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:
58 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
59 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
60 #define TVN_SIZE (1 << TVN_BITS)
61 #define TVR_SIZE (1 << TVR_BITS)
62 #define TVN_MASK (TVN_SIZE - 1)
63 #define TVR_MASK (TVR_SIZE - 1)
67 struct timer_list
*running_timer
;
70 typedef struct tvec_s
{
71 struct list_head vec
[TVN_SIZE
];
74 typedef struct tvec_root_s
{
75 struct list_head vec
[TVR_SIZE
];
78 struct tvec_t_base_s
{
79 struct timer_base_s t_base
;
80 unsigned long timer_jiffies
;
86 } ____cacheline_aligned_in_smp
;
88 typedef struct tvec_t_base_s tvec_base_t
;
89 static DEFINE_PER_CPU(tvec_base_t
, tvec_bases
);
91 static inline void set_running_timer(tvec_base_t
*base
,
92 struct timer_list
*timer
)
95 base
->t_base
.running_timer
= timer
;
99 static void internal_add_timer(tvec_base_t
*base
, struct timer_list
*timer
)
101 unsigned long expires
= timer
->expires
;
102 unsigned long idx
= expires
- base
->timer_jiffies
;
103 struct list_head
*vec
;
105 if (idx
< TVR_SIZE
) {
106 int i
= expires
& TVR_MASK
;
107 vec
= base
->tv1
.vec
+ i
;
108 } else if (idx
< 1 << (TVR_BITS
+ TVN_BITS
)) {
109 int i
= (expires
>> TVR_BITS
) & TVN_MASK
;
110 vec
= base
->tv2
.vec
+ i
;
111 } else if (idx
< 1 << (TVR_BITS
+ 2 * TVN_BITS
)) {
112 int i
= (expires
>> (TVR_BITS
+ TVN_BITS
)) & TVN_MASK
;
113 vec
= base
->tv3
.vec
+ i
;
114 } else if (idx
< 1 << (TVR_BITS
+ 3 * TVN_BITS
)) {
115 int i
= (expires
>> (TVR_BITS
+ 2 * TVN_BITS
)) & TVN_MASK
;
116 vec
= base
->tv4
.vec
+ i
;
117 } else if ((signed long) idx
< 0) {
119 * Can happen if you add a timer with expires == jiffies,
120 * or you set a timer to go off in the past
122 vec
= base
->tv1
.vec
+ (base
->timer_jiffies
& TVR_MASK
);
125 /* If the timeout is larger than 0xffffffff on 64-bit
126 * architectures then we use the maximum timeout:
128 if (idx
> 0xffffffffUL
) {
130 expires
= idx
+ base
->timer_jiffies
;
132 i
= (expires
>> (TVR_BITS
+ 3 * TVN_BITS
)) & TVN_MASK
;
133 vec
= base
->tv5
.vec
+ i
;
138 list_add_tail(&timer
->entry
, vec
);
141 typedef struct timer_base_s timer_base_t
;
143 * Used by TIMER_INITIALIZER, we can't use per_cpu(tvec_bases)
144 * at compile time, and we need timer->base to lock the timer.
146 timer_base_t __init_timer_base
147 ____cacheline_aligned_in_smp
= { .lock
= SPIN_LOCK_UNLOCKED
};
148 EXPORT_SYMBOL(__init_timer_base
);
151 * init_timer - initialize a timer.
152 * @timer: the timer to be initialized
154 * init_timer() must be done to a timer prior calling *any* of the
155 * other timer functions.
157 void fastcall
init_timer(struct timer_list
*timer
)
159 timer
->entry
.next
= NULL
;
160 timer
->base
= &per_cpu(tvec_bases
, raw_smp_processor_id()).t_base
;
162 EXPORT_SYMBOL(init_timer
);
164 static inline void detach_timer(struct timer_list
*timer
,
167 struct list_head
*entry
= &timer
->entry
;
169 __list_del(entry
->prev
, entry
->next
);
172 entry
->prev
= LIST_POISON2
;
176 * We are using hashed locking: holding per_cpu(tvec_bases).t_base.lock
177 * means that all timers which are tied to this base via timer->base are
178 * locked, and the base itself is locked too.
180 * So __run_timers/migrate_timers can safely modify all timers which could
181 * be found on ->tvX lists.
183 * When the timer's base is locked, and the timer removed from list, it is
184 * possible to set timer->base = NULL and drop the lock: the timer remains
187 static timer_base_t
*lock_timer_base(struct timer_list
*timer
,
188 unsigned long *flags
)
194 if (likely(base
!= NULL
)) {
195 spin_lock_irqsave(&base
->lock
, *flags
);
196 if (likely(base
== timer
->base
))
198 /* The timer has migrated to another CPU */
199 spin_unlock_irqrestore(&base
->lock
, *flags
);
205 int __mod_timer(struct timer_list
*timer
, unsigned long expires
)
208 tvec_base_t
*new_base
;
212 BUG_ON(!timer
->function
);
214 base
= lock_timer_base(timer
, &flags
);
216 if (timer_pending(timer
)) {
217 detach_timer(timer
, 0);
221 new_base
= &__get_cpu_var(tvec_bases
);
223 if (base
!= &new_base
->t_base
) {
225 * We are trying to schedule the timer on the local CPU.
226 * However we can't change timer's base while it is running,
227 * otherwise del_timer_sync() can't detect that the timer's
228 * handler yet has not finished. This also guarantees that
229 * the timer is serialized wrt itself.
