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>
37 #include <asm/uaccess.h>
38 #include <asm/unistd.h>
39 #include <asm/div64.h>
40 #include <asm/timex.h>
43 #ifdef CONFIG_TIME_INTERPOLATION
44 static void time_interpolator_update(long delta_nsec
);
46 #define time_interpolator_update(x)
50 * per-CPU timer vector definitions:
53 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
54 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
55 #define TVN_SIZE (1 << TVN_BITS)
56 #define TVR_SIZE (1 << TVR_BITS)
57 #define TVN_MASK (TVN_SIZE - 1)
58 #define TVR_MASK (TVR_SIZE - 1)
62 struct timer_list
*running_timer
;
65 typedef struct tvec_s
{
66 struct list_head vec
[TVN_SIZE
];
69 typedef struct tvec_root_s
{
70 struct list_head vec
[TVR_SIZE
];
73 struct tvec_t_base_s
{
74 struct timer_base_s t_base
;
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
);
86 static inline void set_running_timer(tvec_base_t
*base
,
87 struct timer_list
*timer
)
90 base
->t_base
.running_timer
= timer
;
94 static void check_timer_failed(struct timer_list
*timer
)
96 static int whine_count
;
97 if (whine_count
< 16) {
99 printk("Uninitialised timer!\n");
100 printk("This is just a warning. Your computer is OK\n");
101 printk("function=0x%p, data=0x%lx\n",
102 timer
->function
, timer
->data
);
108 timer
->magic
= TIMER_MAGIC
;
111 static inline void check_timer(struct timer_list
*timer
)
113 if (timer
->magic
!= TIMER_MAGIC
)
114 check_timer_failed(timer
);
118 static void internal_add_timer(tvec_base_t
*base
, struct timer_list
*timer
)
120 unsigned long expires
= timer
->expires
;
121 unsigned long idx
= expires
- base
->timer_jiffies
;
122 struct list_head
*vec
;
124 if (idx
< TVR_SIZE
) {
125 int i
= expires
& TVR_MASK
;
126 vec
= base
->tv1
.vec
+ i
;
127 } else if (idx
< 1 << (TVR_BITS
+ TVN_BITS
)) {
128 int i
= (expires
>> TVR_BITS
) & TVN_MASK
;
129 vec
= base
->tv2
.vec
+ i
;
130 } else if (idx
< 1 << (TVR_BITS
+ 2 * TVN_BITS
)) {
131 int i
= (expires
>> (TVR_BITS
+ TVN_BITS
)) & TVN_MASK
;
132 vec
= base
->tv3
.vec
+ i
;
133 } else if (idx
< 1 << (TVR_BITS
+ 3 * TVN_BITS
)) {
134 int i
= (expires
>> (TVR_BITS
+ 2 * TVN_BITS
)) & TVN_MASK
;
135 vec
= base
->tv4
.vec
+ i
;
136 } else if ((signed long) idx
< 0) {
138 * Can happen if you add a timer with expires == jiffies,
139 * or you set a timer to go off in the past
141 vec
= base
->tv1
.vec
+ (base
->timer_jiffies
& TVR_MASK
);
144 /* If the timeout is larger than 0xffffffff on 64-bit
145 * architectures then we use the maximum timeout:
147 if (idx
> 0xffffffffUL
) {
149 expires
= idx
+ base
->timer_jiffies
;
151 i
= (expires
>> (TVR_BITS
+ 3 * TVN_BITS
)) & TVN_MASK
;
152 vec
= base
->tv5
.vec
+ i
;
157 list_add_tail(&timer
->entry
, vec
);
160 typedef struct timer_base_s timer_base_t
;
162 * Used by TIMER_INITIALIZER, we can't use per_cpu(tvec_bases)
163 * at compile time, and we need timer->base to lock the timer.
165 timer_base_t __init_timer_base
166 ____cacheline_aligned_in_smp
= { .lock
= SPIN_LOCK_UNLOCKED
};
167 EXPORT_SYMBOL(__init_timer_base
);
170 * init_timer - initialize a timer.
171 * @timer: the timer to be initialized
173 * init_timer() must be done to a timer prior calling *any* of the
174 * other timer functions.
176 void fastcall
init_timer(struct timer_list
*timer
)
178 timer
->entry
.next
= NULL
;
179 timer
->base
= &per_cpu(tvec_bases
, raw_smp_processor_id()).t_base
;
180 timer
->magic
= TIMER_MAGIC
;
182 EXPORT_SYMBOL(init_timer
);
184 static inline void detach_timer(struct timer_list
*timer
,
187 struct list_head
*entry
= &timer
->entry
;
189 __list_del(entry
->prev
, entry
->next
);
192 entry
->prev
= LIST_POISON2
;
196 * We are using hashed locking: holding per_cpu(tvec_bases).t_base.lock
197 * means that all timers which are tied to this base via timer->base are
198 * locked, and the base itself is locked too.
200 * So __run_timers/migrate_timers can safely modify all timers which could
201 * be found on ->tvX lists.
203 * When the timer's base is locked, and the timer removed from list, it is
204 * possible to set timer->base = NULL and drop the lock: the timer remains
207 static timer_base_t
*lock_timer_base(struct timer_list
*timer
,
208 unsigned long *flags
)
214 if (likely(base
!= NULL
)) {
215 spin_lock_irqsave(&base
->lock
, *flags
);
216 if (likely(base
== timer
->base
))
218 /* The timer has migrated to another CPU */
219 spin_unlock_irqrestore(&base
->lock
, *flags
);
225 int __mod_timer(struct timer_list
*timer
, unsigned long expires
)
228 tvec_base_t
*new_base
;
232 BUG_ON(!timer
->function
);
235 base
= lock_timer_base(timer
, &flags
);
237 if (timer_pending(timer
)) {
238 detach_timer(timer
, 0);
242 new_base
= &__get_cpu_var(tvec_bases
);
244 if (base
!= &new_base
->t_base
) {
246 * We are trying to schedule the timer on the local CPU.
