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
= per_cpu(tvec_bases
, raw_smp_processor_id());
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 list_head
*head
, *curr
;
388 head
= tv
->vec
+ index
;
391 * We are removing _all_ timers from the list, so we don't have to
392 * detach them individually, just clear the list afterwards.
394 while (curr
!= head
) {
395 struct timer_list
*tmp
;
397 tmp
= list_entry(curr
, struct timer_list
, entry
);
398 BUG_ON(tmp
->base
!= base
);
400 internal_add_timer(base
, tmp
);
402 INIT_LIST_HEAD(head
);
408 * __run_timers - run all expired timers (if any) on this CPU.
409 * @base: the timer vector to be processed.
411 * This function cascades all vectors and executes all expired timer
414 #define INDEX(N) (base->timer_jiffies >> (TVR_BITS + N * TVN_BITS)) & TVN_MASK
416 static inline void __run_timers(tvec_base_t
*base
)
418 struct timer_list
*timer
;
420 spin_lock_irq(&base
->lock
);
421 while (time_after_eq(jiffies
, base
->timer_jiffies
)) {
422 struct list_head work_list
= LIST_HEAD_INIT(work_list
);
423 struct list_head
*head
= &work_list
;
424 int index
= base
->timer_jiffies
& TVR_MASK
;
430 (!cascade(base
, &base
->tv2
, INDEX(0))) &&
431 (!cascade(base
, &base
->tv3
, INDEX(1))) &&
432 !cascade(base
, &base
->tv4
, INDEX(2)))
433 cascade(base
, &base
->tv5
, INDEX(3));
434 ++base
->timer_jiffies
;
435 list_splice_init(base
->tv1
.vec
+ index
, &work_list
);
436 while (!list_empty(head
)) {
437 void (*fn
)(unsigned long);
440 timer
= list_entry(head
->next
,struct timer_list
,entry
);
441 fn
= timer
->function
;
444 set_running_timer(base
, timer
);
445 detach_timer(timer
, 1);
446 spin_unlock_irq(&base
->lock
);
448 int preempt_count
= preempt_count();
450 if (preempt_count
!= preempt_count()) {
451 printk(KERN_WARNING
"huh, entered %p "
452 "with preempt_count %08x, exited"
459 spin_lock_irq(&base
->lock
);
462 set_running_timer(base
, NULL
);
463 spin_unlock_irq(&base
->lock
);
466 #ifdef CONFIG_NO_IDLE_HZ
468 * Find out when the next timer event is due to happen. This
469 * is used on S/390 to stop all activity when a cpus is idle.
470 * This functions needs to be called disabled.
472 unsigned long next_timer_interrupt(void)
475 struct list_head
*list
;
476 struct timer_list
*nte
;
477 unsigned long expires
;
478 unsigned long hr_expires
= MAX_JIFFY_OFFSET
;
483 hr_delta
= hrtimer_get_next_event();
484 if (hr_delta
.tv64
!= KTIME_MAX
) {
485 struct timespec tsdelta
;
486 tsdelta
= ktime_to_timespec(hr_delta
);
487 hr_expires
= timespec_to_jiffies(&tsdelta
);
489 return hr_expires
+ jiffies
;
491 hr_expires
+= jiffies
;
493 base
= __get_cpu_var(tvec_bases
);
494 spin_lock(&base
->lock
);
495 expires
= base
->timer_jiffies
+ (LONG_MAX
>> 1);
498 /* Look for timer events in tv1. */
499 j
= base
->timer_jiffies
& TVR_MASK
;
501 list_for_each_entry(nte
, base
->tv1
.vec
+ j
, entry
) {
502 expires
= nte
->expires
;
503 if (j
< (base
->timer_jiffies
& TVR_MASK
))
504 list
= base
->tv2
.vec
+ (INDEX(0));
507 j
= (j
+ 1) & TVR_MASK
;
508 } while (j
!= (base
->timer_jiffies
& TVR_MASK
));
511 varray
[0] = &base
->tv2
;
512 varray
[1] = &base
->tv3
;
513 varray
[2] = &base
->tv4
;
514 varray
[3] = &base
->tv5
;
515 for (i
= 0; i
< 4; i
++) {
518 if (list_empty(varray
[i
]->vec
+ j
)) {
519 j
= (j
+ 1) & TVN_MASK
;
522 list_for_each_entry(nte
, varray
[i
]->vec
+ j
, entry
)
523 if (time_before(nte
->expires
, expires
))
524 expires
= nte
->expires
;
525 if (j
< (INDEX(i
)) && i
< 3)
526 list
= varray
[i
+ 1]->vec
+ (INDEX(i
+ 1));
528 } while (j
!= (INDEX(i
)));
533 * The search wrapped. We need to look at the next list
534 * from next tv element that would cascade into tv element
535 * where we found the timer element.