231 if (unlikely(base
->running_timer
== timer
)) {
232 /* The timer remains on a former base */
233 new_base
= container_of(base
, tvec_base_t
, t_base
);
235 /* See the comment in lock_timer_base() */
237 spin_unlock(&base
->lock
);
238 spin_lock(&new_base
->t_base
.lock
);
239 timer
->base
= &new_base
->t_base
;
243 timer
->expires
= expires
;
244 internal_add_timer(new_base
, timer
);
245 spin_unlock_irqrestore(&new_base
->t_base
.lock
, flags
);
250 EXPORT_SYMBOL(__mod_timer
);
253 * add_timer_on - start a timer on a particular CPU
254 * @timer: the timer to be added
255 * @cpu: the CPU to start it on
257 * This is not very scalable on SMP. Double adds are not possible.
259 void add_timer_on(struct timer_list
*timer
, int cpu
)
261 tvec_base_t
*base
= &per_cpu(tvec_bases
, cpu
);
264 BUG_ON(timer_pending(timer
) || !timer
->function
);
265 spin_lock_irqsave(&base
->t_base
.lock
, flags
);
266 timer
->base
= &base
->t_base
;
267 internal_add_timer(base
, timer
);
268 spin_unlock_irqrestore(&base
->t_base
.lock
, flags
);
273 * mod_timer - modify a timer's timeout
274 * @timer: the timer to be modified
276 * mod_timer is a more efficient way to update the expire field of an
277 * active timer (if the timer is inactive it will be activated)
279 * mod_timer(timer, expires) is equivalent to:
281 * del_timer(timer); timer->expires = expires; add_timer(timer);
283 * Note that if there are multiple unserialized concurrent users of the
284 * same timer, then mod_timer() is the only safe way to modify the timeout,
285 * since add_timer() cannot modify an already running timer.
287 * The function returns whether it has modified a pending timer or not.
288 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
289 * active timer returns 1.)
291 int mod_timer(struct timer_list
*timer
, unsigned long expires
)
293 BUG_ON(!timer
->function
);
296 * This is a common optimization triggered by the
297 * networking code - if the timer is re-modified
298 * to be the same thing then just return:
300 if (timer
->expires
== expires
&& timer_pending(timer
))
303 return __mod_timer(timer
, expires
);
306 EXPORT_SYMBOL(mod_timer
);
309 * del_timer - deactive a timer.
310 * @timer: the timer to be deactivated
312 * del_timer() deactivates a timer - this works on both active and inactive
315 * The function returns whether it has deactivated a pending timer or not.
316 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
317 * active timer returns 1.)
319 int del_timer(struct timer_list
*timer
)
325 if (timer_pending(timer
)) {
326 base
= lock_timer_base(timer
, &flags
);
327 if (timer_pending(timer
)) {
328 detach_timer(timer
, 1);
331 spin_unlock_irqrestore(&base
->lock
, flags
);
337 EXPORT_SYMBOL(del_timer
);
341 * This function tries to deactivate a timer. Upon successful (ret >= 0)
342 * exit the timer is not queued and the handler is not running on any CPU.
344 * It must not be called from interrupt contexts.
346 int try_to_del_timer_sync(struct timer_list
*timer
)
352 base
= lock_timer_base(timer
, &flags
);
354 if (base
->running_timer
== timer
)
358 if (timer_pending(timer
)) {
359 detach_timer(timer
, 1);
363 spin_unlock_irqrestore(&base
->lock
, flags
);
369 * del_timer_sync - deactivate a timer and wait for the handler to finish.
370 * @timer: the timer to be deactivated
372 * This function only differs from del_timer() on SMP: besides deactivating
373 * the timer it also makes sure the handler has finished executing on other
376 * Synchronization rules: callers must prevent restarting of the timer,
377 * otherwise this function is meaningless. It must not be called from
378 * interrupt contexts. The caller must not hold locks which would prevent
379 * completion of the timer's handler. The timer's handler must not call
380 * add_timer_on(). Upon exit the timer is not queued and the handler is
381 * not running on any CPU.
383 * The function returns whether it has deactivated a pending timer or not.
385 int del_timer_sync(struct timer_list
*timer
)
388 int ret
= try_to_del_timer_sync(timer
);
394 EXPORT_SYMBOL(del_timer_sync
);
397 static int cascade(tvec_base_t
*base
, tvec_t
*tv
, int index
)
399 /* cascade all the timers from tv up one level */
400 struct list_head
*head
, *curr
;
402 head
= tv
->vec
+ index
;
405 * We are removing _all_ timers from the list, so we don't have to
406 * detach them individually, just clear the list afterwards.
408 while (curr
!= head
) {
409 struct timer_list
*tmp
;
411 tmp
= list_entry(curr
, struct timer_list
, entry
);
412 BUG_ON(tmp
->base
!= &base
->t_base
);
414 internal_add_timer(base
, tmp
);
416 INIT_LIST_HEAD(head
);
422 * __run_timers - run all expired timers (if any) on this CPU.
423 * @base: the timer vector to be processed.
425 * This function cascades all vectors and executes all expired timer
428 #define INDEX(N) (base->timer_jiffies >> (TVR_BITS + N * TVN_BITS)) & TVN_MASK
430 static inline void __run_timers(tvec_base_t
*base
)
432 struct timer_list
*timer
;
434 spin_lock_irq(&base
->t_base
.lock
);
435 while (time_after_eq(jiffies
, base
->timer_jiffies
)) {
436 struct list_head work_list
= LIST_HEAD_INIT(work_list
);
437 struct list_head
*head
= &work_list
;
438 int index
= base
->timer_jiffies
& TVR_MASK
;
444 (!cascade(base
, &base
->tv2
, INDEX(0))) &&
445 (!cascade(base
, &base
->tv3
, INDEX(1))) &&
446 !cascade(base
, &base
->tv4
, INDEX(2)))
447 cascade(base
, &base
->tv5
, INDEX(3));
448 ++base
->timer_jiffies
;
449 list_splice_init(base
->tv1
.vec
+ index
, &work_list
);
450 while (!list_empty(head
)) {
451 void (*fn
)(unsigned long);
454 timer
= list_entry(head
->next
,struct timer_list
,entry
);
455 fn
= timer
->function
;
458 set_running_timer(base
, timer
);
459 detach_timer(timer
, 1);
460 spin_unlock_irq(&base
->t_base
.lock
);
462 int preempt_count
= preempt_count();
464 if (preempt_count
!= preempt_count()) {
465 printk(KERN_WARNING
"huh, entered %p "
466 "with preempt_count %08x, exited"
473 spin_lock_irq(&base
->t_base
.lock
);
476 set_running_timer(base
, NULL
);
477 spin_unlock_irq(&base
->t_base
.lock
);
480 #ifdef CONFIG_NO_IDLE_HZ
482 * Find out when the next timer event is due to happen. This
483 * is used on S/390 to stop all activity when a cpus is idle.
