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/sched.h>
17 #include <linux/syscore_ops.h>
18 #include <linux/clocksource.h>
19 #include <linux/jiffies.h>
20 #include <linux/time.h>
21 #include <linux/tick.h>
22 #include <linux/stop_machine.h>
24 /* Structure holding internal timekeeping values. */
26 /* Current clocksource used for timekeeping. */
27 struct clocksource
*clock
;
28 /* NTP adjusted clock multiplier */
30 /* The shift value of the current clocksource. */
33 /* Number of clock cycles in one NTP interval. */
34 cycle_t cycle_interval
;
35 /* Number of clock shifted nano seconds in one NTP interval. */
37 /* shifted nano seconds left over when rounding cycle_interval */
39 /* Raw nano seconds accumulated per NTP interval. */
42 /* Clock shifted nano seconds remainder not stored in xtime.tv_nsec. */
44 /* Difference between accumulated time and NTP time in ntp
45 * shifted nano seconds. */
47 /* Shift conversion between clock shifted nano seconds and
48 * ntp shifted nano seconds. */
51 /* The current time */
52 struct timespec xtime
;
54 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
55 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
56 * at zero at system boot time, so wall_to_monotonic will be negative,
57 * however, we will ALWAYS keep the tv_nsec part positive so we can use
58 * the usual normalization.
60 * wall_to_monotonic is moved after resume from suspend for the
61 * monotonic time not to jump. We need to add total_sleep_time to
62 * wall_to_monotonic to get the real boot based time offset.
64 * - wall_to_monotonic is no longer the boot time, getboottime must be
67 struct timespec wall_to_monotonic
;
68 /* time spent in suspend */
69 struct timespec total_sleep_time
;
70 /* The raw monotonic time for the CLOCK_MONOTONIC_RAW posix clock. */
71 struct timespec raw_time
;
73 /* Offset clock monotonic -> clock realtime */
76 /* Offset clock monotonic -> clock boottime */
79 /* Seqlock for all timekeeper values */
83 static struct timekeeper timekeeper
;
86 * This read-write spinlock protects us from races in SMP while
89 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(xtime_lock
);
92 /* flag for if timekeeping is suspended */
93 int __read_mostly timekeeping_suspended
;
98 * timekeeper_setup_internals - Set up internals to use clocksource clock.
100 * @clock: Pointer to clocksource.
102 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
103 * pair and interval request.
105 * Unless you're the timekeeping code, you should not be using this!
107 static void timekeeper_setup_internals(struct clocksource
*clock
)
110 u64 tmp
, ntpinterval
;
112 timekeeper
.clock
= clock
;
113 clock
->cycle_last
= clock
->read(clock
);
115 /* Do the ns -> cycle conversion first, using original mult */
116 tmp
= NTP_INTERVAL_LENGTH
;
117 tmp
<<= clock
->shift
;
119 tmp
+= clock
->mult
/2;
120 do_div(tmp
, clock
->mult
);
124 interval
= (cycle_t
) tmp
;
125 timekeeper
.cycle_interval
= interval
;
127 /* Go back from cycles -> shifted ns */
128 timekeeper
.xtime_interval
= (u64
) interval
* clock
->mult
;
129 timekeeper
.xtime_remainder
= ntpinterval
- timekeeper
.xtime_interval
;
130 timekeeper
.raw_interval
=
131 ((u64
) interval
* clock
->mult
) >> clock
->shift
;
133 timekeeper
.xtime_nsec
= 0;
134 timekeeper
.shift
= clock
->shift
;
136 timekeeper
.ntp_error
= 0;
137 timekeeper
.ntp_error_shift
= NTP_SCALE_SHIFT
- clock
->shift
;
140 * The timekeeper keeps its own mult values for the currently
141 * active clocksource. These value will be adjusted via NTP
142 * to counteract clock drifting.
144 timekeeper
.mult
= clock
->mult
;
147 /* Timekeeper helper functions. */
148 static inline s64
timekeeping_get_ns(void)
150 cycle_t cycle_now
, cycle_delta
;
151 struct clocksource
*clock
;
153 /* read clocksource: */
154 clock
= timekeeper
.clock
;
155 cycle_now
= clock
->read(clock
);
157 /* calculate the delta since the last update_wall_time: */
158 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
160 /* return delta convert to nanoseconds using ntp adjusted mult. */
161 return clocksource_cyc2ns(cycle_delta
, timekeeper
.mult
,
165 static inline s64
timekeeping_get_ns_raw(void)
167 cycle_t cycle_now
, cycle_delta
;
168 struct clocksource
*clock
;
170 /* read clocksource: */
171 clock
= timekeeper
.clock
;
172 cycle_now
= clock
->read(clock
);
174 /* calculate the delta since the last update_wall_time: */
175 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
177 /* return delta convert to nanoseconds. */
178 return clocksource_cyc2ns(cycle_delta
, clock
->mult
, clock
->shift
);
181 static void update_rt_offset(void)
183 struct timespec tmp
, *wtm
= &timekeeper
.wall_to_monotonic
;
185 set_normalized_timespec(&tmp
, -wtm
->tv_sec
, -wtm
->tv_nsec
);
186 timekeeper
.offs_real
= timespec_to_ktime(tmp
);
189 /* must hold write on timekeeper.lock */
190 static void timekeeping_update(bool clearntp
)
193 timekeeper
.ntp_error
= 0;
197 update_vsyscall(&timekeeper
.xtime
, &timekeeper
.wall_to_monotonic
,
198 timekeeper
.clock
, timekeeper
.mult
);
203 * timekeeping_forward_now - update clock to the current time
205 * Forward the current clock to update its state since the last call to
206 * update_wall_time(). This is useful before significant clock changes,
207 * as it avoids having to deal with this time offset explicitly.
