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;
29 * This read-write spinlock protects us from races in SMP while
32 __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
34 /* flag for if timekeeping is suspended */
35 int __read_mostly timekeeping_suspended;
37 static inline void tk_normalize_xtime(struct timekeeper *tk)
39 while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) {
40 tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift;
45 static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts)
47 tk->xtime_sec = ts->tv_sec;
48 tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift;
51 static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts)
53 tk->xtime_sec += ts->tv_sec;
54 tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift;
55 tk_normalize_xtime(tk);
58 static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm)
63 * Verify consistency of: offset_real = -wall_to_monotonic
64 * before modifying anything
66 set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec,
67 -tk->wall_to_monotonic.tv_nsec);
68 WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64);
69 tk->wall_to_monotonic = wtm;
70 set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
71 tk->offs_real = timespec_to_ktime(tmp);
74 static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t)
76 /* Verify consistency before modifying */
77 WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64);
79 tk->total_sleep_time = t;
80 tk->offs_boot = timespec_to_ktime(t);
84 * timekeeper_setup_internals - Set up internals to use clocksource clock.
86 * @clock: Pointer to clocksource.
88 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
89 * pair and interval request.
91 * Unless you're the timekeeping code, you should not be using this!
93 static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
97 struct clocksource *old_clock;
99 old_clock = tk->clock;
101 clock->cycle_last = clock->read(clock);
103 /* Do the ns -> cycle conversion first, using original mult */
104 tmp = NTP_INTERVAL_LENGTH;
105 tmp <<= clock->shift;
107 tmp += clock->mult/2;
108 do_div(tmp, clock->mult);
112 interval = (cycle_t) tmp;
113 tk->cycle_interval = interval;
115 /* Go back from cycles -> shifted ns */
116 tk->xtime_interval = (u64) interval * clock->mult;
117 tk->xtime_remainder = ntpinterval - tk->xtime_interval;
119 ((u64) interval * clock->mult) >> clock->shift;
121 /* if changing clocks, convert xtime_nsec shift units */
123 int shift_change = clock->shift - old_clock->shift;
124 if (shift_change < 0)
125 tk->xtime_nsec >>= -shift_change;
127 tk->xtime_nsec <<= shift_change;
129 tk->shift = clock->shift;
132 tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
135 * The timekeeper keeps its own mult values for the currently
136 * active clocksource. These value will be adjusted via NTP
137 * to counteract clock drifting.
139 tk->mult = clock->mult;
142 /* Timekeeper helper functions. */
143 static inline s64 timekeeping_get_ns(struct timekeeper *tk)
145 cycle_t cycle_now, cycle_delta;
146 struct clocksource *clock;
149 /* read clocksource: */
151 cycle_now = clock->read(clock);
153 /* calculate the delta since the last update_wall_time: */
154 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
156 nsec = cycle_delta * tk->mult + tk->xtime_nsec;
159 /* If arch requires, add in gettimeoffset() */
160 return nsec + arch_gettimeoffset();
163 static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
165 cycle_t cycle_now, cycle_delta;
166 struct clocksource *clock;
169 /* read clocksource: */
171 cycle_now = clock->read(clock);
173 /* calculate the delta since the last update_wall_time: */
174 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
176 /* convert delta to nanoseconds. */
177 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
179 /* If arch requires, add in gettimeoffset() */
180 return nsec + arch_gettimeoffset();
183 /* must hold write on timekeeper.lock */
184 static void timekeeping_update(struct timekeeper *tk, bool clearntp)
194 * timekeeping_forward_now - update clock to the current time
196 * Forward the current clock to update its state since the last call to
197 * update_wall_time(). This is useful before significant clock changes,
198 * as it avoids having to deal with this time offset explicitly.
200 static void timekeeping_forward_now(struct timekeeper *tk)
202 cycle_t cycle_now, cycle_delta;
203 struct clocksource *clock;
207 cycle_now = clock->read(clock);
208 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
209 clock->cycle_last = cycle_now;
211 tk->xtime_nsec += cycle_delta * tk->mult;
213 /* If arch requires, add in gettimeoffset() */
214 tk->xtime_nsec += (u64)arch_gettimeoffset() << tk->shift;
216 tk_normalize_xtime(tk);
218 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
219 timespec_add_ns(&tk->raw_time, nsec);
223 * getnstimeofday - Returns the time of day in a timespec
224 * @ts: pointer to the timespec to be set
226 * Returns the time of day in a timespec.
