*base
)
602 unsigned long timer_jiffies
= base
->timer_jiffies
;
603 unsigned long expires
= timer_jiffies
+ (LONG_MAX
>> 1);
604 int index
, slot
, array
, found
= 0;
605 struct timer_list
*nte
;
608 /* Look for timer events in tv1. */
609 index
= slot
= timer_jiffies
& TVR_MASK
;
611 list_for_each_entry(nte
, base
->tv1
.vec
+ slot
, entry
) {
613 expires
= nte
->expires
;
614 /* Look at the cascade bucket(s)? */
615 if (!index
|| slot
< index
)
619 slot
= (slot
+ 1) & TVR_MASK
;
620 } while (slot
!= index
);
623 /* Calculate the next cascade event */
625 timer_jiffies
+= TVR_SIZE
- index
;
626 timer_jiffies
>>= TVR_BITS
;
629 varray
[0] = &base
->tv2
;
630 varray
[1] = &base
->tv3
;
631 varray
[2] = &base
->tv4
;
632 varray
[3] = &base
->tv5
;
634 for (array
= 0; array
< 4; array
++) {
635 tvec_t
*varp
= varray
[array
];
637 index
= slot
= timer_jiffies
& TVN_MASK
;
639 list_for_each_entry(nte
, varp
->vec
+ slot
, entry
) {
641 if (time_before(nte
->expires
, expires
))
642 expires
= nte
->expires
;
645 * Do we still search for the first timer or are
646 * we looking up the cascade buckets ?
649 /* Look at the cascade bucket(s)? */
650 if (!index
|| slot
< index
)
654 slot
= (slot
+ 1) & TVN_MASK
;
655 } while (slot
!= index
);
658 timer_jiffies
+= TVN_SIZE
- index
;
659 timer_jiffies
>>= TVN_BITS
;
665 * Check, if the next hrtimer event is before the next timer wheel
668 static unsigned long cmp_next_hrtimer_event(unsigned long now
,
669 unsigned long expires
)
671 ktime_t hr_delta
= hrtimer_get_next_event();
672 struct timespec tsdelta
;
674 if (hr_delta
.tv64
== KTIME_MAX
)
677 if (hr_delta
.tv64
<= TICK_NSEC
)
680 tsdelta
= ktime_to_timespec(hr_delta
);
681 now
+= timespec_to_jiffies(&tsdelta
);
682 if (time_before(now
, expires
))
688 * next_timer_interrupt - return the jiffy of the next pending timer
690 unsigned long next_timer_interrupt(void)
692 tvec_base_t
*base
= __get_cpu_var(tvec_bases
);
693 unsigned long expires
, now
= jiffies
;
695 spin_lock(&base
->lock
);
696 expires
= __next_timer_interrupt(base
);
697 spin_unlock(&base
->lock
);
699 if (time_before_eq(expires
, now
))
702 return cmp_next_hrtimer_event(now
, expires
);
706 /******************************************************************/
710 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
711 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
712 * at zero at system boot time, so wall_to_monotonic will be negative,
713 * however, we will ALWAYS keep the tv_nsec part positive so we can use
714 * the usual normalization.
716 struct timespec xtime
__attribute__ ((aligned (16)));
717 struct timespec wall_to_monotonic
__attribute__ ((aligned (16)));
719 EXPORT_SYMBOL(xtime
);
722 /* XXX - all of this timekeeping code should be later moved to time.c */
723 #include <linux/clocksource.h>
724 static struct clocksource
*clock
; /* pointer to current clocksource */
726 #ifdef CONFIG_GENERIC_TIME
728 * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook
730 * private function, must hold xtime_lock lock when being
731 * called. Returns the number of nanoseconds since the
732 * last call to update_wall_time() (adjusted by NTP scaling)
734 static inline s64
__get_nsec_offset(void)
736 cycle_t cycle_now
, cycle_delta
;
739 /* read clocksource: */
740 cycle_now
= clocksource_read(clock
);
742 /* calculate the delta since the last update_wall_time: */
743 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
745 /* convert to nanoseconds: */
746 ns_offset
= cyc2ns(clock
, cycle_delta
);
752 * __get_realtime_clock_ts - Returns the time of day in a timespec
753 * @ts: pointer to the timespec to be set
755 * Returns the time of day in a timespec. Used by
756 * do_gettimeofday() and get_realtime_clock_ts().
758 static inline void __get_realtime_clock_ts(struct timespec
*ts
)
764 seq
= read_seqbegin(&xtime_lock
);
767 nsecs
= __get_nsec_offset();
769 } while (read_seqretry(&xtime_lock
, seq
));
771 timespec_add_ns(ts
, nsecs
);
775 * getnstimeofday - Returns the time of day in a timespec
776 * @ts: pointer to the timespec to be set
778 * Returns the time of day in a timespec.
