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/timekeeper_internal.h>
12 #include <linux/module.h>
13 #include <linux/interrupt.h>
14 #include <linux/percpu.h>
15 #include <linux/init.h>
17 #include <linux/sched.h>
18 #include <linux/syscore_ops.h>
19 #include <linux/clocksource.h>
20 #include <linux/jiffies.h>
21 #include <linux/time.h>
22 #include <linux/tick.h>
23 #include <linux/stop_machine.h>
26 static struct timekeeper timekeeper;
28 /* flag for if timekeeping is suspended */
29 int __read_mostly timekeeping_suspended;
31 static inline void tk_normalize_xtime(struct timekeeper *tk)
33 while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) {
34 tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift;
39 static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts)
41 tk->xtime_sec = ts->tv_sec;
42 tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift;
45 static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts)
47 tk->xtime_sec += ts->tv_sec;
48 tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift;
49 tk_normalize_xtime(tk);
52 static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm)
57 * Verify consistency of: offset_real = -wall_to_monotonic
58 * before modifying anything
60 set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec,
61 -tk->wall_to_monotonic.tv_nsec);
62 WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64);
63 tk->wall_to_monotonic = wtm;
64 set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
65 tk->offs_real = timespec_to_ktime(tmp);
68 static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t)
70 /* Verify consistency before modifying */
71 WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64);
73 tk->total_sleep_time = t;
74 tk->offs_boot = timespec_to_ktime(t);
78 * timekeeper_setup_internals - Set up internals to use clocksource clock.
80 * @clock: Pointer to clocksource.
82 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
83 * pair and interval request.
85 * Unless you're the timekeeping code, you should not be using this!
87 static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
91 struct clocksource *old_clock;
93 old_clock = tk->clock;
95 clock->cycle_last = clock->read(clock);
97 /* Do the ns -> cycle conversion first, using original mult */
98 tmp = NTP_INTERVAL_LENGTH;
101 tmp += clock->mult/2;
102 do_div(tmp, clock->mult);
106 interval = (cycle_t) tmp;
107 tk->cycle_interval = interval;
109 /* Go back from cycles -> shifted ns */
110 tk->xtime_interval = (u64) interval * clock->mult;
111 tk->xtime_remainder = ntpinterval - tk->xtime_interval;
113 ((u64) interval * clock->mult) >> clock->shift;
115 /* if changing clocks, convert xtime_nsec shift units */
117 int shift_change = clock->shift - old_clock->shift;
118 if (shift_change < 0)
119 tk->xtime_nsec >>= -shift_change;
121 tk->xtime_nsec <<= shift_change;
123 tk->shift = clock->shift;
126 tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
129 * The timekeeper keeps its own mult values for the currently
130 * active clocksource. These value will be adjusted via NTP
131 * to counteract clock drifting.
133 tk->mult = clock->mult;
136 /* Timekeeper helper functions. */
137 static inline s64 timekeeping_get_ns(struct timekeeper *tk)
139 cycle_t cycle_now, cycle_delta;
140 struct clocksource *clock;
143 /* read clocksource: */
145 cycle_now = clock->read(clock);
147 /* calculate the delta since the last update_wall_time: */
148 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
150 nsec = cycle_delta * tk->mult + tk->xtime_nsec;
153 /* If arch requires, add in gettimeoffset() */
154 return nsec + arch_gettimeoffset();
157 static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
159 cycle_t cycle_now, cycle_delta;
160 struct clocksource *clock;
163 /* read clocksource: */
165 cycle_now = clock->read(clock);
167 /* calculate the delta since the last update_wall_time: */
168 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
170 /* convert delta to nanoseconds. */
171 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
173 /* If arch requires, add in gettimeoffset() */
174 return nsec + arch_gettimeoffset();
177 /* must hold write on timekeeper.lock */
178 static void timekeeping_update(struct timekeeper *tk, bool clearntp)
188 * timekeeping_forward_now - update clock to the current time
190 * Forward the current clock to update its state since the last call to
191 * update_wall_time(). This is useful before significant clock changes,
192 * as it avoids having to deal with this time offset explicitly.