209 static void timekeeping_forward_now(void)
211 cycle_t cycle_now
, cycle_delta
;
212 struct clocksource
*clock
;
215 clock
= timekeeper
.clock
;
216 cycle_now
= clock
->read(clock
);
217 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
218 clock
->cycle_last
= cycle_now
;
220 nsec
= clocksource_cyc2ns(cycle_delta
, timekeeper
.mult
,
223 /* If arch requires, add in gettimeoffset() */
224 nsec
+= arch_gettimeoffset();
226 timespec_add_ns(&timekeeper
.xtime
, nsec
);
228 nsec
= clocksource_cyc2ns(cycle_delta
, clock
->mult
, clock
->shift
);
229 timespec_add_ns(&timekeeper
.raw_time
, nsec
);
233 * getnstimeofday - Returns the time of day in a timespec
234 * @ts: pointer to the timespec to be set
236 * Returns the time of day in a timespec.
238 void getnstimeofday(struct timespec
*ts
)
243 WARN_ON(timekeeping_suspended
);
246 seq
= read_seqbegin(&timekeeper
.lock
);
248 *ts
= timekeeper
.xtime
;
249 nsecs
= timekeeping_get_ns();
251 /* If arch requires, add in gettimeoffset() */
252 nsecs
+= arch_gettimeoffset();
254 } while (read_seqretry(&timekeeper
.lock
, seq
));
256 timespec_add_ns(ts
, nsecs
);
258 EXPORT_SYMBOL(getnstimeofday
);
260 ktime_t
ktime_get(void)
265 WARN_ON(timekeeping_suspended
);
268 seq
= read_seqbegin(&timekeeper
.lock
);
269 secs
= timekeeper
.xtime
.tv_sec
+
270 timekeeper
.wall_to_monotonic
.tv_sec
;
271 nsecs
= timekeeper
.xtime
.tv_nsec
+
272 timekeeper
.wall_to_monotonic
.tv_nsec
;
273 nsecs
+= timekeeping_get_ns();
274 /* If arch requires, add in gettimeoffset() */
275 nsecs
+= arch_gettimeoffset();
277 } while (read_seqretry(&timekeeper
.lock
, seq
));
279 * Use ktime_set/ktime_add_ns to create a proper ktime on
280 * 32-bit architectures without CONFIG_KTIME_SCALAR.
282 return ktime_add_ns(ktime_set(secs
, 0), nsecs
);
284 EXPORT_SYMBOL_GPL(ktime_get
);
287 * ktime_get_ts - get the monotonic clock in timespec format
288 * @ts: pointer to timespec variable
290 * The function calculates the monotonic clock from the realtime
291 * clock and the wall_to_monotonic offset and stores the result
292 * in normalized timespec format in the variable pointed to by @ts.
294 void ktime_get_ts(struct timespec
*ts
)
296 struct timespec tomono
;
300 WARN_ON(timekeeping_suspended
);
303 seq
= read_seqbegin(&timekeeper
.lock
);
304 *ts
= timekeeper
.xtime
;
305 tomono
= timekeeper
.wall_to_monotonic
;
306 nsecs
= timekeeping_get_ns();
307 /* If arch requires, add in gettimeoffset() */
308 nsecs
+= arch_gettimeoffset();
310 } while (read_seqretry(&timekeeper
.lock
, seq
));
312 set_normalized_timespec(ts
, ts
->tv_sec
+ tomono
.tv_sec
,
313 ts
->tv_nsec
+ tomono
.tv_nsec
+ nsecs
);
315 EXPORT_SYMBOL_GPL(ktime_get_ts
);
317 #ifdef CONFIG_NTP_PPS
320 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
321 * @ts_raw: pointer to the timespec to be set to raw monotonic time
322 * @ts_real: pointer to the timespec to be set to the time of day
324 * This function reads both the time of day and raw monotonic time at the
325 * same time atomically and stores the resulting timestamps in timespec
328 void getnstime_raw_and_real(struct timespec
*ts_raw
, struct timespec
*ts_real
)
331 s64 nsecs_raw
, nsecs_real
;
333 WARN_ON_ONCE(timekeeping_suspended
);
338 seq
= read_seqbegin(&timekeeper
.lock
);
340 *ts_raw
= timekeeper
.raw_time
;
341 *ts_real
= timekeeper
.xtime
;
343 nsecs_raw
= timekeeping_get_ns_raw();
344 nsecs_real
= timekeeping_get_ns();
346 /* If arch requires, add in gettimeoffset() */
347 arch_offset
= arch_gettimeoffset();
348 nsecs_raw
+= arch_offset
;
349 nsecs_real
+= arch_offset
;
351 } while (read_seqretry(&timekeeper
.lock
, seq
));
353 timespec_add_ns(ts_raw
, nsecs_raw
);
354 timespec_add_ns(ts_real
, nsecs_real
);
356 EXPORT_SYMBOL(getnstime_raw_and_real
);
358 #endif /* CONFIG_NTP_PPS */
361 * do_gettimeofday - Returns the time of day in a timeval
362 * @tv: pointer to the timeval to be set
364 * NOTE: Users should be converted to using getnstimeofday()
366 void do_gettimeofday(struct timeval
