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
;
84 static DEFINE_PER_CPU(tvec_base_t
*, tvec_bases
);
85 tvec_base_t boot_tvec_bases
;
86 EXPORT_SYMBOL(boot_tvec_bases
);
88 static inline void set_running_timer(tvec_base_t
*base
,
89 struct timer_list
*timer
)
92 base
->running_timer
= timer
;
96 static void internal_add_timer(tvec_base_t
*base
, struct timer_list
*timer
)
98 unsigned long expires
= timer
->expires
;
99 unsigned long idx
= expires
- base
->timer_jiffies
;
100 struct list_head
*vec
;
102 if (idx
< TVR_SIZE
) {
103 int i
= expires
& TVR_MASK
;
104 vec
= base
->tv1
.vec
+ i
;
105 } else if (idx
< 1 << (TVR_BITS
+ TVN_BITS
)) {
106 int i
= (expires
>> TVR_BITS
) & TVN_MASK
;
107 vec
= base
->tv2
.vec
+ i
;
108 } else if (idx
< 1 << (TVR_BITS
+ 2 * TVN_BITS
)) {
109 int i
= (expires
>> (TVR_BITS
+ TVN_BITS
)) & TVN_MASK
;
110 vec
= base
->tv3
.vec
+ i
;
111 } else if (idx
< 1 << (TVR_BITS
+ 3 * TVN_BITS
)) {
112 int i
= (expires
>> (TVR_BITS
+ 2 * TVN_BITS
)) & TVN_MASK
;
113 vec
= base
->tv4
.vec
+ i
;
114 } else if ((signed long) idx
< 0) {
116 * Can happen if you add a timer with expires == jiffies,
117 * or you set a timer to go off in the past
119 vec
= base
->tv1
.vec
+ (base
->timer_jiffies
& TVR_MASK
);
122 /* If the timeout is larger than 0xffffffff on 64-bit
123 * architectures then we use the maximum timeout:
125 if (idx
> 0xffffffffUL
) {
127 expires
= idx
+ base
->timer_jiffies
;
129 i
= (expires
>> (TVR_BITS
+ 3 * TVN_BITS
)) & TVN_MASK
;
130 vec
= base
->tv5
.vec
+ i
;
135 list_add_tail(&timer
->entry
, vec
);
139 * init_timer - initialize a timer.
140 * @timer: the timer to be initialized
142 * init_timer() must be done to a timer prior calling *any* of the
143 * other timer functions.
145 void fastcall
init_timer(struct timer_list
*timer
)
147 timer
->entry
.next
= NULL
;
148 timer
->base
= per_cpu(tvec_bases
, raw_smp_processor_id());
150 EXPORT_SYMBOL(init_timer
);
152 static inline void detach_timer(struct timer_list
*timer
,
155 struct list_head
*entry
= &timer
->entry
;
157 __list_del(entry
->prev
, entry
->next
);
160 entry
->prev
= LIST_POISON2
;
164 * We are using hashed locking: holding per_cpu(tvec_bases).lock
165 * means that all timers which are tied to this base via timer->base are
166 * locked, and the base itself is locked too.
168 * So __run_timers/migrate_timers can safely modify all timers which could
169 * be found on ->tvX lists.
171 * When the timer's base is locked, and the timer removed from list, it is
172 * possible to set timer->base = NULL and drop the lock: the timer remains
175 static tvec_base_t
*lock_timer_base(struct timer_list
*timer
,
176 unsigned long *flags
)
182 if (likely(base
!= NULL
)) {
183 spin_lock_irqsave(&base
->lock
, *flags
);
184 if (likely(base
== timer
->base
))
186 /* The timer has migrated to another CPU */
187 spin_unlock_irqrestore(&base
->lock
, *flags
);
193 int __mod_timer(struct timer_list
*timer
, unsigned long expires
)
195 tvec_base_t
*base
, *new_base
;
199 BUG_ON(!timer
->function
);
201 base
= lock_timer_base(timer
, &flags
);
203 if (timer_pending(timer
)) {
204 detach_timer(timer
, 0);
208 new_base
= __get_cpu_var(tvec_bases
);
210 if (base
!= new_base
) {
212 * We are trying to schedule the timer on the local CPU.
213 * However we can't change timer's base while it is running,
214 * otherwise del_timer_sync() can't detect that the timer's
215 * handler yet has not finished. This also guarantees that
216 * the timer is serialized wrt itself.
218 if (unlikely(base
->running_timer
== timer
)) {
219 /* The timer remains on a former base */
222 /* See the comment in lock_timer_base() */
224 spin_unlock(&base
->lock
);
225 spin_lock(&new_base
->lock
);
226 timer
->base
= new_base
;
230 timer
->expires
= expires
;
231 internal_add_timer(new_base
, timer
);
232 spin_unlock_irqrestore(&new_base
->lock
, flags
);
237 EXPORT_SYMBOL(__mod_timer
);
240 * add_timer_on - start a timer on a particular CPU
241 * @timer: the timer to be added
242 * @cpu: the CPU to start it on
244 * This is not very scalable on SMP. Double adds are not possible.
