2 * linux/kernel/time/timekeeping.c
4 * Kernel timekeeping code and accessor functions
6 * This code was moved from linux/kernel/timer.c.
7 * Please see that file for copyright and history logs.
11 #include <linux/module.h>
12 #include <linux/interrupt.h>
13 #include <linux/percpu.h>
14 #include <linux/init.h>
16 #include <linux/sysdev.h>
17 #include <linux/clocksource.h>
18 #include <linux/jiffies.h>
19 #include <linux/time.h>
20 #include <linux/tick.h>
22 /* Structure holding internal timekeeping values. */
24 /* Current clocksource used for timekeeping. */
25 struct clocksource
*clock
;
26 /* The shift value of the current clocksource. */
29 /* Number of clock cycles in one NTP interval. */
30 cycle_t cycle_interval
;
31 /* Number of clock shifted nano seconds in one NTP interval. */
33 /* Raw nano seconds accumulated per NTP interval. */
36 /* Clock shifted nano seconds remainder not stored in xtime.tv_nsec. */
38 /* Difference between accumulated time and NTP time in ntp
39 * shifted nano seconds. */
41 /* Shift conversion between clock shifted nano seconds and
42 * ntp shifted nano seconds. */
44 /* NTP adjusted clock multiplier */
48 struct timekeeper timekeeper
;
51 * timekeeper_setup_internals - Set up internals to use clocksource clock.
53 * @clock: Pointer to clocksource.
55 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
56 * pair and interval request.
58 * Unless you're the timekeeping code, you should not be using this!
60 static void timekeeper_setup_internals(struct clocksource
*clock
)
65 timekeeper
.clock
= clock
;
66 clock
->cycle_last
= clock
->read(clock
);
68 /* Do the ns -> cycle conversion first, using original mult */
69 tmp
= NTP_INTERVAL_LENGTH
;
72 do_div(tmp
, clock
->mult
);
76 interval
= (cycle_t
) tmp
;
77 timekeeper
.cycle_interval
= interval
;
79 /* Go back from cycles -> shifted ns */
80 timekeeper
.xtime_interval
= (u64
) interval
* clock
->mult
;
81 timekeeper
.raw_interval
=
82 ((u64
) interval
* clock
->mult
) >> clock
->shift
;
84 timekeeper
.xtime_nsec
= 0;
85 timekeeper
.shift
= clock
->shift
;
87 timekeeper
.ntp_error
= 0;
88 timekeeper
.ntp_error_shift
= NTP_SCALE_SHIFT
- clock
->shift
;
91 * The timekeeper keeps its own mult values for the currently
92 * active clocksource. These value will be adjusted via NTP
93 * to counteract clock drifting.
95 timekeeper
.mult
= clock
->mult
;
98 /* Timekeeper helper functions. */
99 static inline s64
timekeeping_get_ns(void)
101 cycle_t cycle_now
, cycle_delta
;
102 struct clocksource
*clock
;
104 /* read clocksource: */
105 clock
= timekeeper
.clock
;
106 cycle_now
= clock
->read(clock
);
108 /* calculate the delta since the last update_wall_time: */
109 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
111 /* return delta convert to nanoseconds using ntp adjusted mult. */
112 return clocksource_cyc2ns(cycle_delta
, timekeeper
.mult
,
116 static inline s64
timekeeping_get_ns_raw(void)
118 cycle_t cycle_now
, cycle_delta
;
119 struct clocksource
*clock
;
121 /* read clocksource: */
122 clock
= timekeeper
.clock
;
123 cycle_now
= clock
->read(clock
);
125 /* calculate the delta since the last update_wall_time: */
126 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
128 /* return delta convert to nanoseconds using ntp adjusted mult. */
129 return clocksource_cyc2ns(cycle_delta
, clock
->mult
, clock
->shift
);
133 * This read-write spinlock protects us from races in SMP while
134 * playing with xtime.