484 * This functions needs to be called disabled.
486 unsigned long next_timer_interrupt(void)
489 struct list_head
*list
;
490 struct timer_list
*nte
;
491 unsigned long expires
;
492 unsigned long hr_expires
= MAX_JIFFY_OFFSET
;
497 hr_delta
= hrtimer_get_next_event();
498 if (hr_delta
.tv64
!= KTIME_MAX
) {
499 struct timespec tsdelta
;
500 tsdelta
= ktime_to_timespec(hr_delta
);
501 hr_expires
= timespec_to_jiffies(&tsdelta
);
503 return hr_expires
+ jiffies
;
505 hr_expires
+= jiffies
;
507 base
= &__get_cpu_var(tvec_bases
);
508 spin_lock(&base
->t_base
.lock
);
509 expires
= base
->timer_jiffies
+ (LONG_MAX
>> 1);
512 /* Look for timer events in tv1. */
513 j
= base
->timer_jiffies
& TVR_MASK
;
515 list_for_each_entry(nte
, base
->tv1
.vec
+ j
, entry
) {
516 expires
= nte
->expires
;
517 if (j
< (base
->timer_jiffies
& TVR_MASK
))
518 list
= base
->tv2
.vec
+ (INDEX(0));
521 j
= (j
+ 1) & TVR_MASK
;
522 } while (j
!= (base
->timer_jiffies
& TVR_MASK
));
525 varray
[0] = &base
->tv2
;
526 varray
[1] = &base
->tv3
;
527 varray
[2] = &base
->tv4
;
528 varray
[3] = &base
->tv5
;
529 for (i
= 0; i
< 4; i
++) {
532 if (list_empty(varray
[i
]->vec
+ j
)) {
533 j
= (j
+ 1) & TVN_MASK
;
536 list_for_each_entry(nte
, varray
[i
]->vec
+ j
, entry
)
537 if (time_before(nte
->expires
, expires
))
538 expires
= nte
->expires
;
539 if (j
< (INDEX(i
)) && i
< 3)
540 list
= varray
[i
+ 1]->vec
+ (INDEX(i
+ 1));
542 } while (j
!= (INDEX(i
)));
547 * The search wrapped. We need to look at the next list
548 * from next tv element that would cascade into tv element
549 * where we found the timer element.
551 list_for_each_entry(nte
, list
, entry
) {
552 if (time_before(nte
->expires
, expires
))
553 expires
= nte
->expires
;
556 spin_unlock(&base
->t_base
.lock
);
558 if (time_before(hr_expires
, expires
))
565 /******************************************************************/
568 * Timekeeping variables
570 unsigned long tick_usec
= TICK_USEC
; /* USER_HZ period (usec) */
571 unsigned long tick_nsec
= TICK_NSEC
; /* ACTHZ period (nsec) */
575 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
576 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
577 * at zero at system boot time, so wall_to_monotonic will be negative,
578 * however, we will ALWAYS keep the tv_nsec part positive so we can use
579 * the usual normalization.
581 struct timespec xtime
__attribute__ ((aligned (16)));
582 struct timespec wall_to_monotonic
__attribute__ ((aligned (16)));
584 EXPORT_SYMBOL(xtime
);
586 /* Don't completely fail for HZ > 500. */
587 int tickadj
= 500/HZ
? : 1; /* microsecs */
591 * phase-lock loop variables
593 /* TIME_ERROR prevents overwriting the CMOS clock */
594 int time_state
= TIME_OK
; /* clock synchronization status */
595 int time_status
= STA_UNSYNC
; /* clock status bits */
596 long time_offset
; /* time adjustment (us) */
597 long time_constant
= 2; /* pll time constant */
598 long time_tolerance
= MAXFREQ
; /* frequency tolerance (ppm) */
599 long time_precision
= 1; /* clock precision (us) */
600 long time_maxerror
= NTP_PHASE_LIMIT
; /* maximum error (us) */
601 long time_esterror
= NTP_PHASE_LIMIT
; /* estimated error (us) */
602 static long time_phase
; /* phase offset (scaled us) */
603 long time_freq
= (((NSEC_PER_SEC
+ HZ
/2) % HZ
- HZ
/2) << SHIFT_USEC
) / NSEC_PER_USEC
;
604 /* frequency offset (scaled ppm)*/
605 static long time_adj
; /* tick adjust (scaled 1 / HZ) */
606 long time_reftime
; /* time at last adjustment (s) */
608 long time_next_adjust
;
611 * this routine handles the overflow of the microsecond field
613 * The tricky bits of code to handle the accurate clock support
614 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
615 * They were originally developed for SUN and DEC kernels.
616 * All the kudos should go to Dave for this stuff.