194 static void timekeeping_forward_now(struct timekeeper *tk)
196 cycle_t cycle_now, cycle_delta;
197 struct clocksource *clock;
201 cycle_now = clock->read(clock);
202 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
203 clock->cycle_last = cycle_now;
205 tk->xtime_nsec += cycle_delta * tk->mult;
207 /* If arch requires, add in gettimeoffset() */
208 tk->xtime_nsec += (u64)arch_gettimeoffset() << tk->shift;
210 tk_normalize_xtime(tk);
212 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
213 timespec_add_ns(&tk->raw_time, nsec);
217 * getnstimeofday - Returns the time of day in a timespec
218 * @ts: pointer to the timespec to be set
220 * Returns the time of day in a timespec.
222 void getnstimeofday(struct timespec *ts)
224 struct timekeeper *tk = &timekeeper;
228 WARN_ON(timekeeping_suspended);
231 seq = read_seqbegin(&tk->lock);
233 ts->tv_sec = tk->xtime_sec;
234 nsecs = timekeeping_get_ns(tk);
236 } while (read_seqretry(&tk->lock, seq));
239 timespec_add_ns(ts, nsecs);
241 EXPORT_SYMBOL(getnstimeofday);
243 ktime_t ktime_get(void)
245 struct timekeeper *tk = &timekeeper;
249 WARN_ON(timekeeping_suspended);
252 seq = read_seqbegin(&tk->lock);
253 secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec;
254 nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec;
256 } while (read_seqretry(&tk->lock, seq));
258 * Use ktime_set/ktime_add_ns to create a proper ktime on
259 * 32-bit architectures without CONFIG_KTIME_SCALAR.
261 return ktime_add_ns(ktime_set(secs, 0), nsecs);
263 EXPORT_SYMBOL_GPL(ktime_get);
266 * ktime_get_ts - get the monotonic clock in timespec format
267 * @ts: pointer to timespec variable
269 * The function calculates the monotonic clock from the realtime
270 * clock and the wall_to_monotonic offset and stores the result
271 * in normalized timespec format in the variable pointed to by @ts.
273 void ktime_get_ts(struct timespec *ts)
275 struct timekeeper *tk = &timekeeper;
276 struct timespec tomono;
280 WARN_ON(timekeeping_suspended);
283 seq = read_seqbegin(&tk->lock);
284 ts->tv_sec = tk->xtime_sec;
285 nsec = timekeeping_get_ns(tk);
286 tomono = tk->wall_to_monotonic;
288 } while (read_seqretry(&tk->lock, seq));
290 ts->tv_sec += tomono.tv_sec;
292 timespec_add_ns(ts, nsec + tomono.tv_nsec);
294 EXPORT_SYMBOL_GPL(ktime_get_ts);
296 #ifdef CONFIG_NTP_PPS
299 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
300 * @ts_raw: pointer to the timespec to be set to raw monotonic time
301 * @ts_real: pointer to the timespec to be set to the time of day
303 * This function reads both the time of day and raw monotonic time at the
304 * same time atomically and stores the resulting timestamps in timespec
307 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
309 struct timekeeper *tk = &timekeeper;
311 s64 nsecs_raw, nsecs_real;
313 WARN_ON_ONCE(timekeeping_suspended);
316 seq = read_seqbegin(&tk->lock);
318 *ts_raw = tk->raw_time;
319 ts_real->tv_sec = tk->xtime_sec;
320 ts_real->tv_nsec = 0;
322 nsecs_raw = timekeeping_get_ns_raw(tk);
323 nsecs_real = timekeeping_get_ns(tk);
325 } while (read_seqretry(&tk->lock, seq));
327 timespec_add_ns(ts_raw, nsecs_raw);
328 timespec_add_ns(ts_real, nsecs_real);
330 EXPORT_SYMBOL(getnstime_raw_and_real);
332 #endif /* CONFIG_NTP_PPS */
335 * do_gettimeofday - Returns the time of day in a timeval
336 * @tv: pointer to the timeval to be set
338 * NOTE: Users should be converted to using getnstimeofday()
340 void do_gettimeofday(struct timeval *tv)
344 getnstimeofday(&now);
345 tv->tv_sec = now.tv_sec;
346 tv->tv_usec = now.tv_nsec/1000;
348 EXPORT_SYMBOL(do_gettimeofday);
351 * do_settimeofday - Sets the time of day
352 * @tv: pointer to the timespec variable containing the new time
354 * Sets the time of day to the new time and update NTP and notify hrtimers
356 int do_settimeofday(const struct timespec *tv)
358 struct timekeeper *tk = &timekeeper;
359 struct timespec ts_delta, xt;
362 if (!timespec_valid_strict(tv))
365 write_seqlock_irqsave(&tk->lock, flags);
367 timekeeping_forward_now(tk);
370 ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
371 ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
373 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta));
375 tk_set_xtime(tk, tv);
377 timekeeping_update(tk, true);
379 write_sequnlock_irqrestore(&tk->lock, flags);
381 /* signal hrtimers about time change */
386 EXPORT_SYMBOL(do_settimeofday);
389 * timekeeping_inject_offset - Adds or subtracts from the current time.