*tv
)
370 getnstimeofday(&now
);
371 tv
->tv_sec
= now
.tv_sec
;
372 tv
->tv_usec
= now
.tv_nsec
/1000;
374 EXPORT_SYMBOL(do_gettimeofday
);
377 * do_settimeofday - Sets the time of day
378 * @tv: pointer to the timespec variable containing the new time
380 * Sets the time of day to the new time and update NTP and notify hrtimers
382 int do_settimeofday(const struct timespec
*tv
)
384 struct timespec ts_delta
;
387 if ((unsigned long)tv
->tv_nsec
>= NSEC_PER_SEC
)
390 write_seqlock_irqsave(&timekeeper
.lock
, flags
);
392 timekeeping_forward_now();
394 ts_delta
.tv_sec
= tv
->tv_sec
- timekeeper
.xtime
.tv_sec
;
395 ts_delta
.tv_nsec
= tv
->tv_nsec
- timekeeper
.xtime
.tv_nsec
;
396 timekeeper
.wall_to_monotonic
=
397 timespec_sub(timekeeper
.wall_to_monotonic
, ts_delta
);
399 timekeeper
.xtime
= *tv
;
400 timekeeping_update(true);
402 write_sequnlock_irqrestore(&timekeeper
.lock
, flags
);
404 /* signal hrtimers about time change */
409 EXPORT_SYMBOL(do_settimeofday
);
413 * timekeeping_inject_offset - Adds or subtracts from the current time.
414 * @tv: pointer to the timespec variable containing the offset
416 * Adds or subtracts an offset value from the current time.
418 int timekeeping_inject_offset(struct timespec
*ts
)
422 if ((unsigned long)ts
->tv_nsec
>= NSEC_PER_SEC
)
425 write_seqlock_irqsave(&timekeeper
.lock
, flags
);
427 timekeeping_forward_now();
429 timekeeper
.xtime
= timespec_add(timekeeper
.xtime
, *ts
);
430 timekeeper
.wall_to_monotonic
=
431 timespec_sub(timekeeper
.wall_to_monotonic
, *ts
);
433 timekeeping_update(true);
435 write_sequnlock_irqrestore(&timekeeper
.lock
, flags
);
437 /* signal hrtimers about time change */
442 EXPORT_SYMBOL(timekeeping_inject_offset
);
445 * change_clocksource - Swaps clocksources if a new one is available
447 * Accumulates current time interval and initializes new clocksource
449 static int change_clocksource(void *data
)
451 struct clocksource
*new, *old
;
454 new = (struct clocksource
*) data
;
456 write_seqlock_irqsave(&timekeeper
.lock
, flags
);
458 timekeeping_forward_now();
459 if (!new->enable
|| new->enable(new) == 0) {
460 old
= timekeeper
.clock
;
461 timekeeper_setup_internals(new);
465 timekeeping_update(true);
467 write_sequnlock_irqrestore(&timekeeper
.lock
, flags
);
473 * timekeeping_notify - Install a new clock source
474 * @clock: pointer to the clock source
476 * This function is called from clocksource.c after a new, better clock
477 * source has been registered. The caller holds the clocksource_mutex.
479 void timekeeping_notify(struct clocksource
*clock
)
481 if (timekeeper
.clock
== clock
)
483 stop_machine(change_clocksource
, clock
, NULL
);
488 * ktime_get_real - get the real (wall-) time in ktime_t format
490 * returns the time in ktime_t format
492 ktime_t
ktime_get_real(void)
496 getnstimeofday(&now
);
498 return timespec_to_ktime(now
);
500 EXPORT_SYMBOL_GPL(ktime_get_real
);
503 * getrawmonotonic - Returns the raw monotonic time in a timespec
504 * @ts: pointer to the timespec to be set
506 * Returns the raw monotonic time (completely un-modified by ntp)
508 void getrawmonotonic(struct timespec
*ts
)
514 seq
= read_seqbegin(&timekeeper
.lock
);
515 nsecs
= timekeeping_get_ns_raw();
516 *ts
= timekeeper
.raw_time
;
518 } while (read_seqretry(&timekeeper
.lock
, seq
));
520 timespec_add_ns(ts
, nsecs
);
522 EXPORT_SYMBOL(getrawmonotonic
);
526 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
528 int timekeeping_valid_for_hres(void)
534 seq
= read_seqbegin(&timekeeper
.lock
);
536 ret
= timekeeper
.clock
->flags
& CLOCK_SOURCE_VALID_FOR_HRES
;
538 } while (read_seqretry(&timekeeper
.lock
, seq
));
544 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
546 u64
timekeeping_max_deferment(void)
551 seq
= read_seqbegin(&timekeeper
.lock
);
553 ret
= timekeeper
.clock
->max_idle_ns
;
555 } while (read_seqretry(&timekeeper
.lock
, seq
));
561 * read_persistent_clock - Return time from the persistent clock.