136 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(xtime_lock
);
141 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
142 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
143 * at zero at system boot time, so wall_to_monotonic will be negative,
144 * however, we will ALWAYS keep the tv_nsec part positive so we can use
145 * the usual normalization.
147 * wall_to_monotonic is moved after resume from suspend for the monotonic
148 * time not to jump. We need to add total_sleep_time to wall_to_monotonic
149 * to get the real boot based time offset.
151 * - wall_to_monotonic is no longer the boot time, getboottime must be
154 struct timespec xtime
__attribute__ ((aligned (16)));
155 struct timespec wall_to_monotonic
__attribute__ ((aligned (16)));
156 static unsigned long total_sleep_time
; /* seconds */
159 * The raw monotonic time for the CLOCK_MONOTONIC_RAW posix clock.
161 struct timespec raw_time
;
163 /* flag for if timekeeping is suspended */
164 int __read_mostly timekeeping_suspended
;
166 static struct timespec xtime_cache
__attribute__ ((aligned (16)));
167 void update_xtime_cache(u64 nsec
)
170 timespec_add_ns(&xtime_cache
, nsec
);
173 /* must hold xtime_lock */
174 void timekeeping_leap_insert(int leapsecond
)
176 xtime
.tv_sec
+= leapsecond
;
177 wall_to_monotonic
.tv_sec
-= leapsecond
;
178 update_vsyscall(&xtime
, timekeeper
.clock
);
181 #ifdef CONFIG_GENERIC_TIME
183 * timekeeping_forward_now - update clock to the current time
185 * Forward the current clock to update its state since the last call to
186 * update_wall_time(). This is useful before significant clock changes,
187 * as it avoids having to deal with this time offset explicitly.
189 static void timekeeping_forward_now(void)
191 cycle_t cycle_now
, cycle_delta
;
192 struct clocksource
*clock
;
195 clock
= timekeeper
.clock
;
196 cycle_now
= clock
->read(clock
);
197 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
198 clock
->cycle_last
= cycle_now
;
200 nsec
= clocksource_cyc2ns(cycle_delta
, timekeeper
.mult
,
203 /* If arch requires, add in gettimeoffset() */
204 nsec
+= arch_gettimeoffset();
206 timespec_add_ns(&xtime
, nsec
);
208 nsec
= clocksource_cyc2ns(cycle_delta
, clock
->mult
, clock
->shift
);
209 timespec_add_ns(&raw_time
, nsec
);
213 * getnstimeofday - Returns the time of day in a timespec
214 * @ts: pointer to the timespec to be set
216 * Returns the time of day in a timespec.
218 void getnstimeofday(struct timespec
*ts
)
223 WARN_ON(timekeeping_suspended
);
226 seq
= read_seqbegin(&xtime_lock
);
229 nsecs
= timekeeping_get_ns();
231 /* If arch requires, add in gettimeoffset() */
232 nsecs
+= arch_gettimeoffset();
234 } while (read_seqretry(&xtime_lock
, seq
));
236 timespec_add_ns(ts
, nsecs
);
239 EXPORT_SYMBOL(getnstimeofday
);
241 ktime_t
ktime_get(void)
246 WARN_ON(timekeeping_suspended
);
249 seq
= read_seqbegin(&xtime_lock
);
250 secs
= xtime
.tv_sec
+ wall_to_monotonic
.tv_sec
;
251 nsecs
= xtime
.tv_nsec
+ wall_to_monotonic
.tv_nsec
;
252 nsecs
+= timekeeping_get_ns();
254 } while (read_seqretry(&xtime_lock
, seq
));
256 * Use ktime_set/ktime_add_ns to create a proper ktime on
257 * 32-bit architectures without CONFIG_KTIME_SCALAR.
259 return ktime_add_ns(ktime_set(secs
, 0), nsecs
);
261 EXPORT_SYMBOL_GPL(ktime_get
);
264 * ktime_get_ts - get the monotonic clock in timespec format
265 * @ts: pointer to timespec variable
267 * The function calculates the monotonic clock from the realtime
268 * clock and the wall_to_monotonic offset and stores the result
269 * in normalized timespec format in the variable pointed to by @ts.