780 void getnstimeofday(struct timespec
*ts
)
782 __get_realtime_clock_ts(ts
);
785 EXPORT_SYMBOL(getnstimeofday
);
788 * do_gettimeofday - Returns the time of day in a timeval
789 * @tv: pointer to the timeval to be set
791 * NOTE: Users should be converted to using get_realtime_clock_ts()
793 void do_gettimeofday(struct timeval
*tv
)
797 __get_realtime_clock_ts(&now
);
798 tv
->tv_sec
= now
.tv_sec
;
799 tv
->tv_usec
= now
.tv_nsec
/1000;
802 EXPORT_SYMBOL(do_gettimeofday
);
804 * do_settimeofday - Sets the time of day
805 * @tv: pointer to the timespec variable containing the new time
807 * Sets the time of day to the new time and update NTP and notify hrtimers
809 int do_settimeofday(struct timespec
*tv
)
812 time_t wtm_sec
, sec
= tv
->tv_sec
;
813 long wtm_nsec
, nsec
= tv
->tv_nsec
;
815 if ((unsigned long)tv
->tv_nsec
>= NSEC_PER_SEC
)
818 write_seqlock_irqsave(&xtime_lock
, flags
);
820 nsec
-= __get_nsec_offset();
822 wtm_sec
= wall_to_monotonic
.tv_sec
+ (xtime
.tv_sec
- sec
);
823 wtm_nsec
= wall_to_monotonic
.tv_nsec
+ (xtime
.tv_nsec
- nsec
);
825 set_normalized_timespec(&xtime
, sec
, nsec
);
826 set_normalized_timespec(&wall_to_monotonic
, wtm_sec
, wtm_nsec
);
831 write_sequnlock_irqrestore(&xtime_lock
, flags
);
833 /* signal hrtimers about time change */
839 EXPORT_SYMBOL(do_settimeofday
);
842 * change_clocksource - Swaps clocksources if a new one is available
844 * Accumulates current time interval and initializes new clocksource
846 static void change_clocksource(void)
848 struct clocksource
*new;
852 new = clocksource_get_next();
857 now
= clocksource_read(new);
858 nsec
= __get_nsec_offset();
859 timespec_add_ns(&xtime
, nsec
);
862 clock
->cycle_last
= now
;
865 clock
->xtime_nsec
= 0;
866 clocksource_calculate_interval(clock
, NTP_INTERVAL_LENGTH
);
868 printk(KERN_INFO
"Time: %s clocksource has been installed.\n",
872 static inline void change_clocksource(void) { }
876 * timeofday_is_continuous - check to see if timekeeping is free running
878 int timekeeping_is_continuous(void)
884 seq
= read_seqbegin(&xtime_lock
);
886 ret
= clock
->flags
& CLOCK_SOURCE_VALID_FOR_HRES
;
888 } while (read_seqretry(&xtime_lock
, seq
));
894 * read_persistent_clock - Return time in seconds from the persistent clock.
896 * Weak dummy function for arches that do not yet support it.
897 * Returns seconds from epoch using the battery backed persistent clock.
898 * Returns zero if unsupported.
900 * XXX - Do be sure to remove it once all arches implement it.
902 unsigned long __attribute__((weak
)) read_persistent_clock(void)
908 * timekeeping_init - Initializes the clocksource and common timekeeping values
910 void __init
timekeeping_init(void)
913 unsigned long sec
= read_persistent_clock();
915 write_seqlock_irqsave(&xtime_lock
, flags
);
919 clock
= clocksource_get_next();
920 clocksource_calculate_interval(clock
, NTP_INTERVAL_LENGTH
);
921 clock
->cycle_last
= clocksource_read(clock
);
925 set_normalized_timespec(&wall_to_monotonic
,
926 -xtime
.tv_sec
, -xtime
.tv_nsec
);
928 write_sequnlock_irqrestore(&xtime_lock
, flags
);
932 /* flag for if timekeeping is suspended */
933 static int timekeeping_suspended
;
934 /* time in seconds when suspend began */
935 static unsigned long timekeeping_suspend_time
;
938 * timekeeping_resume - Resumes the generic timekeeping subsystem.
941 * This is for the generic clocksource timekeeping.
942 * xtime/wall_to_monotonic/jiffies/etc are
943 * still managed by arch specific suspend/resume code.