247 * However we can't change timer's base while it is running,
248 * otherwise del_timer_sync() can't detect that the timer's
249 * handler yet has not finished. This also guarantees that
250 * the timer is serialized wrt itself.
252 if (unlikely(base
->running_timer
== timer
)) {
253 /* The timer remains on a former base */
254 new_base
= container_of(base
, tvec_base_t
, t_base
);
256 /* See the comment in lock_timer_base() */
258 spin_unlock(&base
->lock
);
259 spin_lock(&new_base
->t_base
.lock
);
260 timer
->base
= &new_base
->t_base
;
264 timer
->expires
= expires
;
265 internal_add_timer(new_base
, timer
);
266 spin_unlock_irqrestore(&new_base
->t_base
.lock
, flags
);
271 EXPORT_SYMBOL(__mod_timer
);
274 * add_timer_on - start a timer on a particular CPU
275 * @timer: the timer to be added
276 * @cpu: the CPU to start it on
278 * This is not very scalable on SMP. Double adds are not possible.
280 void add_timer_on(struct timer_list
*timer
, int cpu
)
282 tvec_base_t
*base
= &per_cpu(tvec_bases
, cpu
);
285 BUG_ON(timer_pending(timer
) || !timer
->function
);
289 spin_lock_irqsave(&base
->t_base
.lock
, flags
);
290 timer
->base
= &base
->t_base
;
291 internal_add_timer(base
, timer
);
292 spin_unlock_irqrestore(&base
->t_base
.lock
, flags
);
297 * mod_timer - modify a timer's timeout
298 * @timer: the timer to be modified
300 * mod_timer is a more efficient way to update the expire field of an
301 * active timer (if the timer is inactive it will be activated)
303 * mod_timer(timer, expires) is equivalent to:
305 * del_timer(timer); timer->expires = expires; add_timer(timer);
307 * Note that if there are multiple unserialized concurrent users of the
308 * same timer, then mod_timer() is the only safe way to modify the timeout,
309 * since add_timer() cannot modify an already running timer.
311 * The function returns whether it has modified a pending timer or not.
312 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
313 * active timer returns 1.)
315 int mod_timer(struct timer_list
*timer
, unsigned long expires
)
317 BUG_ON(!timer
->function
);
322 * This is a common optimization triggered by the
323 * networking code - if the timer is re-modified
324 * to be the same thing then just return:
326 if (timer
->expires
== expires
&& timer_pending(timer
))
329 return __mod_timer(timer
, expires
);
332 EXPORT_SYMBOL(mod_timer
);
335 * del_timer - deactive a timer.
336 * @timer: the timer to be deactivated
338 * del_timer() deactivates a timer - this works on both active and inactive
341 * The function returns whether it has deactivated a pending timer or not.
342 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
343 * active timer returns 1.)
345 int del_timer(struct timer_list
*timer
)
353 if (timer_pending(timer
)) {
354 base
= lock_timer_base(timer
, &flags
);
355 if (timer_pending(timer
)) {
356 detach_timer(timer
, 1);
359 spin_unlock_irqrestore(&base
->lock
, flags
);
365 EXPORT_SYMBOL(del_timer
);
369 * This function tries to deactivate a timer. Upon successful (ret >= 0)
370 * exit the timer is not queued and the handler is not running on any CPU.
372 * It must not be called from interrupt contexts.
374 int try_to_del_timer_sync(struct timer_list
*timer
)
380 base
= lock_timer_base(timer
, &flags
);
382 if (base
->running_timer
== timer
)
386 if (timer_pending(timer
)) {
387 detach_timer(timer
, 1);
391 spin_unlock_irqrestore(&base
->lock
, flags
);
397 * del_timer_sync - deactivate a timer and wait for the handler to finish.
398 * @timer: the timer to be deactivated
400 * This function only differs from del_timer() on SMP: besides deactivating
401 * the timer it also makes sure the handler has finished executing on other
404 * Synchronization rules: callers must prevent restarting of the timer,
405 * otherwise this function is meaningless. It must not be called from
406 * interrupt contexts. The caller must not hold locks which would prevent
407 * completion of the timer's handler. The timer's handler must not call
408 * add_timer_on(). Upon exit the timer is not queued and the handler is
409 * not running on any CPU.
411 * The function returns whether it has deactivated a pending timer or not.
413 int del_timer_sync(struct timer_list
*timer
)
418 int ret
= try_to_del_timer_sync(timer
);
424 EXPORT_SYMBOL(del_timer_sync
);
427 static int cascade(tvec_base_t
*base
, tvec_t
*tv
, int index
)
429 /* cascade all the timers from tv up one level */
430 struct list_head
*head
, *curr
;
432 head
= tv
->vec
+ index
;
435 * We are removing _all_ timers from the list, so we don't have to
436 * detach them individually, just clear the list afterwards.
438 while (curr
!= head
) {
439 struct timer_list
*tmp
;
441 tmp
= list_entry(curr
, struct timer_list
, entry
);
442 BUG_ON(tmp
->base
!= &base
->t_base
);
444 internal_add_timer(base
, tmp
);
446 INIT_LIST_HEAD(head
);
452 * __run_timers - run all expired timers (if any) on this CPU.
453 * @base: the timer vector to be processed.