228 void getnstimeofday(struct timespec *ts)
230 struct timekeeper *tk = &timekeeper;
234 WARN_ON(timekeeping_suspended);
237 seq = read_seqbegin(&tk->lock);
239 ts->tv_sec = tk->xtime_sec;
240 nsecs = timekeeping_get_ns(tk);
242 } while (read_seqretry(&tk->lock, seq));
245 timespec_add_ns(ts, nsecs);
247 EXPORT_SYMBOL(getnstimeofday);
249 ktime_t ktime_get(void)
251 struct timekeeper *tk = &timekeeper;
255 WARN_ON(timekeeping_suspended);
258 seq = read_seqbegin(&tk->lock);
259 secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec;
260 nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec;
262 } while (read_seqretry(&tk->lock, seq));
264 * Use ktime_set/ktime_add_ns to create a proper ktime on
265 * 32-bit architectures without CONFIG_KTIME_SCALAR.
267 return ktime_add_ns(ktime_set(secs, 0), nsecs);
269 EXPORT_SYMBOL_GPL(ktime_get);
272 * ktime_get_ts - get the monotonic clock in timespec format
273 * @ts: pointer to timespec variable
275 * The function calculates the monotonic clock from the realtime
276 * clock and the wall_to_monotonic offset and stores the result
277 * in normalized timespec format in the variable pointed to by @ts.
279 void ktime_get_ts(struct timespec *ts)
281 struct timekeeper *tk = &timekeeper;
282 struct timespec tomono;
286 WARN_ON(timekeeping_suspended);
289 seq = read_seqbegin(&tk->lock);
290 ts->tv_sec = tk->xtime_sec;
291 nsec = timekeeping_get_ns(tk);
292 tomono = tk->wall_to_monotonic;
294 } while (read_seqretry(&tk->lock, seq));
296 ts->tv_sec += tomono.tv_sec;
298 timespec_add_ns(ts, nsec + tomono.tv_nsec);
300 EXPORT_SYMBOL_GPL(ktime_get_ts);
302 #ifdef CONFIG_NTP_PPS
305 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
306 * @ts_raw: pointer to the timespec to be set to raw monotonic time
307 * @ts_real: pointer to the timespec to be set to the time of day
309 * This function reads both the time of day and raw monotonic time at the
310 * same time atomically and stores the resulting timestamps in timespec
313 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
315 struct timekeeper *tk = &timekeeper;
317 s64 nsecs_raw, nsecs_real;
319 WARN_ON_ONCE(timekeeping_suspended);
322 seq = read_seqbegin(&tk->lock);
324 *ts_raw = tk->raw_time;
325 ts_real->tv_sec = tk->xtime_sec;
326 ts_real->tv_nsec = 0;
328 nsecs_raw = timekeeping_get_ns_raw(tk);
329 nsecs_real = timekeeping_get_ns(tk);
331 } while (read_seqretry(&tk->lock, seq));
333 timespec_add_ns(ts_raw, nsecs_raw);
334 timespec_add_ns(ts_real, nsecs_real);
336 EXPORT_SYMBOL(getnstime_raw_and_real);
338 #endif /* CONFIG_NTP_PPS */
341 * do_gettimeofday - Returns the time of day in a timeval
342 * @tv: pointer to the timeval to be set
344 * NOTE: Users should be converted to using getnstimeofday()
346 void do_gettimeofday(struct timeval *tv)
350 getnstimeofday(&now);
351 tv->tv_sec = now.tv_sec;
352 tv->tv_usec = now.tv_nsec/1000;
354 EXPORT_SYMBOL(do_gettimeofday);
357 * do_settimeofday - Sets the time of day
358 * @tv: pointer to the timespec variable containing the new time
360 * Sets the time of day to the new time and update NTP and notify hrtimers
362 int do_settimeofday(const struct timespec *tv)
364 struct timekeeper *tk = &timekeeper;
365 struct timespec ts_delta, xt;
368 if (!timespec_valid_strict(tv))
371 write_seqlock_irqsave(&tk->lock, flags);
373 timekeeping_forward_now(tk);
376 ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
377 ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
379 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta));
381 tk_set_xtime(tk, tv);
383 timekeeping_update(tk, true);
385 write_sequnlock_irqrestore(&tk->lock, flags);
387 /* signal hrtimers about time change */
392 EXPORT_SYMBOL(do_settimeofday);
395 * timekeeping_inject_offset - Adds or subtracts from the current time.
