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 #ifdef CONFIG_TIME_INTERPOLATION
45 static void time_interpolator_update(long delta_nsec
);
47 #define time_interpolator_update(x)
50 u64 jiffies_64 __cacheline_aligned_in_smp
= INITIAL_JIFFIES
;
52 EXPORT_SYMBOL(jiffies_64
);
55 * per-CPU timer vector definitions:
57 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
58 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
59 #define TVN_SIZE (1 << TVN_BITS)
60 #define TVR_SIZE (1 << TVR_BITS)
61 #define TVN_MASK (TVN_SIZE - 1)
62 #define TVR_MASK (TVR_SIZE - 1)
64 typedef struct tvec_s
{
65 struct list_head vec
[TVN_SIZE
];
68 typedef struct tvec_root_s
{
69 struct list_head vec
[TVR_SIZE
];
72 struct tvec_t_base_s
{
74 struct timer_list
*running_timer
;
75 unsigned long timer_jiffies
;
81 } ____cacheline_aligned_in_smp
;
83 typedef struct tvec_t_base_s tvec_base_t
;
85 tvec_base_t boot_tvec_bases
;
86 EXPORT_SYMBOL(boot_tvec_bases
);
87 static DEFINE_PER_CPU(tvec_base_t
*, tvec_bases
) = { &boot_tvec_bases
};
89 static inline void set_running_timer(tvec_base_t
*base
,
90 struct timer_list
*timer
)
93 base
->running_timer
= timer
;
97 static void internal_add_timer(tvec_base_t
*base
, struct timer_list
*timer
)
99 unsigned long expires
= timer
->expires
;
100 unsigned long idx
= expires
- base
->timer_jiffies
;
101 struct list_head
*vec
;
103 if (idx
< TVR_SIZE
) {
104 int i
= expires
& TVR_MASK
;
105 vec
= base
->tv1
.vec
+ i
;
106 } else if (idx
< 1 << (TVR_BITS
+ TVN_BITS
)) {
107 int i
= (expires
>> TVR_BITS
) & TVN_MASK
;
108 vec
= base
->tv2
.vec
+ i
;
109 } else if (idx
< 1 << (TVR_BITS
+ 2 * TVN_BITS
)) {
110 int i
= (expires
>> (TVR_BITS
+ TVN_BITS
)) & TVN_MASK
;
111 vec
= base
->tv3
.vec
+ i
;
112 } else if (idx
< 1 << (TVR_BITS
+ 3 * TVN_BITS
)) {
113 int i
= (expires
>> (TVR_BITS
+ 2 * TVN_BITS
)) & TVN_MASK
;
114 vec
= base
->tv4
.vec
+ i
;
115 } else if ((signed long) idx
< 0) {
117 * Can happen if you add a timer with expires == jiffies,
118 * or you set a timer to go off in the past
120 vec
= base
->tv1
.vec
+ (base
->timer_jiffies
& TVR_MASK
);
123 /* If the timeout is larger than 0xffffffff on 64-bit
124 * architectures then we use the maximum timeout:
126 if (idx
> 0xffffffffUL
) {
128 expires
= idx
+ base
->timer_jiffies
;
130 i
= (expires
>> (TVR_BITS
+ 3 * TVN_BITS
)) & TVN_MASK
;
131 vec
= base
->tv5
.vec
+ i
;
136 list_add_tail(&timer
->entry
, vec
);
140 * init_timer - initialize a timer.
141 * @timer: the timer to be initialized
143 * init_timer() must be done to a timer prior calling *any* of the
144 * other timer functions.
146 void fastcall
init_timer(struct timer_list
*timer
)
148 timer
->entry
.next
= NULL
;
149 timer
->base
= __raw_get_cpu_var(tvec_bases
);
151 EXPORT_SYMBOL(init_timer
);
153 static inline void detach_timer(struct timer_list
*timer
,
156 struct list_head
*entry
= &timer
->entry
;
158 __list_del(entry
->prev
, entry
->next
);
161 entry
->prev
= LIST_POISON2
;
165 * We are using hashed locking: holding per_cpu(tvec_bases).lock
166 * means that all timers which are tied to this base via timer->base are
167 * locked, and the base itself is locked too.
169 * So __run_timers/migrate_timers can safely modify all timers which could
170 * be found on ->tvX lists.
172 * When the timer's base is locked, and the timer removed from list, it is
173 * possible to set timer->base = NULL and drop the lock: the timer remains
176 static tvec_base_t
*lock_timer_base(struct timer_list
*timer
,
177 unsigned long *flags
)
183 if (likely(base
!= NULL
)) {
184 spin_lock_irqsave(&base
->lock
, *flags
);
185 if (likely(base
== timer
->base
))
187 /* The timer has migrated to another CPU */
188 spin_unlock_irqrestore(&base
->lock
, *flags
);
194 int __mod_timer(struct timer_list
*timer
, unsigned long expires
)
196 tvec_base_t
*base
, *new_base
;
200 BUG_ON(!timer
->function
);
202 base
= lock_timer_base(timer
, &flags
);
204 if (timer_pending(timer
)) {
205 detach_timer(timer
, 0);
209 new_base
= __get_cpu_var(tvec_bases
);
211 if (base
!= new_base
) {
213 * We are trying to schedule the timer on the local CPU.
214 * However we can't change timer's base while it is running,
215 * otherwise del_timer_sync() can't detect that the timer's
216 * handler yet has not finished. This also guarantees that
217 * the timer is serialized wrt itself.
219 if (likely(base
->running_timer
!= timer
)) {
220 /* See the comment in lock_timer_base() */
222 spin_unlock(&base
->lock
);
224 spin_lock(&base
->lock
);
229 timer
->expires
= expires
;
230 internal_add_timer(base
, timer
);
231 spin_unlock_irqrestore(&base
->lock
, flags
);
236 EXPORT_SYMBOL(__mod_timer
);
239 * add_timer_on - start a timer on a particular CPU
240 * @timer: the timer to be added
241 * @cpu: the CPU to start it on
243 * This is not very scalable on SMP. Double adds are not possible.
245 void add_timer_on(struct timer_list
*timer
, int cpu
)
247 tvec_base_t
*base
= per_cpu(tvec_bases
, cpu
);
250 BUG_ON(timer_pending(timer
) || !timer
->function
);
251 spin_lock_irqsave(&base
->lock
, flags
);
253 internal_add_timer(base
, timer
);
254 spin_unlock_irqrestore(&base
->lock
, flags
);
259 * mod_timer - modify a timer's timeout
260 * @timer: the timer to be modified
262 * mod_timer is a more efficient way to update the expire field of an
263 * active timer (if the timer is inactive it will be activated)
265 * mod_timer(timer, expires) is equivalent to:
267 * del_timer(timer); timer->expires = expires; add_timer(timer);
269 * Note that if there are multiple unserialized concurrent users of the
270 * same timer, then mod_timer() is the only safe way to modify the timeout,
271 * since add_timer() cannot modify an already running timer.