246 void add_timer_on(struct timer_list
*timer
, int cpu
)
248 tvec_base_t
*base
= per_cpu(tvec_bases
, cpu
);
251 BUG_ON(timer_pending(timer
) || !timer
->function
);
252 spin_lock_irqsave(&base
->lock
, flags
);
254 internal_add_timer(base
, timer
);
255 spin_unlock_irqrestore(&base
->lock
, flags
);
260 * mod_timer - modify a timer's timeout
261 * @timer: the timer to be modified
263 * mod_timer is a more efficient way to update the expire field of an
264 * active timer (if the timer is inactive it will be activated)
266 * mod_timer(timer, expires) is equivalent to:
268 * del_timer(timer); timer->expires = expires; add_timer(timer);
270 * Note that if there are multiple unserialized concurrent users of the
271 * same timer, then mod_timer() is the only safe way to modify the timeout,
272 * since add_timer() cannot modify an already running timer.
274 * The function returns whether it has modified a pending timer or not.
275 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
276 * active timer returns 1.)
278 int mod_timer(struct timer_list
*timer
, unsigned long expires
)
280 BUG_ON(!timer
->function
);
283 * This is a common optimization triggered by the
284 * networking code - if the timer is re-modified
285 * to be the same thing then just return:
287 if (timer
->expires
== expires
&& timer_pending(timer
))
290 return __mod_timer(timer
, expires
);
293 EXPORT_SYMBOL(mod_timer
);
296 * del_timer - deactive a timer.
297 * @timer: the timer to be deactivated
299 * del_timer() deactivates a timer - this works on both active and inactive
302 * The function returns whether it has deactivated a pending timer or not.
303 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
304 * active timer returns 1.)
306 int del_timer(struct timer_list
*timer
)
312 if (timer_pending(timer
)) {
313 base
= lock_timer_base(timer
, &flags
);
314 if (timer_pending(timer
)) {
315 detach_timer(timer
, 1);
318 spin_unlock_irqrestore(&base
->lock
, flags
);
324 EXPORT_SYMBOL(del_timer
);
328 * This function tries to deactivate a timer. Upon successful (ret >= 0)
329 * exit the timer is not queued and the handler is not running on any CPU.
331 * It must not be called from interrupt contexts.
333 int try_to_del_timer_sync(struct timer_list
*timer
)
339 base
= lock_timer_base(timer
, &flags
);
341 if (base
->running_timer
== timer
)
345 if (timer_pending(timer
)) {
346 detach_timer(timer
, 1);
350 spin_unlock_irqrestore(&base
->lock
, flags
);
356 * del_timer_sync - deactivate a timer and wait for the handler to finish.
357 * @timer: the timer to be deactivated
359 * This function only differs from del_timer() on SMP: besides deactivating
360 * the timer it also makes sure the handler has finished executing on other
363 * Synchronization rules: callers must prevent restarting of the timer,
364 * otherwise this function is meaningless. It must not be called from
365 * interrupt contexts. The caller must not hold locks which would prevent
366 * completion of the timer's handler. The timer's handler must not call
367 * add_timer_on(). Upon exit the timer is not queued and the handler is
368 * not running on any CPU.
370 * The function returns whether it has deactivated a pending timer or not.
372 int del_timer_sync(struct timer_list
*timer
)
375 int ret
= try_to_del_timer_sync(timer
);
381 EXPORT_SYMBOL(del_timer_sync
);
384 static int cascade(tvec_base_t
*base
, tvec_t
*tv
, int index
)
386 /* cascade all the timers from tv up one level */
387 struct list_head
*head
, *curr
;
389 head
= tv
->vec
+ index
;
392 * We are removing _all_ timers from the list, so we don't have to
393 * detach them individually, just clear the list afterwards.
395 while (curr
!= head
) {
396 struct timer_list
*tmp
;
398 tmp
= list_entry(curr
, struct timer_list
, entry
);
399 BUG_ON(tmp
->base
!= base
);
401 internal_add_timer(base
, tmp
);
403 INIT_LIST_HEAD(head
);
409 * __run_timers - run all expired timers (if any) on this CPU.
410 * @base: the timer vector to be processed.
412 * This function cascades all vectors and executes all expired timer
415 #define INDEX(N) (base->timer_jiffies >> (TVR_BITS + N * TVN_BITS)) & TVN_MASK
417 static inline void __run_timers(tvec_base_t
*base
)
419 struct timer_list
*timer
;
421 spin_lock_irq(&base
->lock
);
422 while (time_after_eq(jiffies
, base
->timer_jiffies
)) {
423 struct list_head work_list
= LIST_HEAD_INIT(work_list
);
424 struct list_head
*head
= &work_list
;
425 int index
= base
->timer_jiffies
& TVR_MASK
;
431 (!cascade(base
, &base
->tv2
, INDEX(0))) &&
432 (!cascade(base
, &base
->tv3
, INDEX(1))) &&
433 !cascade(base
, &base
->tv4
, INDEX(2)))
434 cascade(base
, &base
->tv5
, INDEX(3));
435 ++base
->timer_jiffies
;
436 list_splice_init(base
->tv1
.vec
+ index
, &work_list
);
437 while (!list_empty(head
)) {
438 void (*fn
)(unsigned long);
441 timer
= list_entry(head
->next
,struct timer_list
,entry
);
442 fn
= timer
->function
;
445 set_running_timer(base
, timer
);
446 detach_timer(timer
, 1);
447 spin_unlock_irq(&base
->lock
);
449 int preempt_count
= preempt_count();
451 if (preempt_count
!= preempt_count()) {
452 printk(KERN_WARNING
"huh, entered %p "
453 "with preempt_count %08x, exited"
460 spin_lock_irq(&base
->lock
);
463 set_running_timer(base
, NULL
);
464 spin_unlock_irq(&base
->lock
);
467 #ifdef CONFIG_NO_IDLE_HZ
469 * Find out when the next timer event is due to happen. This
470 * is used on S/390 to stop all activity when a cpus is idle.