390 * @tv: pointer to the timespec variable containing the offset
392 * Adds or subtracts an offset value from the current time.
394 int timekeeping_inject_offset(struct timespec *ts)
396 struct timekeeper *tk = &timekeeper;
401 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
404 write_seqlock_irqsave(&tk->lock, flags);
406 timekeeping_forward_now(tk);
408 /* Make sure the proposed value is valid */
409 tmp = timespec_add(tk_xtime(tk), *ts);
410 if (!timespec_valid_strict(&tmp)) {
415 tk_xtime_add(tk, ts);
416 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts));
418 error: /* even if we error out, we forwarded the time, so call update */
419 timekeeping_update(tk, true);
421 write_sequnlock_irqrestore(&tk->lock, flags);
423 /* signal hrtimers about time change */
428 EXPORT_SYMBOL(timekeeping_inject_offset);
431 * change_clocksource - Swaps clocksources if a new one is available
433 * Accumulates current time interval and initializes new clocksource
435 static int change_clocksource(void *data)
437 struct timekeeper *tk = &timekeeper;
438 struct clocksource *new, *old;
441 new = (struct clocksource *) data;
443 write_seqlock_irqsave(&tk->lock, flags);
445 timekeeping_forward_now(tk);
446 if (!new->enable || new->enable(new) == 0) {
448 tk_setup_internals(tk, new);
452 timekeeping_update(tk, true);
454 write_sequnlock_irqrestore(&tk->lock, flags);
460 * timekeeping_notify - Install a new clock source
461 * @clock: pointer to the clock source
463 * This function is called from clocksource.c after a new, better clock
464 * source has been registered. The caller holds the clocksource_mutex.
466 void timekeeping_notify(struct clocksource *clock)
468 struct timekeeper *tk = &timekeeper;
470 if (tk->clock == clock)
472 stop_machine(change_clocksource, clock, NULL);
477 * ktime_get_real - get the real (wall-) time in ktime_t format
479 * returns the time in ktime_t format
481 ktime_t ktime_get_real(void)
485 getnstimeofday(&now);
487 return timespec_to_ktime(now);
489 EXPORT_SYMBOL_GPL(ktime_get_real);
492 * getrawmonotonic - Returns the raw monotonic time in a timespec
493 * @ts: pointer to the timespec to be set
495 * Returns the raw monotonic time (completely un-modified by ntp)
497 void getrawmonotonic(struct timespec *ts)
499 struct timekeeper *tk = &timekeeper;
504 seq = read_seqbegin(&tk->lock);
505 nsecs = timekeeping_get_ns_raw(tk);
508 } while (read_seqretry(&tk->lock, seq));
510 timespec_add_ns(ts, nsecs);
512 EXPORT_SYMBOL(getrawmonotonic);
515 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
517 int timekeeping_valid_for_hres(void)
519 struct timekeeper *tk = &timekeeper;
524 seq = read_seqbegin(&tk->lock);
526 ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
528 } while (read_seqretry(&tk->lock, seq));
534 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
536 u64 timekeeping_max_deferment(void)
538 struct timekeeper *tk = &timekeeper;
543 seq = read_seqbegin(&tk->lock);
545 ret = tk->clock->max_idle_ns;
547 } while (read_seqretry(&tk->lock, seq));
553 * read_persistent_clock - Return time from the persistent clock.
555 * Weak dummy function for arches that do not yet support it.
556 * Reads the time from the battery backed persistent clock.
557 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
559 * XXX - Do be sure to remove it once all arches implement it.
561 void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
568 * read_boot_clock - Return time of the system start.