537 list_for_each_entry(nte
, list
, entry
) {
538 if (time_before(nte
->expires
, expires
))
539 expires
= nte
->expires
;
542 spin_unlock(&base
->lock
);
544 if (time_before(hr_expires
, expires
))
551 /******************************************************************/
554 * Timekeeping variables
556 unsigned long tick_usec
= TICK_USEC
; /* USER_HZ period (usec) */
557 unsigned long tick_nsec
= TICK_NSEC
; /* ACTHZ period (nsec) */
561 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
562 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
563 * at zero at system boot time, so wall_to_monotonic will be negative,
564 * however, we will ALWAYS keep the tv_nsec part positive so we can use
565 * the usual normalization.
567 struct timespec xtime
__attribute__ ((aligned (16)));
568 struct timespec wall_to_monotonic
__attribute__ ((aligned (16)));
570 EXPORT_SYMBOL(xtime
);
572 /* Don't completely fail for HZ > 500. */
573 int tickadj
= 500/HZ
? : 1; /* microsecs */
577 * phase-lock loop variables
579 /* TIME_ERROR prevents overwriting the CMOS clock */
580 int time_state
= TIME_OK
; /* clock synchronization status */
581 int time_status
= STA_UNSYNC
; /* clock status bits */
582 long time_offset
; /* time adjustment (us) */
583 long time_constant
= 2; /* pll time constant */
584 long time_tolerance
= MAXFREQ
; /* frequency tolerance (ppm) */
585 long time_precision
= 1; /* clock precision (us) */
586 long time_maxerror
= NTP_PHASE_LIMIT
; /* maximum error (us) */
587 long time_esterror
= NTP_PHASE_LIMIT
; /* estimated error (us) */
588 static long time_phase
; /* phase offset (scaled us) */
589 long time_freq
= (((NSEC_PER_SEC
+ HZ
/2) % HZ
- HZ
/2) << SHIFT_USEC
) / NSEC_PER_USEC
;
590 /* frequency offset (scaled ppm)*/
591 static long time_adj
; /* tick adjust (scaled 1 / HZ) */
592 long time_reftime
; /* time at last adjustment (s) */
594 long time_next_adjust
;
597 * this routine handles the overflow of the microsecond field
599 * The tricky bits of code to handle the accurate clock support
600 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
601 * They were originally developed for SUN and DEC kernels.
602 * All the kudos should go to Dave for this stuff.
605 static void second_overflow(void)
609 /* Bump the maxerror field */
610 time_maxerror
+= time_tolerance
>> SHIFT_USEC
;
611 if (time_maxerror
> NTP_PHASE_LIMIT
) {
612 time_maxerror
= NTP_PHASE_LIMIT
;
613 time_status
|= STA_UNSYNC
;
617 * Leap second processing. If in leap-insert state at the end of the
618 * day, the system clock is set back one second; if in leap-delete
619 * state, the system clock is set ahead one second. The microtime()
620 * routine or external clock driver will insure that reported time is
621 * always monotonic. The ugly divides should be replaced.
623 switch (time_state
) {
625 if (time_status
& STA_INS
)
626 time_state
= TIME_INS
;
627 else if (time_status
& STA_DEL
)
628 time_state
= TIME_DEL
;
631 if (xtime
.tv_sec
% 86400 == 0) {
633 wall_to_monotonic
.tv_sec
++;
635 * The timer interpolator will make time change
636 * gradually instead of an immediate jump by one second
638 time_interpolator_update(-NSEC_PER_SEC
);
639 time_state
= TIME_OOP
;
641 printk(KERN_NOTICE
"Clock: inserting leap second "
646 if ((xtime
.tv_sec
+ 1) % 86400 == 0) {
648 wall_to_monotonic
.tv_sec
--;
650 * Use of time interpolator for a gradual change of
653 time_interpolator_update(NSEC_PER_SEC
);
654 time_state
= TIME_WAIT
;
656 printk(KERN_NOTICE
"Clock: deleting leap second "
661 time_state
= TIME_WAIT
;
664 if (!(time_status
& (STA_INS
| STA_DEL
)))
665 time_state
= TIME_OK
;
669 * Compute the phase adjustment for the next second. In PLL mode, the
670 * offset is reduced by a fixed factor times the time constant. In FLL
671 * mode the offset is used directly. In either mode, the maximum phase
672 * adjustment for each second is clamped so as to spread the adjustment
673 * over not more than the number of seconds between updates.