471 * This functions needs to be called disabled.
473 unsigned long next_timer_interrupt(void)
476 struct list_head
*list
;
477 struct timer_list
*nte
;
478 unsigned long expires
;
479 unsigned long hr_expires
= MAX_JIFFY_OFFSET
;
484 hr_delta
= hrtimer_get_next_event();
485 if (hr_delta
.tv64
!= KTIME_MAX
) {
486 struct timespec tsdelta
;
487 tsdelta
= ktime_to_timespec(hr_delta
);
488 hr_expires
= timespec_to_jiffies(&tsdelta
);
490 return hr_expires
+ jiffies
;
492 hr_expires
+= jiffies
;
494 base
= __get_cpu_var(tvec_bases
);
495 spin_lock(&base
->lock
);
496 expires
= base
->timer_jiffies
+ (LONG_MAX
>> 1);
499 /* Look for timer events in tv1. */
500 j
= base
->timer_jiffies
& TVR_MASK
;
502 list_for_each_entry(nte
, base
->tv1
.vec
+ j
, entry
) {
503 expires
= nte
->expires
;
504 if (j
< (base
->timer_jiffies
& TVR_MASK
))
505 list
= base
->tv2
.vec
+ (INDEX(0));
508 j
= (j
+ 1) & TVR_MASK
;
509 } while (j
!= (base
->timer_jiffies
& TVR_MASK
));
512 varray
[0] = &base
->tv2
;
513 varray
[1] = &base
->tv3
;
514 varray
[2] = &base
->tv4
;
515 varray
[3] = &base
->tv5
;
516 for (i
= 0; i
< 4; i
++) {
519 if (list_empty(varray
[i
]->vec
+ j
)) {
520 j
= (j
+ 1) & TVN_MASK
;
523 list_for_each_entry(nte
, varray
[i
]->vec
+ j
, entry
)
524 if (time_before(nte
->expires
, expires
))
525 expires
= nte
->expires
;
526 if (j
< (INDEX(i
)) && i
< 3)
527 list
= varray
[i
+ 1]->vec
+ (INDEX(i
+ 1));
529 } while (j
!= (INDEX(i
)));
534 * The search wrapped. We need to look at the next list
535 * from next tv element that would cascade into tv element
536 * where we found the timer element.
538 list_for_each_entry(nte
, list
, entry
) {
539 if (time_before(nte
->expires
, expires
))
540 expires
= nte
->expires
;
543 spin_unlock(&base
->lock
);
545 if (time_before(hr_expires
, expires
))
552 /******************************************************************/
555 * Timekeeping variables
557 unsigned long tick_usec
= TICK_USEC
; /* USER_HZ period (usec) */
558 unsigned long tick_nsec
= TICK_NSEC
; /* ACTHZ period (nsec) */
562 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
563 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
564 * at zero at system boot time, so wall_to_monotonic will be negative,
565 * however, we will ALWAYS keep the tv_nsec part positive so we can use
566 * the usual normalization.
568 struct timespec xtime
__attribute__ ((aligned (16)));
569 struct timespec wall_to_monotonic
__attribute__ ((aligned (16)));
571 EXPORT_SYMBOL(xtime
);
573 /* Don't completely fail for HZ > 500. */
574 int tickadj
= 500/HZ
? : 1; /* microsecs */
578 * phase-lock loop variables
580 /* TIME_ERROR prevents overwriting the CMOS clock */
581 int time_state
= TIME_OK
; /* clock synchronization status */
582 int time_status
= STA_UNSYNC
; /* clock status bits */
583 long time_offset
; /* time adjustment (us) */
584 long time_constant
= 2; /* pll time constant */
585 long time_tolerance
= MAXFREQ
; /* frequency tolerance (ppm) */
586 long time_precision
= 1; /* clock precision (us) */
587 long time_maxerror
= NTP_PHASE_LIMIT
; /* maximum error (us) */
588 long time_esterror
= NTP_PHASE_LIMIT
; /* estimated error (us) */
589 static long time_phase
; /* phase offset (scaled us) */
590 long time_freq
= (((NSEC_PER_SEC
+ HZ
/2) % HZ
- HZ
/2) << SHIFT_USEC
) / NSEC_PER_USEC
;
591 /* frequency offset (scaled ppm)*/
592 static long time_adj
; /* tick adjust (scaled 1 / HZ) */
593 long time_reftime
; /* time at last adjustment (s) */
595 long time_next_adjust
;
598 * this routine handles the overflow of the microsecond field
600 * The tricky bits of code to handle the accurate clock support
601 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
602 * They were originally developed for SUN and DEC kernels.
603 * All the kudos should go to Dave for this stuff.
606 static void second_overflow(void)
610 /* Bump the maxerror field */
611 time_maxerror
+= time_tolerance
>> SHIFT_USEC
;
612 if (time_maxerror
> NTP_PHASE_LIMIT
) {
613 time_maxerror
= NTP_PHASE_LIMIT
;
614 time_status
|= STA_UNSYNC
;
618 * Leap second processing. If in leap-insert state at the end of the
619 * day, the system clock is set back one second; if in leap-delete
620 * state, the system clock is set ahead one second. The microtime()
621 * routine or external clock driver will insure that reported time is
622 * always monotonic. The ugly divides should be replaced.