945 static int timekeeping_resume(struct sys_device
*dev
)
948 unsigned long now
= read_persistent_clock();
950 write_seqlock_irqsave(&xtime_lock
, flags
);
952 if (now
&& (now
> timekeeping_suspend_time
)) {
953 unsigned long sleep_length
= now
- timekeeping_suspend_time
;
955 xtime
.tv_sec
+= sleep_length
;
956 wall_to_monotonic
.tv_sec
-= sleep_length
;
958 /* re-base the last cycle value */
959 clock
->cycle_last
= clocksource_read(clock
);
961 timekeeping_suspended
= 0;
962 write_sequnlock_irqrestore(&xtime_lock
, flags
);
964 touch_softlockup_watchdog();
965 hrtimer_notify_resume();
970 static int timekeeping_suspend(struct sys_device
*dev
, pm_message_t state
)
974 write_seqlock_irqsave(&xtime_lock
, flags
);
975 timekeeping_suspended
= 1;
976 timekeeping_suspend_time
= read_persistent_clock();
977 write_sequnlock_irqrestore(&xtime_lock
, flags
);
981 /* sysfs resume/suspend bits for timekeeping */
982 static struct sysdev_class timekeeping_sysclass
= {
983 .resume
= timekeeping_resume
,
984 .suspend
= timekeeping_suspend
,
985 set_kset_name("timekeeping"),
988 static struct sys_device device_timer
= {
990 .cls
= &timekeeping_sysclass
,
993 static int __init
timekeeping_init_device(void)
995 int error
= sysdev_class_register(&timekeeping_sysclass
);
997 error
= sysdev_register(&device_timer
);
1001 device_initcall(timekeeping_init_device
);
1004 * If the error is already larger, we look ahead even further
1005 * to compensate for late or lost adjustments.
1007 static __always_inline
int clocksource_bigadjust(s64 error
, s64
*interval
,
1011 u32 look_ahead
, adj
;
1015 * Use the current error value to determine how much to look ahead.
1016 * The larger the error the slower we adjust for it to avoid problems
1017 * with losing too many ticks, otherwise we would overadjust and
1018 * produce an even larger error. The smaller the adjustment the
1019 * faster we try to adjust for it, as lost ticks can do less harm
1020 * here. This is tuned so that an error of about 1 msec is adusted
1021 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1023 error2
= clock
->error
>> (TICK_LENGTH_SHIFT
+ 22 - 2 * SHIFT_HZ
);
1024 error2
= abs(error2
);
1025 for (look_ahead
= 0; error2
> 0; look_ahead
++)
1029 * Now calculate the error in (1 << look_ahead) ticks, but first
1030 * remove the single look ahead already included in the error.
1032 tick_error
= current_tick_length() >>
1033 (TICK_LENGTH_SHIFT
- clock
->shift
+ 1);
1034 tick_error
-= clock
->xtime_interval
>> 1;
1035 error
= ((error
- tick_error
) >> look_ahead
) + tick_error
;
1037 /* Finally calculate the adjustment shift value. */
1042 *interval
= -*interval
;
1046 for (adj
= 0; error
> i
; adj
++)
1055 * Adjust the multiplier to reduce the error value,
1056 * this is optimized for the most common adjustments of -1,0,1,
1057 * for other values we can do a bit more work.
1059 static void clocksource_adjust(struct clocksource
*clock
, s64 offset
)
1061 s64 error
, interval
= clock
->cycle_interval
;
1064 error
= clock
->error
>> (TICK_LENGTH_SHIFT
- clock
->shift
- 1);
1065 if (error
> interval
) {
1067 if (likely(error
<= interval
))
1070 adj
= clocksource_bigadjust(error
, &interval
, &offset
);
1071 } else if (error
< -interval
) {
1073 if (likely(error
>= -interval
)) {
1075 interval
= -interval
;
1078 adj
= clocksource_bigadjust(error
, &interval
, &offset
);
1083 clock
->xtime_interval
+= interval
;
1084 clock
->xtime_nsec
-= offset
;
1085 clock
->error
-= (interval
- offset
) <<
1086 (TICK_LENGTH_SHIFT
- clock
->shift
);
1090 * update_wall_time - Uses the current clocksource to increment the wall time
1092 * Called from the timer interrupt, must hold a write on xtime_lock.
1094 static void update_wall_time(void)
1098 /* Make sure we're fully resumed: */
1099 if (unlikely(timekeeping_suspended
))
1102 #ifdef CONFIG_GENERIC_TIME
1103 offset
= (clocksource_read(clock
) - clock
->cycle_last
) & clock
->mask
;
1105 offset
= clock
->cycle_interval
;
1107 clock
->xtime_nsec
+= (s64
)xtime
.tv_nsec
<< clock
->shift
;
1109 /* normally this loop will run just once, however in the
1110 * case of lost or late ticks, it will accumulate correctly.