619 static void second_overflow(void)
623 /* Bump the maxerror field */
624 time_maxerror
+= time_tolerance
>> SHIFT_USEC
;
625 if (time_maxerror
> NTP_PHASE_LIMIT
) {
626 time_maxerror
= NTP_PHASE_LIMIT
;
627 time_status
|= STA_UNSYNC
;
631 * Leap second processing. If in leap-insert state at the end of the
632 * day, the system clock is set back one second; if in leap-delete
633 * state, the system clock is set ahead one second. The microtime()
634 * routine or external clock driver will insure that reported time is
635 * always monotonic. The ugly divides should be replaced.
637 switch (time_state
) {
639 if (time_status
& STA_INS
)
640 time_state
= TIME_INS
;
641 else if (time_status
& STA_DEL
)
642 time_state
= TIME_DEL
;
645 if (xtime
.tv_sec
% 86400 == 0) {
647 wall_to_monotonic
.tv_sec
++;
649 * The timer interpolator will make time change
650 * gradually instead of an immediate jump by one second
652 time_interpolator_update(-NSEC_PER_SEC
);
653 time_state
= TIME_OOP
;
655 printk(KERN_NOTICE
"Clock: inserting leap second "
660 if ((xtime
.tv_sec
+ 1) % 86400 == 0) {
662 wall_to_monotonic
.tv_sec
--;
664 * Use of time interpolator for a gradual change of
667 time_interpolator_update(NSEC_PER_SEC
);
668 time_state
= TIME_WAIT
;
670 printk(KERN_NOTICE
"Clock: deleting leap second "
675 time_state
= TIME_WAIT
;
678 if (!(time_status
& (STA_INS
| STA_DEL
)))
679 time_state
= TIME_OK
;
683 * Compute the phase adjustment for the next second. In PLL mode, the
684 * offset is reduced by a fixed factor times the time constant. In FLL
685 * mode the offset is used directly. In either mode, the maximum phase
686 * adjustment for each second is clamped so as to spread the adjustment
687 * over not more than the number of seconds between updates.
690 if (!(time_status
& STA_FLL
))
691 ltemp
= shift_right(ltemp
, SHIFT_KG
+ time_constant
);
692 ltemp
= min(ltemp
, (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
);
693 ltemp
= max(ltemp
, -(MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
);
694 time_offset
-= ltemp
;
695 time_adj
= ltemp
<< (SHIFT_SCALE
- SHIFT_HZ
- SHIFT_UPDATE
);
698 * Compute the frequency estimate and additional phase adjustment due
699 * to frequency error for the next second. When the PPS signal is
700 * engaged, gnaw on the watchdog counter and update the frequency
701 * computed by the pll and the PPS signal.
704 if (pps_valid
== PPS_VALID
) { /* PPS signal lost */
705 pps_jitter
= MAXTIME
;
706 pps_stabil
= MAXFREQ
;
707 time_status
&= ~(STA_PPSSIGNAL
| STA_PPSJITTER
|
708 STA_PPSWANDER
| STA_PPSERROR
);
710 ltemp
= time_freq
+ pps_freq
;
711 time_adj
+= shift_right(ltemp
,(SHIFT_USEC
+ SHIFT_HZ
- SHIFT_SCALE
));
715 * Compensate for (HZ==100) != (1 << SHIFT_HZ). Add 25% and 3.125% to
716 * get 128.125; => only 0.125% error (p. 14)
718 time_adj
+= shift_right(time_adj
, 2) + shift_right(time_adj
, 5);
722 * Compensate for (HZ==250) != (1 << SHIFT_HZ). Add 1.5625% and
723 * 0.78125% to get 255.85938; => only 0.05% error (p. 14)
725 time_adj
+= shift_right(time_adj
, 6) + shift_right(time_adj
, 7);
729 * Compensate for (HZ==1000) != (1 << SHIFT_HZ). Add 1.5625% and
730 * 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
732 time_adj
+= shift_right(time_adj
, 6) + shift_right(time_adj
, 7);
737 * Returns how many microseconds we need to add to xtime this tick
738 * in doing an adjustment requested with adjtime.
740 static long adjtime_adjustment(void)
742 long time_adjust_step
;
744 time_adjust_step
= time_adjust
;
745 if (time_adjust_step
) {
747 * We are doing an adjtime thing. Prepare time_adjust_step to
748 * be within bounds. Note that a positive time_adjust means we
749 * want the clock to run faster.
751 * Limit the amount of the step to be in the range
752 * -tickadj .. +tickadj
754 time_adjust_step
= min(time_adjust_step
, (long)tickadj
);
755 time_adjust_step
= max(time_adjust_step
, (long)-tickadj
);
757 return time_adjust_step
;
760 /* in the NTP reference this is called "hardclock()" */
761 static void update_wall_time_one_tick(void)
763 long time_adjust_step
, delta_nsec
;
765 time_adjust_step
= adjtime_adjustment();
766 if (time_adjust_step
)
767 /* Reduce by this step the amount of time left */
768 time_adjust
-= time_adjust_step
;
769 delta_nsec
= tick_nsec
+ time_adjust_step
* 1000;
771 * Advance the phase, once it gets to one microsecond, then
772 * advance the tick more.
774 time_phase
+= time_adj
;
775 if ((time_phase
>= FINENSEC
) || (time_phase
<= -FINENSEC
)) {
776 long ltemp
= shift_right(time_phase
, (SHIFT_SCALE
- 10));
777 time_phase
-= ltemp
<< (SHIFT_SCALE
- 10);
780 xtime
.tv_nsec
+= delta_nsec
;
781 time_interpolator_update(delta_nsec
);
783 /* Changes by adjtime() do not take effect till next tick. */
784 if (time_next_adjust
!= 0) {
785 time_adjust
= time_next_adjust
;
786 time_next_adjust
= 0;
791 * Return how long ticks are at the moment, that is, how much time
792 * update_wall_time_one_tick will add to xtime next time we call it
793 * (assuming no calls to do_adjtimex in the meantime).