273 * The function returns whether it has modified a pending timer or not.
274 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
275 * active timer returns 1.)
277 int mod_timer(struct timer_list
*timer
, unsigned long expires
)
279 BUG_ON(!timer
->function
);
282 * This is a common optimization triggered by the
283 * networking code - if the timer is re-modified
284 * to be the same thing then just return:
286 if (timer
->expires
== expires
&& timer_pending(timer
))
289 return __mod_timer(timer
, expires
);
292 EXPORT_SYMBOL(mod_timer
);
295 * del_timer - deactive a timer.
296 * @timer: the timer to be deactivated
298 * del_timer() deactivates a timer - this works on both active and inactive
301 * The function returns whether it has deactivated a pending timer or not.
302 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
303 * active timer returns 1.)
305 int del_timer(struct timer_list
*timer
)
311 if (timer_pending(timer
)) {
312 base
= lock_timer_base(timer
, &flags
);
313 if (timer_pending(timer
)) {
314 detach_timer(timer
, 1);
317 spin_unlock_irqrestore(&base
->lock
, flags
);
323 EXPORT_SYMBOL(del_timer
);
327 * This function tries to deactivate a timer. Upon successful (ret >= 0)
328 * exit the timer is not queued and the handler is not running on any CPU.
330 * It must not be called from interrupt contexts.
332 int try_to_del_timer_sync(struct timer_list
*timer
)
338 base
= lock_timer_base(timer
, &flags
);
340 if (base
->running_timer
== timer
)
344 if (timer_pending(timer
)) {
345 detach_timer(timer
, 1);
349 spin_unlock_irqrestore(&base
->lock
, flags
);
355 * del_timer_sync - deactivate a timer and wait for the handler to finish.
356 * @timer: the timer to be deactivated
358 * This function only differs from del_timer() on SMP: besides deactivating
359 * the timer it also makes sure the handler has finished executing on other
362 * Synchronization rules: callers must prevent restarting of the timer,
363 * otherwise this function is meaningless. It must not be called from
364 * interrupt contexts. The caller must not hold locks which would prevent
365 * completion of the timer's handler. The timer's handler must not call
366 * add_timer_on(). Upon exit the timer is not queued and the handler is
367 * not running on any CPU.
369 * The function returns whether it has deactivated a pending timer or not.
371 int del_timer_sync(struct timer_list
*timer
)
374 int ret
= try_to_del_timer_sync(timer
);
380 EXPORT_SYMBOL(del_timer_sync
);
383 static int cascade(tvec_base_t
*base
, tvec_t
*tv
, int index
)
385 /* cascade all the timers from tv up one level */
386 struct timer_list
*timer
, *tmp
;
387 struct list_head tv_list
;
389 list_replace_init(tv
->vec
+ index
, &tv_list
);
392 * We are removing _all_ timers from the list, so we
393 * don't have to detach them individually.
395 list_for_each_entry_safe(timer
, tmp
, &tv_list
, entry
) {
396 BUG_ON(timer
->base
!= base
);
397 internal_add_timer(base
, timer
);
404 * __run_timers - run all expired timers (if any) on this CPU.
405 * @base: the timer vector to be processed.
407 * This function cascades all vectors and executes all expired timer
410 #define INDEX(N) (base->timer_jiffies >> (TVR_BITS + N * TVN_BITS)) & TVN_MASK
412 static inline void __run_timers(tvec_base_t
*base
)
414 struct timer_list
*timer
;
416 spin_lock_irq(&base
->lock
);
417 while (time_after_eq(jiffies
, base
->timer_jiffies
)) {
418 struct list_head work_list
;
419 struct list_head
*head
= &work_list
;
420 int index
= base
->timer_jiffies
& TVR_MASK
;
426 (!cascade(base
, &base
->tv2
, INDEX(0))) &&
427 (!cascade(base
, &base
->tv3
, INDEX(1))) &&
428 !cascade(base
, &base
->tv4
, INDEX(2)))
429 cascade(base
, &base
->tv5
, INDEX(3));
430 ++base
->timer_jiffies
;
431 list_replace_init(base
->tv1
.vec
+ index
, &work_list
);
432 while (!list_empty(head
)) {
433 void (*fn
)(unsigned long);
436 timer
= list_entry(head
->next
,struct timer_list
,entry
);
437 fn
= timer
->function
;
440 set_running_timer(base
, timer
);
441 detach_timer(timer
, 1);
442 spin_unlock_irq(&base
->lock
);
444 int preempt_count
= preempt_count();
446 if (preempt_count
!= preempt_count()) {
447 printk(KERN_WARNING
"huh, entered %p "
448 "with preempt_count %08x, exited"
455 spin_lock_irq(&base
->lock
);
458 set_running_timer(base
, NULL
);
459 spin_unlock_irq(&base
->lock
);
462 #ifdef CONFIG_NO_IDLE_HZ
464 * Find out when the next timer event is due to happen. This
465 * is used on S/390 to stop all activity when a cpus is idle.
466 * This functions needs to be called disabled.
468 unsigned long next_timer_interrupt(void)
471 struct list_head
*list
;
472 struct timer_list
*nte
;
473 unsigned long expires
;
474 unsigned long hr_expires
= MAX_JIFFY_OFFSET
;
479 hr_delta
= hrtimer_get_next_event();
480 if (hr_delta
.tv64
!= KTIME_MAX
) {
481 struct timespec tsdelta
;
482 tsdelta
= ktime_to_timespec(hr_delta
);
483 hr_expires
= timespec_to_jiffies(&tsdelta
);
485 return hr_expires
+ jiffies
;
487 hr_expires
+= jiffies
;
489 base
= __get_cpu_var(tvec_bases
);
490 spin_lock(&base
->lock
);
491 expires
= base
->timer_jiffies
+ (LONG_MAX
>> 1);
494 /* Look for timer events in tv1. */
495 j
= base
->timer_jiffies
& TVR_MASK
;
497 list_for_each_entry(nte
, base
->tv1
.vec
+ j
, entry
) {
498 expires
= nte
->expires
;
499 if (j
< (base
->timer_jiffies
& TVR_MASK
))
500 list
= base
->tv2
.vec
+ (INDEX(0));
503 j
= (j
+ 1) & TVR_MASK
;
504 } while (j
!= (base
->timer_jiffies
& TVR_MASK
));
507 varray
[0] = &base
->tv2
;
508 varray
[1] = &base
->tv3
;
509 varray
[2] = &base
->tv4
;
510 varray
[3] = &base
->tv5
;
511 for (i
= 0; i
< 4; i
++) {
514 if (list_empty(varray
[i
]->vec
+ j
)) {
515 j
= (j
+ 1) & TVN_MASK
;
518 list_for_each_entry(nte
, varray
[i
]->vec
+ j
, entry
)
519 if (time_before(nte
->expires
, expires
))
520 expires
= nte
->expires
;
521 if (j
< (INDEX(i
)) && i
< 3)
522 list
= varray
[i
+ 1]->vec
+ (INDEX(i
+ 1));
524 } while (j
!= (INDEX(i
)));
529 * The search wrapped. We need to look at the next list
530 * from next tv element that would cascade into tv element
531 * where we found the timer element.