455 * This function cascades all vectors and executes all expired timer
458 #define INDEX(N) (base->timer_jiffies >> (TVR_BITS + N * TVN_BITS)) & TVN_MASK
460 static inline void __run_timers(tvec_base_t
*base
)
462 struct timer_list
*timer
;
464 spin_lock_irq(&base
->t_base
.lock
);
465 while (time_after_eq(jiffies
, base
->timer_jiffies
)) {
466 struct list_head work_list
= LIST_HEAD_INIT(work_list
);
467 struct list_head
*head
= &work_list
;
468 int index
= base
->timer_jiffies
& TVR_MASK
;
474 (!cascade(base
, &base
->tv2
, INDEX(0))) &&
475 (!cascade(base
, &base
->tv3
, INDEX(1))) &&
476 !cascade(base
, &base
->tv4
, INDEX(2)))
477 cascade(base
, &base
->tv5
, INDEX(3));
478 ++base
->timer_jiffies
;
479 list_splice_init(base
->tv1
.vec
+ index
, &work_list
);
480 while (!list_empty(head
)) {
481 void (*fn
)(unsigned long);
484 timer
= list_entry(head
->next
,struct timer_list
,entry
);
485 fn
= timer
->function
;
488 set_running_timer(base
, timer
);
489 detach_timer(timer
, 1);
490 spin_unlock_irq(&base
->t_base
.lock
);
492 int preempt_count
= preempt_count();
494 if (preempt_count
!= preempt_count()) {
495 printk(KERN_WARNING
"huh, entered %p "
496 "with preempt_count %08x, exited"
503 spin_lock_irq(&base
->t_base
.lock
);
506 set_running_timer(base
, NULL
);
507 spin_unlock_irq(&base
->t_base
.lock
);
510 #ifdef CONFIG_NO_IDLE_HZ
512 * Find out when the next timer event is due to happen. This
513 * is used on S/390 to stop all activity when a cpus is idle.
514 * This functions needs to be called disabled.
516 unsigned long next_timer_interrupt(void)
519 struct list_head
*list
;
520 struct timer_list
*nte
;
521 unsigned long expires
;
525 base
= &__get_cpu_var(tvec_bases
);
526 spin_lock(&base
->t_base
.lock
);
527 expires
= base
->timer_jiffies
+ (LONG_MAX
>> 1);
530 /* Look for timer events in tv1. */
531 j
= base
->timer_jiffies
& TVR_MASK
;
533 list_for_each_entry(nte
, base
->tv1
.vec
+ j
, entry
) {
534 expires
= nte
->expires
;
535 if (j
< (base
->timer_jiffies
& TVR_MASK
))
536 list
= base
->tv2
.vec
+ (INDEX(0));
539 j
= (j
+ 1) & TVR_MASK
;
540 } while (j
!= (base
->timer_jiffies
& TVR_MASK
));
543 varray
[0] = &base
->tv2
;
544 varray
[1] = &base
->tv3
;
545 varray
[2] = &base
->tv4
;
546 varray
[3] = &base
->tv5
;
547 for (i
= 0; i
< 4; i
++) {
550 if (list_empty(varray
[i
]->vec
+ j
)) {
551 j
= (j
+ 1) & TVN_MASK
;
554 list_for_each_entry(nte
, varray
[i
]->vec
+ j
, entry
)
555 if (time_before(nte
->expires
, expires
))
556 expires
= nte
->expires
;
557 if (j
< (INDEX(i
)) && i
< 3)
558 list
= varray
[i
+ 1]->vec
+ (INDEX(i
+ 1));
560 } while (j
!= (INDEX(i
)));
565 * The search wrapped. We need to look at the next list
566 * from next tv element that would cascade into tv element
567 * where we found the timer element.
569 list_for_each_entry(nte
, list
, entry
) {
570 if (time_before(nte
->expires
, expires
))
571 expires
= nte
->expires
;
574 spin_unlock(&base
->t_base
.lock
);
579 /******************************************************************/
582 * Timekeeping variables
584 unsigned long tick_usec
= TICK_USEC
; /* USER_HZ period (usec) */
585 unsigned long tick_nsec
= TICK_NSEC
; /* ACTHZ period (nsec) */
589 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
590 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
591 * at zero at system boot time, so wall_to_monotonic will be negative,
592 * however, we will ALWAYS keep the tv_nsec part positive so we can use
593 * the usual normalization.
595 struct timespec xtime
__attribute__ ((aligned (16)));
596 struct timespec wall_to_monotonic
__attribute__ ((aligned (16)));
598 EXPORT_SYMBOL(xtime
);
600 /* Don't completely fail for HZ > 500. */
601 int tickadj
= 500/HZ
? : 1; /* microsecs */
605 * phase-lock loop variables
607 /* TIME_ERROR prevents overwriting the CMOS clock */
608 int time_state
= TIME_OK
; /* clock synchronization status */
609 int time_status
= STA_UNSYNC
; /* clock status bits */
610 long time_offset
; /* time adjustment (us) */
611 long time_constant
= 2; /* pll time constant */
612 long time_tolerance
= MAXFREQ
; /* frequency tolerance (ppm) */
613 long time_precision
= 1; /* clock precision (us) */
614 long time_maxerror
= NTP_PHASE_LIMIT
; /* maximum error (us) */
615 long time_esterror
= NTP_PHASE_LIMIT
; /* estimated error (us) */
616 static long time_phase
; /* phase offset (scaled us) */
617 long time_freq
= (((NSEC_PER_SEC
+ HZ
/2) % HZ
- HZ
/2) << SHIFT_USEC
) / NSEC_PER_USEC
;
618 /* frequency offset (scaled ppm)*/
619 static long time_adj
; /* tick adjust (scaled 1 / HZ) */
620 long time_reftime
; /* time at last adjustment (s) */
622 long time_next_adjust
;
625 * this routine handles the overflow of the microsecond field
627 * The tricky bits of code to handle the accurate clock support
628 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
629 * They were originally developed for SUN and DEC kernels.
630 * All the kudos should go to Dave for this stuff.