563 * Weak dummy function for arches that do not yet support it.
564 * Reads the time from the battery backed persistent clock.
565 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
567 * XXX - Do be sure to remove it once all arches implement it.
569 void __attribute__((weak
)) read_persistent_clock(struct timespec
*ts
)
576 * read_boot_clock - Return time of the system start.
578 * Weak dummy function for arches that do not yet support it.
579 * Function to read the exact time the system has been started.
580 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
582 * XXX - Do be sure to remove it once all arches implement it.
584 void __attribute__((weak
)) read_boot_clock(struct timespec
*ts
)
591 * timekeeping_init - Initializes the clocksource and common timekeeping values
593 void __init
timekeeping_init(void)
595 struct clocksource
*clock
;
597 struct timespec now
, boot
;
599 read_persistent_clock(&now
);
600 read_boot_clock(&boot
);
602 seqlock_init(&timekeeper
.lock
);
606 write_seqlock_irqsave(&timekeeper
.lock
, flags
);
607 clock
= clocksource_default_clock();
609 clock
->enable(clock
);
610 timekeeper_setup_internals(clock
);
612 timekeeper
.xtime
.tv_sec
= now
.tv_sec
;
613 timekeeper
.xtime
.tv_nsec
= now
.tv_nsec
;
614 timekeeper
.raw_time
.tv_sec
= 0;
615 timekeeper
.raw_time
.tv_nsec
= 0;
616 if (boot
.tv_sec
== 0 && boot
.tv_nsec
== 0) {
617 boot
.tv_sec
= timekeeper
.xtime
.tv_sec
;
618 boot
.tv_nsec
= timekeeper
.xtime
.tv_nsec
;
620 set_normalized_timespec(&timekeeper
.wall_to_monotonic
,
621 -boot
.tv_sec
, -boot
.tv_nsec
);
623 timekeeper
.total_sleep_time
.tv_sec
= 0;
624 timekeeper
.total_sleep_time
.tv_nsec
= 0;
625 write_sequnlock_irqrestore(&timekeeper
.lock
, flags
);
628 /* time in seconds when suspend began */
629 static struct timespec timekeeping_suspend_time
;
631 static void update_sleep_time(struct timespec t
)
633 timekeeper
.total_sleep_time
= t
;
634 timekeeper
.offs_boot
= timespec_to_ktime(t
);
638 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
639 * @delta: pointer to a timespec delta value
641 * Takes a timespec offset measuring a suspend interval and properly
642 * adds the sleep offset to the timekeeping variables.
644 static void __timekeeping_inject_sleeptime(struct timespec
*delta
)
646 if (!timespec_valid(delta
)) {
647 printk(KERN_WARNING
"__timekeeping_inject_sleeptime: Invalid "
648 "sleep delta value!\n");
652 timekeeper
.xtime
= timespec_add(timekeeper
.xtime
, *delta
);
653 timekeeper
.wall_to_monotonic
=
654 timespec_sub(timekeeper
.wall_to_monotonic
, *delta
);
655 update_sleep_time(timespec_add(timekeeper
.total_sleep_time
, *delta
));
660 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
661 * @delta: pointer to a timespec delta value
663 * This hook is for architectures that cannot support read_persistent_clock
664 * because their RTC/persistent clock is only accessible when irqs are enabled.
666 * This function should only be called by rtc_resume(), and allows
667 * a suspend offset to be injected into the timekeeping values.
669 void timekeeping_inject_sleeptime(struct timespec
*delta
)
674 /* Make sure we don't set the clock twice */
675 read_persistent_clock(&ts
);
676 if (!(ts
.tv_sec
== 0 && ts
.tv_nsec
== 0))
679 write_seqlock_irqsave(&timekeeper
.lock
, flags
);
681 timekeeping_forward_now();
683 __timekeeping_inject_sleeptime(delta
);
685 timekeeping_update(true);
687 write_sequnlock_irqrestore(&timekeeper
.lock
, flags
);
689 /* signal hrtimers about time change */
695 * timekeeping_resume - Resumes the generic timekeeping subsystem.
697 * This is for the generic clocksource timekeeping.
698 * xtime/wall_to_monotonic/jiffies/etc are
699 * still managed by arch specific suspend/resume code.