533 list_for_each_entry(nte
, list
, entry
) {
534 if (time_before(nte
->expires
, expires
))
535 expires
= nte
->expires
;
538 spin_unlock(&base
->lock
);
541 * It can happen that other CPUs service timer IRQs and increment
542 * jiffies, but we have not yet got a local timer tick to process
543 * the timer wheels. In that case, the expiry time can be before
544 * jiffies, but since the high-resolution timer here is relative to
545 * jiffies, the default expression when high-resolution timers are
548 * time_before(MAX_JIFFY_OFFSET + jiffies, expires)
550 * would falsely evaluate to true. If that is the case, just
551 * return jiffies so that we can immediately fire the local timer
553 if (time_before(expires
, jiffies
))
556 if (time_before(hr_expires
, expires
))
563 /******************************************************************/
566 * Timekeeping variables
568 unsigned long tick_usec
= TICK_USEC
; /* USER_HZ period (usec) */
569 unsigned long tick_nsec
= TICK_NSEC
; /* ACTHZ period (nsec) */
573 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
574 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
575 * at zero at system boot time, so wall_to_monotonic will be negative,
576 * however, we will ALWAYS keep the tv_nsec part positive so we can use
577 * the usual normalization.
579 struct timespec xtime
__attribute__ ((aligned (16)));
580 struct timespec wall_to_monotonic
__attribute__ ((aligned (16)));
582 EXPORT_SYMBOL(xtime
);
584 /* Don't completely fail for HZ > 500. */
585 int tickadj
= 500/HZ
? : 1; /* microsecs */
589 * phase-lock loop variables
591 /* TIME_ERROR prevents overwriting the CMOS clock */
592 int time_state
= TIME_OK
; /* clock synchronization status */
593 int time_status
= STA_UNSYNC
; /* clock status bits */
594 long time_offset
; /* time adjustment (us) */
595 long time_constant
= 2; /* pll time constant */
596 long time_tolerance
= MAXFREQ
; /* frequency tolerance (ppm) */
597 long time_precision
= 1; /* clock precision (us) */
598 long time_maxerror
= NTP_PHASE_LIMIT
; /* maximum error (us) */
599 long time_esterror
= NTP_PHASE_LIMIT
; /* estimated error (us) */
600 long time_freq
= (((NSEC_PER_SEC
+ HZ
/2) % HZ
- HZ
/2) << SHIFT_USEC
) / NSEC_PER_USEC
;
601 /* frequency offset (scaled ppm)*/
602 static long time_adj
; /* tick adjust (scaled 1 / HZ) */
603 long time_reftime
; /* time at last adjustment (s) */
605 long time_next_adjust
;
608 * this routine handles the overflow of the microsecond field
610 * The tricky bits of code to handle the accurate clock support
611 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
612 * They were originally developed for SUN and DEC kernels.
613 * All the kudos should go to Dave for this stuff.
616 static void second_overflow(void)
620 /* Bump the maxerror field */
621 time_maxerror
+= time_tolerance
>> SHIFT_USEC
;
622 if (time_maxerror
> NTP_PHASE_LIMIT
) {
623 time_maxerror
= NTP_PHASE_LIMIT
;
624 time_status
|= STA_UNSYNC
;
628 * Leap second processing. If in leap-insert state at the end of the
629 * day, the system clock is set back one second; if in leap-delete
630 * state, the system clock is set ahead one second. The microtime()
631 * routine or external clock driver will insure that reported time is
632 * always monotonic. The ugly divides should be replaced.
634 switch (time_state
) {
636 if (time_status
& STA_INS
)
637 time_state
= TIME_INS
;
638 else if (time_status
& STA_DEL
)
639 time_state
= TIME_DEL
;
642 if (xtime
.tv_sec
% 86400 == 0) {
644 wall_to_monotonic
.tv_sec
++;
646 * The timer interpolator will make time change
647 * gradually instead of an immediate jump by one second
649 time_interpolator_update(-NSEC_PER_SEC
);
650 time_state
= TIME_OOP
;
652 printk(KERN_NOTICE
"Clock: inserting leap second "
657 if ((xtime
.tv_sec
+ 1) % 86400 == 0) {
659 wall_to_monotonic
.tv_sec
--;
661 * Use of time interpolator for a gradual change of
664 time_interpolator_update(NSEC_PER_SEC
);
665 time_state
= TIME_WAIT
;
667 printk(KERN_NOTICE
"Clock: deleting leap second "
672 time_state
= TIME_WAIT
;
675 if (!(time_status
& (STA_INS
| STA_DEL
)))
676 time_state
= TIME_OK
;
680 * Compute the phase adjustment for the next second. In PLL mode, the
681 * offset is reduced by a fixed factor times the time constant. In FLL
682 * mode the offset is used directly. In either mode, the maximum phase
683 * adjustment for each second is clamped so as to spread the adjustment
684 * over not more than the number of seconds between updates.
687 if (!(time_status
& STA_FLL
))
688 ltemp
= shift_right(ltemp
, SHIFT_KG
+ time_constant
);
689 ltemp
= min(ltemp
, (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
);
690 ltemp
= max(ltemp
, -(MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
);
691 time_offset
-= ltemp
;
692 time_adj
= ltemp
<< (SHIFT_SCALE
- SHIFT_HZ
- SHIFT_UPDATE
);
695 * Compute the frequency estimate and additional phase adjustment due
696 * to frequency error for the next second.
699 time_adj
+= shift_right(ltemp
,(SHIFT_USEC
+ SHIFT_HZ
- SHIFT_SCALE
));
703 * Compensate for (HZ==100) != (1 << SHIFT_HZ). Add 25% and 3.125% to
704 * get 128.125; => only 0.125% error (p. 14)
706 time_adj
+= shift_right(time_adj
, 2) + shift_right(time_adj
, 5);
710 * Compensate for (HZ==250) != (1 << SHIFT_HZ). Add 1.5625% and
711 * 0.78125% to get 255.85938; => only 0.05% error (p. 14)
713 time_adj
+= shift_right(time_adj
, 6) + shift_right(time_adj
, 7);
717 * Compensate for (HZ==1000) != (1 << SHIFT_HZ). Add 1.5625% and
718 * 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
720 time_adj
+= shift_right(time_adj
, 6) + shift_right(time_adj
, 7);
725 * Returns how many microseconds we need to add to xtime this tick
726 * in doing an adjustment requested with adjtime.