396 * @tv: pointer to the timespec variable containing the offset
398 * Adds or subtracts an offset value from the current time.
400 int timekeeping_inject_offset(struct timespec *ts)
402 struct timekeeper *tk = &timekeeper;
407 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
410 write_seqlock_irqsave(&tk->lock, flags);
412 timekeeping_forward_now(tk);
414 /* Make sure the proposed value is valid */
415 tmp = timespec_add(tk_xtime(tk), *ts);
416 if (!timespec_valid_strict(&tmp)) {
421 tk_xtime_add(tk, ts);
422 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts));
424 error: /* even if we error out, we forwarded the time, so call update */
425 timekeeping_update(tk, true);
427 write_sequnlock_irqrestore(&tk->lock, flags);
429 /* signal hrtimers about time change */
434 EXPORT_SYMBOL(timekeeping_inject_offset);
437 * change_clocksource - Swaps clocksources if a new one is available
439 * Accumulates current time interval and initializes new clocksource
441 static int change_clocksource(void *data)
443 struct timekeeper *tk = &timekeeper;
444 struct clocksource *new, *old;
447 new = (struct clocksource *) data;
449 write_seqlock_irqsave(&tk->lock, flags);
451 timekeeping_forward_now(tk);
452 if (!new->enable || new->enable(new) == 0) {
454 tk_setup_internals(tk, new);
458 timekeeping_update(tk, true);
460 write_sequnlock_irqrestore(&tk->lock, flags);
466 * timekeeping_notify - Install a new clock source
467 * @clock: pointer to the clock source
469 * This function is called from clocksource.c after a new, better clock
470 * source has been registered. The caller holds the clocksource_mutex.
472 void timekeeping_notify(struct clocksource *clock)
474 struct timekeeper *tk = &timekeeper;
476 if (tk->clock == clock)
478 stop_machine(change_clocksource, clock, NULL);
483 * ktime_get_real - get the real (wall-) time in ktime_t format
485 * returns the time in ktime_t format
487 ktime_t ktime_get_real(void)
491 getnstimeofday(&now);
493 return timespec_to_ktime(now);
495 EXPORT_SYMBOL_GPL(ktime_get_real);
498 * getrawmonotonic - Returns the raw monotonic time in a timespec
499 * @ts: pointer to the timespec to be set
501 * Returns the raw monotonic time (completely un-modified by ntp)
503 void getrawmonotonic(struct timespec *ts)
505 struct timekeeper *tk = &timekeeper;
510 seq = read_seqbegin(&tk->lock);
511 nsecs = timekeeping_get_ns_raw(tk);
514 } while (read_seqretry(&tk->lock, seq));
516 timespec_add_ns(ts, nsecs);
518 EXPORT_SYMBOL(getrawmonotonic);
521 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
523 int timekeeping_valid_for_hres(void)
525 struct timekeeper *tk = &timekeeper;
530 seq = read_seqbegin(&tk->lock);
532 ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
534 } while (read_seqretry(&tk->lock, seq));
540 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
542 u64 timekeeping_max_deferment(void)
544 struct timekeeper *tk = &timekeeper;
549 seq = read_seqbegin(&tk->lock);
551 ret = tk->clock->max_idle_ns;
553 } while (read_seqretry(&tk->lock, seq));
559 * read_persistent_clock - Return time from the persistent clock.
561 * Weak dummy function for arches that do not yet support it.
562 * Reads the time from the battery backed persistent clock.
563 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
565 * XXX - Do be sure to remove it once all arches implement it.
567 void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
574 * read_boot_clock - Return time of the system start.
576 * Weak dummy function for arches that do not yet support it.
577 * Function to read the exact time the system has been started.
578 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
580 * XXX - Do be sure to remove it once all arches implement it.