624 switch (time_state
) {
626 if (time_status
& STA_INS
)
627 time_state
= TIME_INS
;
628 else if (time_status
& STA_DEL
)
629 time_state
= TIME_DEL
;
632 if (xtime
.tv_sec
% 86400 == 0) {
634 wall_to_monotonic
.tv_sec
++;
636 * The timer interpolator will make time change
637 * gradually instead of an immediate jump by one second
639 time_interpolator_update(-NSEC_PER_SEC
);
640 time_state
= TIME_OOP
;
642 printk(KERN_NOTICE
"Clock: inserting leap second "
647 if ((xtime
.tv_sec
+ 1) % 86400 == 0) {
649 wall_to_monotonic
.tv_sec
--;
651 * Use of time interpolator for a gradual change of
654 time_interpolator_update(NSEC_PER_SEC
);
655 time_state
= TIME_WAIT
;
657 printk(KERN_NOTICE
"Clock: deleting leap second "
662 time_state
= TIME_WAIT
;
665 if (!(time_status
& (STA_INS
| STA_DEL
)))
666 time_state
= TIME_OK
;
670 * Compute the phase adjustment for the next second. In PLL mode, the
671 * offset is reduced by a fixed factor times the time constant. In FLL
672 * mode the offset is used directly. In either mode, the maximum phase
673 * adjustment for each second is clamped so as to spread the adjustment
674 * over not more than the number of seconds between updates.
677 if (!(time_status
& STA_FLL
))
678 ltemp
= shift_right(ltemp
, SHIFT_KG
+ time_constant
);
679 ltemp
= min(ltemp
, (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
);
680 ltemp
= max(ltemp
, -(MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
);
681 time_offset
-= ltemp
;
682 time_adj
= ltemp
<< (SHIFT_SCALE
- SHIFT_HZ
- SHIFT_UPDATE
);
685 * Compute the frequency estimate and additional phase adjustment due
686 * to frequency error for the next second.
689 time_adj
+= shift_right(ltemp
,(SHIFT_USEC
+ SHIFT_HZ
- SHIFT_SCALE
));
693 * Compensate for (HZ==100) != (1 << SHIFT_HZ). Add 25% and 3.125% to
694 * get 128.125; => only 0.125% error (p. 14)
696 time_adj
+= shift_right(time_adj
, 2) + shift_right(time_adj
, 5);
700 * Compensate for (HZ==250) != (1 << SHIFT_HZ). Add 1.5625% and
701 * 0.78125% to get 255.85938; => only 0.05% error (p. 14)
703 time_adj
+= shift_right(time_adj
, 6) + shift_right(time_adj
, 7);
707 * Compensate for (HZ==1000) != (1 << SHIFT_HZ). Add 1.5625% and
708 * 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
710 time_adj
+= shift_right(time_adj
, 6) + shift_right(time_adj
, 7);
715 * Returns how many microseconds we need to add to xtime this tick
716 * in doing an adjustment requested with adjtime.
718 static long adjtime_adjustment(void)
720 long time_adjust_step
;
722 time_adjust_step
= time_adjust
;
723 if (time_adjust_step
) {
725 * We are doing an adjtime thing. Prepare time_adjust_step to
726 * be within bounds. Note that a positive time_adjust means we
727 * want the clock to run faster.
729 * Limit the amount of the step to be in the range
730 * -tickadj .. +tickadj
732 time_adjust_step
= min(time_adjust_step
, (long)tickadj
);
733 time_adjust_step
= max(time_adjust_step
, (long)-tickadj
);
735 return time_adjust_step
;
738 /* in the NTP reference this is called "hardclock()" */
739 static void update_wall_time_one_tick(void)
741 long time_adjust_step
, delta_nsec
;
743 time_adjust_step
= adjtime_adjustment();
744 if (time_adjust_step
)
745 /* Reduce by this step the amount of time left */
746 time_adjust
-= time_adjust_step
;
747 delta_nsec
= tick_nsec
+ time_adjust_step
* 1000;
749 * Advance the phase, once it gets to one microsecond, then
750 * advance the tick more.
752 time_phase
+= time_adj
;
753 if ((time_phase
>= FINENSEC
) || (time_phase
<= -FINENSEC
)) {
754 long ltemp
= shift_right(time_phase
, (SHIFT_SCALE
- 10));
755 time_phase
-= ltemp
<< (SHIFT_SCALE
- 10);
758 xtime
.tv_nsec
+= delta_nsec
;
759 time_interpolator_update(delta_nsec
);
761 /* Changes by adjtime() do not take effect till next tick. */
762 if (time_next_adjust
!= 0) {
763 time_adjust
= time_next_adjust
;
764 time_next_adjust
= 0;
769 * Return how long ticks are at the moment, that is, how much time
770 * update_wall_time_one_tick will add to xtime next time we call it
771 * (assuming no calls to do_adjtimex in the meantime).
772 * The return value is in fixed-point nanoseconds with SHIFT_SCALE-10
773 * bits to the right of the binary point.
774 * This function has no side-effects.
776 u64
current_tick_length(void)
780 delta_nsec
= tick_nsec
+ adjtime_adjustment() * 1000;
781 return ((u64
) delta_nsec
<< (SHIFT_SCALE
- 10)) + time_adj
;
785 * Using a loop looks inefficient, but "ticks" is
786 * usually just one (we shouldn't be losing ticks,
787 * we're doing this this way mainly for interrupt
788 * latency reasons, not because we think we'll
789 * have lots of lost timer ticks
791 static void update_wall_time(unsigned long ticks
)
795 update_wall_time_one_tick();
796 if (xtime
.tv_nsec
>= 1000000000) {
797 xtime
.tv_nsec
-= 1000000000;
805 * Called from the timer interrupt handler to charge one tick to the current
806 * process. user_tick is 1 if the tick is user time, 0 for system.