676 if (!(time_status
& STA_FLL
))
677 ltemp
= shift_right(ltemp
, SHIFT_KG
+ time_constant
);
678 ltemp
= min(ltemp
, (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
);
679 ltemp
= max(ltemp
, -(MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
);
680 time_offset
-= ltemp
;
681 time_adj
= ltemp
<< (SHIFT_SCALE
- SHIFT_HZ
- SHIFT_UPDATE
);
684 * Compute the frequency estimate and additional phase adjustment due
685 * to frequency error for the next second.
688 time_adj
+= shift_right(ltemp
,(SHIFT_USEC
+ SHIFT_HZ
- SHIFT_SCALE
));
692 * Compensate for (HZ==100) != (1 << SHIFT_HZ). Add 25% and 3.125% to
693 * get 128.125; => only 0.125% error (p. 14)
695 time_adj
+= shift_right(time_adj
, 2) + shift_right(time_adj
, 5);
699 * Compensate for (HZ==250) != (1 << SHIFT_HZ). Add 1.5625% and
700 * 0.78125% to get 255.85938; => only 0.05% error (p. 14)
702 time_adj
+= shift_right(time_adj
, 6) + shift_right(time_adj
, 7);
706 * Compensate for (HZ==1000) != (1 << SHIFT_HZ). Add 1.5625% and
707 * 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
709 time_adj
+= shift_right(time_adj
, 6) + shift_right(time_adj
, 7);
714 * Returns how many microseconds we need to add to xtime this tick
715 * in doing an adjustment requested with adjtime.
717 static long adjtime_adjustment(void)
719 long time_adjust_step
;
721 time_adjust_step
= time_adjust
;
722 if (time_adjust_step
) {
724 * We are doing an adjtime thing. Prepare time_adjust_step to
725 * be within bounds. Note that a positive time_adjust means we
726 * want the clock to run faster.
728 * Limit the amount of the step to be in the range
729 * -tickadj .. +tickadj
731 time_adjust_step
= min(time_adjust_step
, (long)tickadj
);
732 time_adjust_step
= max(time_adjust_step
, (long)-tickadj
);
734 return time_adjust_step
;
737 /* in the NTP reference this is called "hardclock()" */
738 static void update_wall_time_one_tick(void)
740 long time_adjust_step
, delta_nsec
;
742 time_adjust_step
= adjtime_adjustment();
743 if (time_adjust_step
)
744 /* Reduce by this step the amount of time left */
745 time_adjust
-= time_adjust_step
;
746 delta_nsec
= tick_nsec
+ time_adjust_step
* 1000;
748 * Advance the phase, once it gets to one microsecond, then
749 * advance the tick more.
751 time_phase
+= time_adj
;
752 if ((time_phase
>= FINENSEC
) || (time_phase
<= -FINENSEC
)) {
753 long ltemp
= shift_right(time_phase
, (SHIFT_SCALE
- 10));
754 time_phase
-= ltemp
<< (SHIFT_SCALE
- 10);
757 xtime
.tv_nsec
+= delta_nsec
;
758 time_interpolator_update(delta_nsec
);
760 /* Changes by adjtime() do not take effect till next tick. */
761 if (time_next_adjust
!= 0) {
762 time_adjust
= time_next_adjust
;
763 time_next_adjust
= 0;
768 * Return how long ticks are at the moment, that is, how much time
769 * update_wall_time_one_tick will add to xtime next time we call it
770 * (assuming no calls to do_adjtimex in the meantime).
771 * The return value is in fixed-point nanoseconds with SHIFT_SCALE-10
772 * bits to the right of the binary point.
773 * This function has no side-effects.
775 u64
current_tick_length(void)
779 delta_nsec
= tick_nsec
+ adjtime_adjustment() * 1000;
780 return ((u64
) delta_nsec
<< (SHIFT_SCALE
- 10)) + time_adj
;
784 * Using a loop looks inefficient, but "ticks" is
785 * usually just one (we shouldn't be losing ticks,
786 * we're doing this this way mainly for interrupt
787 * latency reasons, not because we think we'll
788 * have lots of lost timer ticks
790 static void update_wall_time(unsigned long ticks
)
794 update_wall_time_one_tick();
795 if (xtime
.tv_nsec
>= 1000000000) {
796 xtime
.tv_nsec
-= 1000000000;
804 * Called from the timer interrupt handler to charge one tick to the current
805 * process. user_tick is 1 if the tick is user time, 0 for system.