701 static void timekeeping_resume(void)
706 read_persistent_clock(&ts
);
708 clocksource_resume();
710 write_seqlock_irqsave(&timekeeper
.lock
, flags
);
712 if (timespec_compare(&ts
, &timekeeping_suspend_time
) > 0) {
713 ts
= timespec_sub(ts
, timekeeping_suspend_time
);
714 __timekeeping_inject_sleeptime(&ts
);
716 /* re-base the last cycle value */
717 timekeeper
.clock
->cycle_last
= timekeeper
.clock
->read(timekeeper
.clock
);
718 timekeeper
.ntp_error
= 0;
719 timekeeping_suspended
= 0;
720 write_sequnlock_irqrestore(&timekeeper
.lock
, flags
);
722 touch_softlockup_watchdog();
724 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME
, NULL
);
726 /* Resume hrtimers */
730 static int timekeeping_suspend(void)
733 struct timespec delta
, delta_delta
;
734 static struct timespec old_delta
;
736 read_persistent_clock(&timekeeping_suspend_time
);
738 write_seqlock_irqsave(&timekeeper
.lock
, flags
);
739 timekeeping_forward_now();
740 timekeeping_suspended
= 1;
743 * To avoid drift caused by repeated suspend/resumes,
744 * which each can add ~1 second drift error,
745 * try to compensate so the difference in system time
746 * and persistent_clock time stays close to constant.
748 delta
= timespec_sub(timekeeper
.xtime
, timekeeping_suspend_time
);
749 delta_delta
= timespec_sub(delta
, old_delta
);
750 if (abs(delta_delta
.tv_sec
) >= 2) {
752 * if delta_delta is too large, assume time correction
753 * has occured and set old_delta to the current delta.
757 /* Otherwise try to adjust old_system to compensate */
758 timekeeping_suspend_time
=
759 timespec_add(timekeeping_suspend_time
, delta_delta
);
761 write_sequnlock_irqrestore(&timekeeper
.lock
, flags
);
763 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND
, NULL
);
764 clocksource_suspend();
769 /* sysfs resume/suspend bits for timekeeping */
770 static struct syscore_ops timekeeping_syscore_ops
= {
771 .resume
= timekeeping_resume
,
772 .suspend
= timekeeping_suspend
,
775 static int __init
timekeeping_init_ops(void)
777 register_syscore_ops(&timekeeping_syscore_ops
);
781 device_initcall(timekeeping_init_ops
);
784 * If the error is already larger, we look ahead even further
785 * to compensate for late or lost adjustments.
787 static __always_inline
int timekeeping_bigadjust(s64 error
, s64
*interval
,
795 * Use the current error value to determine how much to look ahead.
796 * The larger the error the slower we adjust for it to avoid problems
797 * with losing too many ticks, otherwise we would overadjust and
798 * produce an even larger error. The smaller the adjustment the
799 * faster we try to adjust for it, as lost ticks can do less harm
800 * here. This is tuned so that an error of about 1 msec is adjusted
801 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
803 error2
= timekeeper
.ntp_error
>> (NTP_SCALE_SHIFT
+ 22 - 2 * SHIFT_HZ
);
804 error2
= abs(error2
);
805 for (look_ahead
= 0; error2
> 0; look_ahead
++)
809 * Now calculate the error in (1 << look_ahead) ticks, but first
810 * remove the single look ahead already included in the error.
812 tick_error
= ntp_tick_length() >> (timekeeper
.ntp_error_shift
+ 1);
813 tick_error
-= timekeeper
.xtime_interval
>> 1;
814 error
= ((error
- tick_error
) >> look_ahead
) + tick_error
;
816 /* Finally calculate the adjustment shift value. */
821 *interval
= -*interval
;
825 for (adj
= 0; error
> i
; adj
++)
834 * Adjust the multiplier to reduce the error value,
835 * this is optimized for the most common adjustments of -1,0,1,
836 * for other values we can do a bit more work.
838 static void timekeeping_adjust(s64 offset
)
840 s64 error
, interval
= timekeeper
.cycle_interval
;
844 * The point of this is to check if the error is greater than half
847 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
849 * Note we subtract one in the shift, so that error is really error*2.
850 * This "saves" dividing(shifting) interval twice, but keeps the
851 * (error > interval) comparison as still measuring if error is
852 * larger than half an interval.
854 * Note: It does not "save" on aggravation when reading the code.
856 error
= timekeeper
.ntp_error
>> (timekeeper
.ntp_error_shift
- 1);
857 if (error
> interval
) {
859 * We now divide error by 4(via shift), which checks if
860 * the error is greater than twice the interval.
861 * If it is greater, we need a bigadjust, if its smaller,
862 * we can adjust by 1.
866 * XXX - In update_wall_time, we round up to the next
867 * nanosecond, and store the amount rounded up into
868 * the error. This causes the likely below to be unlikely.
870 * The proper fix is to avoid rounding up by using
871 * the high precision timekeeper.xtime_nsec instead of
872 * xtime.tv_nsec everywhere. Fixing this will take some
875 if (likely(error
<= interval
))
878 adj
= timekeeping_bigadjust(error
, &interval
, &offset
);
879 } else if (error
< -interval
) {
880 /* See comment above, this is just switched for the negative */
882 if (likely(error
>= -interval
)) {
884 interval
= -interval
;
887 adj
= timekeeping_bigadjust(error
, &interval
, &offset
);
888 } else /* No adjustment needed */
891 if (unlikely(timekeeper
.clock
->maxadj
&&
892 (timekeeper
.mult
+ adj
>
893 timekeeper
.clock
->mult
+ timekeeper
.clock
->maxadj
))) {
894 printk_once(KERN_WARNING
895 "Adjusting %s more than 11%% (%ld vs %ld)\n",
896 timekeeper
.clock
->name
, (long)timekeeper
.mult
+ adj
,
897 (long)timekeeper
.clock
->mult
+
898 timekeeper
.clock
->maxadj
);
901 * So the following can be confusing.