808 void update_process_times(int user_tick
)
810 struct task_struct
*p
= current
;
811 int cpu
= smp_processor_id();
813 /* Note: this timer irq context must be accounted for as well. */
815 account_user_time(p
, jiffies_to_cputime(1));
817 account_system_time(p
, HARDIRQ_OFFSET
, jiffies_to_cputime(1));
819 if (rcu_pending(cpu
))
820 rcu_check_callbacks(cpu
, user_tick
);
822 run_posix_cpu_timers(p
);
826 * Nr of active tasks - counted in fixed-point numbers
828 static unsigned long count_active_tasks(void)
830 return (nr_running() + nr_uninterruptible()) * FIXED_1
;
834 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
835 * imply that avenrun[] is the standard name for this kind of thing.
836 * Nothing else seems to be standardized: the fractional size etc
837 * all seem to differ on different machines.
839 * Requires xtime_lock to access.
841 unsigned long avenrun
[3];
843 EXPORT_SYMBOL(avenrun
);
846 * calc_load - given tick count, update the avenrun load estimates.
847 * This is called while holding a write_lock on xtime_lock.
849 static inline void calc_load(unsigned long ticks
)
851 unsigned long active_tasks
; /* fixed-point */
852 static int count
= LOAD_FREQ
;
857 active_tasks
= count_active_tasks();
858 CALC_LOAD(avenrun
[0], EXP_1
, active_tasks
);
859 CALC_LOAD(avenrun
[1], EXP_5
, active_tasks
);
860 CALC_LOAD(avenrun
[2], EXP_15
, active_tasks
);
864 /* jiffies at the most recent update of wall time */
865 unsigned long wall_jiffies
= INITIAL_JIFFIES
;
868 * This read-write spinlock protects us from races in SMP while
869 * playing with xtime and avenrun.
871 #ifndef ARCH_HAVE_XTIME_LOCK
872 seqlock_t xtime_lock __cacheline_aligned_in_smp
= SEQLOCK_UNLOCKED
;
874 EXPORT_SYMBOL(xtime_lock
);
878 * This function runs timers and the timer-tq in bottom half context.
880 static void run_timer_softirq(struct softirq_action
*h
)
882 tvec_base_t
*base
= __get_cpu_var(tvec_bases
);
884 hrtimer_run_queues();
885 if (time_after_eq(jiffies
, base
->timer_jiffies
))
890 * Called by the local, per-CPU timer interrupt on SMP.
892 void run_local_timers(void)
894 raise_softirq(TIMER_SOFTIRQ
);
899 * Called by the timer interrupt. xtime_lock must already be taken
902 static inline void update_times(void)
906 ticks
= jiffies
- wall_jiffies
;
908 wall_jiffies
+= ticks
;
909 update_wall_time(ticks
);
915 * The 64-bit jiffies value is not atomic - you MUST NOT read it
916 * without sampling the sequence number in xtime_lock.
917 * jiffies is defined in the linker script...
920 void do_timer(struct pt_regs
*regs
)
923 /* prevent loading jiffies before storing new jiffies_64 value. */
928 #ifdef __ARCH_WANT_SYS_ALARM
931 * For backwards compatibility? This can be done in libc so Alpha
932 * and all newer ports shouldn't need it.
934 asmlinkage
unsigned long sys_alarm(unsigned int seconds
)
936 return alarm_setitimer(seconds
);
944 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
945 * should be moved into arch/i386 instead?
949 * sys_getpid - return the thread group id of the current process
951 * Note, despite the name, this returns the tgid not the pid. The tgid and
952 * the pid are identical unless CLONE_THREAD was specified on clone() in
953 * which case the tgid is the same in all threads of the same group.
955 * This is SMP safe as current->tgid does not change.
957 asmlinkage
long sys_getpid(void)
959 return current
->tgid
;
963 * Accessing ->group_leader->real_parent is not SMP-safe, it could
964 * change from under us. However, rather than getting any lock
965 * we can use an optimistic algorithm: get the parent
966 * pid, and go back and check that the parent is still
967 * the same. If it has changed (which is extremely unlikely
968 * indeed), we just try again..
970 * NOTE! This depends on the fact that even if we _do_
971 * get an old value of "parent", we can happily dereference
972 * the pointer (it was and remains a dereferencable kernel pointer
973 * no matter what): we just can't necessarily trust the result
974 * until we know that the parent pointer is valid.
976 * NOTE2: ->group_leader never changes from under us.
978 asmlinkage
long sys_getppid(void)
981 struct task_struct
*me
= current
;
982 struct task_struct
*parent
;
984 parent
= me
->group_leader
->real_parent
;
987 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
989 struct task_struct
*old
= parent
;
992 * Make sure we read the pid before re-reading the
996 parent
= me
->group_leader
->real_parent
;
1006 asmlinkage
long sys_getuid(void)
1008 /* Only we change this so SMP safe */
1009 return current
->uid
;
1012 asmlinkage
long sys_geteuid(void)
1014 /* Only we change this so SMP safe */
1015 return current
->euid
;
1018 asmlinkage
long sys_getgid(void)
1020 /* Only we change this so SMP safe */
1021 return current
->gid
;
1024 asmlinkage
long sys_getegid(void)
1026 /* Only we change this so SMP safe */
1027 return current
->egid
;
1032 static void process_timeout(unsigned long __data
)
1034 wake_up_process((task_t
*)__data
);
1038 * schedule_timeout - sleep until timeout
1039 * @timeout: timeout value in jiffies
1041 * Make the current task sleep until @timeout jiffies have
1042 * elapsed. The routine will return immediately unless
1043 * the current task state has been set (see set_current_state()).