728 static long adjtime_adjustment(void)
730 long time_adjust_step
;
732 time_adjust_step
= time_adjust
;
733 if (time_adjust_step
) {
735 * We are doing an adjtime thing. Prepare time_adjust_step to
736 * be within bounds. Note that a positive time_adjust means we
737 * want the clock to run faster.
739 * Limit the amount of the step to be in the range
740 * -tickadj .. +tickadj
742 time_adjust_step
= min(time_adjust_step
, (long)tickadj
);
743 time_adjust_step
= max(time_adjust_step
, (long)-tickadj
);
745 return time_adjust_step
;
748 /* in the NTP reference this is called "hardclock()" */
749 static void update_ntp_one_tick(void)
751 long time_adjust_step
;
753 time_adjust_step
= adjtime_adjustment();
754 if (time_adjust_step
)
755 /* Reduce by this step the amount of time left */
756 time_adjust
-= time_adjust_step
;
758 /* Changes by adjtime() do not take effect till next tick. */
759 if (time_next_adjust
!= 0) {
760 time_adjust
= time_next_adjust
;
761 time_next_adjust
= 0;
766 * Return how long ticks are at the moment, that is, how much time
767 * update_wall_time_one_tick will add to xtime next time we call it
768 * (assuming no calls to do_adjtimex in the meantime).
769 * The return value is in fixed-point nanoseconds shifted by the
770 * specified number of bits to the right of the binary point.
771 * This function has no side-effects.
773 u64
current_tick_length(void)
778 /* calculate the finest interval NTP will allow.
779 * ie: nanosecond value shifted by (SHIFT_SCALE - 10)
781 delta_nsec
= tick_nsec
+ adjtime_adjustment() * 1000;
782 ret
= (u64
)delta_nsec
<< TICK_LENGTH_SHIFT
;
783 ret
+= (s64
)time_adj
<< (TICK_LENGTH_SHIFT
- (SHIFT_SCALE
- 10));
788 /* XXX - all of this timekeeping code should be later moved to time.c */
789 #include <linux/clocksource.h>
790 static struct clocksource
*clock
; /* pointer to current clocksource */
792 #ifdef CONFIG_GENERIC_TIME
794 * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook
796 * private function, must hold xtime_lock lock when being
797 * called. Returns the number of nanoseconds since the
798 * last call to update_wall_time() (adjusted by NTP scaling)
800 static inline s64
__get_nsec_offset(void)
802 cycle_t cycle_now
, cycle_delta
;
805 /* read clocksource: */
806 cycle_now
= clocksource_read(clock
);
808 /* calculate the delta since the last update_wall_time: */
809 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
811 /* convert to nanoseconds: */
812 ns_offset
= cyc2ns(clock
, cycle_delta
);
818 * __get_realtime_clock_ts - Returns the time of day in a timespec
819 * @ts: pointer to the timespec to be set
821 * Returns the time of day in a timespec. Used by
822 * do_gettimeofday() and get_realtime_clock_ts().
824 static inline void __get_realtime_clock_ts(struct timespec
*ts
)
830 seq
= read_seqbegin(&xtime_lock
);
833 nsecs
= __get_nsec_offset();
835 } while (read_seqretry(&xtime_lock
, seq
));
837 timespec_add_ns(ts
, nsecs
);
841 * getnstimeofday - Returns the time of day in a timespec
842 * @ts: pointer to the timespec to be set
844 * Returns the time of day in a timespec.
846 void getnstimeofday(struct timespec
*ts
)
848 __get_realtime_clock_ts(ts
);
851 EXPORT_SYMBOL(getnstimeofday
);
854 * do_gettimeofday - Returns the time of day in a timeval
855 * @tv: pointer to the timeval to be set
857 * NOTE: Users should be converted to using get_realtime_clock_ts()
859 void do_gettimeofday(struct timeval
*tv
)
863 __get_realtime_clock_ts(&now
);
864 tv
->tv_sec
= now
.tv_sec
;
865 tv
->tv_usec
= now
.tv_nsec
/1000;
868 EXPORT_SYMBOL(do_gettimeofday
);
870 * do_settimeofday - Sets the time of day
871 * @tv: pointer to the timespec variable containing the new time
873 * Sets the time of day to the new time and update NTP and notify hrtimers
875 int do_settimeofday(struct timespec
*tv
)
878 time_t wtm_sec
, sec
= tv
->tv_sec
;
879 long wtm_nsec
, nsec
= tv
->tv_nsec
;
881 if ((unsigned long)tv
->tv_nsec
>= NSEC_PER_SEC
)
884 write_seqlock_irqsave(&xtime_lock
, flags
);
886 nsec
-= __get_nsec_offset();
888 wtm_sec
= wall_to_monotonic
.tv_sec
+ (xtime
.tv_sec
- sec
);
889 wtm_nsec
= wall_to_monotonic
.tv_nsec
+ (xtime
.tv_nsec
- nsec
);
891 set_normalized_timespec(&xtime
, sec
, nsec
);
892 set_normalized_timespec(&wall_to_monotonic
, wtm_sec
, wtm_nsec
);
896 write_sequnlock_irqrestore(&xtime_lock
, flags
);
898 /* signal hrtimers about time change */
904 EXPORT_SYMBOL(do_settimeofday
);
907 * change_clocksource - Swaps clocksources if a new one is available
909 * Accumulates current time interval and initializes new clocksource
911 static int change_clocksource(void)
913 struct clocksource
*new;
916 new = clocksource_get_next();
918 now
= clocksource_read(new);
919 nsec
= __get_nsec_offset();
920 timespec_add_ns(&xtime
, nsec
);
923 clock
->cycle_last
= now
;
924 printk(KERN_INFO
"Time: %s clocksource has been installed.\n",
927 } else if (clock
->update_callback
) {
928 return clock
->update_callback();
933 #define change_clocksource() (0)
937 * timeofday_is_continuous - check to see if timekeeping is free running
939 int timekeeping_is_continuous(void)
945 seq
= read_seqbegin(&xtime_lock
);
947 ret
= clock
->is_continuous
;
949 } while (read_seqretry(&xtime_lock
, seq
));
955 * timekeeping_init - Initializes the clocksource and common timekeeping values
957 void __init
timekeeping_init(void)
961 write_seqlock_irqsave(&xtime_lock
, flags
);
962 clock
= clocksource_get_next();
963 clocksource_calculate_interval(clock
, tick_nsec
);
964 clock
->cycle_last
= clocksource_read(clock
);
966 write_sequnlock_irqrestore(&xtime_lock
, flags
);
971 * timekeeping_resume - Resumes the generic timekeeping subsystem.
974 * This is for the generic clocksource timekeeping.