271 void ktime_get_ts(struct timespec
*ts
)
273 struct timespec tomono
;
277 WARN_ON(timekeeping_suspended
);
280 seq
= read_seqbegin(&xtime_lock
);
282 tomono
= wall_to_monotonic
;
283 nsecs
= timekeeping_get_ns();
285 } while (read_seqretry(&xtime_lock
, seq
));
287 set_normalized_timespec(ts
, ts
->tv_sec
+ tomono
.tv_sec
,
288 ts
->tv_nsec
+ tomono
.tv_nsec
+ nsecs
);
290 EXPORT_SYMBOL_GPL(ktime_get_ts
);
293 * do_gettimeofday - Returns the time of day in a timeval
294 * @tv: pointer to the timeval to be set
296 * NOTE: Users should be converted to using getnstimeofday()
298 void do_gettimeofday(struct timeval
*tv
)
302 getnstimeofday(&now
);
303 tv
->tv_sec
= now
.tv_sec
;
304 tv
->tv_usec
= now
.tv_nsec
/1000;
307 EXPORT_SYMBOL(do_gettimeofday
);
309 * do_settimeofday - Sets the time of day
310 * @tv: pointer to the timespec variable containing the new time
312 * Sets the time of day to the new time and update NTP and notify hrtimers
314 int do_settimeofday(struct timespec
*tv
)
316 struct timespec ts_delta
;
319 if ((unsigned long)tv
->tv_nsec
>= NSEC_PER_SEC
)
322 write_seqlock_irqsave(&xtime_lock
, flags
);
324 timekeeping_forward_now();
326 ts_delta
.tv_sec
= tv
->tv_sec
- xtime
.tv_sec
;
327 ts_delta
.tv_nsec
= tv
->tv_nsec
- xtime
.tv_nsec
;
328 wall_to_monotonic
= timespec_sub(wall_to_monotonic
, ts_delta
);
332 update_xtime_cache(0);
334 timekeeper
.ntp_error
= 0;
337 update_vsyscall(&xtime
, timekeeper
.clock
);
339 write_sequnlock_irqrestore(&xtime_lock
, flags
);
341 /* signal hrtimers about time change */
347 EXPORT_SYMBOL(do_settimeofday
);
350 * change_clocksource - Swaps clocksources if a new one is available
352 * Accumulates current time interval and initializes new clocksource
354 static void change_clocksource(void)
356 struct clocksource
*new, *old
;
358 new = clocksource_get_next();
360 if (!new || timekeeper
.clock
== new)
363 timekeeping_forward_now();
365 if (new->enable
&& !new->enable(new))
368 old
= timekeeper
.clock
;
369 timekeeper_setup_internals(new);
376 #else /* GENERIC_TIME */
377 static inline void timekeeping_forward_now(void) { }
378 static inline void change_clocksource(void) { }
381 * ktime_get - get the monotonic time in ktime_t format
383 * returns the time in ktime_t format
385 ktime_t
ktime_get(void)
391 return timespec_to_ktime(now
);
393 EXPORT_SYMBOL_GPL(ktime_get
);
396 * ktime_get_ts - get the monotonic clock in timespec format
397 * @ts: pointer to timespec variable
399 * The function calculates the monotonic clock from the realtime
400 * clock and the wall_to_monotonic offset and stores the result
401 * in normalized timespec format in the variable pointed to by @ts.