903 * To keep things simple, lets assume adj == 1 for now.
905 * When adj != 1, remember that the interval and offset values
906 * have been appropriately scaled so the math is the same.
908 * The basic idea here is that we're increasing the multiplier
909 * by one, this causes the xtime_interval to be incremented by
910 * one cycle_interval. This is because:
911 * xtime_interval = cycle_interval * mult
912 * So if mult is being incremented by one:
913 * xtime_interval = cycle_interval * (mult + 1)
915 * xtime_interval = (cycle_interval * mult) + cycle_interval
916 * Which can be shortened to:
917 * xtime_interval += cycle_interval
919 * So offset stores the non-accumulated cycles. Thus the current
920 * time (in shifted nanoseconds) is:
921 * now = (offset * adj) + xtime_nsec
922 * Now, even though we're adjusting the clock frequency, we have
923 * to keep time consistent. In other words, we can't jump back
924 * in time, and we also want to avoid jumping forward in time.
926 * So given the same offset value, we need the time to be the same
927 * both before and after the freq adjustment.
928 * now = (offset * adj_1) + xtime_nsec_1
929 * now = (offset * adj_2) + xtime_nsec_2
931 * (offset * adj_1) + xtime_nsec_1 =
932 * (offset * adj_2) + xtime_nsec_2
936 * (offset * adj_1) + xtime_nsec_1 =
937 * (offset * (adj_1+1)) + xtime_nsec_2
938 * (offset * adj_1) + xtime_nsec_1 =
939 * (offset * adj_1) + offset + xtime_nsec_2
940 * Canceling the sides:
941 * xtime_nsec_1 = offset + xtime_nsec_2
943 * xtime_nsec_2 = xtime_nsec_1 - offset
944 * Which simplfies to:
945 * xtime_nsec -= offset
947 * XXX - TODO: Doc ntp_error calculation.
949 timekeeper
.mult
+= adj
;
950 timekeeper
.xtime_interval
+= interval
;
951 timekeeper
.xtime_nsec
-= offset
;
952 timekeeper
.ntp_error
-= (interval
- offset
) <<
953 timekeeper
.ntp_error_shift
;
958 * logarithmic_accumulation - shifted accumulation of cycles
960 * This functions accumulates a shifted interval of cycles into
961 * into a shifted interval nanoseconds. Allows for O(log) accumulation
964 * Returns the unconsumed cycles.
966 static cycle_t
logarithmic_accumulation(cycle_t offset
, int shift
)
968 u64 nsecps
= (u64
)NSEC_PER_SEC
<< timekeeper
.shift
;
971 /* If the offset is smaller than a shifted interval, do nothing */
972 if (offset
< timekeeper
.cycle_interval
<<shift
)
975 /* Accumulate one shifted interval */
976 offset
-= timekeeper
.cycle_interval
<< shift
;
977 timekeeper
.clock
->cycle_last
+= timekeeper
.cycle_interval
<< shift
;
979 timekeeper
.xtime_nsec
+= timekeeper
.xtime_interval
<< shift
;
980 while (timekeeper
.xtime_nsec
>= nsecps
) {
982 timekeeper
.xtime_nsec
-= nsecps
;
983 timekeeper
.xtime
.tv_sec
++;
984 leap
= second_overflow(timekeeper
.xtime
.tv_sec
);
985 timekeeper
.xtime
.tv_sec
+= leap
;
986 timekeeper
.wall_to_monotonic
.tv_sec
-= leap
;
988 clock_was_set_delayed();
991 /* Accumulate raw time */
992 raw_nsecs
= timekeeper
.raw_interval
<< shift
;
993 raw_nsecs
+= timekeeper
.raw_time
.tv_nsec
;
994 if (raw_nsecs
>= NSEC_PER_SEC
) {
995 u64 raw_secs
= raw_nsecs
;
996 raw_nsecs
= do_div(raw_secs
, NSEC_PER_SEC
);
997 timekeeper
.raw_time
.tv_sec
+= raw_secs
;
999 timekeeper
.raw_time
.tv_nsec
= raw_nsecs
;
1001 /* Accumulate error between NTP and clock interval */
1002 timekeeper
.ntp_error
+= ntp_tick_length() << shift
;
1003 timekeeper
.ntp_error
-=
1004 (timekeeper
.xtime_interval
+ timekeeper
.xtime_remainder
) <<
1005 (timekeeper
.ntp_error_shift
+ shift
);
1012 * update_wall_time - Uses the current clocksource to increment the wall time
1015 static void update_wall_time(void)
1017 struct clocksource
*clock
;
1019 int shift
= 0, maxshift
;
1020 unsigned long flags
;
1022 write_seqlock_irqsave(&timekeeper
.lock
, flags
);
1024 /* Make sure we're fully resumed: */
1025 if (unlikely(timekeeping_suspended
))
1028 clock
= timekeeper
.clock
;
1030 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1031 offset
= timekeeper
.cycle_interval
;
1033 offset
= (clock
->read(clock
) - clock
->cycle_last
) & clock
->mask
;
1035 timekeeper
.xtime_nsec
= (s64
)timekeeper
.xtime
.tv_nsec
<<
1039 * With NO_HZ we may have to accumulate many cycle_intervals
1040 * (think "ticks") worth of time at once. To do this efficiently,
1041 * we calculate the largest doubling multiple of cycle_intervals
1042 * that is smaller than the offset. We then accumulate that
1043 * chunk in one go, and then try to consume the next smaller
1046 shift
= ilog2(offset
) - ilog2(timekeeper
.cycle_interval
);
1047 shift
= max(0, shift
);
1048 /* Bound shift to one less than what overflows tick_length */
1049 maxshift
= (64 - (ilog2(ntp_tick_length())+1)) - 1;
1050 shift
= min(shift
, maxshift
);
1051 while (offset
>= timekeeper
.cycle_interval
) {
1052 offset
= logarithmic_accumulation(offset
, shift
);
1053 if(offset
< timekeeper
.cycle_interval
<<shift
)
1057 /* correct the clock when NTP error is too big */
1058 timekeeping_adjust(offset
);
1061 * Since in the loop above, we accumulate any amount of time
1062 * in xtime_nsec over a second into xtime.tv_sec, its possible for
1063 * xtime_nsec to be fairly small after the loop. Further, if we're
1064 * slightly speeding the clocksource up in timekeeping_adjust(),
1065 * its possible the required corrective factor to xtime_nsec could
1066 * cause it to underflow.