633 static void second_overflow(void)
637 /* Bump the maxerror field */
638 time_maxerror
+= time_tolerance
>> SHIFT_USEC
;
639 if ( time_maxerror
> NTP_PHASE_LIMIT
) {
640 time_maxerror
= NTP_PHASE_LIMIT
;
641 time_status
|= STA_UNSYNC
;
645 * Leap second processing. If in leap-insert state at
646 * the end of the day, the system clock is set back one
647 * second; if in leap-delete state, the system clock is
648 * set ahead one second. The microtime() routine or
649 * external clock driver will insure that reported time
650 * is always monotonic. The ugly divides should be
653 switch (time_state
) {
656 if (time_status
& STA_INS
)
657 time_state
= TIME_INS
;
658 else if (time_status
& STA_DEL
)
659 time_state
= TIME_DEL
;
663 if (xtime
.tv_sec
% 86400 == 0) {
665 wall_to_monotonic
.tv_sec
++;
666 /* The timer interpolator will make time change gradually instead
667 * of an immediate jump by one second.
669 time_interpolator_update(-NSEC_PER_SEC
);
670 time_state
= TIME_OOP
;
672 printk(KERN_NOTICE
"Clock: inserting leap second 23:59:60 UTC\n");
677 if ((xtime
.tv_sec
+ 1) % 86400 == 0) {
679 wall_to_monotonic
.tv_sec
--;
680 /* Use of time interpolator for a gradual change of time */
681 time_interpolator_update(NSEC_PER_SEC
);
682 time_state
= TIME_WAIT
;
684 printk(KERN_NOTICE
"Clock: deleting leap second 23:59:59 UTC\n");
689 time_state
= TIME_WAIT
;
693 if (!(time_status
& (STA_INS
| STA_DEL
)))
694 time_state
= TIME_OK
;
698 * Compute the phase adjustment for the next second. In
699 * PLL mode, the offset is reduced by a fixed factor
700 * times the time constant. In FLL mode the offset is
701 * used directly. In either mode, the maximum phase
702 * adjustment for each second is clamped so as to spread
703 * the adjustment over not more than the number of
704 * seconds between updates.
706 if (time_offset
< 0) {
707 ltemp
= -time_offset
;
708 if (!(time_status
& STA_FLL
))
709 ltemp
>>= SHIFT_KG
+ time_constant
;
710 if (ltemp
> (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
)
711 ltemp
= (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
;
712 time_offset
+= ltemp
;
713 time_adj
= -ltemp
<< (SHIFT_SCALE
- SHIFT_HZ
- SHIFT_UPDATE
);
716 if (!(time_status
& STA_FLL
))
717 ltemp
>>= SHIFT_KG
+ time_constant
;
718 if (ltemp
> (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
)
719 ltemp
= (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
;
720 time_offset
-= ltemp
;
721 time_adj
= ltemp
<< (SHIFT_SCALE
- SHIFT_HZ
- SHIFT_UPDATE
);
725 * Compute the frequency estimate and additional phase
726 * adjustment due to frequency error for the next
727 * second. When the PPS signal is engaged, gnaw on the
728 * watchdog counter and update the frequency computed by
729 * the pll and the PPS signal.
732 if (pps_valid
== PPS_VALID
) { /* PPS signal lost */
733 pps_jitter
= MAXTIME
;
734 pps_stabil
= MAXFREQ
;
735 time_status
&= ~(STA_PPSSIGNAL
| STA_PPSJITTER
|
736 STA_PPSWANDER
| STA_PPSERROR
);
738 ltemp
= time_freq
+ pps_freq
;
740 time_adj
-= -ltemp
>>
741 (SHIFT_USEC
+ SHIFT_HZ
- SHIFT_SCALE
);
744 (SHIFT_USEC
+ SHIFT_HZ
- SHIFT_SCALE
);
747 /* Compensate for (HZ==100) != (1 << SHIFT_HZ).
748 * Add 25% and 3.125% to get 128.125; => only 0.125% error (p. 14)
751 time_adj
-= (-time_adj
>> 2) + (-time_adj
>> 5);
753 time_adj
+= (time_adj
>> 2) + (time_adj
>> 5);
756 /* Compensate for (HZ==1000) != (1 << SHIFT_HZ).
757 * Add 1.5625% and 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
760 time_adj
-= (-time_adj
>> 6) + (-time_adj
>> 7);
762 time_adj
+= (time_adj
>> 6) + (time_adj
>> 7);
766 /* in the NTP reference this is called "hardclock()" */
767 static void update_wall_time_one_tick(void)
769 long time_adjust_step
, delta_nsec
;
771 if ( (time_adjust_step
= time_adjust
) != 0 ) {
772 /* We are doing an adjtime thing.
774 * Prepare time_adjust_step to be within bounds.
775 * Note that a positive time_adjust means we want the clock
778 * Limit the amount of the step to be in the range
779 * -tickadj .. +tickadj
781 if (time_adjust
> tickadj
)
782 time_adjust_step
= tickadj
;
783 else if (time_adjust
< -tickadj
)
784 time_adjust_step
= -tickadj
;
786 /* Reduce by this step the amount of time left */
787 time_adjust
-= time_adjust_step
;
789 delta_nsec
= tick_nsec
+ time_adjust_step
* 1000;
791 * Advance the phase, once it gets to one microsecond, then
792 * advance the tick more.
794 time_phase
+= time_adj
;
795 if (time_phase
<= -FINENSEC
) {
796 long ltemp
= -time_phase
>> (SHIFT_SCALE
- 10);
797 time_phase
+= ltemp
<< (SHIFT_SCALE
- 10);
800 else if (time_phase
>= FINENSEC
) {
801 long ltemp
= time_phase
>> (SHIFT_SCALE
- 10);
802 time_phase
-= ltemp
<< (SHIFT_SCALE
- 10);
805 xtime
.tv_nsec
+= delta_nsec
;
806 time_interpolator_update(delta_nsec
);
808 /* Changes by adjtime() do not take effect till next tick. */
809 if (time_next_adjust
!= 0) {
810 time_adjust
= time_next_adjust
;
811 time_next_adjust
= 0;
816 * Using a loop looks inefficient, but "ticks" is
817 * usually just one (we shouldn't be losing ticks,
818 * we're doing this this way mainly for interrupt
819 * latency reasons, not because we think we'll
820 * have lots of lost timer ticks
822 static void update_wall_time(unsigned long ticks
)
826 update_wall_time_one_tick();
827 if (xtime
.tv_nsec
>= 1000000000) {
828 xtime
.tv_nsec
-= 1000000000;
836 * Called from the timer interrupt handler to charge one tick to the current
837 * process. user_tick is 1 if the tick is user time, 0 for system.