794 * The return value is in fixed-point nanoseconds with SHIFT_SCALE-10
795 * bits to the right of the binary point.
796 * This function has no side-effects.
798 u64
current_tick_length(void)
802 delta_nsec
= tick_nsec
+ adjtime_adjustment() * 1000;
803 return ((u64
) delta_nsec
<< (SHIFT_SCALE
- 10)) + time_adj
;
807 * Using a loop looks inefficient, but "ticks" is
808 * usually just one (we shouldn't be losing ticks,
809 * we're doing this this way mainly for interrupt
810 * latency reasons, not because we think we'll
811 * have lots of lost timer ticks
813 static void update_wall_time(unsigned long ticks
)
817 update_wall_time_one_tick();
818 if (xtime
.tv_nsec
>= 1000000000) {
819 xtime
.tv_nsec
-= 1000000000;
827 * Called from the timer interrupt handler to charge one tick to the current
828 * process. user_tick is 1 if the tick is user time, 0 for system.
830 void update_process_times(int user_tick
)
832 struct task_struct
*p
= current
;
833 int cpu
= smp_processor_id();
835 /* Note: this timer irq context must be accounted for as well. */
837 account_user_time(p
, jiffies_to_cputime(1));
839 account_system_time(p
, HARDIRQ_OFFSET
, jiffies_to_cputime(1));
841 if (rcu_pending(cpu
))
842 rcu_check_callbacks(cpu
, user_tick
);
844 run_posix_cpu_timers(p
);
848 * Nr of active tasks - counted in fixed-point numbers
850 static unsigned long count_active_tasks(void)
852 return (nr_running() + nr_uninterruptible()) * FIXED_1
;
856 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
857 * imply that avenrun[] is the standard name for this kind of thing.
858 * Nothing else seems to be standardized: the fractional size etc
859 * all seem to differ on different machines.
861 * Requires xtime_lock to access.
863 unsigned long avenrun
[3];
865 EXPORT_SYMBOL(avenrun
);
868 * calc_load - given tick count, update the avenrun load estimates.
869 * This is called while holding a write_lock on xtime_lock.
871 static inline void calc_load(unsigned long ticks
)
873 unsigned long active_tasks
; /* fixed-point */
874 static int count
= LOAD_FREQ
;
879 active_tasks
= count_active_tasks();
880 CALC_LOAD(avenrun
[0], EXP_1
, active_tasks
);
881 CALC_LOAD(avenrun
[1], EXP_5
, active_tasks
);
882 CALC_LOAD(avenrun
[2], EXP_15
, active_tasks
);
886 /* jiffies at the most recent update of wall time */
887 unsigned long wall_jiffies
= INITIAL_JIFFIES
;
890 * This read-write spinlock protects us from races in SMP while
891 * playing with xtime and avenrun.
893 #ifndef ARCH_HAVE_XTIME_LOCK
894 seqlock_t xtime_lock __cacheline_aligned_in_smp
= SEQLOCK_UNLOCKED
;
896 EXPORT_SYMBOL(xtime_lock
);
900 * This function runs timers and the timer-tq in bottom half context.
902 static void run_timer_softirq(struct softirq_action
*h
)
904 tvec_base_t
*base
= &__get_cpu_var(tvec_bases
);
906 hrtimer_run_queues();
907 if (time_after_eq(jiffies
, base
->timer_jiffies
))
912 * Called by the local, per-CPU timer interrupt on SMP.
914 void run_local_timers(void)
916 raise_softirq(TIMER_SOFTIRQ
);
920 * Called by the timer interrupt. xtime_lock must already be taken
923 static inline void update_times(void)
927 ticks
= jiffies
- wall_jiffies
;
929 wall_jiffies
+= ticks
;
930 update_wall_time(ticks
);
936 * The 64-bit jiffies value is not atomic - you MUST NOT read it
937 * without sampling the sequence number in xtime_lock.
938 * jiffies is defined in the linker script...
941 void do_timer(struct pt_regs
*regs
)
944 /* prevent loading jiffies before storing new jiffies_64 value. */
947 softlockup_tick(regs
);
950 #ifdef __ARCH_WANT_SYS_ALARM
953 * For backwards compatibility? This can be done in libc so Alpha
954 * and all newer ports shouldn't need it.
956 asmlinkage
unsigned long sys_alarm(unsigned int seconds
)
958 struct itimerval it_new
, it_old
;
959 unsigned int oldalarm
;
961 it_new
.it_interval
.tv_sec
= it_new
.it_interval
.tv_usec
= 0;
962 it_new
.it_value
.tv_sec
= seconds
;
963 it_new
.it_value
.tv_usec
= 0;
964 do_setitimer(ITIMER_REAL
, &it_new
, &it_old
);
965 oldalarm
= it_old
.it_value
.tv_sec
;
966 /* ehhh.. We can't return 0 if we have an alarm pending.. */
967 /* And we'd better return too much than too little anyway */
968 if ((!oldalarm
&& it_old
.it_value
.tv_usec
) || it_old
.it_value
.tv_usec
>= 500000)
978 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
979 * should be moved into arch/i386 instead?
983 * sys_getpid - return the thread group id of the current process
985 * Note, despite the name, this returns the tgid not the pid. The tgid and
986 * the pid are identical unless CLONE_THREAD was specified on clone() in
987 * which case the tgid is the same in all threads of the same group.
989 * This is SMP safe as current->tgid does not change.
991 asmlinkage
long sys_getpid(void)
993 return current
->tgid
;
997 * Accessing ->group_leader->real_parent is not SMP-safe, it could
998 * change from under us. However, rather than getting any lock
999 * we can use an optimistic algorithm: get the parent
1000 * pid, and go back and check that the parent is still
1001 * the same. If it has changed (which is extremely unlikely
1002 * indeed), we just try again..