1112 while (offset
>= clock
->cycle_interval
) {
1113 /* accumulate one interval */
1114 clock
->xtime_nsec
+= clock
->xtime_interval
;
1115 clock
->cycle_last
+= clock
->cycle_interval
;
1116 offset
-= clock
->cycle_interval
;
1118 if (clock
->xtime_nsec
>= (u64
)NSEC_PER_SEC
<< clock
->shift
) {
1119 clock
->xtime_nsec
-= (u64
)NSEC_PER_SEC
<< clock
->shift
;
1124 /* interpolator bits */
1125 time_interpolator_update(clock
->xtime_interval
1128 /* accumulate error between NTP and clock interval */
1129 clock
->error
+= current_tick_length();
1130 clock
->error
-= clock
->xtime_interval
<< (TICK_LENGTH_SHIFT
- clock
->shift
);
1133 /* correct the clock when NTP error is too big */
1134 clocksource_adjust(clock
, offset
);
1136 /* store full nanoseconds into xtime */
1137 xtime
.tv_nsec
= (s64
)clock
->xtime_nsec
>> clock
->shift
;
1138 clock
->xtime_nsec
-= (s64
)xtime
.tv_nsec
<< clock
->shift
;
1140 /* check to see if there is a new clocksource to use */
1141 change_clocksource();
1145 * Called from the timer interrupt handler to charge one tick to the current
1146 * process. user_tick is 1 if the tick is user time, 0 for system.
1148 void update_process_times(int user_tick
)
1150 struct task_struct
*p
= current
;
1151 int cpu
= smp_processor_id();
1153 /* Note: this timer irq context must be accounted for as well. */
1155 account_user_time(p
, jiffies_to_cputime(1));
1157 account_system_time(p
, HARDIRQ_OFFSET
, jiffies_to_cputime(1));
1159 if (rcu_pending(cpu
))
1160 rcu_check_callbacks(cpu
, user_tick
);
1162 run_posix_cpu_timers(p
);
1166 * Nr of active tasks - counted in fixed-point numbers
1168 static unsigned long count_active_tasks(void)
1170 return nr_active() * FIXED_1
;
1174 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1175 * imply that avenrun[] is the standard name for this kind of thing.
1176 * Nothing else seems to be standardized: the fractional size etc
1177 * all seem to differ on different machines.
1179 * Requires xtime_lock to access.
1181 unsigned long avenrun
[3];
1183 EXPORT_SYMBOL(avenrun
);
1186 * calc_load - given tick count, update the avenrun load estimates.
1187 * This is called while holding a write_lock on xtime_lock.
1189 static inline void calc_load(unsigned long ticks
)
1191 unsigned long active_tasks
; /* fixed-point */
1192 static int count
= LOAD_FREQ
;
1195 if (unlikely(count
< 0)) {
1196 active_tasks
= count_active_tasks();
1198 CALC_LOAD(avenrun
[0], EXP_1
, active_tasks
);
1199 CALC_LOAD(avenrun
[1], EXP_5
, active_tasks
);
1200 CALC_LOAD(avenrun
[2], EXP_15
, active_tasks
);
1202 } while (count
< 0);
1207 * This read-write spinlock protects us from races in SMP while
1208 * playing with xtime and avenrun.
1210 __attribute__((weak
)) __cacheline_aligned_in_smp
DEFINE_SEQLOCK(xtime_lock
);
1212 EXPORT_SYMBOL(xtime_lock
);
1215 * This function runs timers and the timer-tq in bottom half context.
1217 static void run_timer_softirq(struct softirq_action
*h
)
1219 tvec_base_t
*base
= __get_cpu_var(tvec_bases
);
1221 hrtimer_run_queues();
1222 if (time_after_eq(jiffies
, base
->timer_jiffies
))
1227 * Called by the local, per-CPU timer interrupt on SMP.
1229 void run_local_timers(void)
1231 raise_softirq(TIMER_SOFTIRQ
);
1236 * Called by the timer interrupt. xtime_lock must already be taken
1239 static inline void update_times(unsigned long ticks
)
1246 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1247 * without sampling the sequence number in xtime_lock.
1248 * jiffies is defined in the linker script...
1251 void do_timer(unsigned long ticks
)
1253 jiffies_64
+= ticks
;
1254 update_times(ticks
);
1257 #ifdef __ARCH_WANT_SYS_ALARM
1260 * For backwards compatibility? This can be done in libc so Alpha
1261 * and all newer ports shouldn't need it.