1045 * You can set the task state as follows -
1047 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1048 * pass before the routine returns. The routine will return 0
1050 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1051 * delivered to the current task. In this case the remaining time
1052 * in jiffies will be returned, or 0 if the timer expired in time
1054 * The current task state is guaranteed to be TASK_RUNNING when this
1057 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1058 * the CPU away without a bound on the timeout. In this case the return
1059 * value will be %MAX_SCHEDULE_TIMEOUT.
1061 * In all cases the return value is guaranteed to be non-negative.
1063 fastcall
signed long __sched
schedule_timeout(signed long timeout
)
1065 struct timer_list timer
;
1066 unsigned long expire
;
1070 case MAX_SCHEDULE_TIMEOUT
:
1072 * These two special cases are useful to be comfortable
1073 * in the caller. Nothing more. We could take
1074 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1075 * but I' d like to return a valid offset (>=0) to allow
1076 * the caller to do everything it want with the retval.
1082 * Another bit of PARANOID. Note that the retval will be
1083 * 0 since no piece of kernel is supposed to do a check
1084 * for a negative retval of schedule_timeout() (since it
1085 * should never happens anyway). You just have the printk()
1086 * that will tell you if something is gone wrong and where.
1090 printk(KERN_ERR
"schedule_timeout: wrong timeout "
1091 "value %lx from %p\n", timeout
,
1092 __builtin_return_address(0));
1093 current
->state
= TASK_RUNNING
;
1098 expire
= timeout
+ jiffies
;
1100 setup_timer(&timer
, process_timeout
, (unsigned long)current
);
1101 __mod_timer(&timer
, expire
);
1103 del_singleshot_timer_sync(&timer
);
1105 timeout
= expire
- jiffies
;
1108 return timeout
< 0 ? 0 : timeout
;
1110 EXPORT_SYMBOL(schedule_timeout
);
1113 * We can use __set_current_state() here because schedule_timeout() calls
1114 * schedule() unconditionally.
1116 signed long __sched
schedule_timeout_interruptible(signed long timeout
)
1118 __set_current_state(TASK_INTERRUPTIBLE
);
1119 return schedule_timeout(timeout
);
1121 EXPORT_SYMBOL(schedule_timeout_interruptible
);
1123 signed long __sched
schedule_timeout_uninterruptible(signed long timeout
)
1125 __set_current_state(TASK_UNINTERRUPTIBLE
);
1126 return schedule_timeout(timeout
);
1128 EXPORT_SYMBOL(schedule_timeout_uninterruptible
);
1130 /* Thread ID - the internal kernel "pid" */
1131 asmlinkage
long sys_gettid(void)
1133 return current
->pid
;
1137 * sys_sysinfo - fill in sysinfo struct
1139 asmlinkage
long sys_sysinfo(struct sysinfo __user
*info
)
1142 unsigned long mem_total
, sav_total
;
1143 unsigned int mem_unit
, bitcount
;
1146 memset((char *)&val
, 0, sizeof(struct sysinfo
));
1150 seq
= read_seqbegin(&xtime_lock
);
1153 * This is annoying. The below is the same thing
1154 * posix_get_clock_monotonic() does, but it wants to
1155 * take the lock which we want to cover the loads stuff
1159 getnstimeofday(&tp
);
1160 tp
.tv_sec
+= wall_to_monotonic
.tv_sec
;
1161 tp
.tv_nsec
+= wall_to_monotonic
.tv_nsec
;
1162 if (tp
.tv_nsec
- NSEC_PER_SEC
>= 0) {
1163 tp
.tv_nsec
= tp
.tv_nsec
- NSEC_PER_SEC
;
1166 val
.uptime
= tp
.tv_sec
+ (tp
.tv_nsec
? 1 : 0);
1168 val
.loads
[0] = avenrun
[0] << (SI_LOAD_SHIFT
- FSHIFT
);
1169 val
.loads
[1] = avenrun
[1] << (SI_LOAD_SHIFT
- FSHIFT
);
1170 val
.loads
[2] = avenrun
[2] << (SI_LOAD_SHIFT
- FSHIFT
);
1172 val
.procs
= nr_threads
;
1173 } while (read_seqretry(&xtime_lock
, seq
));
1179 * If the sum of all the available memory (i.e. ram + swap)
1180 * is less than can be stored in a 32 bit unsigned long then
1181 * we can be binary compatible with 2.2.x kernels. If not,
1182 * well, in that case 2.2.x was broken anyways...