807 void update_process_times(int user_tick
)
809 struct task_struct
*p
= current
;
810 int cpu
= smp_processor_id();
812 /* Note: this timer irq context must be accounted for as well. */
814 account_user_time(p
, jiffies_to_cputime(1));
816 account_system_time(p
, HARDIRQ_OFFSET
, jiffies_to_cputime(1));
818 if (rcu_pending(cpu
))
819 rcu_check_callbacks(cpu
, user_tick
);
821 run_posix_cpu_timers(p
);
825 * Nr of active tasks - counted in fixed-point numbers
827 static unsigned long count_active_tasks(void)
829 return nr_active() * FIXED_1
;
833 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
834 * imply that avenrun[] is the standard name for this kind of thing.
835 * Nothing else seems to be standardized: the fractional size etc
836 * all seem to differ on different machines.
838 * Requires xtime_lock to access.
840 unsigned long avenrun
[3];
842 EXPORT_SYMBOL(avenrun
);
845 * calc_load - given tick count, update the avenrun load estimates.
846 * This is called while holding a write_lock on xtime_lock.
848 static inline void calc_load(unsigned long ticks
)
850 unsigned long active_tasks
; /* fixed-point */
851 static int count
= LOAD_FREQ
;
856 active_tasks
= count_active_tasks();
857 CALC_LOAD(avenrun
[0], EXP_1
, active_tasks
);
858 CALC_LOAD(avenrun
[1], EXP_5
, active_tasks
);
859 CALC_LOAD(avenrun
[2], EXP_15
, active_tasks
);
863 /* jiffies at the most recent update of wall time */
864 unsigned long wall_jiffies
= INITIAL_JIFFIES
;
867 * This read-write spinlock protects us from races in SMP while
868 * playing with xtime and avenrun.
870 #ifndef ARCH_HAVE_XTIME_LOCK
871 seqlock_t xtime_lock __cacheline_aligned_in_smp
= SEQLOCK_UNLOCKED
;
873 EXPORT_SYMBOL(xtime_lock
);
877 * This function runs timers and the timer-tq in bottom half context.
879 static void run_timer_softirq(struct softirq_action
*h
)
881 tvec_base_t
*base
= __get_cpu_var(tvec_bases
);
883 hrtimer_run_queues();
884 if (time_after_eq(jiffies
, base
->timer_jiffies
))
889 * Called by the local, per-CPU timer interrupt on SMP.
891 void run_local_timers(void)
893 raise_softirq(TIMER_SOFTIRQ
);
898 * Called by the timer interrupt. xtime_lock must already be taken
901 static inline void update_times(void)
905 ticks
= jiffies
- wall_jiffies
;
907 wall_jiffies
+= ticks
;
908 update_wall_time(ticks
);
914 * The 64-bit jiffies value is not atomic - you MUST NOT read it
915 * without sampling the sequence number in xtime_lock.
916 * jiffies is defined in the linker script...
919 void do_timer(struct pt_regs
*regs
)
922 /* prevent loading jiffies before storing new jiffies_64 value. */
927 #ifdef __ARCH_WANT_SYS_ALARM
930 * For backwards compatibility? This can be done in libc so Alpha
931 * and all newer ports shouldn't need it.
933 asmlinkage
unsigned long sys_alarm(unsigned int seconds
)
935 return alarm_setitimer(seconds
);
943 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
944 * should be moved into arch/i386 instead?
948 * sys_getpid - return the thread group id of the current process
950 * Note, despite the name, this returns the tgid not the pid. The tgid and
951 * the pid are identical unless CLONE_THREAD was specified on clone() in
952 * which case the tgid is the same in all threads of the same group.
954 * This is SMP safe as current->tgid does not change.
956 asmlinkage
long sys_getpid(void)
958 return current
->tgid
;
962 * Accessing ->group_leader->real_parent is not SMP-safe, it could
963 * change from under us. However, rather than getting any lock
964 * we can use an optimistic algorithm: get the parent
965 * pid, and go back and check that the parent is still
966 * the same. If it has changed (which is extremely unlikely
967 * indeed), we just try again..