1004 * NOTE! This depends on the fact that even if we _do_
1005 * get an old value of "parent", we can happily dereference
1006 * the pointer (it was and remains a dereferencable kernel pointer
1007 * no matter what): we just can't necessarily trust the result
1008 * until we know that the parent pointer is valid.
1010 * NOTE2: ->group_leader never changes from under us.
1012 asmlinkage
long sys_getppid(void)
1015 struct task_struct
*me
= current
;
1016 struct task_struct
*parent
;
1018 parent
= me
->group_leader
->real_parent
;
1021 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1023 struct task_struct
*old
= parent
;
1026 * Make sure we read the pid before re-reading the
1030 parent
= me
->group_leader
->real_parent
;
1040 asmlinkage
long sys_getuid(void)
1042 /* Only we change this so SMP safe */
1043 return current
->uid
;
1046 asmlinkage
long sys_geteuid(void)
1048 /* Only we change this so SMP safe */
1049 return current
->euid
;
1052 asmlinkage
long sys_getgid(void)
1054 /* Only we change this so SMP safe */
1055 return current
->gid
;
1058 asmlinkage
long sys_getegid(void)
1060 /* Only we change this so SMP safe */
1061 return current
->egid
;
1066 static void process_timeout(unsigned long __data
)
1068 wake_up_process((task_t
*)__data
);
1072 * schedule_timeout - sleep until timeout
1073 * @timeout: timeout value in jiffies
1075 * Make the current task sleep until @timeout jiffies have
1076 * elapsed. The routine will return immediately unless
1077 * the current task state has been set (see set_current_state()).
1079 * You can set the task state as follows -
1081 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1082 * pass before the routine returns. The routine will return 0
1084 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1085 * delivered to the current task. In this case the remaining time
1086 * in jiffies will be returned, or 0 if the timer expired in time
1088 * The current task state is guaranteed to be TASK_RUNNING when this
1091 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1092 * the CPU away without a bound on the timeout. In this case the return
1093 * value will be %MAX_SCHEDULE_TIMEOUT.
1095 * In all cases the return value is guaranteed to be non-negative.
1097 fastcall
signed long __sched
schedule_timeout(signed long timeout
)
1099 struct timer_list timer
;
1100 unsigned long expire
;
1104 case MAX_SCHEDULE_TIMEOUT
:
1106 * These two special cases are useful to be comfortable
1107 * in the caller. Nothing more. We could take
1108 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1109 * but I' d like to return a valid offset (>=0) to allow
1110 * the caller to do everything it want with the retval.
1116 * Another bit of PARANOID. Note that the retval will be
1117 * 0 since no piece of kernel is supposed to do a check
1118 * for a negative retval of schedule_timeout() (since it
1119 * should never happens anyway). You just have the printk()
1120 * that will tell you if something is gone wrong and where.
1124 printk(KERN_ERR
"schedule_timeout: wrong timeout "
1125 "value %lx from %p\n", timeout
,
1126 __builtin_return_address(0));
1127 current
->state
= TASK_RUNNING
;
1132 expire
= timeout
+ jiffies
;
1134 setup_timer(&timer
, process_timeout
, (unsigned long)current
);
1135 __mod_timer(&timer
, expire
);
1137 del_singleshot_timer_sync(&timer
);
1139 timeout
= expire
- jiffies
;
1142 return timeout
< 0 ? 0 : timeout
;
1144 EXPORT_SYMBOL(schedule_timeout
);
1147 * We can use __set_current_state() here because schedule_timeout() calls
1148 * schedule() unconditionally.
1150 signed long __sched
schedule_timeout_interruptible(signed long timeout
)
1152 __set_current_state(TASK_INTERRUPTIBLE
);
1153 return schedule_timeout(timeout
);
1155 EXPORT_SYMBOL(schedule_timeout_interruptible
);
1157 signed long __sched
schedule_timeout_uninterruptible(signed long timeout
)
1159 __set_current_state(TASK_UNINTERRUPTIBLE
);
1160 return schedule_timeout(timeout
);
1162 EXPORT_SYMBOL(schedule_timeout_uninterruptible
);
1164 /* Thread ID - the internal kernel "pid" */
1165 asmlinkage
long sys_gettid(void)
1167 return current
->pid
;
1171 * sys_sysinfo - fill in sysinfo struct
1173 asmlinkage
long sys_sysinfo(struct sysinfo __user
*info
)
1176 unsigned long mem_total
, sav_total
;
1177 unsigned int mem_unit
, bitcount
;
1180 memset((char *)&val
, 0, sizeof(struct sysinfo
));
1184 seq
= read_seqbegin(&xtime_lock
);
1187 * This is annoying. The below is the same thing
1188 * posix_get_clock_monotonic() does, but it wants to
1189 * take the lock which we want to cover the loads stuff
1193 getnstimeofday(&tp
);
1194 tp
.tv_sec
+= wall_to_monotonic
.tv_sec
;
1195 tp
.tv_nsec
+= wall_to_monotonic
.tv_nsec
;
1196 if (tp
.tv_nsec
- NSEC_PER_SEC
>= 0) {
1197 tp
.tv_nsec
= tp
.tv_nsec
- NSEC_PER_SEC
;
1200 val
.uptime
= tp
.tv_sec
+ (tp
.tv_nsec
? 1 : 0);
1202 val
.loads
[0] = avenrun
[0] << (SI_LOAD_SHIFT
- FSHIFT
);
1203 val
.loads
[1] = avenrun
[1] << (SI_LOAD_SHIFT
- FSHIFT
);
1204 val
.loads
[2] = avenrun
[2] << (SI_LOAD_SHIFT
- FSHIFT
);
1206 val
.procs
= nr_threads
;
1207 } while (read_seqretry(&xtime_lock
, seq
));
1213 * If the sum of all the available memory (i.e. ram + swap)
1214 * is less than can be stored in a 32 bit unsigned long then
1215 * we can be binary compatible with 2.2.x kernels. If not,
1216 * well, in that case 2.2.x was broken anyways...