839 void update_process_times(int user_tick
)
841 struct task_struct
*p
= current
;
842 int cpu
= smp_processor_id();
844 /* Note: this timer irq context must be accounted for as well. */
846 account_user_time(p
, jiffies_to_cputime(1));
848 account_system_time(p
, HARDIRQ_OFFSET
, jiffies_to_cputime(1));
850 if (rcu_pending(cpu
))
851 rcu_check_callbacks(cpu
, user_tick
);
853 run_posix_cpu_timers(p
);
857 * Nr of active tasks - counted in fixed-point numbers
859 static unsigned long count_active_tasks(void)
861 return (nr_running() + nr_uninterruptible()) * FIXED_1
;
865 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
866 * imply that avenrun[] is the standard name for this kind of thing.
867 * Nothing else seems to be standardized: the fractional size etc
868 * all seem to differ on different machines.
870 * Requires xtime_lock to access.
872 unsigned long avenrun
[3];
874 EXPORT_SYMBOL(avenrun
);
877 * calc_load - given tick count, update the avenrun load estimates.
878 * This is called while holding a write_lock on xtime_lock.
880 static inline void calc_load(unsigned long ticks
)
882 unsigned long active_tasks
; /* fixed-point */
883 static int count
= LOAD_FREQ
;
888 active_tasks
= count_active_tasks();
889 CALC_LOAD(avenrun
[0], EXP_1
, active_tasks
);
890 CALC_LOAD(avenrun
[1], EXP_5
, active_tasks
);
891 CALC_LOAD(avenrun
[2], EXP_15
, active_tasks
);
895 /* jiffies at the most recent update of wall time */
896 unsigned long wall_jiffies
= INITIAL_JIFFIES
;
899 * This read-write spinlock protects us from races in SMP while
900 * playing with xtime and avenrun.
902 #ifndef ARCH_HAVE_XTIME_LOCK
903 seqlock_t xtime_lock __cacheline_aligned_in_smp
= SEQLOCK_UNLOCKED
;
905 EXPORT_SYMBOL(xtime_lock
);
909 * This function runs timers and the timer-tq in bottom half context.
911 static void run_timer_softirq(struct softirq_action
*h
)
913 tvec_base_t
*base
= &__get_cpu_var(tvec_bases
);
915 if (time_after_eq(jiffies
, base
->timer_jiffies
))
920 * Called by the local, per-CPU timer interrupt on SMP.
922 void run_local_timers(void)
924 raise_softirq(TIMER_SOFTIRQ
);
928 * Called by the timer interrupt. xtime_lock must already be taken
931 static inline void update_times(void)
935 ticks
= jiffies
- wall_jiffies
;
937 wall_jiffies
+= ticks
;
938 update_wall_time(ticks
);
944 * The 64-bit jiffies value is not atomic - you MUST NOT read it
945 * without sampling the sequence number in xtime_lock.
946 * jiffies is defined in the linker script...
949 void do_timer(struct pt_regs
*regs
)
955 #ifdef __ARCH_WANT_SYS_ALARM
958 * For backwards compatibility? This can be done in libc so Alpha
959 * and all newer ports shouldn't need it.
961 asmlinkage
unsigned long sys_alarm(unsigned int seconds
)
963 struct itimerval it_new
, it_old
;
964 unsigned int oldalarm
;
966 it_new
.it_interval
.tv_sec
= it_new
.it_interval
.tv_usec
= 0;
967 it_new
.it_value
.tv_sec
= seconds
;
968 it_new
.it_value
.tv_usec
= 0;
969 do_setitimer(ITIMER_REAL
, &it_new
, &it_old
);
970 oldalarm
= it_old
.it_value
.tv_sec
;
971 /* ehhh.. We can't return 0 if we have an alarm pending.. */
972 /* And we'd better return too much than too little anyway */
973 if ((!oldalarm
&& it_old
.it_value
.tv_usec
) || it_old
.it_value
.tv_usec
>= 500000)
983 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
984 * should be moved into arch/i386 instead?
988 * sys_getpid - return the thread group id of the current process
990 * Note, despite the name, this returns the tgid not the pid. The tgid and
991 * the pid are identical unless CLONE_THREAD was specified on clone() in
992 * which case the tgid is the same in all threads of the same group.
994 * This is SMP safe as current->tgid does not change.
996 asmlinkage
long sys_getpid(void)
998 return current
->tgid
;
1002 * Accessing ->group_leader->real_parent is not SMP-safe, it could
1003 * change from under us. However, rather than getting any lock
1004 * we can use an optimistic algorithm: get the parent
1005 * pid, and go back and check that the parent is still
1006 * the same. If it has changed (which is extremely unlikely
1007 * indeed), we just try again..
1009 * NOTE! This depends on the fact that even if we _do_
1010 * get an old value of "parent", we can happily dereference
1011 * the pointer (it was and remains a dereferencable kernel pointer
1012 * no matter what): we just can't necessarily trust the result
1013 * until we know that the parent pointer is valid.
1015 * NOTE2: ->group_leader never changes from under us.