570 * Weak dummy function for arches that do not yet support it.
571 * Function to read the exact time the system has been started.
572 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
574 * XXX - Do be sure to remove it once all arches implement it.
576 void __attribute__((weak)) read_boot_clock(struct timespec *ts)
583 * timekeeping_init - Initializes the clocksource and common timekeeping values
585 void __init timekeeping_init(void)
587 struct timekeeper *tk = &timekeeper;
588 struct clocksource *clock;
590 struct timespec now, boot, tmp;
592 read_persistent_clock(&now);
593 if (!timespec_valid_strict(&now)) {
594 pr_warn("WARNING: Persistent clock returned invalid value!\n"
595 " Check your CMOS/BIOS settings.\n");
600 read_boot_clock(&boot);
601 if (!timespec_valid_strict(&boot)) {
602 pr_warn("WARNING: Boot clock returned invalid value!\n"
603 " Check your CMOS/BIOS settings.\n");
608 seqlock_init(&tk->lock);
612 write_seqlock_irqsave(&tk->lock, flags);
613 clock = clocksource_default_clock();
615 clock->enable(clock);
616 tk_setup_internals(tk, clock);
618 tk_set_xtime(tk, &now);
619 tk->raw_time.tv_sec = 0;
620 tk->raw_time.tv_nsec = 0;
621 if (boot.tv_sec == 0 && boot.tv_nsec == 0)
624 set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec);
625 tk_set_wall_to_mono(tk, tmp);
629 tk_set_sleep_time(tk, tmp);
631 write_sequnlock_irqrestore(&tk->lock, flags);
634 /* time in seconds when suspend began */
635 static struct timespec timekeeping_suspend_time;
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 timekeeper *tk,
645 struct timespec *delta)
647 if (!timespec_valid_strict(delta)) {
648 printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid "
649 "sleep delta value!\n");
652 tk_xtime_add(tk, delta);
653 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta));
654 tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta));
658 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
659 * @delta: pointer to a timespec delta value
661 * This hook is for architectures that cannot support read_persistent_clock
662 * because their RTC/persistent clock is only accessible when irqs are enabled.
664 * This function should only be called by rtc_resume(), and allows
665 * a suspend offset to be injected into the timekeeping values.
667 void timekeeping_inject_sleeptime(struct timespec *delta)
669 struct timekeeper *tk = &timekeeper;
673 /* Make sure we don't set the clock twice */
674 read_persistent_clock(&ts);
675 if (!(ts.tv_sec == 0 && ts.tv_nsec == 0))
678 write_seqlock_irqsave(&tk->lock, flags);
680 timekeeping_forward_now(tk);
682 __timekeeping_inject_sleeptime(tk, delta);
684 timekeeping_update(tk, true);
686 write_sequnlock_irqrestore(&tk->lock, flags);
688 /* signal hrtimers about time change */
693 * timekeeping_resume - Resumes the generic timekeeping subsystem.
695 * This is for the generic clocksource timekeeping.
696 * xtime/wall_to_monotonic/jiffies/etc are
697 * still managed by arch specific suspend/resume code.
699 static void timekeeping_resume(void)
701 struct timekeeper *tk = &timekeeper;
705 read_persistent_clock(&ts);
707 clockevents_resume();
708 clocksource_resume();
710 write_seqlock_irqsave(&tk->lock, flags);
712 if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) {
713 ts = timespec_sub(ts, timekeeping_suspend_time);
714 __timekeeping_inject_sleeptime(tk, &ts);
716 /* re-base the last cycle value */
717 tk->clock->cycle_last = tk->clock->read(tk->clock);
719 timekeeping_suspended = 0;
720 timekeeping_update(tk, false);
721 write_sequnlock_irqrestore(&tk->lock, flags);
723 touch_softlockup_watchdog();
725 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
727 /* Resume hrtimers */
731 static int timekeeping_suspend(void)
733 struct timekeeper *tk = &timekeeper;
735 struct timespec delta, delta_delta;
736 static struct timespec old_delta;
738 read_persistent_clock(&timekeeping_suspend_time);
740 write_seqlock_irqsave(&tk->lock, flags);
741 timekeeping_forward_now(tk);
742 timekeeping_suspended = 1;
745 * To avoid drift caused by repeated suspend/resumes,
746 * which each can add ~1 second drift error,
747 * try to compensate so the difference in system time
748 * and persistent_clock time stays close to constant.