969 * NOTE! This depends on the fact that even if we _do_
970 * get an old value of "parent", we can happily dereference
971 * the pointer (it was and remains a dereferencable kernel pointer
972 * no matter what): we just can't necessarily trust the result
973 * until we know that the parent pointer is valid.
975 * NOTE2: ->group_leader never changes from under us.
977 asmlinkage
long sys_getppid(void)
980 struct task_struct
*me
= current
;
981 struct task_struct
*parent
;
983 parent
= me
->group_leader
->real_parent
;
986 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
988 struct task_struct
*old
= parent
;
991 * Make sure we read the pid before re-reading the
995 parent
= me
->group_leader
->real_parent
;
1005 asmlinkage
long sys_getuid(void)
1007 /* Only we change this so SMP safe */
1008 return current
->uid
;
1011 asmlinkage
long sys_geteuid(void)
1013 /* Only we change this so SMP safe */
1014 return current
->euid
;
1017 asmlinkage
long sys_getgid(void)
1019 /* Only we change this so SMP safe */
1020 return current
->gid
;
1023 asmlinkage
long sys_getegid(void)
1025 /* Only we change this so SMP safe */
1026 return current
->egid
;
1031 static void process_timeout(unsigned long __data
)
1033 wake_up_process((task_t
*)__data
);
1037 * schedule_timeout - sleep until timeout
1038 * @timeout: timeout value in jiffies
1040 * Make the current task sleep until @timeout jiffies have
1041 * elapsed. The routine will return immediately unless
1042 * the current task state has been set (see set_current_state()).
1044 * You can set the task state as follows -
1046 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1047 * pass before the routine returns. The routine will return 0
1049 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1050 * delivered to the current task. In this case the remaining time
1051 * in jiffies will be returned, or 0 if the timer expired in time
1053 * The current task state is guaranteed to be TASK_RUNNING when this
1056 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1057 * the CPU away without a bound on the timeout. In this case the return
1058 * value will be %MAX_SCHEDULE_TIMEOUT.
1060 * In all cases the return value is guaranteed to be non-negative.
1062 fastcall
signed long __sched
schedule_timeout(signed long timeout
)
1064 struct timer_list timer
;
1065 unsigned long expire
;
1069 case MAX_SCHEDULE_TIMEOUT
:
1071 * These two special cases are useful to be comfortable
1072 * in the caller. Nothing more. We could take
1073 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1074 * but I' d like to return a valid offset (>=0) to allow
1075 * the caller to do everything it want with the retval.
1081 * Another bit of PARANOID. Note that the retval will be
1082 * 0 since no piece of kernel is supposed to do a check
1083 * for a negative retval of schedule_timeout() (since it
1084 * should never happens anyway). You just have the printk()
1085 * that will tell you if something is gone wrong and where.
1089 printk(KERN_ERR
"schedule_timeout: wrong timeout "
1090 "value %lx from %p\n", timeout
,
1091 __builtin_return_address(0));
1092 current
->state
= TASK_RUNNING
;
1097 expire
= timeout
+ jiffies
;
1099 setup_timer(&timer
, process_timeout
, (unsigned long)current
);
1100 __mod_timer(&timer
, expire
);
1102 del_singleshot_timer_sync(&timer
);
1104 timeout
= expire
- jiffies
;
1107 return timeout
< 0 ? 0 : timeout
;
1109 EXPORT_SYMBOL(schedule_timeout
);
1112 * We can use __set_current_state() here because schedule_timeout() calls
1113 * schedule() unconditionally.