975 * xtime/wall_to_monotonic/jiffies/wall_jiffies/etc are
976 * still managed by arch specific suspend/resume code.
978 static int timekeeping_resume(struct sys_device
*dev
)
982 write_seqlock_irqsave(&xtime_lock
, flags
);
983 /* restart the last cycle value */
984 clock
->cycle_last
= clocksource_read(clock
);
985 write_sequnlock_irqrestore(&xtime_lock
, flags
);
989 /* sysfs resume/suspend bits for timekeeping */
990 static struct sysdev_class timekeeping_sysclass
= {
991 .resume
= timekeeping_resume
,
992 set_kset_name("timekeeping"),
995 static struct sys_device device_timer
= {
997 .cls
= &timekeeping_sysclass
,
1000 static int __init
timekeeping_init_device(void)
1002 int error
= sysdev_class_register(&timekeeping_sysclass
);
1004 error
= sysdev_register(&device_timer
);
1008 device_initcall(timekeeping_init_device
);
1011 * If the error is already larger, we look ahead another tick,
1012 * to compensate for late or lost adjustments.
1014 static __always_inline
int clocksource_bigadjust(int sign
, s64 error
, s64
*interval
, s64
*offset
)
1019 * As soon as the machine is synchronized to the external time
1020 * source this should be the common case.
1023 if (likely(sign
> 0 ? error
<= *interval
: error
>= *interval
))
1027 * An extra look ahead dampens the effect of the current error,
1028 * which can grow quite large with continously late updates, as
1029 * it would dominate the adjustment value and can lead to
1032 error
+= current_tick_length() >> (TICK_LENGTH_SHIFT
- clock
->shift
+ 1);
1033 error
-= clock
->xtime_interval
>> 1;
1038 if (sign
> 0 ? error
<= *interval
: error
>= *interval
)
1044 * Add the current adjustments to the error and take the offset
1045 * into account, the latter can cause the error to be hardly
1046 * reduced at the next tick. Check the error again if there's
1047 * room for another adjustment, thus further reducing the error
1048 * which otherwise had to be corrected at the next update.
1050 error
= (error
<< 1) - *interval
+ *offset
;
1051 if (sign
> 0 ? error
> *interval
: error
< *interval
)
1060 * Adjust the multiplier to reduce the error value,
1061 * this is optimized for the most common adjustments of -1,0,1,
1062 * for other values we can do a bit more work.
1064 static void clocksource_adjust(struct clocksource
*clock
, s64 offset
)
1066 s64 error
, interval
= clock
->cycle_interval
;
1069 error
= clock
->error
>> (TICK_LENGTH_SHIFT
- clock
->shift
- 1);
1070 if (error
> interval
) {
1071 adj
= clocksource_bigadjust(1, error
, &interval
, &offset
);
1072 } else if (error
< -interval
) {
1073 interval
= -interval
;
1075 adj
= clocksource_bigadjust(-1, error
, &interval
, &offset
);
1080 clock
->xtime_interval
+= interval
;
1081 clock
->xtime_nsec
-= offset
;
1082 clock
->error
-= (interval
- offset
) << (TICK_LENGTH_SHIFT
- clock
->shift
);
1086 * update_wall_time - Uses the current clocksource to increment the wall time
1088 * Called from the timer interrupt, must hold a write on xtime_lock.
1090 static void update_wall_time(void)
1094 clock
->xtime_nsec
+= (s64
)xtime
.tv_nsec
<< clock
->shift
;
1096 #ifdef CONFIG_GENERIC_TIME
1097 offset
= (clocksource_read(clock
) - clock
->cycle_last
) & clock
->mask
;
1099 offset
= clock
->cycle_interval
;
1102 /* normally this loop will run just once, however in the
1103 * case of lost or late ticks, it will accumulate correctly.
1105 while (offset
>= clock
->cycle_interval
) {
1106 /* accumulate one interval */
1107 clock
->xtime_nsec
+= clock
->xtime_interval
;
1108 clock
->cycle_last
+= clock
->cycle_interval
;
1109 offset
-= clock
->cycle_interval
;
1111 if (clock
->xtime_nsec
>= (u64
)NSEC_PER_SEC
<< clock
->shift
) {
1112 clock
->xtime_nsec
-= (u64
)NSEC_PER_SEC
<< clock
->shift
;
1117 /* interpolator bits */
1118 time_interpolator_update(clock
->xtime_interval
1120 /* increment the NTP state machine */
1121 update_ntp_one_tick();
1123 /* accumulate error between NTP and clock interval */
1124 clock
->error
+= current_tick_length();
1125 clock
->error
-= clock
->xtime_interval
<< (TICK_LENGTH_SHIFT
- clock
->shift
);
1128 /* correct the clock when NTP error is too big */
1129 clocksource_adjust(clock
, offset
);
1131 /* store full nanoseconds into xtime */
1132 xtime
.tv_nsec
= clock
->xtime_nsec
>> clock
->shift
;
1133 clock
->xtime_nsec
-= (s64
)xtime
.tv_nsec
<< clock
->shift
;
1135 /* check to see if there is a new clocksource to use */
1136 if (change_clocksource()) {
1138 clock
->xtime_nsec
= 0;
1139 clocksource_calculate_interval(clock
, tick_nsec
);
1144 * Called from the timer interrupt handler to charge one tick to the current
1145 * process. user_tick is 1 if the tick is user time, 0 for system.
1147 void update_process_times(int user_tick
)
1149 struct task_struct
*p
= current
;
1150 int cpu
= smp_processor_id();
1152 /* Note: this timer irq context must be accounted for as well. */
1154 account_user_time(p
, jiffies_to_cputime(1));
1156 account_system_time(p
, HARDIRQ_OFFSET
, jiffies_to_cputime(1));
1158 if (rcu_pending(cpu
))
1159 rcu_check_callbacks(cpu
, user_tick
);
1161 run_posix_cpu_timers(p
);
1165 * Nr of active tasks - counted in fixed-point numbers
1167 static unsigned long count_active_tasks(void)
1169 return nr_active() * FIXED_1
;
1173 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1174 * imply that avenrun[] is the standard name for this kind of thing.
1175 * Nothing else seems to be standardized: the fractional size etc
1176 * all seem to differ on different machines.
1178 * Requires xtime_lock to access.
1180 unsigned long avenrun
[3];
1182 EXPORT_SYMBOL(avenrun
);
1185 * calc_load - given tick count, update the avenrun load estimates.
1186 * This is called while holding a write_lock on xtime_lock.