750 delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time);
751 delta_delta = timespec_sub(delta, old_delta);
752 if (abs(delta_delta.tv_sec) >= 2) {
754 * if delta_delta is too large, assume time correction
755 * has occured and set old_delta to the current delta.
759 /* Otherwise try to adjust old_system to compensate */
760 timekeeping_suspend_time =
761 timespec_add(timekeeping_suspend_time, delta_delta);
763 write_sequnlock_irqrestore(&tk->lock, flags);
765 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
766 clocksource_suspend();
767 clockevents_suspend();
772 /* sysfs resume/suspend bits for timekeeping */
773 static struct syscore_ops timekeeping_syscore_ops = {
774 .resume = timekeeping_resume,
775 .suspend = timekeeping_suspend,
778 static int __init timekeeping_init_ops(void)
780 register_syscore_ops(&timekeeping_syscore_ops);
784 device_initcall(timekeeping_init_ops);
787 * If the error is already larger, we look ahead even further
788 * to compensate for late or lost adjustments.
790 static __always_inline int timekeeping_bigadjust(struct timekeeper *tk,
791 s64 error, s64 *interval,
799 * Use the current error value to determine how much to look ahead.
800 * The larger the error the slower we adjust for it to avoid problems
801 * with losing too many ticks, otherwise we would overadjust and
802 * produce an even larger error. The smaller the adjustment the
803 * faster we try to adjust for it, as lost ticks can do less harm
804 * here. This is tuned so that an error of about 1 msec is adjusted
805 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
807 error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
808 error2 = abs(error2);
809 for (look_ahead = 0; error2 > 0; look_ahead++)
813 * Now calculate the error in (1 << look_ahead) ticks, but first
814 * remove the single look ahead already included in the error.
816 tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1);
817 tick_error -= tk->xtime_interval >> 1;
818 error = ((error - tick_error) >> look_ahead) + tick_error;
820 /* Finally calculate the adjustment shift value. */
825 *interval = -*interval;
829 for (adj = 0; error > i; adj++)
838 * Adjust the multiplier to reduce the error value,
839 * this is optimized for the most common adjustments of -1,0,1,
840 * for other values we can do a bit more work.
842 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
844 s64 error, interval = tk->cycle_interval;
848 * The point of this is to check if the error is greater than half
851 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
853 * Note we subtract one in the shift, so that error is really error*2.
854 * This "saves" dividing(shifting) interval twice, but keeps the
855 * (error > interval) comparison as still measuring if error is
856 * larger than half an interval.
858 * Note: It does not "save" on aggravation when reading the code.
860 error = tk->ntp_error >> (tk->ntp_error_shift - 1);
861 if (error > interval) {
863 * We now divide error by 4(via shift), which checks if
864 * the error is greater than twice the interval.
865 * If it is greater, we need a bigadjust, if its smaller,
866 * we can adjust by 1.
870 * XXX - In update_wall_time, we round up to the next
871 * nanosecond, and store the amount rounded up into
872 * the error. This causes the likely below to be unlikely.
874 * The proper fix is to avoid rounding up by using
875 * the high precision tk->xtime_nsec instead of
876 * xtime.tv_nsec everywhere. Fixing this will take some
879 if (likely(error <= interval))
882 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
884 if (error < -interval) {
885 /* See comment above, this is just switched for the negative */
887 if (likely(error >= -interval)) {
889 interval = -interval;
892 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
899 if (unlikely(tk->clock->maxadj &&
900 (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) {
901 printk_once(KERN_WARNING
902 "Adjusting %s more than 11%% (%ld vs %ld)\n",
903 tk->clock->name, (long)tk->mult + adj,
904 (long)tk->clock->mult + tk->clock->maxadj);
907 * So the following can be confusing.
909 * To keep things simple, lets assume adj == 1 for now.
911 * When adj != 1, remember that the interval and offset values
912 * have been appropriately scaled so the math is the same.