582 void __attribute__((weak)) read_boot_clock(struct timespec *ts)
589 * timekeeping_init - Initializes the clocksource and common timekeeping values
591 void __init timekeeping_init(void)
593 struct timekeeper *tk = &timekeeper;
594 struct clocksource *clock;
596 struct timespec now, boot, tmp;
598 read_persistent_clock(&now);
599 if (!timespec_valid_strict(&now)) {
600 pr_warn("WARNING: Persistent clock returned invalid value!\n"
601 " Check your CMOS/BIOS settings.\n");
606 read_boot_clock(&boot);
607 if (!timespec_valid_strict(&boot)) {
608 pr_warn("WARNING: Boot clock returned invalid value!\n"
609 " Check your CMOS/BIOS settings.\n");
614 seqlock_init(&tk->lock);
618 write_seqlock_irqsave(&tk->lock, flags);
619 clock = clocksource_default_clock();
621 clock->enable(clock);
622 tk_setup_internals(tk, clock);
624 tk_set_xtime(tk, &now);
625 tk->raw_time.tv_sec = 0;
626 tk->raw_time.tv_nsec = 0;
627 if (boot.tv_sec == 0 && boot.tv_nsec == 0)
630 set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec);
631 tk_set_wall_to_mono(tk, tmp);
635 tk_set_sleep_time(tk, tmp);
637 write_sequnlock_irqrestore(&tk->lock, flags);
640 /* time in seconds when suspend began */
641 static struct timespec timekeeping_suspend_time;
644 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
645 * @delta: pointer to a timespec delta value
647 * Takes a timespec offset measuring a suspend interval and properly
648 * adds the sleep offset to the timekeeping variables.
650 static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
651 struct timespec *delta)
653 if (!timespec_valid_strict(delta)) {
654 printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid "
655 "sleep delta value!\n");
658 tk_xtime_add(tk, delta);
659 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta));
660 tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta));
664 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
665 * @delta: pointer to a timespec delta value
667 * This hook is for architectures that cannot support read_persistent_clock
668 * because their RTC/persistent clock is only accessible when irqs are enabled.
670 * This function should only be called by rtc_resume(), and allows
671 * a suspend offset to be injected into the timekeeping values.
673 void timekeeping_inject_sleeptime(struct timespec *delta)
675 struct timekeeper *tk = &timekeeper;
679 /* Make sure we don't set the clock twice */
680 read_persistent_clock(&ts);
681 if (!(ts.tv_sec == 0 && ts.tv_nsec == 0))
684 write_seqlock_irqsave(&tk->lock, flags);
686 timekeeping_forward_now(tk);
688 __timekeeping_inject_sleeptime(tk, delta);
690 timekeeping_update(tk, true);
692 write_sequnlock_irqrestore(&tk->lock, flags);
694 /* signal hrtimers about time change */
699 * timekeeping_resume - Resumes the generic timekeeping subsystem.
701 * This is for the generic clocksource timekeeping.
702 * xtime/wall_to_monotonic/jiffies/etc are
703 * still managed by arch specific suspend/resume code.
705 static void timekeeping_resume(void)
707 struct timekeeper *tk = &timekeeper;
711 read_persistent_clock(&ts);
713 clockevents_resume();
714 clocksource_resume();
716 write_seqlock_irqsave(&tk->lock, flags);
718 if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) {
719 ts = timespec_sub(ts, timekeeping_suspend_time);
720 __timekeeping_inject_sleeptime(tk, &ts);
722 /* re-base the last cycle value */
723 tk->clock->cycle_last = tk->clock->read(tk->clock);
725 timekeeping_suspended = 0;
726 timekeeping_update(tk, false);
727 write_sequnlock_irqrestore(&tk->lock, flags);
729 touch_softlockup_watchdog();
731 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
733 /* Resume hrtimers */
737 static int timekeeping_suspend(void)
739 struct timekeeper *tk = &timekeeper;
741 struct timespec delta, delta_delta;
742 static struct timespec old_delta;
744 read_persistent_clock(&timekeeping_suspend_time);
746 write_seqlock_irqsave(&tk->lock, flags);
747 timekeeping_forward_now(tk);
748 timekeeping_suspended = 1;
751 * To avoid drift caused by repeated suspend/resumes,
752 * which each can add ~1 second drift error,
753 * try to compensate so the difference in system time
754 * and persistent_clock time stays close to constant.