1068 * Now, we cannot simply roll the accumulated second back, since
1069 * the NTP subsystem has been notified via second_overflow. So
1070 * instead we push xtime_nsec forward by the amount we underflowed,
1071 * and add that amount into the error.
1073 * We'll correct this error next time through this function, when
1074 * xtime_nsec is not as small.
1076 if (unlikely((s64
)timekeeper
.xtime_nsec
< 0)) {
1077 s64 neg
= -(s64
)timekeeper
.xtime_nsec
;
1078 timekeeper
.xtime_nsec
= 0;
1079 timekeeper
.ntp_error
+= neg
<< timekeeper
.ntp_error_shift
;
1084 * Store full nanoseconds into xtime after rounding it up and
1085 * add the remainder to the error difference.
1087 timekeeper
.xtime
.tv_nsec
= ((s64
)timekeeper
.xtime_nsec
>>
1088 timekeeper
.shift
) + 1;
1089 timekeeper
.xtime_nsec
-= (s64
)timekeeper
.xtime
.tv_nsec
<<
1091 timekeeper
.ntp_error
+= timekeeper
.xtime_nsec
<<
1092 timekeeper
.ntp_error_shift
;
1095 * Finally, make sure that after the rounding
1096 * xtime.tv_nsec isn't larger than NSEC_PER_SEC
1098 if (unlikely(timekeeper
.xtime
.tv_nsec
>= NSEC_PER_SEC
)) {
1100 timekeeper
.xtime
.tv_nsec
-= NSEC_PER_SEC
;
1101 timekeeper
.xtime
.tv_sec
++;
1102 leap
= second_overflow(timekeeper
.xtime
.tv_sec
);
1103 timekeeper
.xtime
.tv_sec
+= leap
;
1104 timekeeper
.wall_to_monotonic
.tv_sec
-= leap
;
1106 clock_was_set_delayed();
1109 timekeeping_update(false);
1112 write_sequnlock_irqrestore(&timekeeper
.lock
, flags
);
1117 * getboottime - Return the real time of system boot.
1118 * @ts: pointer to the timespec to be set
1120 * Returns the wall-time of boot in a timespec.
1122 * This is based on the wall_to_monotonic offset and the total suspend
1123 * time. Calls to settimeofday will affect the value returned (which
1124 * basically means that however wrong your real time clock is at boot time,
1125 * you get the right time here).
1127 void getboottime(struct timespec
*ts
)
1129 struct timespec boottime
= {
1130 .tv_sec
= timekeeper
.wall_to_monotonic
.tv_sec
+
1131 timekeeper
.total_sleep_time
.tv_sec
,
1132 .tv_nsec
= timekeeper
.wall_to_monotonic
.tv_nsec
+
1133 timekeeper
.total_sleep_time
.tv_nsec
1136 set_normalized_timespec(ts
, -boottime
.tv_sec
, -boottime
.tv_nsec
);
1138 EXPORT_SYMBOL_GPL(getboottime
);
1142 * get_monotonic_boottime - Returns monotonic time since boot
1143 * @ts: pointer to the timespec to be set
1145 * Returns the monotonic time since boot in a timespec.
1147 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1148 * includes the time spent in suspend.