1218 * -Erik Andersen <andersee@debian.org>
1221 mem_total
= val
.totalram
+ val
.totalswap
;
1222 if (mem_total
< val
.totalram
|| mem_total
< val
.totalswap
)
1225 mem_unit
= val
.mem_unit
;
1226 while (mem_unit
> 1) {
1229 sav_total
= mem_total
;
1231 if (mem_total
< sav_total
)
1236 * If mem_total did not overflow, multiply all memory values by
1237 * val.mem_unit and set it to 1. This leaves things compatible
1238 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1243 val
.totalram
<<= bitcount
;
1244 val
.freeram
<<= bitcount
;
1245 val
.sharedram
<<= bitcount
;
1246 val
.bufferram
<<= bitcount
;
1247 val
.totalswap
<<= bitcount
;
1248 val
.freeswap
<<= bitcount
;
1249 val
.totalhigh
<<= bitcount
;
1250 val
.freehigh
<<= bitcount
;
1253 if (copy_to_user(info
, &val
, sizeof(struct sysinfo
)))
1259 static void __devinit
init_timers_cpu(int cpu
)
1264 base
= &per_cpu(tvec_bases
, cpu
);
1265 spin_lock_init(&base
->t_base
.lock
);
1266 for (j
= 0; j
< TVN_SIZE
; j
++) {
1267 INIT_LIST_HEAD(base
->tv5
.vec
+ j
);
1268 INIT_LIST_HEAD(base
->tv4
.vec
+ j
);
1269 INIT_LIST_HEAD(base
->tv3
.vec
+ j
);
1270 INIT_LIST_HEAD(base
->tv2
.vec
+ j
);
1272 for (j
= 0; j
< TVR_SIZE
; j
++)
1273 INIT_LIST_HEAD(base
->tv1
.vec
+ j
);
1275 base
->timer_jiffies
= jiffies
;
1278 #ifdef CONFIG_HOTPLUG_CPU
1279 static void migrate_timer_list(tvec_base_t
*new_base
, struct list_head
*head
)
1281 struct timer_list
*timer
;
1283 while (!list_empty(head
)) {
1284 timer
= list_entry(head
->next
, struct timer_list
, entry
);
1285 detach_timer(timer
, 0);
1286 timer
->base
= &new_base
->t_base
;
1287 internal_add_timer(new_base
, timer
);
1291 static void __devinit
migrate_timers(int cpu
)
1293 tvec_base_t
*old_base
;
1294 tvec_base_t
*new_base
;
1297 BUG_ON(cpu_online(cpu
));
1298 old_base
= &per_cpu(tvec_bases
, cpu
);
1299 new_base
= &get_cpu_var(tvec_bases
);
1301 local_irq_disable();
1302 spin_lock(&new_base
->t_base
.lock
);
1303 spin_lock(&old_base
->t_base
.lock
);
1305 if (old_base
->t_base
.running_timer
)
1307 for (i
= 0; i
< TVR_SIZE
; i
++)
1308 migrate_timer_list(new_base
, old_base
->tv1
.vec
+ i
);
1309 for (i
= 0; i
< TVN_SIZE
; i
++) {
1310 migrate_timer_list(new_base
, old_base
->tv2
.vec
+ i
);
1311 migrate_timer_list(new_base
, old_base
->tv3
.vec
+ i
);
1312 migrate_timer_list(new_base
, old_base
->tv4
.vec
+ i
);
1313 migrate_timer_list(new_base
, old_base
->tv5
.vec
+ i
);
1316 spin_unlock(&old_base
->t_base
.lock
);
1317 spin_unlock(&new_base
->t_base
.lock
);
1319 put_cpu_var(tvec_bases
);
1321 #endif /* CONFIG_HOTPLUG_CPU */
1323 static int __devinit
timer_cpu_notify(struct notifier_block
*self
,
1324 unsigned long action
, void *hcpu
)
1326 long cpu
= (long)hcpu
;
1328 case CPU_UP_PREPARE
:
1329 init_timers_cpu(cpu
);
1331 #ifdef CONFIG_HOTPLUG_CPU
1333 migrate_timers(cpu
);
1342 static struct notifier_block __devinitdata timers_nb
= {
1343 .notifier_call
= timer_cpu_notify
,
1347 void __init
init_timers(void)
1349 timer_cpu_notify(&timers_nb
, (unsigned long)CPU_UP_PREPARE
,
1350 (void *)(long)smp_processor_id());
1351 register_cpu_notifier(&timers_nb
);
1352 open_softirq(TIMER_SOFTIRQ
, run_timer_softirq
, NULL
);
1355 #ifdef CONFIG_TIME_INTERPOLATION
1357 struct time_interpolator
*time_interpolator __read_mostly
;
1358 static struct time_interpolator
*time_interpolator_list __read_mostly
;
1359 static DEFINE_SPINLOCK(time_interpolator_lock
);
1361 static inline u64
time_interpolator_get_cycles(unsigned int src
)
1363 unsigned long (*x
)(void);
1367 case TIME_SOURCE_FUNCTION
:
1368 x
= time_interpolator
->addr
;
1371 case TIME_SOURCE_MMIO64
:
1372 return readq_relaxed((void __iomem
*)time_interpolator
->addr
);
1374 case TIME_SOURCE_MMIO32
:
1375 return readl_relaxed((void __iomem
*)time_interpolator
->addr
);
1377 default: return get_cycles();
1381 static inline u64
time_interpolator_get_counter(int writelock
)
1383 unsigned int src
= time_interpolator
->source
;
1385 if (time_interpolator
->jitter
)
1391 lcycle
= time_interpolator
->last_cycle
;
1392 now
= time_interpolator_get_cycles(src
);
1393 if (lcycle
&& time_after(lcycle
, now
))
1396 /* When holding the xtime write lock, there's no need
1397 * to add the overhead of the cmpxchg. Readers are
1398 * force to retry until the write lock is released.