403 void ktime_get_ts(struct timespec
*ts
)
405 struct timespec tomono
;
409 seq
= read_seqbegin(&xtime_lock
);
411 tomono
= wall_to_monotonic
;
413 } while (read_seqretry(&xtime_lock
, seq
));
415 set_normalized_timespec(ts
, ts
->tv_sec
+ tomono
.tv_sec
,
416 ts
->tv_nsec
+ tomono
.tv_nsec
);
418 EXPORT_SYMBOL_GPL(ktime_get_ts
);
419 #endif /* !GENERIC_TIME */
422 * ktime_get_real - get the real (wall-) time in ktime_t format
424 * returns the time in ktime_t format
426 ktime_t
ktime_get_real(void)
430 getnstimeofday(&now
);
432 return timespec_to_ktime(now
);
434 EXPORT_SYMBOL_GPL(ktime_get_real
);
437 * getrawmonotonic - Returns the raw monotonic time in a timespec
438 * @ts: pointer to the timespec to be set
440 * Returns the raw monotonic time (completely un-modified by ntp)
442 void getrawmonotonic(struct timespec
*ts
)
448 seq
= read_seqbegin(&xtime_lock
);
449 nsecs
= timekeeping_get_ns_raw();
452 } while (read_seqretry(&xtime_lock
, seq
));
454 timespec_add_ns(ts
, nsecs
);
456 EXPORT_SYMBOL(getrawmonotonic
);
460 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
462 int timekeeping_valid_for_hres(void)
468 seq
= read_seqbegin(&xtime_lock
);
470 ret
= timekeeper
.clock
->flags
& CLOCK_SOURCE_VALID_FOR_HRES
;
472 } while (read_seqretry(&xtime_lock
, seq
));
478 * read_persistent_clock - Return time in seconds from the persistent clock.
480 * Weak dummy function for arches that do not yet support it.
481 * Returns seconds from epoch using the battery backed persistent clock.
482 * Returns zero if unsupported.
484 * XXX - Do be sure to remove it once all arches implement it.
486 unsigned long __attribute__((weak
)) read_persistent_clock(void)
492 * timekeeping_init - Initializes the clocksource and common timekeeping values
494 void __init
timekeeping_init(void)
496 struct clocksource
*clock
;
498 unsigned long sec
= read_persistent_clock();
500 write_seqlock_irqsave(&xtime_lock
, flags
);
504 clock
= clocksource_default_clock();
506 clock
->enable(clock
);
507 timekeeper_setup_internals(clock
);
512 raw_time
.tv_nsec
= 0;
513 set_normalized_timespec(&wall_to_monotonic
,
514 -xtime
.tv_sec
, -xtime
.tv_nsec
);
515 update_xtime_cache(0);
516 total_sleep_time
= 0;
517 write_sequnlock_irqrestore(&xtime_lock
, flags
);
520 /* time in seconds when suspend began */
521 static unsigned long timekeeping_suspend_time
;
524 * timekeeping_resume - Resumes the generic timekeeping subsystem.
527 * This is for the generic clocksource timekeeping.
528 * xtime/wall_to_monotonic/jiffies/etc are
529 * still managed by arch specific suspend/resume code.
531 static int timekeeping_resume(struct sys_device
*dev
)
534 unsigned long now
= read_persistent_clock();
536 clocksource_resume();
538 write_seqlock_irqsave(&xtime_lock
, flags
);
540 if (now
&& (now
> timekeeping_suspend_time
)) {
541 unsigned long sleep_length
= now
- timekeeping_suspend_time
;
543 xtime
.tv_sec
+= sleep_length
;
544 wall_to_monotonic
.tv_sec
-= sleep_length
;
545 total_sleep_time
+= sleep_length
;
547 update_xtime_cache(0);
548 /* re-base the last cycle value */
549 timekeeper
.clock
->cycle_last
= timekeeper
.clock
->read(timekeeper
.clock
);
550 timekeeper
.ntp_error
= 0;
551 timekeeping_suspended
= 0;
552 write_sequnlock_irqrestore(&xtime_lock
, flags
);
554 touch_softlockup_watchdog();
556 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME
, NULL
);
558 /* Resume hrtimers */
559 hres_timers_resume();
564 static int timekeeping_suspend(struct sys_device
*dev
, pm_message_t state
)
568 timekeeping_suspend_time
= read_persistent_clock();
570 write_seqlock_irqsave(&xtime_lock
, flags
);
571 timekeeping_forward_now();
572 timekeeping_suspended
= 1;
573 write_sequnlock_irqrestore(&xtime_lock
, flags
);
575 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND
, NULL
);
580 /* sysfs resume/suspend bits for timekeeping */
581 static struct sysdev_class timekeeping_sysclass
= {
582 .name
= "timekeeping",
583 .resume
= timekeeping_resume
,
584 .suspend
= timekeeping_suspend
,
587 static struct sys_device device_timer
= {
589 .cls
= &timekeeping_sysclass
,
592 static int __init
timekeeping_init_device(void)
594 int error
= sysdev_class_register(&timekeeping_sysclass
);
596 error
= sysdev_register(&device_timer
);
600 device_initcall(timekeeping_init_device
);
603 * If the error is already larger, we look ahead even further
604 * to compensate for late or lost adjustments.