1017 asmlinkage
long sys_getppid(void)
1020 struct task_struct
*me
= current
;
1021 struct task_struct
*parent
;
1023 parent
= me
->group_leader
->real_parent
;
1028 struct task_struct
*old
= parent
;
1031 * Make sure we read the pid before re-reading the
1035 parent
= me
->group_leader
->real_parent
;
1045 asmlinkage
long sys_getuid(void)
1047 /* Only we change this so SMP safe */
1048 return current
->uid
;
1051 asmlinkage
long sys_geteuid(void)
1053 /* Only we change this so SMP safe */
1054 return current
->euid
;
1057 asmlinkage
long sys_getgid(void)
1059 /* Only we change this so SMP safe */
1060 return current
->gid
;
1063 asmlinkage
long sys_getegid(void)
1065 /* Only we change this so SMP safe */
1066 return current
->egid
;
1071 static void process_timeout(unsigned long __data
)
1073 wake_up_process((task_t
*)__data
);
1077 * schedule_timeout - sleep until timeout
1078 * @timeout: timeout value in jiffies
1080 * Make the current task sleep until @timeout jiffies have
1081 * elapsed. The routine will return immediately unless
1082 * the current task state has been set (see set_current_state()).
1084 * You can set the task state as follows -
1086 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1087 * pass before the routine returns. The routine will return 0
1089 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1090 * delivered to the current task. In this case the remaining time
1091 * in jiffies will be returned, or 0 if the timer expired in time
1093 * The current task state is guaranteed to be TASK_RUNNING when this
1096 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1097 * the CPU away without a bound on the timeout. In this case the return
1098 * value will be %MAX_SCHEDULE_TIMEOUT.
1100 * In all cases the return value is guaranteed to be non-negative.
1102 fastcall
signed long __sched
schedule_timeout(signed long timeout
)
1104 struct timer_list timer
;
1105 unsigned long expire
;
1109 case MAX_SCHEDULE_TIMEOUT
:
1111 * These two special cases are useful to be comfortable
1112 * in the caller. Nothing more. We could take
1113 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1114 * but I' d like to return a valid offset (>=0) to allow
1115 * the caller to do everything it want with the retval.
1121 * Another bit of PARANOID. Note that the retval will be
1122 * 0 since no piece of kernel is supposed to do a check
1123 * for a negative retval of schedule_timeout() (since it
1124 * should never happens anyway). You just have the printk()
1125 * that will tell you if something is gone wrong and where.
1129 printk(KERN_ERR
"schedule_timeout: wrong timeout "
1130 "value %lx from %p\n", timeout
,
1131 __builtin_return_address(0));
1132 current
->state
= TASK_RUNNING
;
1137 expire
= timeout
+ jiffies
;
1140 timer
.expires
= expire
;
1141 timer
.data
= (unsigned long) current
;
1142 timer
.function
= process_timeout
;
1146 del_singleshot_timer_sync(&timer
);
1148 timeout
= expire
- jiffies
;
1151 return timeout
< 0 ? 0 : timeout
;
1154 EXPORT_SYMBOL(schedule_timeout
);
1156 /* Thread ID - the internal kernel "pid" */
1157 asmlinkage
long sys_gettid(void)
1159 return current
->pid
;
1162 static long __sched
nanosleep_restart(struct restart_block
*restart
)
1164 unsigned long expire
= restart
->arg0
, now
= jiffies
;
1165 struct timespec __user
*rmtp
= (struct timespec __user
*) restart
->arg1
;
1168 /* Did it expire while we handled signals? */
1169 if (!time_after(expire
, now
))
1172 current
->state
= TASK_INTERRUPTIBLE
;
1173 expire
= schedule_timeout(expire
- now
);
1178 jiffies_to_timespec(expire
, &t
);
1180 ret
= -ERESTART_RESTARTBLOCK
;
1181 if (rmtp
&& copy_to_user(rmtp
, &t
, sizeof(t
)))
1183 /* The 'restart' block is already filled in */
1188 asmlinkage
long sys_nanosleep(struct timespec __user
*rqtp
, struct timespec __user
*rmtp
)
1191 unsigned long expire
;
1194 if (copy_from_user(&t
, rqtp
, sizeof(t
)))
1197 if ((t
.tv_nsec
>= 1000000000L) || (t
.tv_nsec
< 0) || (t
.tv_sec
< 0))
1200 expire
= timespec_to_jiffies(&t
) + (t
.tv_sec
|| t
.tv_nsec
);
1201 current
->state
= TASK_INTERRUPTIBLE
;
1202 expire
= schedule_timeout(expire
);
1206 struct restart_block
*restart
;
1207 jiffies_to_timespec(expire
, &t
);
1208 if (rmtp
&& copy_to_user(rmtp
, &t
, sizeof(t
)))
1211 restart
= ¤t_thread_info()->restart_block
;
1212 restart
->fn
= nanosleep_restart
;
1213 restart
->arg0
= jiffies
+ expire
;
1214 restart
->arg1
= (unsigned long) rmtp
;
1215 ret
= -ERESTART_RESTARTBLOCK
;
1221 * sys_sysinfo - fill in sysinfo struct
1223 asmlinkage
long sys_sysinfo(struct sysinfo __user
*info
)
1226 unsigned long mem_total
, sav_total
;
1227 unsigned int mem_unit
, bitcount
;
1230 memset((char *)&val
, 0, sizeof(struct sysinfo
));
1234 seq
= read_seqbegin(&xtime_lock
);
1237 * This is annoying. The below is the same thing
1238 * posix_get_clock_monotonic() does, but it wants to
1239 * take the lock which we want to cover the loads stuff
1243 getnstimeofday(&tp
);
1244 tp
.tv_sec
+= wall_to_monotonic
.tv_sec
;
1245 tp
.tv_nsec
+= wall_to_monotonic
.tv_nsec
;
1246 if (tp
.tv_nsec
- NSEC_PER_SEC
>= 0) {
1247 tp
.tv_nsec
= tp
.tv_nsec
- NSEC_PER_SEC
;
1250 val
.uptime
= tp
.tv_sec
+ (tp
.tv_nsec
? 1 : 0);
1252 val
.loads
[0] = avenrun
[0] << (SI_LOAD_SHIFT
- FSHIFT
);
1253 val
.loads
[1] = avenrun
[1] << (SI_LOAD_SHIFT
- FSHIFT
);
1254 val
.loads
[2] = avenrun
[2] << (SI_LOAD_SHIFT
- FSHIFT
);
1256 val
.procs
= nr_threads
;
1257 } while (read_seqretry(&xtime_lock
, seq
));
1263 * If the sum of all the available memory (i.e. ram + swap)
1264 * is less than can be stored in a 32 bit unsigned long then
1265 * we can be binary compatible with 2.2.x kernels. If not,
1266 * well, in that case 2.2.x was broken anyways...