1263 asmlinkage
unsigned long sys_alarm(unsigned int seconds
)
1265 return alarm_setitimer(seconds
);
1273 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1274 * should be moved into arch/i386 instead?
1278 * sys_getpid - return the thread group id of the current process
1280 * Note, despite the name, this returns the tgid not the pid. The tgid and
1281 * the pid are identical unless CLONE_THREAD was specified on clone() in
1282 * which case the tgid is the same in all threads of the same group.
1284 * This is SMP safe as current->tgid does not change.
1286 asmlinkage
long sys_getpid(void)
1288 return current
->tgid
;
1292 * Accessing ->real_parent is not SMP-safe, it could
1293 * change from under us. However, we can use a stale
1294 * value of ->real_parent under rcu_read_lock(), see
1295 * release_task()->call_rcu(delayed_put_task_struct).
1297 asmlinkage
long sys_getppid(void)
1302 pid
= rcu_dereference(current
->real_parent
)->tgid
;
1308 asmlinkage
long sys_getuid(void)
1310 /* Only we change this so SMP safe */
1311 return current
->uid
;
1314 asmlinkage
long sys_geteuid(void)
1316 /* Only we change this so SMP safe */
1317 return current
->euid
;
1320 asmlinkage
long sys_getgid(void)
1322 /* Only we change this so SMP safe */
1323 return current
->gid
;
1326 asmlinkage
long sys_getegid(void)
1328 /* Only we change this so SMP safe */
1329 return current
->egid
;
1334 static void process_timeout(unsigned long __data
)
1336 wake_up_process((struct task_struct
*)__data
);
1340 * schedule_timeout - sleep until timeout
1341 * @timeout: timeout value in jiffies
1343 * Make the current task sleep until @timeout jiffies have
1344 * elapsed. The routine will return immediately unless
1345 * the current task state has been set (see set_current_state()).
1347 * You can set the task state as follows -
1349 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1350 * pass before the routine returns. The routine will return 0
1352 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1353 * delivered to the current task. In this case the remaining time
1354 * in jiffies will be returned, or 0 if the timer expired in time
1356 * The current task state is guaranteed to be TASK_RUNNING when this
1359 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1360 * the CPU away without a bound on the timeout. In this case the return
1361 * value will be %MAX_SCHEDULE_TIMEOUT.
1363 * In all cases the return value is guaranteed to be non-negative.
1365 fastcall
signed long __sched
schedule_timeout(signed long timeout
)
1367 struct timer_list timer
;
1368 unsigned long expire
;
1372 case MAX_SCHEDULE_TIMEOUT
:
1374 * These two special cases are useful to be comfortable
1375 * in the caller. Nothing more. We could take
1376 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1377 * but I' d like to return a valid offset (>=0) to allow
1378 * the caller to do everything it want with the retval.
1384 * Another bit of PARANOID. Note that the retval will be
1385 * 0 since no piece of kernel is supposed to do a check
1386 * for a negative retval of schedule_timeout() (since it
1387 * should never happens anyway). You just have the printk()
1388 * that will tell you if something is gone wrong and where.
1391 printk(KERN_ERR
"schedule_timeout: wrong timeout "
1392 "value %lx\n", timeout
);
1394 current
->state
= TASK_RUNNING
;
1399 expire
= timeout
+ jiffies
;
1401 setup_timer(&timer
, process_timeout
, (unsigned long)current
);
1402 __mod_timer(&timer
, expire
);
1404 del_singleshot_timer_sync(&timer
);
1406 timeout
= expire
- jiffies
;
1409 return timeout
< 0 ? 0 : timeout
;
1411 EXPORT_SYMBOL(schedule_timeout
);
1414 * We can use __set_current_state() here because schedule_timeout() calls
1415 * schedule() unconditionally.