756 delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time);
757 delta_delta = timespec_sub(delta, old_delta);
758 if (abs(delta_delta.tv_sec) >= 2) {
760 * if delta_delta is too large, assume time correction
761 * has occured and set old_delta to the current delta.
765 /* Otherwise try to adjust old_system to compensate */
766 timekeeping_suspend_time =
767 timespec_add(timekeeping_suspend_time, delta_delta);
769 write_sequnlock_irqrestore(&tk->lock, flags);
771 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
772 clocksource_suspend();
773 clockevents_suspend();
778 /* sysfs resume/suspend bits for timekeeping */
779 static struct syscore_ops timekeeping_syscore_ops = {
780 .resume = timekeeping_resume,
781 .suspend = timekeeping_suspend,
784 static int __init timekeeping_init_ops(void)
786 register_syscore_ops(&timekeeping_syscore_ops);
790 device_initcall(timekeeping_init_ops);
793 * If the error is already larger, we look ahead even further
794 * to compensate for late or lost adjustments.
796 static __always_inline int timekeeping_bigadjust(struct timekeeper *tk,
797 s64 error, s64 *interval,
805 * Use the current error value to determine how much to look ahead.
806 * The larger the error the slower we adjust for it to avoid problems
807 * with losing too many ticks, otherwise we would overadjust and
808 * produce an even larger error. The smaller the adjustment the
809 * faster we try to adjust for it, as lost ticks can do less harm
810 * here. This is tuned so that an error of about 1 msec is adjusted
811 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
813 error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
814 error2 = abs(error2);
815 for (look_ahead = 0; error2 > 0; look_ahead++)
819 * Now calculate the error in (1 << look_ahead) ticks, but first
820 * remove the single look ahead already included in the error.
822 tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1);
823 tick_error -= tk->xtime_interval >> 1;
824 error = ((error - tick_error) >> look_ahead) + tick_error;
826 /* Finally calculate the adjustment shift value. */
831 *interval = -*interval;
835 for (adj = 0; error > i; adj++)
844 * Adjust the multiplier to reduce the error value,
845 * this is optimized for the most common adjustments of -1,0,1,
846 * for other values we can do a bit more work.
848 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
850 s64 error, interval = tk->cycle_interval;
854 * The point of this is to check if the error is greater than half
857 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
859 * Note we subtract one in the shift, so that error is really error*2.
860 * This "saves" dividing(shifting) interval twice, but keeps the
861 * (error > interval) comparison as still measuring if error is
862 * larger than half an interval.
864 * Note: It does not "save" on aggravation when reading the code.
866 error = tk->ntp_error >> (tk->ntp_error_shift - 1);
867 if (error > interval) {
869 * We now divide error by 4(via shift), which checks if
870 * the error is greater than twice the interval.
871 * If it is greater, we need a bigadjust, if its smaller,
872 * we can adjust by 1.
876 * XXX - In update_wall_time, we round up to the next
877 * nanosecond, and store the amount rounded up into
878 * the error. This causes the likely below to be unlikely.
880 * The proper fix is to avoid rounding up by using
881 * the high precision tk->xtime_nsec instead of
882 * xtime.tv_nsec everywhere. Fixing this will take some
885 if (likely(error <= interval))
888 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
890 if (error < -interval) {
891 /* See comment above, this is just switched for the negative */
893 if (likely(error >= -interval)) {
895 interval = -interval;
898 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
905 if (unlikely(tk->clock->maxadj &&
906 (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) {
907 printk_once(KERN_WARNING
908 "Adjusting %s more than 11%% (%ld vs %ld)\n",
909 tk->clock->name, (long)tk->mult + adj,
910 (long)tk->clock->mult + tk->clock->maxadj);
913 * So the following can be confusing.
915 * To keep things simple, lets assume adj == 1 for now.
917 * When adj != 1, remember that the interval and offset values
918 * have been appropriately scaled so the math is the same.