1268 * -Erik Andersen <andersee@debian.org>
1271 mem_total
= val
.totalram
+ val
.totalswap
;
1272 if (mem_total
< val
.totalram
|| mem_total
< val
.totalswap
)
1275 mem_unit
= val
.mem_unit
;
1276 while (mem_unit
> 1) {
1279 sav_total
= mem_total
;
1281 if (mem_total
< sav_total
)
1286 * If mem_total did not overflow, multiply all memory values by
1287 * val.mem_unit and set it to 1. This leaves things compatible
1288 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1293 val
.totalram
<<= bitcount
;
1294 val
.freeram
<<= bitcount
;
1295 val
.sharedram
<<= bitcount
;
1296 val
.bufferram
<<= bitcount
;
1297 val
.totalswap
<<= bitcount
;
1298 val
.freeswap
<<= bitcount
;
1299 val
.totalhigh
<<= bitcount
;
1300 val
.freehigh
<<= bitcount
;
1303 if (copy_to_user(info
, &val
, sizeof(struct sysinfo
)))
1309 static void __devinit
init_timers_cpu(int cpu
)
1314 base
= &per_cpu(tvec_bases
, cpu
);
1315 spin_lock_init(&base
->t_base
.lock
);
1316 for (j
= 0; j
< TVN_SIZE
; j
++) {
1317 INIT_LIST_HEAD(base
->tv5
.vec
+ j
);
1318 INIT_LIST_HEAD(base
->tv4
.vec
+ j
);
1319 INIT_LIST_HEAD(base
->tv3
.vec
+ j
);
1320 INIT_LIST_HEAD(base
->tv2
.vec
+ j
);
1322 for (j
= 0; j
< TVR_SIZE
; j
++)
1323 INIT_LIST_HEAD(base
->tv1
.vec
+ j
);
1325 base
->timer_jiffies
= jiffies
;
1328 #ifdef CONFIG_HOTPLUG_CPU
1329 static void migrate_timer_list(tvec_base_t
*new_base
, struct list_head
*head
)
1331 struct timer_list
*timer
;
1333 while (!list_empty(head
)) {
1334 timer
= list_entry(head
->next
, struct timer_list
, entry
);
1335 detach_timer(timer
, 0);
1336 timer
->base
= &new_base
->t_base
;
1337 internal_add_timer(new_base
, timer
);
1341 static void __devinit
migrate_timers(int cpu
)
1343 tvec_base_t
*old_base
;
1344 tvec_base_t
*new_base
;
1347 BUG_ON(cpu_online(cpu
));
1348 old_base
= &per_cpu(tvec_bases
, cpu
);
1349 new_base
= &get_cpu_var(tvec_bases
);
1351 local_irq_disable();
1352 spin_lock(&new_base
->t_base
.lock
);
1353 spin_lock(&old_base
->t_base
.lock
);
1355 if (old_base
->t_base
.running_timer
)
1357 for (i
= 0; i
< TVR_SIZE
; i
++)
1358 migrate_timer_list(new_base
, old_base
->tv1
.vec
+ i
);
1359 for (i
= 0; i
< TVN_SIZE
; i
++) {
1360 migrate_timer_list(new_base
, old_base
->tv2
.vec
+ i
);
1361 migrate_timer_list(new_base
, old_base
->tv3
.vec
+ i
);
1362 migrate_timer_list(new_base
, old_base
->tv4
.vec
+ i
);
1363 migrate_timer_list(new_base
, old_base
->tv5
.vec
+ i
);
1366 spin_unlock(&old_base
->t_base
.lock
);
1367 spin_unlock(&new_base
->t_base
.lock
);
1369 put_cpu_var(tvec_bases
);
1371 #endif /* CONFIG_HOTPLUG_CPU */
1373 static int __devinit
timer_cpu_notify(struct notifier_block
*self
,
1374 unsigned long action
, void *hcpu
)
1376 long cpu
= (long)hcpu
;
1378 case CPU_UP_PREPARE
:
1379 init_timers_cpu(cpu
);
1381 #ifdef CONFIG_HOTPLUG_CPU
1383 migrate_timers(cpu
);
1392 static struct notifier_block __devinitdata timers_nb
= {
1393 .notifier_call
= timer_cpu_notify
,
1397 void __init
init_timers(void)
1399 timer_cpu_notify(&timers_nb
, (unsigned long)CPU_UP_PREPARE
,
1400 (void *)(long)smp_processor_id());
1401 register_cpu_notifier(&timers_nb
);
1402 open_softirq(TIMER_SOFTIRQ
, run_timer_softirq
, NULL
);
1405 #ifdef CONFIG_TIME_INTERPOLATION
1407 struct time_interpolator
*time_interpolator
;
1408 static struct time_interpolator
*time_interpolator_list
;
1409 static DEFINE_SPINLOCK(time_interpolator_lock
);
1411 static inline u64
time_interpolator_get_cycles(unsigned int src
)
1413 unsigned long (*x
)(void);
1417 case TIME_SOURCE_FUNCTION
:
1418 x
= time_interpolator
->addr
;
1421 case TIME_SOURCE_MMIO64
:
1422 return readq((void __iomem
*) time_interpolator
->addr
);
1424 case TIME_SOURCE_MMIO32
:
1425 return readl((void __iomem
*) time_interpolator
->addr
);
1427 default: return get_cycles();
1431 static inline u64
time_interpolator_get_counter(void)
1433 unsigned int src
= time_interpolator
->source
;
1435 if (time_interpolator
->jitter
)
1441 lcycle
= time_interpolator
->last_cycle
;
1442 now
= time_interpolator_get_cycles(src
);
1443 if (lcycle
&& time_after(lcycle
, now
))
1445 /* Keep track of the last timer value returned. The use of cmpxchg here
1446 * will cause contention in an SMP environment.