1115 signed long __sched
schedule_timeout_interruptible(signed long timeout
)
1117 __set_current_state(TASK_INTERRUPTIBLE
);
1118 return schedule_timeout(timeout
);
1120 EXPORT_SYMBOL(schedule_timeout_interruptible
);
1122 signed long __sched
schedule_timeout_uninterruptible(signed long timeout
)
1124 __set_current_state(TASK_UNINTERRUPTIBLE
);
1125 return schedule_timeout(timeout
);
1127 EXPORT_SYMBOL(schedule_timeout_uninterruptible
);
1129 /* Thread ID - the internal kernel "pid" */
1130 asmlinkage
long sys_gettid(void)
1132 return current
->pid
;
1136 * sys_sysinfo - fill in sysinfo struct
1138 asmlinkage
long sys_sysinfo(struct sysinfo __user
*info
)
1141 unsigned long mem_total
, sav_total
;
1142 unsigned int mem_unit
, bitcount
;
1145 memset((char *)&val
, 0, sizeof(struct sysinfo
));
1149 seq
= read_seqbegin(&xtime_lock
);
1152 * This is annoying. The below is the same thing
1153 * posix_get_clock_monotonic() does, but it wants to
1154 * take the lock which we want to cover the loads stuff
1158 getnstimeofday(&tp
);
1159 tp
.tv_sec
+= wall_to_monotonic
.tv_sec
;
1160 tp
.tv_nsec
+= wall_to_monotonic
.tv_nsec
;
1161 if (tp
.tv_nsec
- NSEC_PER_SEC
>= 0) {
1162 tp
.tv_nsec
= tp
.tv_nsec
- NSEC_PER_SEC
;
1165 val
.uptime
= tp
.tv_sec
+ (tp
.tv_nsec
? 1 : 0);
1167 val
.loads
[0] = avenrun
[0] << (SI_LOAD_SHIFT
- FSHIFT
);
1168 val
.loads
[1] = avenrun
[1] << (SI_LOAD_SHIFT
- FSHIFT
);
1169 val
.loads
[2] = avenrun
[2] << (SI_LOAD_SHIFT
- FSHIFT
);
1171 val
.procs
= nr_threads
;
1172 } while (read_seqretry(&xtime_lock
, seq
));
1178 * If the sum of all the available memory (i.e. ram + swap)
1179 * is less than can be stored in a 32 bit unsigned long then
1180 * we can be binary compatible with 2.2.x kernels. If not,
1181 * well, in that case 2.2.x was broken anyways...
1183 * -Erik Andersen <andersee@debian.org>
1186 mem_total
= val
.totalram
+ val
.totalswap
;
1187 if (mem_total
< val
.totalram
|| mem_total
< val
.totalswap
)
1190 mem_unit
= val
.mem_unit
;
1191 while (mem_unit
> 1) {
1194 sav_total
= mem_total
;
1196 if (mem_total
< sav_total
)
1201 * If mem_total did not overflow, multiply all memory values by
1202 * val.mem_unit and set it to 1. This leaves things compatible
1203 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1208 val
.totalram
<<= bitcount
;
1209 val
.freeram
<<= bitcount
;
1210 val
.sharedram
<<= bitcount
;
1211 val
.bufferram
<<= bitcount
;
1212 val
.totalswap
<<= bitcount
;
1213 val
.freeswap
<<= bitcount
;
1214 val
.totalhigh
<<= bitcount
;
1215 val
.freehigh
<<= bitcount
;
1218 if (copy_to_user(info
, &val
, sizeof(struct sysinfo
)))
1224 static int __devinit
init_timers_cpu(int cpu
)
1228 static char __devinitdata tvec_base_done
[NR_CPUS
];
1230 if (!tvec_base_done
[cpu
]) {
1231 static char boot_done
;
1235 * The APs use this path later in boot
1237 base
= kmalloc_node(sizeof(*base
), GFP_KERNEL
,
1241 memset(base
, 0, sizeof(*base
));
1242 per_cpu(tvec_bases
, cpu
) = base
;
1245 * This is for the boot CPU - we use compile-time
1246 * static initialisation because per-cpu memory isn't
1247 * ready yet and because the memory allocators are not
1248 * initialised either.