1150 void get_monotonic_boottime(struct timespec
*ts
)
1152 struct timespec tomono
, sleep
;
1156 WARN_ON(timekeeping_suspended
);
1159 seq
= read_seqbegin(&timekeeper
.lock
);
1160 *ts
= timekeeper
.xtime
;
1161 tomono
= timekeeper
.wall_to_monotonic
;
1162 sleep
= timekeeper
.total_sleep_time
;
1163 nsecs
= timekeeping_get_ns();
1165 } while (read_seqretry(&timekeeper
.lock
, seq
));
1167 set_normalized_timespec(ts
, ts
->tv_sec
+ tomono
.tv_sec
+ sleep
.tv_sec
,
1168 ts
->tv_nsec
+ tomono
.tv_nsec
+ sleep
.tv_nsec
+ nsecs
);
1170 EXPORT_SYMBOL_GPL(get_monotonic_boottime
);
1173 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1175 * Returns the monotonic time since boot in a ktime
1177 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1178 * includes the time spent in suspend.
1180 ktime_t
ktime_get_boottime(void)
1184 get_monotonic_boottime(&ts
);
1185 return timespec_to_ktime(ts
);
1187 EXPORT_SYMBOL_GPL(ktime_get_boottime
);
1190 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1191 * @ts: pointer to the timespec to be converted
1193 void monotonic_to_bootbased(struct timespec
*ts
)
1195 *ts
= timespec_add(*ts
, timekeeper
.total_sleep_time
);
1197 EXPORT_SYMBOL_GPL(monotonic_to_bootbased
);
1199 unsigned long get_seconds(void)
1201 return timekeeper
.xtime
.tv_sec
;
1203 EXPORT_SYMBOL(get_seconds
);
1205 struct timespec
__current_kernel_time(void)
1207 return timekeeper
.xtime
;
1210 struct timespec
current_kernel_time(void)
1212 struct timespec now
;
1216 seq
= read_seqbegin(&timekeeper
.lock
);
1218 now
= timekeeper
.xtime
;
1219 } while (read_seqretry(&timekeeper
.lock
, seq
));
1223 EXPORT_SYMBOL(current_kernel_time
);
1225 struct timespec
get_monotonic_coarse(void)
1227 struct timespec now
, mono
;
1231 seq
= read_seqbegin(&timekeeper
.lock
);
1233 now
= timekeeper
.xtime
;
1234 mono
= timekeeper
.wall_to_monotonic
;
1235 } while (read_seqretry(&timekeeper
.lock
, seq
));
1237 set_normalized_timespec(&now
, now
.tv_sec
+ mono
.tv_sec
,
1238 now
.tv_nsec
+ mono
.tv_nsec
);
1243 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1244 * without sampling the sequence number in xtime_lock.
1245 * jiffies is defined in the linker script...
1247 void do_timer(unsigned long ticks
)
1249 jiffies_64
+= ticks
;
1251 calc_global_load(ticks
);
1255 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1256 * and sleep offsets.
1257 * @xtim: pointer to timespec to be set with xtime
1258 * @wtom: pointer to timespec to be set with wall_to_monotonic
1259 * @sleep: pointer to timespec to be set with time in suspend
1261 void get_xtime_and_monotonic_and_sleep_offset(struct timespec
*xtim
,
1262 struct timespec
*wtom
, struct timespec
*sleep
)
1267 seq
= read_seqbegin(&timekeeper
.lock
);
1268 *xtim
= timekeeper
.xtime
;
1269 *wtom
= timekeeper
.wall_to_monotonic
;
1270 *sleep
= timekeeper
.total_sleep_time
;
1271 } while (read_seqretry(&timekeeper
.lock
, seq
));
1274 #ifdef CONFIG_HIGH_RES_TIMERS
1276 * ktime_get_update_offsets - hrtimer helper
1277 * @offs_real: pointer to storage for monotonic -> realtime offset
1278 * @offs_boot: pointer to storage for monotonic -> boottime offset
1280 * Returns current monotonic time and updates the offsets
1281 * Called from hrtimer_interupt() or retrigger_next_event()
1283 ktime_t
ktime_get_update_offsets(ktime_t
*offs_real
, ktime_t
*offs_boot
)
1290 seq
= read_seqbegin(&timekeeper
.lock
);
1292 secs
= timekeeper
.xtime
.tv_sec
;
1293 nsecs
= timekeeper
.xtime
.tv_nsec
;
1294 nsecs
+= timekeeping_get_ns();
1295 /* If arch requires, add in gettimeoffset() */
1296 nsecs
+= arch_gettimeoffset();
1298 *offs_real
= timekeeper
.offs_real
;
1299 *offs_boot
= timekeeper
.offs_boot
;
1300 } while (read_seqretry(&timekeeper
.lock
, seq
));
1302 now
= ktime_add_ns(ktime_set(secs
, 0), nsecs
);
1303 now
= ktime_sub(now
, *offs_real
);
1309 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1311 ktime_t
ktime_get_monotonic_offset(void)
1314 struct timespec wtom
;
1317 seq
= read_seqbegin(&timekeeper
.lock
);
1318 wtom
= timekeeper
.wall_to_monotonic
;
1319 } while (read_seqretry(&timekeeper
.lock
, seq
));
1321 return timespec_to_ktime(wtom
);
1323 EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset
);
1327 * xtime_update() - advances the timekeeping infrastructure
1328 * @ticks: number of ticks, that have elapsed since the last call.
1330 * Must be called with interrupts disabled.
1332 void xtime_update(unsigned long ticks
)
1334 write_seqlock(&xtime_lock
);
1336 write_sequnlock(&xtime_lock
);