1417 signed long __sched
schedule_timeout_interruptible(signed long timeout
)
1419 __set_current_state(TASK_INTERRUPTIBLE
);
1420 return schedule_timeout(timeout
);
1422 EXPORT_SYMBOL(schedule_timeout_interruptible
);
1424 signed long __sched
schedule_timeout_uninterruptible(signed long timeout
)
1426 __set_current_state(TASK_UNINTERRUPTIBLE
);
1427 return schedule_timeout(timeout
);
1429 EXPORT_SYMBOL(schedule_timeout_uninterruptible
);
1431 /* Thread ID - the internal kernel "pid" */
1432 asmlinkage
long sys_gettid(void)
1434 return current
->pid
;
1438 * do_sysinfo - fill in sysinfo struct
1439 * @info: pointer to buffer to fill
1441 int do_sysinfo(struct sysinfo
*info
)
1443 unsigned long mem_total
, sav_total
;
1444 unsigned int mem_unit
, bitcount
;
1447 memset(info
, 0, sizeof(struct sysinfo
));
1451 seq
= read_seqbegin(&xtime_lock
);
1454 * This is annoying. The below is the same thing
1455 * posix_get_clock_monotonic() does, but it wants to
1456 * take the lock which we want to cover the loads stuff
1460 getnstimeofday(&tp
);
1461 tp
.tv_sec
+= wall_to_monotonic
.tv_sec
;
1462 tp
.tv_nsec
+= wall_to_monotonic
.tv_nsec
;
1463 if (tp
.tv_nsec
- NSEC_PER_SEC
>= 0) {
1464 tp
.tv_nsec
= tp
.tv_nsec
- NSEC_PER_SEC
;
1467 info
->uptime
= tp
.tv_sec
+ (tp
.tv_nsec
? 1 : 0);
1469 info
->loads
[0] = avenrun
[0] << (SI_LOAD_SHIFT
- FSHIFT
);
1470 info
->loads
[1] = avenrun
[1] << (SI_LOAD_SHIFT
- FSHIFT
);
1471 info
->loads
[2] = avenrun
[2] << (SI_LOAD_SHIFT
- FSHIFT
);
1473 info
->procs
= nr_threads
;
1474 } while (read_seqretry(&xtime_lock
, seq
));
1480 * If the sum of all the available memory (i.e. ram + swap)
1481 * is less than can be stored in a 32 bit unsigned long then
1482 * we can be binary compatible with 2.2.x kernels. If not,
1483 * well, in that case 2.2.x was broken anyways...
1485 * -Erik Andersen <andersee@debian.org>
1488 mem_total
= info
->totalram
+ info
->totalswap
;
1489 if (mem_total
< info
->totalram
|| mem_total
< info
->totalswap
)
1492 mem_unit
= info
->mem_unit
;
1493 while (mem_unit
> 1) {
1496 sav_total
= mem_total
;
1498 if (mem_total
< sav_total
)
1503 * If mem_total did not overflow, multiply all memory values by
1504 * info->mem_unit and set it to 1. This leaves things compatible
1505 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1510 info
->totalram
<<= bitcount
;
1511 info
->freeram
<<= bitcount
;
1512 info
->sharedram
<<= bitcount
;
1513 info
->bufferram
<<= bitcount
;
1514 info
->totalswap
<<= bitcount
;
1515 info
->freeswap
<<= bitcount
;
1516 info
->totalhigh
<<= bitcount
;
1517 info
->freehigh
<<= bitcount
;
1523 asmlinkage
long sys_sysinfo(struct sysinfo __user
*info
)
1529 if (copy_to_user(info
, &val
, sizeof(struct sysinfo
)))
1536 * lockdep: we want to track each per-CPU base as a separate lock-class,
1537 * but timer-bases are kmalloc()-ed, so we need to attach separate
1540 static struct lock_class_key base_lock_keys
[NR_CPUS
];
1542 static int __devinit
init_timers_cpu(int cpu
)
1546 static char __devinitdata tvec_base_done
[NR_CPUS
];
1548 if (!tvec_base_done
[cpu
]) {
1549 static char boot_done
;
1553 * The APs use this path later in boot
1555 base
= kmalloc_node(sizeof(*base
), GFP_KERNEL
,
1559 memset(base
, 0, sizeof(*base
));
1560 per_cpu(tvec_bases
, cpu
) = base
;
1563 * This is for the boot CPU - we use compile-time
1564 * static initialisation because per-cpu memory isn't
1565 * ready yet and because the memory allocators are not
1566 * initialised either.