1448 } while (unlikely(cmpxchg(&time_interpolator
->last_cycle
, lcycle
, now
) != lcycle
));
1452 return time_interpolator_get_cycles(src
);
1455 void time_interpolator_reset(void)
1457 time_interpolator
->offset
= 0;
1458 time_interpolator
->last_counter
= time_interpolator_get_counter();
1461 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1463 unsigned long time_interpolator_get_offset(void)
1465 /* If we do not have a time interpolator set up then just return zero */
1466 if (!time_interpolator
)
1469 return time_interpolator
->offset
+
1470 GET_TI_NSECS(time_interpolator_get_counter(), time_interpolator
);
1473 #define INTERPOLATOR_ADJUST 65536
1474 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1476 static void time_interpolator_update(long delta_nsec
)
1479 unsigned long offset
;
1481 /* If there is no time interpolator set up then do nothing */
1482 if (!time_interpolator
)
1485 /* The interpolator compensates for late ticks by accumulating
1486 * the late time in time_interpolator->offset. A tick earlier than
1487 * expected will lead to a reset of the offset and a corresponding
1488 * jump of the clock forward. Again this only works if the
1489 * interpolator clock is running slightly slower than the regular clock
1490 * and the tuning logic insures that.
1493 counter
= time_interpolator_get_counter();
1494 offset
= time_interpolator
->offset
+ GET_TI_NSECS(counter
, time_interpolator
);
1496 if (delta_nsec
< 0 || (unsigned long) delta_nsec
< offset
)
1497 time_interpolator
->offset
= offset
- delta_nsec
;
1499 time_interpolator
->skips
++;
1500 time_interpolator
->ns_skipped
+= delta_nsec
- offset
;
1501 time_interpolator
->offset
= 0;
1503 time_interpolator
->last_counter
= counter
;
1505 /* Tuning logic for time interpolator invoked every minute or so.
1506 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1507 * Increase interpolator clock speed if we skip too much time.
1509 if (jiffies
% INTERPOLATOR_ADJUST
== 0)
1511 if (time_interpolator
->skips
== 0 && time_interpolator
->offset
> TICK_NSEC
)
1512 time_interpolator
->nsec_per_cyc
--;
1513 if (time_interpolator
->ns_skipped
> INTERPOLATOR_MAX_SKIP
&& time_interpolator
->offset
== 0)
1514 time_interpolator
->nsec_per_cyc
++;
1515 time_interpolator
->skips
= 0;
1516 time_interpolator
->ns_skipped
= 0;
1521 is_better_time_interpolator(struct time_interpolator
*new)
1523 if (!time_interpolator
)
1525 return new->frequency
> 2*time_interpolator
->frequency
||
1526 (unsigned long)new->drift
< (unsigned long)time_interpolator
->drift
;
1530 register_time_interpolator(struct time_interpolator
*ti
)
1532 unsigned long flags
;
1535 if (ti
->frequency
== 0 || ti
->mask
== 0)
1538 ti
->nsec_per_cyc
= ((u64
)NSEC_PER_SEC
<< ti
->shift
) / ti
->frequency
;
1539 spin_lock(&time_interpolator_lock
);
1540 write_seqlock_irqsave(&xtime_lock
, flags
);
1541 if (is_better_time_interpolator(ti
)) {
1542 time_interpolator
= ti
;
1543 time_interpolator_reset();
1545 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1547 ti
->next
= time_interpolator_list
;
1548 time_interpolator_list
= ti
;
1549 spin_unlock(&time_interpolator_lock
);
1553 unregister_time_interpolator(struct time_interpolator
*ti
)
1555 struct time_interpolator
*curr
, **prev
;
1556 unsigned long flags
;
1558 spin_lock(&time_interpolator_lock
);
1559 prev
= &time_interpolator_list
;
1560 for (curr
= *prev
; curr
; curr
= curr
->next
) {
1568 write_seqlock_irqsave(&xtime_lock
, flags
);
1569 if (ti
== time_interpolator
) {
1570 /* we lost the best time-interpolator: */
1571 time_interpolator
= NULL
;
1572 /* find the next-best interpolator */
1573 for (curr
= time_interpolator_list
; curr
; curr
= curr
->next
)
1574 if (is_better_time_interpolator(curr
))
1575 time_interpolator
= curr
;
1576 time_interpolator_reset();
1578 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1579 spin_unlock(&time_interpolator_lock
);
1581 #endif /* CONFIG_TIME_INTERPOLATION */
1584 * msleep - sleep safely even with waitqueue interruptions
1585 * @msecs: Time in milliseconds to sleep for
1587 void msleep(unsigned int msecs
)
1589 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1592 set_current_state(TASK_UNINTERRUPTIBLE
);
1593 timeout
= schedule_timeout(timeout
);
1597 EXPORT_SYMBOL(msleep
);
1600 * msleep_interruptible - sleep waiting for waitqueue interruptions
1601 * @msecs: Time in milliseconds to sleep for
1603 unsigned long msleep_interruptible(unsigned int msecs
)
1605 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1607 while (timeout
&& !signal_pending(current
)) {
1608 set_current_state(TASK_INTERRUPTIBLE
);
1609 timeout
= schedule_timeout(timeout
);
1611 return jiffies_to_msecs(timeout
);
1614 EXPORT_SYMBOL(msleep_interruptible
);