1184 * -Erik Andersen <andersee@debian.org>
1187 mem_total
= val
.totalram
+ val
.totalswap
;
1188 if (mem_total
< val
.totalram
|| mem_total
< val
.totalswap
)
1191 mem_unit
= val
.mem_unit
;
1192 while (mem_unit
> 1) {
1195 sav_total
= mem_total
;
1197 if (mem_total
< sav_total
)
1202 * If mem_total did not overflow, multiply all memory values by
1203 * val.mem_unit and set it to 1. This leaves things compatible
1204 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1209 val
.totalram
<<= bitcount
;
1210 val
.freeram
<<= bitcount
;
1211 val
.sharedram
<<= bitcount
;
1212 val
.bufferram
<<= bitcount
;
1213 val
.totalswap
<<= bitcount
;
1214 val
.freeswap
<<= bitcount
;
1215 val
.totalhigh
<<= bitcount
;
1216 val
.freehigh
<<= bitcount
;
1219 if (copy_to_user(info
, &val
, sizeof(struct sysinfo
)))
1225 static int __devinit
init_timers_cpu(int cpu
)
1230 base
= per_cpu(tvec_bases
, cpu
);
1232 static char boot_done
;
1235 * Cannot do allocation in init_timers as that runs before the
1236 * allocator initializes (and would waste memory if there are
1237 * more possible CPUs than will ever be installed/brought up).
1240 base
= kmalloc_node(sizeof(*base
), GFP_KERNEL
,
1244 memset(base
, 0, sizeof(*base
));
1246 base
= &boot_tvec_bases
;
1249 per_cpu(tvec_bases
, cpu
) = base
;
1251 spin_lock_init(&base
->lock
);
1252 for (j
= 0; j
< TVN_SIZE
; j
++) {
1253 INIT_LIST_HEAD(base
->tv5
.vec
+ j
);
1254 INIT_LIST_HEAD(base
->tv4
.vec
+ j
);
1255 INIT_LIST_HEAD(base
->tv3
.vec
+ j
);
1256 INIT_LIST_HEAD(base
->tv2
.vec
+ j
);
1258 for (j
= 0; j
< TVR_SIZE
; j
++)
1259 INIT_LIST_HEAD(base
->tv1
.vec
+ j
);
1261 base
->timer_jiffies
= jiffies
;
1265 #ifdef CONFIG_HOTPLUG_CPU
1266 static void migrate_timer_list(tvec_base_t
*new_base
, struct list_head
*head
)
1268 struct timer_list
*timer
;
1270 while (!list_empty(head
)) {
1271 timer
= list_entry(head
->next
, struct timer_list
, entry
);
1272 detach_timer(timer
, 0);
1273 timer
->base
= new_base
;
1274 internal_add_timer(new_base
, timer
);
1278 static void __devinit
migrate_timers(int cpu
)
1280 tvec_base_t
*old_base
;
1281 tvec_base_t
*new_base
;
1284 BUG_ON(cpu_online(cpu
));
1285 old_base
= per_cpu(tvec_bases
, cpu
);
1286 new_base
= get_cpu_var(tvec_bases
);
1288 local_irq_disable();
1289 spin_lock(&new_base
->lock
);
1290 spin_lock(&old_base
->lock
);
1292 BUG_ON(old_base
->running_timer
);
1294 for (i
= 0; i
< TVR_SIZE
; i
++)
1295 migrate_timer_list(new_base
, old_base
->tv1
.vec
+ i
);
1296 for (i
= 0; i
< TVN_SIZE
; i
++) {
1297 migrate_timer_list(new_base
, old_base
->tv2
.vec
+ i
);
1298 migrate_timer_list(new_base
, old_base
->tv3
.vec
+ i
);
1299 migrate_timer_list(new_base
, old_base
->tv4
.vec
+ i
);
1300 migrate_timer_list(new_base
, old_base
->tv5
.vec
+ i
);
1303 spin_unlock(&old_base
->lock
);
1304 spin_unlock(&new_base
->lock
);
1306 put_cpu_var(tvec_bases
);
1308 #endif /* CONFIG_HOTPLUG_CPU */
1310 static int __devinit
timer_cpu_notify(struct notifier_block
*self
,
1311 unsigned long action
, void *hcpu
)
1313 long cpu
= (long)hcpu
;
1315 case CPU_UP_PREPARE
:
1316 if (init_timers_cpu(cpu
) < 0)
1319 #ifdef CONFIG_HOTPLUG_CPU
1321 migrate_timers(cpu
);
1330 static struct notifier_block __devinitdata timers_nb
= {
1331 .notifier_call
= timer_cpu_notify
,
1335 void __init
init_timers(void)
1337 timer_cpu_notify(&timers_nb
, (unsigned long)CPU_UP_PREPARE
,
1338 (void *)(long)smp_processor_id());
1339 register_cpu_notifier(&timers_nb
);
1340 open_softirq(TIMER_SOFTIRQ
, run_timer_softirq
, NULL
);
1343 #ifdef CONFIG_TIME_INTERPOLATION
1345 struct time_interpolator
*time_interpolator __read_mostly
;
1346 static struct time_interpolator
*time_interpolator_list __read_mostly
;
1347 static DEFINE_SPINLOCK(time_interpolator_lock
);
1349 static inline u64
time_interpolator_get_cycles(unsigned int src
)
1351 unsigned long (*x
)(void);
1355 case TIME_SOURCE_FUNCTION
:
1356 x
= time_interpolator
->addr
;
1359 case TIME_SOURCE_MMIO64
:
1360 return readq_relaxed((void __iomem
*)time_interpolator
->addr
);
1362 case TIME_SOURCE_MMIO32
:
1363 return readl_relaxed((void __iomem
*)time_interpolator
->addr
);
1365 default: return get_cycles();
1369 static inline u64
time_interpolator_get_counter(int writelock
)
1371 unsigned int src
= time_interpolator
->source
;
1373 if (time_interpolator
->jitter
)
1379 lcycle
= time_interpolator
->last_cycle
;
1380 now
= time_interpolator_get_cycles(src
);
1381 if (lcycle
&& time_after(lcycle
, now
))
1384 /* When holding the xtime write lock, there's no need
1385 * to add the overhead of the cmpxchg. Readers are
1386 * force to retry until the write lock is released.