1569 base
= &boot_tvec_bases
;
1571 tvec_base_done
[cpu
] = 1;
1573 base
= per_cpu(tvec_bases
, cpu
);
1576 spin_lock_init(&base
->lock
);
1577 lockdep_set_class(&base
->lock
, base_lock_keys
+ cpu
);
1579 for (j
= 0; j
< TVN_SIZE
; j
++) {
1580 INIT_LIST_HEAD(base
->tv5
.vec
+ j
);
1581 INIT_LIST_HEAD(base
->tv4
.vec
+ j
);
1582 INIT_LIST_HEAD(base
->tv3
.vec
+ j
);
1583 INIT_LIST_HEAD(base
->tv2
.vec
+ j
);
1585 for (j
= 0; j
< TVR_SIZE
; j
++)
1586 INIT_LIST_HEAD(base
->tv1
.vec
+ j
);
1588 base
->timer_jiffies
= jiffies
;
1592 #ifdef CONFIG_HOTPLUG_CPU
1593 static void migrate_timer_list(tvec_base_t
*new_base
, struct list_head
*head
)
1595 struct timer_list
*timer
;
1597 while (!list_empty(head
)) {
1598 timer
= list_entry(head
->next
, struct timer_list
, entry
);
1599 detach_timer(timer
, 0);
1600 timer
->base
= new_base
;
1601 internal_add_timer(new_base
, timer
);
1605 static void __devinit
migrate_timers(int cpu
)
1607 tvec_base_t
*old_base
;
1608 tvec_base_t
*new_base
;
1611 BUG_ON(cpu_online(cpu
));
1612 old_base
= per_cpu(tvec_bases
, cpu
);
1613 new_base
= get_cpu_var(tvec_bases
);
1615 local_irq_disable();
1616 spin_lock(&new_base
->lock
);
1617 spin_lock(&old_base
->lock
);
1619 BUG_ON(old_base
->running_timer
);
1621 for (i
= 0; i
< TVR_SIZE
; i
++)
1622 migrate_timer_list(new_base
, old_base
->tv1
.vec
+ i
);
1623 for (i
= 0; i
< TVN_SIZE
; i
++) {
1624 migrate_timer_list(new_base
, old_base
->tv2
.vec
+ i
);
1625 migrate_timer_list(new_base
, old_base
->tv3
.vec
+ i
);
1626 migrate_timer_list(new_base
, old_base
->tv4
.vec
+ i
);
1627 migrate_timer_list(new_base
, old_base
->tv5
.vec
+ i
);
1630 spin_unlock(&old_base
->lock
);
1631 spin_unlock(&new_base
->lock
);
1633 put_cpu_var(tvec_bases
);
1635 #endif /* CONFIG_HOTPLUG_CPU */
1637 static int __cpuinit
timer_cpu_notify(struct notifier_block
*self
,
1638 unsigned long action
, void *hcpu
)
1640 long cpu
= (long)hcpu
;
1642 case CPU_UP_PREPARE
:
1643 if (init_timers_cpu(cpu
) < 0)
1646 #ifdef CONFIG_HOTPLUG_CPU
1648 migrate_timers(cpu
);
1657 static struct notifier_block __cpuinitdata timers_nb
= {
1658 .notifier_call
= timer_cpu_notify
,
1662 void __init
init_timers(void)
1664 int err
= timer_cpu_notify(&timers_nb
, (unsigned long)CPU_UP_PREPARE
,
1665 (void *)(long)smp_processor_id());
1667 BUG_ON(err
== NOTIFY_BAD
);
1668 register_cpu_notifier(&timers_nb
);
1669 open_softirq(TIMER_SOFTIRQ
, run_timer_softirq
, NULL
);
1672 #ifdef CONFIG_TIME_INTERPOLATION
1674 struct time_interpolator
*time_interpolator __read_mostly
;
1675 static struct time_interpolator
*time_interpolator_list __read_mostly
;
1676 static DEFINE_SPINLOCK(time_interpolator_lock
);
1678 static inline cycles_t
time_interpolator_get_cycles(unsigned int src
)
1680 unsigned long (*x
)(void);
1684 case TIME_SOURCE_FUNCTION
:
1685 x
= time_interpolator
->addr
;
1688 case TIME_SOURCE_MMIO64
:
1689 return readq_relaxed((void __iomem
*)time_interpolator
->addr
);
1691 case TIME_SOURCE_MMIO32
:
1692 return readl_relaxed((void __iomem
*)time_interpolator
->addr
);
1694 default: return get_cycles();
1698 static inline u64
time_interpolator_get_counter(int writelock
)
1700 unsigned int src
= time_interpolator
->source
;
1702 if (time_interpolator
->jitter
)
1708 lcycle
= time_interpolator
->last_cycle
;
1709 now
= time_interpolator_get_cycles(src
);
1710 if (lcycle
&& time_after(lcycle
, now
))
1713 /* When holding the xtime write lock, there's no need
1714 * to add the overhead of the cmpxchg. Readers are
1715 * force to retry until the write lock is released.
1718 time_interpolator
->last_cycle
= now
;
1721 /* Keep track of the last timer value returned. The use of cmpxchg here
1722 * will cause contention in an SMP environment.