914 * The basic idea here is that we're increasing the multiplier
915 * by one, this causes the xtime_interval to be incremented by
916 * one cycle_interval. This is because:
917 * xtime_interval = cycle_interval * mult
918 * So if mult is being incremented by one:
919 * xtime_interval = cycle_interval * (mult + 1)
921 * xtime_interval = (cycle_interval * mult) + cycle_interval
922 * Which can be shortened to:
923 * xtime_interval += cycle_interval
925 * So offset stores the non-accumulated cycles. Thus the current
926 * time (in shifted nanoseconds) is:
927 * now = (offset * adj) + xtime_nsec
928 * Now, even though we're adjusting the clock frequency, we have
929 * to keep time consistent. In other words, we can't jump back
930 * in time, and we also want to avoid jumping forward in time.
932 * So given the same offset value, we need the time to be the same
933 * both before and after the freq adjustment.
934 * now = (offset * adj_1) + xtime_nsec_1
935 * now = (offset * adj_2) + xtime_nsec_2
937 * (offset * adj_1) + xtime_nsec_1 =
938 * (offset * adj_2) + xtime_nsec_2
942 * (offset * adj_1) + xtime_nsec_1 =
943 * (offset * (adj_1+1)) + xtime_nsec_2
944 * (offset * adj_1) + xtime_nsec_1 =
945 * (offset * adj_1) + offset + xtime_nsec_2
946 * Canceling the sides:
947 * xtime_nsec_1 = offset + xtime_nsec_2
949 * xtime_nsec_2 = xtime_nsec_1 - offset
950 * Which simplfies to:
951 * xtime_nsec -= offset
953 * XXX - TODO: Doc ntp_error calculation.
956 tk->xtime_interval += interval;
957 tk->xtime_nsec -= offset;
958 tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
962 * It may be possible that when we entered this function, xtime_nsec
963 * was very small. Further, if we're slightly speeding the clocksource
964 * in the code above, its possible the required corrective factor to
965 * xtime_nsec could cause it to underflow.
967 * Now, since we already accumulated the second, cannot simply roll
968 * the accumulated second back, since the NTP subsystem has been
969 * notified via second_overflow. So instead we push xtime_nsec forward
970 * by the amount we underflowed, and add that amount into the error.
972 * We'll correct this error next time through this function, when
973 * xtime_nsec is not as small.
975 if (unlikely((s64)tk->xtime_nsec < 0)) {
976 s64 neg = -(s64)tk->xtime_nsec;
978 tk->ntp_error += neg << tk->ntp_error_shift;
984 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
986 * Helper function that accumulates a the nsecs greater then a second
987 * from the xtime_nsec field to the xtime_secs field.
988 * It also calls into the NTP code to handle leapsecond processing.
991 static inline void accumulate_nsecs_to_secs(struct timekeeper *tk)
993 u64 nsecps = (u64)NSEC_PER_SEC << tk->shift;
995 while (tk->xtime_nsec >= nsecps) {
998 tk->xtime_nsec -= nsecps;
1001 /* Figure out if its a leap sec and apply if needed */
1002 leap = second_overflow(tk->xtime_sec);
1003 if (unlikely(leap)) {
1006 tk->xtime_sec += leap;
1010 tk_set_wall_to_mono(tk,
1011 timespec_sub(tk->wall_to_monotonic, ts));
1013 clock_was_set_delayed();
1019 * logarithmic_accumulation - shifted accumulation of cycles
1021 * This functions accumulates a shifted interval of cycles into
1022 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1025 * Returns the unconsumed cycles.
1027 static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1032 /* If the offset is smaller then a shifted interval, do nothing */
1033 if (offset < tk->cycle_interval<<shift)
1036 /* Accumulate one shifted interval */
1037 offset -= tk->cycle_interval << shift;
1038 tk->clock->cycle_last += tk->cycle_interval << shift;
1040 tk->xtime_nsec += tk->xtime_interval << shift;
1041 accumulate_nsecs_to_secs(tk);
1043 /* Accumulate raw time */
1044 raw_nsecs = (u64)tk->raw_interval << shift;
1045 raw_nsecs += tk->raw_time.tv_nsec;
1046 if (raw_nsecs >= NSEC_PER_SEC) {
1047 u64 raw_secs = raw_nsecs;
1048 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1049 tk->raw_time.tv_sec += raw_secs;
1051 tk->raw_time.tv_nsec = raw_nsecs;
1053 /* Accumulate error between NTP and clock interval */
1054 tk->ntp_error += ntp_tick_length() << shift;
1055 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1056 (tk->ntp_error_shift + shift);
1061 #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
1062 static inline void old_vsyscall_fixup(struct timekeeper *tk)
1067 * Store only full nanoseconds into xtime_nsec after rounding
1068 * it up and add the remainder to the error difference.