920 * The basic idea here is that we're increasing the multiplier
921 * by one, this causes the xtime_interval to be incremented by
922 * one cycle_interval. This is because:
923 * xtime_interval = cycle_interval * mult
924 * So if mult is being incremented by one:
925 * xtime_interval = cycle_interval * (mult + 1)
927 * xtime_interval = (cycle_interval * mult) + cycle_interval
928 * Which can be shortened to:
929 * xtime_interval += cycle_interval
931 * So offset stores the non-accumulated cycles. Thus the current
932 * time (in shifted nanoseconds) is:
933 * now = (offset * adj) + xtime_nsec
934 * Now, even though we're adjusting the clock frequency, we have
935 * to keep time consistent. In other words, we can't jump back
936 * in time, and we also want to avoid jumping forward in time.
938 * So given the same offset value, we need the time to be the same
939 * both before and after the freq adjustment.
940 * now = (offset * adj_1) + xtime_nsec_1
941 * now = (offset * adj_2) + xtime_nsec_2
943 * (offset * adj_1) + xtime_nsec_1 =
944 * (offset * adj_2) + xtime_nsec_2
948 * (offset * adj_1) + xtime_nsec_1 =
949 * (offset * (adj_1+1)) + xtime_nsec_2
950 * (offset * adj_1) + xtime_nsec_1 =
951 * (offset * adj_1) + offset + xtime_nsec_2
952 * Canceling the sides:
953 * xtime_nsec_1 = offset + xtime_nsec_2
955 * xtime_nsec_2 = xtime_nsec_1 - offset
956 * Which simplfies to:
957 * xtime_nsec -= offset
959 * XXX - TODO: Doc ntp_error calculation.
962 tk->xtime_interval += interval;
963 tk->xtime_nsec -= offset;
964 tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
968 * It may be possible that when we entered this function, xtime_nsec
969 * was very small. Further, if we're slightly speeding the clocksource
970 * in the code above, its possible the required corrective factor to
971 * xtime_nsec could cause it to underflow.
973 * Now, since we already accumulated the second, cannot simply roll
974 * the accumulated second back, since the NTP subsystem has been
975 * notified via second_overflow. So instead we push xtime_nsec forward
976 * by the amount we underflowed, and add that amount into the error.
978 * We'll correct this error next time through this function, when
979 * xtime_nsec is not as small.
981 if (unlikely((s64)tk->xtime_nsec < 0)) {
982 s64 neg = -(s64)tk->xtime_nsec;
984 tk->ntp_error += neg << tk->ntp_error_shift;
990 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
992 * Helper function that accumulates a the nsecs greater then a second
993 * from the xtime_nsec field to the xtime_secs field.
994 * It also calls into the NTP code to handle leapsecond processing.
997 static inline void accumulate_nsecs_to_secs(struct timekeeper *tk)
999 u64 nsecps = (u64)NSEC_PER_SEC << tk->shift;
1001 while (tk->xtime_nsec >= nsecps) {
1004 tk->xtime_nsec -= nsecps;
1007 /* Figure out if its a leap sec and apply if needed */
1008 leap = second_overflow(tk->xtime_sec);
1009 if (unlikely(leap)) {
1012 tk->xtime_sec += leap;
1016 tk_set_wall_to_mono(tk,
1017 timespec_sub(tk->wall_to_monotonic, ts));
1019 clock_was_set_delayed();
1025 * logarithmic_accumulation - shifted accumulation of cycles
1027 * This functions accumulates a shifted interval of cycles into
1028 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1031 * Returns the unconsumed cycles.
1033 static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1038 /* If the offset is smaller then a shifted interval, do nothing */
1039 if (offset < tk->cycle_interval<<shift)
1042 /* Accumulate one shifted interval */
1043 offset -= tk->cycle_interval << shift;
1044 tk->clock->cycle_last += tk->cycle_interval << shift;
1046 tk->xtime_nsec += tk->xtime_interval << shift;
1047 accumulate_nsecs_to_secs(tk);
1049 /* Accumulate raw time */
1050 raw_nsecs = (u64)tk->raw_interval << shift;
1051 raw_nsecs += tk->raw_time.tv_nsec;
1052 if (raw_nsecs >= NSEC_PER_SEC) {
1053 u64 raw_secs = raw_nsecs;
1054 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1055 tk->raw_time.tv_sec += raw_secs;
1057 tk->raw_time.tv_nsec = raw_nsecs;
1059 /* Accumulate error between NTP and clock interval */
1060 tk->ntp_error += ntp_tick_length() << shift;
1061 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1062 (tk->ntp_error_shift + shift);
1067 #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
1068 static inline void old_vsyscall_fixup(struct timekeeper *tk)
1073 * Store only full nanoseconds into xtime_nsec after rounding
1074 * it up and add the remainder to the error difference.