1389 time_interpolator
->last_cycle
= now
;
1392 /* Keep track of the last timer value returned. The use of cmpxchg here
1393 * will cause contention in an SMP environment.
1395 } while (unlikely(cmpxchg(&time_interpolator
->last_cycle
, lcycle
, now
) != lcycle
));
1399 return time_interpolator_get_cycles(src
);
1402 void time_interpolator_reset(void)
1404 time_interpolator
->offset
= 0;
1405 time_interpolator
->last_counter
= time_interpolator_get_counter(1);
1408 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1410 unsigned long time_interpolator_get_offset(void)
1412 /* If we do not have a time interpolator set up then just return zero */
1413 if (!time_interpolator
)
1416 return time_interpolator
->offset
+
1417 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator
);
1420 #define INTERPOLATOR_ADJUST 65536
1421 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1423 static void time_interpolator_update(long delta_nsec
)
1426 unsigned long offset
;
1428 /* If there is no time interpolator set up then do nothing */
1429 if (!time_interpolator
)
1433 * The interpolator compensates for late ticks by accumulating the late
1434 * time in time_interpolator->offset. A tick earlier than expected will
1435 * lead to a reset of the offset and a corresponding jump of the clock
1436 * forward. Again this only works if the interpolator clock is running
1437 * slightly slower than the regular clock and the tuning logic insures
1441 counter
= time_interpolator_get_counter(1);
1442 offset
= time_interpolator
->offset
+
1443 GET_TI_NSECS(counter
, time_interpolator
);
1445 if (delta_nsec
< 0 || (unsigned long) delta_nsec
< offset
)
1446 time_interpolator
->offset
= offset
- delta_nsec
;
1448 time_interpolator
->skips
++;
1449 time_interpolator
->ns_skipped
+= delta_nsec
- offset
;
1450 time_interpolator
->offset
= 0;
1452 time_interpolator
->last_counter
= counter
;
1454 /* Tuning logic for time interpolator invoked every minute or so.
1455 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1456 * Increase interpolator clock speed if we skip too much time.
1458 if (jiffies
% INTERPOLATOR_ADJUST
== 0)
1460 if (time_interpolator
->skips
== 0 && time_interpolator
->offset
> TICK_NSEC
)
1461 time_interpolator
->nsec_per_cyc
--;
1462 if (time_interpolator
->ns_skipped
> INTERPOLATOR_MAX_SKIP
&& time_interpolator
->offset
== 0)
1463 time_interpolator
->nsec_per_cyc
++;
1464 time_interpolator
->skips
= 0;
1465 time_interpolator
->ns_skipped
= 0;
1470 is_better_time_interpolator(struct time_interpolator
*new)
1472 if (!time_interpolator
)
1474 return new->frequency
> 2*time_interpolator
->frequency
||
1475 (unsigned long)new->drift
< (unsigned long)time_interpolator
->drift
;
1479 register_time_interpolator(struct time_interpolator
*ti
)
1481 unsigned long flags
;
1484 if (ti
->frequency
== 0 || ti
->mask
== 0)
1487 ti
->nsec_per_cyc
= ((u64
)NSEC_PER_SEC
<< ti
->shift
) / ti
->frequency
;
1488 spin_lock(&time_interpolator_lock
);
1489 write_seqlock_irqsave(&xtime_lock
, flags
);
1490 if (is_better_time_interpolator(ti
)) {
1491 time_interpolator
= ti
;
1492 time_interpolator_reset();
1494 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1496 ti
->next
= time_interpolator_list
;
1497 time_interpolator_list
= ti
;
1498 spin_unlock(&time_interpolator_lock
);
1502 unregister_time_interpolator(struct time_interpolator
*ti
)
1504 struct time_interpolator
*curr
, **prev
;
1505 unsigned long flags
;
1507 spin_lock(&time_interpolator_lock
);
1508 prev
= &time_interpolator_list
;
1509 for (curr
= *prev
; curr
; curr
= curr
->next
) {
1517 write_seqlock_irqsave(&xtime_lock
, flags
);
1518 if (ti
== time_interpolator
) {
1519 /* we lost the best time-interpolator: */
1520 time_interpolator
= NULL
;
1521 /* find the next-best interpolator */
1522 for (curr
= time_interpolator_list
; curr
; curr
= curr
->next
)
1523 if (is_better_time_interpolator(curr
))
1524 time_interpolator
= curr
;
1525 time_interpolator_reset();
1527 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1528 spin_unlock(&time_interpolator_lock
);
1530 #endif /* CONFIG_TIME_INTERPOLATION */
1533 * msleep - sleep safely even with waitqueue interruptions
1534 * @msecs: Time in milliseconds to sleep for
1536 void msleep(unsigned int msecs
)
1538 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1541 timeout
= schedule_timeout_uninterruptible(timeout
);
1544 EXPORT_SYMBOL(msleep
);
1547 * msleep_interruptible - sleep waiting for signals
1548 * @msecs: Time in milliseconds to sleep for
1550 unsigned long msleep_interruptible(unsigned int msecs
)
1552 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1554 while (timeout
&& !signal_pending(current
))
1555 timeout
= schedule_timeout_interruptible(timeout
);
1556 return jiffies_to_msecs(timeout
);
1559 EXPORT_SYMBOL(msleep_interruptible
);