1251 base
= &boot_tvec_bases
;
1253 tvec_base_done
[cpu
] = 1;
1255 base
= per_cpu(tvec_bases
, cpu
);
1258 spin_lock_init(&base
->lock
);
1259 for (j
= 0; j
< TVN_SIZE
; j
++) {
1260 INIT_LIST_HEAD(base
->tv5
.vec
+ j
);
1261 INIT_LIST_HEAD(base
->tv4
.vec
+ j
);
1262 INIT_LIST_HEAD(base
->tv3
.vec
+ j
);
1263 INIT_LIST_HEAD(base
->tv2
.vec
+ j
);
1265 for (j
= 0; j
< TVR_SIZE
; j
++)
1266 INIT_LIST_HEAD(base
->tv1
.vec
+ j
);
1268 base
->timer_jiffies
= jiffies
;
1272 #ifdef CONFIG_HOTPLUG_CPU
1273 static void migrate_timer_list(tvec_base_t
*new_base
, struct list_head
*head
)
1275 struct timer_list
*timer
;
1277 while (!list_empty(head
)) {
1278 timer
= list_entry(head
->next
, struct timer_list
, entry
);
1279 detach_timer(timer
, 0);
1280 timer
->base
= new_base
;
1281 internal_add_timer(new_base
, timer
);
1285 static void __devinit
migrate_timers(int cpu
)
1287 tvec_base_t
*old_base
;
1288 tvec_base_t
*new_base
;
1291 BUG_ON(cpu_online(cpu
));
1292 old_base
= per_cpu(tvec_bases
, cpu
);
1293 new_base
= get_cpu_var(tvec_bases
);
1295 local_irq_disable();
1296 spin_lock(&new_base
->lock
);
1297 spin_lock(&old_base
->lock
);
1299 BUG_ON(old_base
->running_timer
);
1301 for (i
= 0; i
< TVR_SIZE
; i
++)
1302 migrate_timer_list(new_base
, old_base
->tv1
.vec
+ i
);
1303 for (i
= 0; i
< TVN_SIZE
; i
++) {
1304 migrate_timer_list(new_base
, old_base
->tv2
.vec
+ i
);
1305 migrate_timer_list(new_base
, old_base
->tv3
.vec
+ i
);
1306 migrate_timer_list(new_base
, old_base
->tv4
.vec
+ i
);
1307 migrate_timer_list(new_base
, old_base
->tv5
.vec
+ i
);
1310 spin_unlock(&old_base
->lock
);
1311 spin_unlock(&new_base
->lock
);
1313 put_cpu_var(tvec_bases
);
1315 #endif /* CONFIG_HOTPLUG_CPU */
1317 static int __devinit
timer_cpu_notify(struct notifier_block
*self
,
1318 unsigned long action
, void *hcpu
)
1320 long cpu
= (long)hcpu
;
1322 case CPU_UP_PREPARE
:
1323 if (init_timers_cpu(cpu
) < 0)
1326 #ifdef CONFIG_HOTPLUG_CPU
1328 migrate_timers(cpu
);
1337 static struct notifier_block timers_nb
= {
1338 .notifier_call
= timer_cpu_notify
,
1342 void __init
init_timers(void)
1344 timer_cpu_notify(&timers_nb
, (unsigned long)CPU_UP_PREPARE
,
1345 (void *)(long)smp_processor_id());
1346 register_cpu_notifier(&timers_nb
);
1347 open_softirq(TIMER_SOFTIRQ
, run_timer_softirq
, NULL
);
1350 #ifdef CONFIG_TIME_INTERPOLATION
1352 struct time_interpolator
*time_interpolator __read_mostly
;
1353 static struct time_interpolator
*time_interpolator_list __read_mostly
;
1354 static DEFINE_SPINLOCK(time_interpolator_lock
);
1356 static inline u64
time_interpolator_get_cycles(unsigned int src
)
1358 unsigned long (*x
)(void);
1362 case TIME_SOURCE_FUNCTION
:
1363 x
= time_interpolator
->addr
;
1366 case TIME_SOURCE_MMIO64
:
1367 return readq_relaxed((void __iomem
*)time_interpolator
->addr
);
1369 case TIME_SOURCE_MMIO32
:
1370 return readl_relaxed((void __iomem
*)time_interpolator
->addr
);
1372 default: return get_cycles();
1376 static inline u64
time_interpolator_get_counter(int writelock
)
1378 unsigned int src
= time_interpolator
->source
;
1380 if (time_interpolator
->jitter
)
1386 lcycle
= time_interpolator
->last_cycle
;
1387 now
= time_interpolator_get_cycles(src
);
1388 if (lcycle
&& time_after(lcycle
, now
))
1391 /* When holding the xtime write lock, there's no need
1392 * to add the overhead of the cmpxchg. Readers are
1393 * force to retry until the write lock is released.
1396 time_interpolator
->last_cycle
= now
;
1399 /* Keep track of the last timer value returned. The use of cmpxchg here
1400 * will cause contention in an SMP environment.