1724 } while (unlikely(cmpxchg(&time_interpolator
->last_cycle
, lcycle
, now
) != lcycle
));
1728 return time_interpolator_get_cycles(src
);
1731 void time_interpolator_reset(void)
1733 time_interpolator
->offset
= 0;
1734 time_interpolator
->last_counter
= time_interpolator_get_counter(1);
1737 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1739 unsigned long time_interpolator_get_offset(void)
1741 /* If we do not have a time interpolator set up then just return zero */
1742 if (!time_interpolator
)
1745 return time_interpolator
->offset
+
1746 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator
);
1749 #define INTERPOLATOR_ADJUST 65536
1750 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1752 void time_interpolator_update(long delta_nsec
)
1755 unsigned long offset
;
1757 /* If there is no time interpolator set up then do nothing */
1758 if (!time_interpolator
)
1762 * The interpolator compensates for late ticks by accumulating the late
1763 * time in time_interpolator->offset. A tick earlier than expected will
1764 * lead to a reset of the offset and a corresponding jump of the clock
1765 * forward. Again this only works if the interpolator clock is running
1766 * slightly slower than the regular clock and the tuning logic insures
1770 counter
= time_interpolator_get_counter(1);
1771 offset
= time_interpolator
->offset
+
1772 GET_TI_NSECS(counter
, time_interpolator
);
1774 if (delta_nsec
< 0 || (unsigned long) delta_nsec
< offset
)
1775 time_interpolator
->offset
= offset
- delta_nsec
;
1777 time_interpolator
->skips
++;
1778 time_interpolator
->ns_skipped
+= delta_nsec
- offset
;
1779 time_interpolator
->offset
= 0;
1781 time_interpolator
->last_counter
= counter
;
1783 /* Tuning logic for time interpolator invoked every minute or so.
1784 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1785 * Increase interpolator clock speed if we skip too much time.
1787 if (jiffies
% INTERPOLATOR_ADJUST
== 0)
1789 if (time_interpolator
->skips
== 0 && time_interpolator
->offset
> tick_nsec
)
1790 time_interpolator
->nsec_per_cyc
--;
1791 if (time_interpolator
->ns_skipped
> INTERPOLATOR_MAX_SKIP
&& time_interpolator
->offset
== 0)
1792 time_interpolator
->nsec_per_cyc
++;
1793 time_interpolator
->skips
= 0;
1794 time_interpolator
->ns_skipped
= 0;
1799 is_better_time_interpolator(struct time_interpolator
*new)
1801 if (!time_interpolator
)
1803 return new->frequency
> 2*time_interpolator
->frequency
||
1804 (unsigned long)new->drift
< (unsigned long)time_interpolator
->drift
;
1808 register_time_interpolator(struct time_interpolator
*ti
)
1810 unsigned long flags
;
1813 BUG_ON(ti
->frequency
== 0 || ti
->mask
== 0);
1815 ti
->nsec_per_cyc
= ((u64
)NSEC_PER_SEC
<< ti
->shift
) / ti
->frequency
;
1816 spin_lock(&time_interpolator_lock
);
1817 write_seqlock_irqsave(&xtime_lock
, flags
);
1818 if (is_better_time_interpolator(ti
)) {
1819 time_interpolator
= ti
;
1820 time_interpolator_reset();
1822 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1824 ti
->next
= time_interpolator_list
;
1825 time_interpolator_list
= ti
;
1826 spin_unlock(&time_interpolator_lock
);
1830 unregister_time_interpolator(struct time_interpolator
*ti
)
1832 struct time_interpolator
*curr
, **prev
;
1833 unsigned long flags
;
1835 spin_lock(&time_interpolator_lock
);
1836 prev
= &time_interpolator_list
;
1837 for (curr
= *prev
; curr
; curr
= curr
->next
) {
1845 write_seqlock_irqsave(&xtime_lock
, flags
);
1846 if (ti
== time_interpolator
) {
1847 /* we lost the best time-interpolator: */
1848 time_interpolator
= NULL
;
1849 /* find the next-best interpolator */
1850 for (curr
= time_interpolator_list
; curr
; curr
= curr
->next
)
1851 if (is_better_time_interpolator(curr
))
1852 time_interpolator
= curr
;
1853 time_interpolator_reset();
1855 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1856 spin_unlock(&time_interpolator_lock
);
1858 #endif /* CONFIG_TIME_INTERPOLATION */
1861 * msleep - sleep safely even with waitqueue interruptions
1862 * @msecs: Time in milliseconds to sleep for
1864 void msleep(unsigned int msecs
)
1866 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1869 timeout
= schedule_timeout_uninterruptible(timeout
);
1872 EXPORT_SYMBOL(msleep
);
1875 * msleep_interruptible - sleep waiting for signals
1876 * @msecs: Time in milliseconds to sleep for
1878 unsigned long msleep_interruptible(unsigned int msecs
)
1880 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1882 while (timeout
&& !signal_pending(current
))
1883 timeout
= schedule_timeout_interruptible(timeout
);
1884 return jiffies_to_msecs(timeout
);
1887 EXPORT_SYMBOL(msleep_interruptible
);