1075 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
1076 * by truncating the remainder in vsyscalls. However, it causes
1077 * additional work to be done in timekeeping_adjust(). Once
1078 * the vsyscall implementations are converted to use xtime_nsec
1079 * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
1080 * users are removed, this can be killed.
1082 remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1);
1083 tk->xtime_nsec -= remainder;
1084 tk->xtime_nsec += 1ULL << tk->shift;
1085 tk->ntp_error += remainder << tk->ntp_error_shift;
1089 #define old_vsyscall_fixup(tk)
1095 * update_wall_time - Uses the current clocksource to increment the wall time
1098 static void update_wall_time(void)
1100 struct clocksource *clock;
1101 struct timekeeper *tk = &timekeeper;
1103 int shift = 0, maxshift;
1104 unsigned long flags;
1106 write_seqlock_irqsave(&tk->lock, flags);
1108 /* Make sure we're fully resumed: */
1109 if (unlikely(timekeeping_suspended))
1114 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1115 offset = tk->cycle_interval;
1117 offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
1120 /* Check if there's really nothing to do */
1121 if (offset < tk->cycle_interval)
1125 * With NO_HZ we may have to accumulate many cycle_intervals
1126 * (think "ticks") worth of time at once. To do this efficiently,
1127 * we calculate the largest doubling multiple of cycle_intervals
1128 * that is smaller than the offset. We then accumulate that
1129 * chunk in one go, and then try to consume the next smaller
1132 shift = ilog2(offset) - ilog2(tk->cycle_interval);
1133 shift = max(0, shift);
1134 /* Bound shift to one less than what overflows tick_length */
1135 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1136 shift = min(shift, maxshift);
1137 while (offset >= tk->cycle_interval) {
1138 offset = logarithmic_accumulation(tk, offset, shift);
1139 if (offset < tk->cycle_interval<<shift)
1143 /* correct the clock when NTP error is too big */
1144 timekeeping_adjust(tk, offset);
1147 * XXX This can be killed once everyone converts
1148 * to the new update_vsyscall.
1150 old_vsyscall_fixup(tk);
1153 * Finally, make sure that after the rounding
1154 * xtime_nsec isn't larger than NSEC_PER_SEC
1156 accumulate_nsecs_to_secs(tk);
1158 timekeeping_update(tk, false);
1161 write_sequnlock_irqrestore(&tk->lock, flags);
1166 * getboottime - Return the real time of system boot.
1167 * @ts: pointer to the timespec to be set
1169 * Returns the wall-time of boot in a timespec.
1171 * This is based on the wall_to_monotonic offset and the total suspend
1172 * time. Calls to settimeofday will affect the value returned (which
1173 * basically means that however wrong your real time clock is at boot time,
1174 * you get the right time here).
1176 void getboottime(struct timespec *ts)
1178 struct timekeeper *tk = &timekeeper;
1179 struct timespec boottime = {
1180 .tv_sec = tk->wall_to_monotonic.tv_sec +
1181 tk->total_sleep_time.tv_sec,
1182 .tv_nsec = tk->wall_to_monotonic.tv_nsec +
1183 tk->total_sleep_time.tv_nsec
1186 set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
1188 EXPORT_SYMBOL_GPL(getboottime);
1191 * get_monotonic_boottime - Returns monotonic time since boot
1192 * @ts: pointer to the timespec to be set
1194 * Returns the monotonic time since boot in a timespec.
1196 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1197 * includes the time spent in suspend.