1069 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
1070 * by truncating the remainder in vsyscalls. However, it causes
1071 * additional work to be done in timekeeping_adjust(). Once
1072 * the vsyscall implementations are converted to use xtime_nsec
1073 * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
1074 * users are removed, this can be killed.
1076 remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1);
1077 tk->xtime_nsec -= remainder;
1078 tk->xtime_nsec += 1ULL << tk->shift;
1079 tk->ntp_error += remainder << tk->ntp_error_shift;
1083 #define old_vsyscall_fixup(tk)
1089 * update_wall_time - Uses the current clocksource to increment the wall time
1092 static void update_wall_time(void)
1094 struct clocksource *clock;
1095 struct timekeeper *tk = &timekeeper;
1097 int shift = 0, maxshift;
1098 unsigned long flags;
1100 write_seqlock_irqsave(&tk->lock, flags);
1102 /* Make sure we're fully resumed: */
1103 if (unlikely(timekeeping_suspended))
1108 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1109 offset = tk->cycle_interval;
1111 offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
1114 /* Check if there's really nothing to do */
1115 if (offset < tk->cycle_interval)
1119 * With NO_HZ we may have to accumulate many cycle_intervals
1120 * (think "ticks") worth of time at once. To do this efficiently,
1121 * we calculate the largest doubling multiple of cycle_intervals
1122 * that is smaller than the offset. We then accumulate that
1123 * chunk in one go, and then try to consume the next smaller
1126 shift = ilog2(offset) - ilog2(tk->cycle_interval);
1127 shift = max(0, shift);
1128 /* Bound shift to one less than what overflows tick_length */
1129 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1130 shift = min(shift, maxshift);
1131 while (offset >= tk->cycle_interval) {
1132 offset = logarithmic_accumulation(tk, offset, shift);
1133 if (offset < tk->cycle_interval<<shift)
1137 /* correct the clock when NTP error is too big */
1138 timekeeping_adjust(tk, offset);
1141 * XXX This can be killed once everyone converts
1142 * to the new update_vsyscall.
1144 old_vsyscall_fixup(tk);
1147 * Finally, make sure that after the rounding
1148 * xtime_nsec isn't larger than NSEC_PER_SEC
1150 accumulate_nsecs_to_secs(tk);
1152 timekeeping_update(tk, false);
1155 write_sequnlock_irqrestore(&tk->lock, flags);
1160 * getboottime - Return the real time of system boot.
1161 * @ts: pointer to the timespec to be set
1163 * Returns the wall-time of boot in a timespec.
1165 * This is based on the wall_to_monotonic offset and the total suspend
1166 * time. Calls to settimeofday will affect the value returned (which
1167 * basically means that however wrong your real time clock is at boot time,
1168 * you get the right time here).
1170 void getboottime(struct timespec *ts)
1172 struct timekeeper *tk = &timekeeper;
1173 struct timespec boottime = {
1174 .tv_sec = tk->wall_to_monotonic.tv_sec +
1175 tk->total_sleep_time.tv_sec,
1176 .tv_nsec = tk->wall_to_monotonic.tv_nsec +
1177 tk->total_sleep_time.tv_nsec
1180 set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
1182 EXPORT_SYMBOL_GPL(getboottime);
1185 * get_monotonic_boottime - Returns monotonic time since boot
1186 * @ts: pointer to the timespec to be set
1188 * Returns the monotonic time since boot in a timespec.
1190 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1191 * includes the time spent in suspend.
1193 void get_monotonic_boottime(struct timespec *ts)
1195 struct timekeeper *tk = &timekeeper;
1196 struct timespec tomono, sleep;
1200 WARN_ON(timekeeping_suspended);
1203 seq = read_seqbegin(&tk->lock);
1204 ts->tv_sec = tk->xtime_sec;
1205 nsec = timekeeping_get_ns(tk);
1206 tomono = tk->wall_to_monotonic;
1207 sleep = tk->total_sleep_time;
1209 } while (read_seqretry(&tk->lock, seq));
1211 ts->tv_sec += tomono.tv_sec + sleep.tv_sec;
1213 timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec);
1215 EXPORT_SYMBOL_GPL(get_monotonic_boottime);
1218 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1220 * Returns the monotonic time since boot in a ktime
1222 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1223 * includes the time spent in suspend.