606 static __always_inline
int timekeeping_bigadjust(s64 error
, s64
*interval
,
614 * Use the current error value to determine how much to look ahead.
615 * The larger the error the slower we adjust for it to avoid problems
616 * with losing too many ticks, otherwise we would overadjust and
617 * produce an even larger error. The smaller the adjustment the
618 * faster we try to adjust for it, as lost ticks can do less harm
619 * here. This is tuned so that an error of about 1 msec is adjusted
620 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
622 error2
= timekeeper
.ntp_error
>> (NTP_SCALE_SHIFT
+ 22 - 2 * SHIFT_HZ
);
623 error2
= abs(error2
);
624 for (look_ahead
= 0; error2
> 0; look_ahead
++)
628 * Now calculate the error in (1 << look_ahead) ticks, but first
629 * remove the single look ahead already included in the error.
631 tick_error
= tick_length
>> (timekeeper
.ntp_error_shift
+ 1);
632 tick_error
-= timekeeper
.xtime_interval
>> 1;
633 error
= ((error
- tick_error
) >> look_ahead
) + tick_error
;
635 /* Finally calculate the adjustment shift value. */
640 *interval
= -*interval
;
644 for (adj
= 0; error
> i
; adj
++)
653 * Adjust the multiplier to reduce the error value,
654 * this is optimized for the most common adjustments of -1,0,1,
655 * for other values we can do a bit more work.
657 static void timekeeping_adjust(s64 offset
)
659 s64 error
, interval
= timekeeper
.cycle_interval
;
662 error
= timekeeper
.ntp_error
>> (timekeeper
.ntp_error_shift
- 1);
663 if (error
> interval
) {
665 if (likely(error
<= interval
))
668 adj
= timekeeping_bigadjust(error
, &interval
, &offset
);
669 } else if (error
< -interval
) {
671 if (likely(error
>= -interval
)) {
673 interval
= -interval
;
676 adj
= timekeeping_bigadjust(error
, &interval
, &offset
);
680 timekeeper
.mult
+= adj
;
681 timekeeper
.xtime_interval
+= interval
;
682 timekeeper
.xtime_nsec
-= offset
;
683 timekeeper
.ntp_error
-= (interval
- offset
) <<
684 timekeeper
.ntp_error_shift
;
688 * update_wall_time - Uses the current clocksource to increment the wall time
690 * Called from the timer interrupt, must hold a write on xtime_lock.
692 void update_wall_time(void)
694 struct clocksource
*clock
;
698 /* Make sure we're fully resumed: */
699 if (unlikely(timekeeping_suspended
))
702 clock
= timekeeper
.clock
;
703 #ifdef CONFIG_GENERIC_TIME
704 offset
= (clock
->read(clock
) - clock
->cycle_last
) & clock
->mask
;
706 offset
= timekeeper
.cycle_interval
;
708 timekeeper
.xtime_nsec
= (s64
)xtime
.tv_nsec
<< timekeeper
.shift
;
710 /* normally this loop will run just once, however in the
711 * case of lost or late ticks, it will accumulate correctly.