1401 time_interpolator
->last_cycle
= now
;
1404 /* Keep track of the last timer value returned. The use of cmpxchg here
1405 * will cause contention in an SMP environment.
1407 } while (unlikely(cmpxchg(&time_interpolator
->last_cycle
, lcycle
, now
) != lcycle
));
1411 return time_interpolator_get_cycles(src
);
1414 void time_interpolator_reset(void)
1416 time_interpolator
->offset
= 0;
1417 time_interpolator
->last_counter
= time_interpolator_get_counter(1);
1420 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1422 unsigned long time_interpolator_get_offset(void)
1424 /* If we do not have a time interpolator set up then just return zero */
1425 if (!time_interpolator
)
1428 return time_interpolator
->offset
+
1429 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator
);
1432 #define INTERPOLATOR_ADJUST 65536
1433 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1435 static void time_interpolator_update(long delta_nsec
)
1438 unsigned long offset
;
1440 /* If there is no time interpolator set up then do nothing */
1441 if (!time_interpolator
)
1445 * The interpolator compensates for late ticks by accumulating the late
1446 * time in time_interpolator->offset. A tick earlier than expected will
1447 * lead to a reset of the offset and a corresponding jump of the clock
1448 * forward. Again this only works if the interpolator clock is running
1449 * slightly slower than the regular clock and the tuning logic insures
1453 counter
= time_interpolator_get_counter(1);
1454 offset
= time_interpolator
->offset
+
1455 GET_TI_NSECS(counter
, time_interpolator
);
1457 if (delta_nsec
< 0 || (unsigned long) delta_nsec
< offset
)
1458 time_interpolator
->offset
= offset
- delta_nsec
;
1460 time_interpolator
->skips
++;
1461 time_interpolator
->ns_skipped
+= delta_nsec
- offset
;
1462 time_interpolator
->offset
= 0;
1464 time_interpolator
->last_counter
= counter
;
1466 /* Tuning logic for time interpolator invoked every minute or so.
1467 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1468 * Increase interpolator clock speed if we skip too much time.
1470 if (jiffies
% INTERPOLATOR_ADJUST
== 0)
1472 if (time_interpolator
->skips
== 0 && time_interpolator
->offset
> TICK_NSEC
)
1473 time_interpolator
->nsec_per_cyc
--;
1474 if (time_interpolator
->ns_skipped
> INTERPOLATOR_MAX_SKIP
&& time_interpolator
->offset
== 0)
1475 time_interpolator
->nsec_per_cyc
++;
1476 time_interpolator
->skips
= 0;
1477 time_interpolator
->ns_skipped
= 0;
1482 is_better_time_interpolator(struct time_interpolator
*new)
1484 if (!time_interpolator
)
1486 return new->frequency
> 2*time_interpolator
->frequency
||
1487 (unsigned long)new->drift
< (unsigned long)time_interpolator
->drift
;
1491 register_time_interpolator(struct time_interpolator
*ti
)
1493 unsigned long flags
;
1496 if (ti
->frequency
== 0 || ti
->mask
== 0)
1499 ti
->nsec_per_cyc
= ((u64
)NSEC_PER_SEC
<< ti
->shift
) / ti
->frequency
;
1500 spin_lock(&time_interpolator_lock
);
1501 write_seqlock_irqsave(&xtime_lock
, flags
);
1502 if (is_better_time_interpolator(ti
)) {
1503 time_interpolator
= ti
;
1504 time_interpolator_reset();
1506 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1508 ti
->next
= time_interpolator_list
;
1509 time_interpolator_list
= ti
;
1510 spin_unlock(&time_interpolator_lock
);
1514 unregister_time_interpolator(struct time_interpolator
*ti
)
1516 struct time_interpolator
*curr
, **prev
;
1517 unsigned long flags
;
1519 spin_lock(&time_interpolator_lock
);
1520 prev
= &time_interpolator_list
;
1521 for (curr
= *prev
; curr
; curr
= curr
->next
) {
1529 write_seqlock_irqsave(&xtime_lock
, flags
);
1530 if (ti
== time_interpolator
) {
1531 /* we lost the best time-interpolator: */
1532 time_interpolator
= NULL
;
1533 /* find the next-best interpolator */
1534 for (curr
= time_interpolator_list
; curr
; curr
= curr
->next
)
1535 if (is_better_time_interpolator(curr
))
1536 time_interpolator
= curr
;
1537 time_interpolator_reset();
1539 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1540 spin_unlock(&time_interpolator_lock
);
1542 #endif /* CONFIG_TIME_INTERPOLATION */
1545 * msleep - sleep safely even with waitqueue interruptions
1546 * @msecs: Time in milliseconds to sleep for
1548 void msleep(unsigned int msecs
)
1550 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1553 timeout
= schedule_timeout_uninterruptible(timeout
);
1556 EXPORT_SYMBOL(msleep
);
1559 * msleep_interruptible - sleep waiting for signals
1560 * @msecs: Time in milliseconds to sleep for
1562 unsigned long msleep_interruptible(unsigned int msecs
)
1564 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1566 while (timeout
&& !signal_pending(current
))
1567 timeout
= schedule_timeout_interruptible(timeout
);
1568 return jiffies_to_msecs(timeout
);
1571 EXPORT_SYMBOL(msleep_interruptible
);