1188 static inline void calc_load(unsigned long ticks
)
1190 unsigned long active_tasks
; /* fixed-point */
1191 static int count
= LOAD_FREQ
;
1196 active_tasks
= count_active_tasks();
1197 CALC_LOAD(avenrun
[0], EXP_1
, active_tasks
);
1198 CALC_LOAD(avenrun
[1], EXP_5
, active_tasks
);
1199 CALC_LOAD(avenrun
[2], EXP_15
, active_tasks
);
1203 /* jiffies at the most recent update of wall time */
1204 unsigned long wall_jiffies
= INITIAL_JIFFIES
;
1207 * This read-write spinlock protects us from races in SMP while
1208 * playing with xtime and avenrun.
1210 #ifndef ARCH_HAVE_XTIME_LOCK
1211 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(xtime_lock
);
1213 EXPORT_SYMBOL(xtime_lock
);
1217 * This function runs timers and the timer-tq in bottom half context.
1219 static void run_timer_softirq(struct softirq_action
*h
)
1221 tvec_base_t
*base
= __get_cpu_var(tvec_bases
);
1223 hrtimer_run_queues();
1224 if (time_after_eq(jiffies
, base
->timer_jiffies
))
1229 * Called by the local, per-CPU timer interrupt on SMP.
1231 void run_local_timers(void)
1233 raise_softirq(TIMER_SOFTIRQ
);
1238 * Called by the timer interrupt. xtime_lock must already be taken
1241 static inline void update_times(void)
1243 unsigned long ticks
;
1245 ticks
= jiffies
- wall_jiffies
;
1246 wall_jiffies
+= ticks
;
1252 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1253 * without sampling the sequence number in xtime_lock.
1254 * jiffies is defined in the linker script...
1257 void do_timer(struct pt_regs
*regs
)
1260 /* prevent loading jiffies before storing new jiffies_64 value. */
1265 #ifdef __ARCH_WANT_SYS_ALARM
1268 * For backwards compatibility? This can be done in libc so Alpha
1269 * and all newer ports shouldn't need it.
1271 asmlinkage
unsigned long sys_alarm(unsigned int seconds
)
1273 return alarm_setitimer(seconds
);
1281 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1282 * should be moved into arch/i386 instead?
1286 * sys_getpid - return the thread group id of the current process
1288 * Note, despite the name, this returns the tgid not the pid. The tgid and
1289 * the pid are identical unless CLONE_THREAD was specified on clone() in
1290 * which case the tgid is the same in all threads of the same group.
1292 * This is SMP safe as current->tgid does not change.
1294 asmlinkage
long sys_getpid(void)
1296 return current
->tgid
;
1300 * Accessing ->group_leader->real_parent is not SMP-safe, it could
1301 * change from under us. However, rather than getting any lock
1302 * we can use an optimistic algorithm: get the parent
1303 * pid, and go back and check that the parent is still
1304 * the same. If it has changed (which is extremely unlikely
1305 * indeed), we just try again..
1307 * NOTE! This depends on the fact that even if we _do_
1308 * get an old value of "parent", we can happily dereference
1309 * the pointer (it was and remains a dereferencable kernel pointer
1310 * no matter what): we just can't necessarily trust the result
1311 * until we know that the parent pointer is valid.
1313 * NOTE2: ->group_leader never changes from under us.
1315 asmlinkage
long sys_getppid(void)
1318 struct task_struct
*me
= current
;
1319 struct task_struct
*parent
;
1321 parent
= me
->group_leader
->real_parent
;
1324 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1326 struct task_struct
*old
= parent
;
1329 * Make sure we read the pid before re-reading the
1333 parent
= me
->group_leader
->real_parent
;
1343 asmlinkage
long sys_getuid(void)
1345 /* Only we change this so SMP safe */
1346 return current
->uid
;
1349 asmlinkage
long sys_geteuid(void)
1351 /* Only we change this so SMP safe */
1352 return current
->euid
;
1355 asmlinkage
long sys_getgid(void)
1357 /* Only we change this so SMP safe */
1358 return current
->gid
;
1361 asmlinkage
long sys_getegid(void)
1363 /* Only we change this so SMP safe */
1364 return current
->egid
;
1369 static void process_timeout(unsigned long __data
)
1371 wake_up_process((task_t
*)__data
);
1375 * schedule_timeout - sleep until timeout
1376 * @timeout: timeout value in jiffies
1378 * Make the current task sleep until @timeout jiffies have
1379 * elapsed. The routine will return immediately unless
1380 * the current task state has been set (see set_current_state()).
1382 * You can set the task state as follows -
1384 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1385 * pass before the routine returns. The routine will return 0
1387 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1388 * delivered to the current task. In this case the remaining time
1389 * in jiffies will be returned, or 0 if the timer expired in time
1391 * The current task state is guaranteed to be TASK_RUNNING when this
1394 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1395 * the CPU away without a bound on the timeout. In this case the return
1396 * value will be %MAX_SCHEDULE_TIMEOUT.
1398 * In all cases the return value is guaranteed to be non-negative.
1400 fastcall
signed long __sched
schedule_timeout(signed long timeout
)
1402 struct timer_list timer
;
1403 unsigned long expire
;
1407 case MAX_SCHEDULE_TIMEOUT
:
1409 * These two special cases are useful to be comfortable
1410 * in the caller. Nothing more. We could take
1411 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1412 * but I' d like to return a valid offset (>=0) to allow
1413 * the caller to do everything it want with the retval.
1419 * Another bit of PARANOID. Note that the retval will be
1420 * 0 since no piece of kernel is supposed to do a check
1421 * for a negative retval of schedule_timeout() (since it
1422 * should never happens anyway). You just have the printk()
1423 * that will tell you if something is gone wrong and where.
1427 printk(KERN_ERR
"schedule_timeout: wrong timeout "
1428 "value %lx from %p\n", timeout
,
1429 __builtin_return_address(0));
1430 current
->state
= TASK_RUNNING
;
1435 expire
= timeout
+ jiffies
;
1437 setup_timer(&timer
, process_timeout
, (unsigned long)current
);
1438 __mod_timer(&timer
, expire
);
1440 del_singleshot_timer_sync(&timer
);
1442 timeout
= expire
- jiffies
;
1445 return timeout
< 0 ? 0 : timeout
;
1447 EXPORT_SYMBOL(schedule_timeout
);
1450 * We can use __set_current_state() here because schedule_timeout() calls
1451 * schedule() unconditionally.