1225 ktime_t ktime_get_boottime(void)
1229 get_monotonic_boottime(&ts);
1230 return timespec_to_ktime(ts);
1232 EXPORT_SYMBOL_GPL(ktime_get_boottime);
1235 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1236 * @ts: pointer to the timespec to be converted
1238 void monotonic_to_bootbased(struct timespec *ts)
1240 struct timekeeper *tk = &timekeeper;
1242 *ts = timespec_add(*ts, tk->total_sleep_time);
1244 EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
1246 unsigned long get_seconds(void)
1248 struct timekeeper *tk = &timekeeper;
1250 return tk->xtime_sec;
1252 EXPORT_SYMBOL(get_seconds);
1254 struct timespec __current_kernel_time(void)
1256 struct timekeeper *tk = &timekeeper;
1258 return tk_xtime(tk);
1261 struct timespec current_kernel_time(void)
1263 struct timekeeper *tk = &timekeeper;
1264 struct timespec now;
1268 seq = read_seqbegin(&tk->lock);
1271 } while (read_seqretry(&tk->lock, seq));
1275 EXPORT_SYMBOL(current_kernel_time);
1277 struct timespec get_monotonic_coarse(void)
1279 struct timekeeper *tk = &timekeeper;
1280 struct timespec now, mono;
1284 seq = read_seqbegin(&tk->lock);
1287 mono = tk->wall_to_monotonic;
1288 } while (read_seqretry(&tk->lock, seq));
1290 set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
1291 now.tv_nsec + mono.tv_nsec);
1296 * Must hold jiffies_lock
1298 void do_timer(unsigned long ticks)
1300 jiffies_64 += ticks;
1302 calc_global_load(ticks);
1306 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1307 * and sleep offsets.
1308 * @xtim: pointer to timespec to be set with xtime
1309 * @wtom: pointer to timespec to be set with wall_to_monotonic
1310 * @sleep: pointer to timespec to be set with time in suspend
1312 void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
1313 struct timespec *wtom, struct timespec *sleep)
1315 struct timekeeper *tk = &timekeeper;
1319 seq = read_seqbegin(&tk->lock);
1320 *xtim = tk_xtime(tk);
1321 *wtom = tk->wall_to_monotonic;
1322 *sleep = tk->total_sleep_time;
1323 } while (read_seqretry(&tk->lock, seq));
1326 #ifdef CONFIG_HIGH_RES_TIMERS
1328 * ktime_get_update_offsets - hrtimer helper
1329 * @offs_real: pointer to storage for monotonic -> realtime offset
1330 * @offs_boot: pointer to storage for monotonic -> boottime offset
1332 * Returns current monotonic time and updates the offsets
1333 * Called from hrtimer_interupt() or retrigger_next_event()
1335 ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot)
1337 struct timekeeper *tk = &timekeeper;
1343 seq = read_seqbegin(&tk->lock);
1345 secs = tk->xtime_sec;
1346 nsecs = timekeeping_get_ns(tk);
1348 *offs_real = tk->offs_real;
1349 *offs_boot = tk->offs_boot;
1350 } while (read_seqretry(&tk->lock, seq));
1352 now = ktime_add_ns(ktime_set(secs, 0), nsecs);
1353 now = ktime_sub(now, *offs_real);
1359 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1361 ktime_t ktime_get_monotonic_offset(void)
1363 struct timekeeper *tk = &timekeeper;
1365 struct timespec wtom;
1368 seq = read_seqbegin(&tk->lock);
1369 wtom = tk->wall_to_monotonic;
1370 } while (read_seqretry(&tk->lock, seq));
1372 return timespec_to_ktime(wtom);
1374 EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset);
1377 * xtime_update() - advances the timekeeping infrastructure
1378 * @ticks: number of ticks, that have elapsed since the last call.
1380 * Must be called with interrupts disabled.
1382 void xtime_update(unsigned long ticks)
1384 write_seqlock(&jiffies_lock);
1386 write_sequnlock(&jiffies_lock);