713 while (offset
>= timekeeper
.cycle_interval
) {
714 u64 nsecps
= (u64
)NSEC_PER_SEC
<< timekeeper
.shift
;
716 /* accumulate one interval */
717 offset
-= timekeeper
.cycle_interval
;
718 clock
->cycle_last
+= timekeeper
.cycle_interval
;
720 timekeeper
.xtime_nsec
+= timekeeper
.xtime_interval
;
721 if (timekeeper
.xtime_nsec
>= nsecps
) {
722 timekeeper
.xtime_nsec
-= nsecps
;
727 raw_time
.tv_nsec
+= timekeeper
.raw_interval
;
728 if (raw_time
.tv_nsec
>= NSEC_PER_SEC
) {
729 raw_time
.tv_nsec
-= NSEC_PER_SEC
;
733 /* accumulate error between NTP and clock interval */
734 timekeeper
.ntp_error
+= tick_length
;
735 timekeeper
.ntp_error
-= timekeeper
.xtime_interval
<<
736 timekeeper
.ntp_error_shift
;
739 /* correct the clock when NTP error is too big */
740 timekeeping_adjust(offset
);
743 * Since in the loop above, we accumulate any amount of time
744 * in xtime_nsec over a second into xtime.tv_sec, its possible for
745 * xtime_nsec to be fairly small after the loop. Further, if we're
746 * slightly speeding the clocksource up in timekeeping_adjust(),
747 * its possible the required corrective factor to xtime_nsec could
748 * cause it to underflow.
750 * Now, we cannot simply roll the accumulated second back, since
751 * the NTP subsystem has been notified via second_overflow. So
752 * instead we push xtime_nsec forward by the amount we underflowed,
753 * and add that amount into the error.
755 * We'll correct this error next time through this function, when
756 * xtime_nsec is not as small.
758 if (unlikely((s64
)timekeeper
.xtime_nsec
< 0)) {
759 s64 neg
= -(s64
)timekeeper
.xtime_nsec
;
760 timekeeper
.xtime_nsec
= 0;
761 timekeeper
.ntp_error
+= neg
<< timekeeper
.ntp_error_shift
;
764 /* store full nanoseconds into xtime after rounding it up and
765 * add the remainder to the error difference.
767 xtime
.tv_nsec
= ((s64
) timekeeper
.xtime_nsec
>> timekeeper
.shift
) + 1;
768 timekeeper
.xtime_nsec
-= (s64
) xtime
.tv_nsec
<< timekeeper
.shift
;
769 timekeeper
.ntp_error
+= timekeeper
.xtime_nsec
<<
770 timekeeper
.ntp_error_shift
;
772 nsecs
= clocksource_cyc2ns(offset
, timekeeper
.mult
, timekeeper
.shift
);
773 update_xtime_cache(nsecs
);
775 /* check to see if there is a new clocksource to use */
776 change_clocksource();
777 update_vsyscall(&xtime
, timekeeper
.clock
);
781 * getboottime - Return the real time of system boot.
782 * @ts: pointer to the timespec to be set
784 * Returns the time of day in a timespec.
786 * This is based on the wall_to_monotonic offset and the total suspend
787 * time. Calls to settimeofday will affect the value returned (which
788 * basically means that however wrong your real time clock is at boot time,
789 * you get the right time here).
791 void getboottime(struct timespec
*ts
)
793 set_normalized_timespec(ts
,
794 - (wall_to_monotonic
.tv_sec
+ total_sleep_time
),
795 - wall_to_monotonic
.tv_nsec
);
799 * monotonic_to_bootbased - Convert the monotonic time to boot based.
800 * @ts: pointer to the timespec to be converted
802 void monotonic_to_bootbased(struct timespec
*ts
)
804 ts
->tv_sec
+= total_sleep_time
;
807 unsigned long get_seconds(void)
809 return xtime_cache
.tv_sec
;
811 EXPORT_SYMBOL(get_seconds
);
814 struct timespec
current_kernel_time(void)
820 seq
= read_seqbegin(&xtime_lock
);
823 } while (read_seqretry(&xtime_lock
, seq
));
827 EXPORT_SYMBOL(current_kernel_time
);