1453 signed long __sched
schedule_timeout_interruptible(signed long timeout
)
1455 __set_current_state(TASK_INTERRUPTIBLE
);
1456 return schedule_timeout(timeout
);
1458 EXPORT_SYMBOL(schedule_timeout_interruptible
);
1460 signed long __sched
schedule_timeout_uninterruptible(signed long timeout
)
1462 __set_current_state(TASK_UNINTERRUPTIBLE
);
1463 return schedule_timeout(timeout
);
1465 EXPORT_SYMBOL(schedule_timeout_uninterruptible
);
1467 /* Thread ID - the internal kernel "pid" */
1468 asmlinkage
long sys_gettid(void)
1470 return current
->pid
;
1474 * sys_sysinfo - fill in sysinfo struct
1476 asmlinkage
long sys_sysinfo(struct sysinfo __user
*info
)
1479 unsigned long mem_total
, sav_total
;
1480 unsigned int mem_unit
, bitcount
;
1483 memset((char *)&val
, 0, sizeof(struct sysinfo
));
1487 seq
= read_seqbegin(&xtime_lock
);
1490 * This is annoying. The below is the same thing
1491 * posix_get_clock_monotonic() does, but it wants to
1492 * take the lock which we want to cover the loads stuff
1496 getnstimeofday(&tp
);
1497 tp
.tv_sec
+= wall_to_monotonic
.tv_sec
;
1498 tp
.tv_nsec
+= wall_to_monotonic
.tv_nsec
;
1499 if (tp
.tv_nsec
- NSEC_PER_SEC
>= 0) {
1500 tp
.tv_nsec
= tp
.tv_nsec
- NSEC_PER_SEC
;
1503 val
.uptime
= tp
.tv_sec
+ (tp
.tv_nsec
? 1 : 0);
1505 val
.loads
[0] = avenrun
[0] << (SI_LOAD_SHIFT
- FSHIFT
);
1506 val
.loads
[1] = avenrun
[1] << (SI_LOAD_SHIFT
- FSHIFT
);
1507 val
.loads
[2] = avenrun
[2] << (SI_LOAD_SHIFT
- FSHIFT
);
1509 val
.procs
= nr_threads
;
1510 } while (read_seqretry(&xtime_lock
, seq
));
1516 * If the sum of all the available memory (i.e. ram + swap)
1517 * is less than can be stored in a 32 bit unsigned long then
1518 * we can be binary compatible with 2.2.x kernels. If not,
1519 * well, in that case 2.2.x was broken anyways...
1521 * -Erik Andersen <andersee@debian.org>
1524 mem_total
= val
.totalram
+ val
.totalswap
;
1525 if (mem_total
< val
.totalram
|| mem_total
< val
.totalswap
)
1528 mem_unit
= val
.mem_unit
;
1529 while (mem_unit
> 1) {
1532 sav_total
= mem_total
;
1534 if (mem_total
< sav_total
)
1539 * If mem_total did not overflow, multiply all memory values by
1540 * val.mem_unit and set it to 1. This leaves things compatible
1541 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1546 val
.totalram
<<= bitcount
;
1547 val
.freeram
<<= bitcount
;
1548 val
.sharedram
<<= bitcount
;
1549 val
.bufferram
<<= bitcount
;
1550 val
.totalswap
<<= bitcount
;
1551 val
.freeswap
<<= bitcount
;
1552 val
.totalhigh
<<= bitcount
;
1553 val
.freehigh
<<= bitcount
;
1556 if (copy_to_user(info
, &val
, sizeof(struct sysinfo
)))
1562 static int __devinit
init_timers_cpu(int cpu
)
1566 static char __devinitdata tvec_base_done
[NR_CPUS
];
1568 if (!tvec_base_done
[cpu
]) {
1569 static char boot_done
;
1573 * The APs use this path later in boot
1575 base
= kmalloc_node(sizeof(*base
), GFP_KERNEL
,
1579 memset(base
, 0, sizeof(*base
));
1580 per_cpu(tvec_bases
, cpu
) = base
;
1583 * This is for the boot CPU - we use compile-time
1584 * static initialisation because per-cpu memory isn't
1585 * ready yet and because the memory allocators are not
1586 * initialised either.
1589 base
= &boot_tvec_bases
;
1591 tvec_base_done
[cpu
] = 1;
1593 base
= per_cpu(tvec_bases
, cpu
);
1596 spin_lock_init(&base
->lock
);
1597 for (j
= 0; j
< TVN_SIZE
; j
++) {
1598 INIT_LIST_HEAD(base
->tv5
.vec
+ j
);
1599 INIT_LIST_HEAD(base
->tv4
.vec
+ j
);
1600 INIT_LIST_HEAD(base
->tv3
.vec
+ j
);
1601 INIT_LIST_HEAD(base
->tv2
.vec
+ j
);
1603 for (j
= 0; j
< TVR_SIZE
; j
++)
1604 INIT_LIST_HEAD(base
->tv1
.vec
+ j
);
1606 base
->timer_jiffies
= jiffies
;
1610 #ifdef CONFIG_HOTPLUG_CPU
1611 static void migrate_timer_list(tvec_base_t
*new_base
, struct list_head
*head
)
1613 struct timer_list
*timer
;
1615 while (!list_empty(head
)) {
1616 timer
= list_entry(head
->next
, struct timer_list
, entry
);
1617 detach_timer(timer
, 0);
1618 timer
->base
= new_base
;
1619 internal_add_timer(new_base
, timer
);
1623 static void __devinit
migrate_timers(int cpu
)
1625 tvec_base_t
*old_base
;
1626 tvec_base_t
*new_base
;
1629 BUG_ON(cpu_online(cpu
));
1630 old_base
= per_cpu(tvec_bases
, cpu
);
1631 new_base
= get_cpu_var(tvec_bases
);
1633 local_irq_disable();
1634 spin_lock(&new_base
->lock
);
1635 spin_lock(&old_base
->lock
);
1637 BUG_ON(old_base
->running_timer
);
1639 for (i
= 0; i
< TVR_SIZE
; i
++)
1640 migrate_timer_list(new_base
, old_base
->tv1
.vec
+ i
);
1641 for (i
= 0; i
< TVN_SIZE
; i
++) {
1642 migrate_timer_list(new_base
, old_base
->tv2
.vec
+ i
);
1643 migrate_timer_list(new_base
, old_base
->tv3
.vec
+ i
);
1644 migrate_timer_list(new_base
, old_base
->tv4
.vec
+ i
);
1645 migrate_timer_list(new_base
, old_base
->tv5
.vec
+ i
);
1648 spin_unlock(&old_base
->lock
);
1649 spin_unlock(&new_base
->lock
);
1651 put_cpu_var(tvec_bases
);
1653 #endif /* CONFIG_HOTPLUG_CPU */
1655 static int __devinit
timer_cpu_notify(struct notifier_block
*self
,
1656 unsigned long action
, void *hcpu
)
1658 long cpu
= (long)hcpu
;
1660 case CPU_UP_PREPARE
:
1661 if (init_timers_cpu(cpu
) < 0)
1664 #ifdef CONFIG_HOTPLUG_CPU
1666 migrate_timers(cpu
);
1675 static struct notifier_block __devinitdata timers_nb
= {
1676 .notifier_call
= timer_cpu_notify
,
1680 void __init
init_timers(void)
1682 timer_cpu_notify(&timers_nb
, (unsigned long)CPU_UP_PREPARE
,
1683 (void *)(long)smp_processor_id());
1684 register_cpu_notifier(&timers_nb
);
1685 open_softirq(TIMER_SOFTIRQ
, run_timer_softirq
, NULL
);
1688 #ifdef CONFIG_TIME_INTERPOLATION
1690 struct time_interpolator
*time_interpolator __read_mostly
;
1691 static struct time_interpolator
*time_interpolator_list __read_mostly
;
1692 static DEFINE_SPINLOCK(time_interpolator_lock
);
1694 static inline u64
time_interpolator_get_cycles(unsigned int src
)
1696 unsigned long (*x
)(void);
1700 case TIME_SOURCE_FUNCTION
:
1701 x
= time_interpolator
->addr
;
1704 case TIME_SOURCE_MMIO64
:
1705 return readq_relaxed((void __iomem
*)time_interpolator
->addr
);
1707 case TIME_SOURCE_MMIO32
:
1708 return readl_relaxed((void __iomem
*)time_interpolator
->addr
);
1710 default: return get_cycles();
1714 static inline u64
time_interpolator_get_counter(int writelock
)
1716 unsigned int src
= time_interpolator
->source
;
1718 if (time_interpolator
->jitter
)
1724 lcycle
= time_interpolator
->last_cycle
;
1725 now
= time_interpolator_get_cycles(src
);
1726 if (lcycle
&& time_after(lcycle
, now
))
1729 /* When holding the xtime write lock, there's no need
1730 * to add the overhead of the cmpxchg. Readers are
1731 * force to retry until the write lock is released.