1402 } while (unlikely(cmpxchg(&time_interpolator
->last_cycle
, lcycle
, now
) != lcycle
));
1406 return time_interpolator_get_cycles(src
);
1409 void time_interpolator_reset(void)
1411 time_interpolator
->offset
= 0;
1412 time_interpolator
->last_counter
= time_interpolator_get_counter(1);
1415 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1417 unsigned long time_interpolator_get_offset(void)
1419 /* If we do not have a time interpolator set up then just return zero */
1420 if (!time_interpolator
)
1423 return time_interpolator
->offset
+
1424 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator
);
1427 #define INTERPOLATOR_ADJUST 65536
1428 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1430 static void time_interpolator_update(long delta_nsec
)
1433 unsigned long offset
;
1435 /* If there is no time interpolator set up then do nothing */
1436 if (!time_interpolator
)
1440 * The interpolator compensates for late ticks by accumulating the late
1441 * time in time_interpolator->offset. A tick earlier than expected will
1442 * lead to a reset of the offset and a corresponding jump of the clock
1443 * forward. Again this only works if the interpolator clock is running
1444 * slightly slower than the regular clock and the tuning logic insures
1448 counter
= time_interpolator_get_counter(1);
1449 offset
= time_interpolator
->offset
+
1450 GET_TI_NSECS(counter
, time_interpolator
);
1452 if (delta_nsec
< 0 || (unsigned long) delta_nsec
< offset
)
1453 time_interpolator
->offset
= offset
- delta_nsec
;
1455 time_interpolator
->skips
++;
1456 time_interpolator
->ns_skipped
+= delta_nsec
- offset
;
1457 time_interpolator
->offset
= 0;
1459 time_interpolator
->last_counter
= counter
;
1461 /* Tuning logic for time interpolator invoked every minute or so.
1462 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1463 * Increase interpolator clock speed if we skip too much time.
1465 if (jiffies
% INTERPOLATOR_ADJUST
== 0)
1467 if (time_interpolator
->skips
== 0 && time_interpolator
->offset
> tick_nsec
)
1468 time_interpolator
->nsec_per_cyc
--;
1469 if (time_interpolator
->ns_skipped
> INTERPOLATOR_MAX_SKIP
&& time_interpolator
->offset
== 0)
1470 time_interpolator
->nsec_per_cyc
++;
1471 time_interpolator
->skips
= 0;
1472 time_interpolator
->ns_skipped
= 0;
1477 is_better_time_interpolator(struct time_interpolator
*new)
1479 if (!time_interpolator
)
1481 return new->frequency
> 2*time_interpolator
->frequency
||
1482 (unsigned long)new->drift
< (unsigned long)time_interpolator
->drift
;
1486 register_time_interpolator(struct time_interpolator
*ti
)
1488 unsigned long flags
;
1491 BUG_ON(ti
->frequency
== 0 || ti
->mask
== 0);
1493 ti
->nsec_per_cyc
= ((u64
)NSEC_PER_SEC
<< ti
->shift
) / ti
->frequency
;
1494 spin_lock(&time_interpolator_lock
);
1495 write_seqlock_irqsave(&xtime_lock
, flags
);
1496 if (is_better_time_interpolator(ti
)) {
1497 time_interpolator
= ti
;
1498 time_interpolator_reset();
1500 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1502 ti
->next
= time_interpolator_list
;
1503 time_interpolator_list
= ti
;
1504 spin_unlock(&time_interpolator_lock
);
1508 unregister_time_interpolator(struct time_interpolator
*ti
)
1510 struct time_interpolator
*curr
, **prev
;
1511 unsigned long flags
;
1513 spin_lock(&time_interpolator_lock
);
1514 prev
= &time_interpolator_list
;
1515 for (curr
= *prev
; curr
; curr
= curr
->next
) {
1523 write_seqlock_irqsave(&xtime_lock
, flags
);
1524 if (ti
== time_interpolator
) {
1525 /* we lost the best time-interpolator: */
1526 time_interpolator
= NULL
;
1527 /* find the next-best interpolator */
1528 for (curr
= time_interpolator_list
; curr
; curr
= curr
->next
)
1529 if (is_better_time_interpolator(curr
))
1530 time_interpolator
= curr
;
1531 time_interpolator_reset();
1533 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1534 spin_unlock(&time_interpolator_lock
);
1536 #endif /* CONFIG_TIME_INTERPOLATION */
1539 * msleep - sleep safely even with waitqueue interruptions
1540 * @msecs: Time in milliseconds to sleep for
1542 void msleep(unsigned int msecs
)
1544 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1547 timeout
= schedule_timeout_uninterruptible(timeout
);
1550 EXPORT_SYMBOL(msleep
);
1553 * msleep_interruptible - sleep waiting for signals
1554 * @msecs: Time in milliseconds to sleep for
1556 unsigned long msleep_interruptible(unsigned int msecs
)
1558 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1560 while (timeout
&& !signal_pending(current
))
1561 timeout
= schedule_timeout_interruptible(timeout
);
1562 return jiffies_to_msecs(timeout
);
1565 EXPORT_SYMBOL(msleep_interruptible
);