1734 time_interpolator
->last_cycle
= now
;
1737 /* Keep track of the last timer value returned. The use of cmpxchg here
1738 * will cause contention in an SMP environment.
1740 } while (unlikely(cmpxchg(&time_interpolator
->last_cycle
, lcycle
, now
) != lcycle
));
1744 return time_interpolator_get_cycles(src
);
1747 void time_interpolator_reset(void)
1749 time_interpolator
->offset
= 0;
1750 time_interpolator
->last_counter
= time_interpolator_get_counter(1);
1753 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1755 unsigned long time_interpolator_get_offset(void)
1757 /* If we do not have a time interpolator set up then just return zero */
1758 if (!time_interpolator
)
1761 return time_interpolator
->offset
+
1762 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator
);
1765 #define INTERPOLATOR_ADJUST 65536
1766 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1768 static void time_interpolator_update(long delta_nsec
)
1771 unsigned long offset
;
1773 /* If there is no time interpolator set up then do nothing */
1774 if (!time_interpolator
)
1778 * The interpolator compensates for late ticks by accumulating the late
1779 * time in time_interpolator->offset. A tick earlier than expected will
1780 * lead to a reset of the offset and a corresponding jump of the clock
1781 * forward. Again this only works if the interpolator clock is running
1782 * slightly slower than the regular clock and the tuning logic insures
1786 counter
= time_interpolator_get_counter(1);
1787 offset
= time_interpolator
->offset
+
1788 GET_TI_NSECS(counter
, time_interpolator
);
1790 if (delta_nsec
< 0 || (unsigned long) delta_nsec
< offset
)
1791 time_interpolator
->offset
= offset
- delta_nsec
;
1793 time_interpolator
->skips
++;
1794 time_interpolator
->ns_skipped
+= delta_nsec
- offset
;
1795 time_interpolator
->offset
= 0;
1797 time_interpolator
->last_counter
= counter
;
1799 /* Tuning logic for time interpolator invoked every minute or so.
1800 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1801 * Increase interpolator clock speed if we skip too much time.
1803 if (jiffies
% INTERPOLATOR_ADJUST
== 0)
1805 if (time_interpolator
->skips
== 0 && time_interpolator
->offset
> tick_nsec
)
1806 time_interpolator
->nsec_per_cyc
--;
1807 if (time_interpolator
->ns_skipped
> INTERPOLATOR_MAX_SKIP
&& time_interpolator
->offset
== 0)
1808 time_interpolator
->nsec_per_cyc
++;
1809 time_interpolator
->skips
= 0;
1810 time_interpolator
->ns_skipped
= 0;
1815 is_better_time_interpolator(struct time_interpolator
*new)
1817 if (!time_interpolator
)
1819 return new->frequency
> 2*time_interpolator
->frequency
||
1820 (unsigned long)new->drift
< (unsigned long)time_interpolator
->drift
;
1824 register_time_interpolator(struct time_interpolator
*ti
)
1826 unsigned long flags
;
1829 BUG_ON(ti
->frequency
== 0 || ti
->mask
== 0);
1831 ti
->nsec_per_cyc
= ((u64
)NSEC_PER_SEC
<< ti
->shift
) / ti
->frequency
;
1832 spin_lock(&time_interpolator_lock
);
1833 write_seqlock_irqsave(&xtime_lock
, flags
);
1834 if (is_better_time_interpolator(ti
)) {
1835 time_interpolator
= ti
;
1836 time_interpolator_reset();
1838 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1840 ti
->next
= time_interpolator_list
;
1841 time_interpolator_list
= ti
;
1842 spin_unlock(&time_interpolator_lock
);
1846 unregister_time_interpolator(struct time_interpolator
*ti
)
1848 struct time_interpolator
*curr
, **prev
;
1849 unsigned long flags
;
1851 spin_lock(&time_interpolator_lock
);
1852 prev
= &time_interpolator_list
;
1853 for (curr
= *prev
; curr
; curr
= curr
->next
) {
1861 write_seqlock_irqsave(&xtime_lock
, flags
);
1862 if (ti
== time_interpolator
) {
1863 /* we lost the best time-interpolator: */
1864 time_interpolator
= NULL
;
1865 /* find the next-best interpolator */
1866 for (curr
= time_interpolator_list
; curr
; curr
= curr
->next
)
1867 if (is_better_time_interpolator(curr
))
1868 time_interpolator
= curr
;
1869 time_interpolator_reset();
1871 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1872 spin_unlock(&time_interpolator_lock
);
1874 #endif /* CONFIG_TIME_INTERPOLATION */
1877 * msleep - sleep safely even with waitqueue interruptions
1878 * @msecs: Time in milliseconds to sleep for
1880 void msleep(unsigned int msecs
)
1882 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1885 timeout
= schedule_timeout_uninterruptible(timeout
);
1888 EXPORT_SYMBOL(msleep
);
1891 * msleep_interruptible - sleep waiting for signals
1892 * @msecs: Time in milliseconds to sleep for
1894 unsigned long msleep_interruptible(unsigned int msecs
)
1896 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1898 while (timeout
&& !signal_pending(current
))
1899 timeout
= schedule_timeout_interruptible(timeout
);
1900 return jiffies_to_msecs(timeout
);
1903 EXPORT_SYMBOL(msleep_interruptible
);