1199 void get_monotonic_boottime(struct timespec *ts)
1201 struct timekeeper *tk = &timekeeper;
1202 struct timespec tomono, sleep;
1206 WARN_ON(timekeeping_suspended);
1209 seq = read_seqbegin(&tk->lock);
1210 ts->tv_sec = tk->xtime_sec;
1211 nsec = timekeeping_get_ns(tk);
1212 tomono = tk->wall_to_monotonic;
1213 sleep = tk->total_sleep_time;
1215 } while (read_seqretry(&tk->lock, seq));
1217 ts->tv_sec += tomono.tv_sec + sleep.tv_sec;
1219 timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec);
1221 EXPORT_SYMBOL_GPL(get_monotonic_boottime);
1224 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1226 * Returns the monotonic time since boot in a ktime
1228 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1229 * includes the time spent in suspend.
1231 ktime_t ktime_get_boottime(void)
1235 get_monotonic_boottime(&ts);
1236 return timespec_to_ktime(ts);
1238 EXPORT_SYMBOL_GPL(ktime_get_boottime);
1241 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1242 * @ts: pointer to the timespec to be converted
1244 void monotonic_to_bootbased(struct timespec *ts)
1246 struct timekeeper *tk = &timekeeper;
1248 *ts = timespec_add(*ts, tk->total_sleep_time);
1250 EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
1252 unsigned long get_seconds(void)
1254 struct timekeeper *tk = &timekeeper;
1256 return tk->xtime_sec;
1258 EXPORT_SYMBOL(get_seconds);
1260 struct timespec __current_kernel_time(void)
1262 struct timekeeper *tk = &timekeeper;
1264 return tk_xtime(tk);
1267 struct timespec current_kernel_time(void)
1269 struct timekeeper *tk = &timekeeper;
1270 struct timespec now;
1274 seq = read_seqbegin(&tk->lock);
1277 } while (read_seqretry(&tk->lock, seq));
1281 EXPORT_SYMBOL(current_kernel_time);
1283 struct timespec get_monotonic_coarse(void)
1285 struct timekeeper *tk = &timekeeper;
1286 struct timespec now, mono;
1290 seq = read_seqbegin(&tk->lock);
1293 mono = tk->wall_to_monotonic;
1294 } while (read_seqretry(&tk->lock, seq));
1296 set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
1297 now.tv_nsec + mono.tv_nsec);
1302 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1303 * without sampling the sequence number in xtime_lock.
1304 * jiffies is defined in the linker script...
1306 void do_timer(unsigned long ticks)
1308 jiffies_64 += ticks;
1310 calc_global_load(ticks);
1314 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1315 * and sleep offsets.
1316 * @xtim: pointer to timespec to be set with xtime
1317 * @wtom: pointer to timespec to be set with wall_to_monotonic
1318 * @sleep: pointer to timespec to be set with time in suspend
1320 void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
1321 struct timespec *wtom, struct timespec *sleep)
1323 struct timekeeper *tk = &timekeeper;
1327 seq = read_seqbegin(&tk->lock);
1328 *xtim = tk_xtime(tk);
1329 *wtom = tk->wall_to_monotonic;
1330 *sleep = tk->total_sleep_time;
1331 } while (read_seqretry(&tk->lock, seq));
1334 #ifdef CONFIG_HIGH_RES_TIMERS
1336 * ktime_get_update_offsets - hrtimer helper
1337 * @offs_real: pointer to storage for monotonic -> realtime offset
1338 * @offs_boot: pointer to storage for monotonic -> boottime offset
1340 * Returns current monotonic time and updates the offsets
1341 * Called from hrtimer_interupt() or retrigger_next_event()
1343 ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot)
1345 struct timekeeper *tk = &timekeeper;
1351 seq = read_seqbegin(&tk->lock);
1353 secs = tk->xtime_sec;
1354 nsecs = timekeeping_get_ns(tk);
1356 *offs_real = tk->offs_real;
1357 *offs_boot = tk->offs_boot;
1358 } while (read_seqretry(&tk->lock, seq));
1360 now = ktime_add_ns(ktime_set(secs, 0), nsecs);
1361 now = ktime_sub(now, *offs_real);
1367 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1369 ktime_t ktime_get_monotonic_offset(void)
1371 struct timekeeper *tk = &timekeeper;
1373 struct timespec wtom;
1376 seq = read_seqbegin(&tk->lock);
1377 wtom = tk->wall_to_monotonic;
1378 } while (read_seqretry(&tk->lock, seq));
1380 return timespec_to_ktime(wtom);
1382 EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset);
1385 * xtime_update() - advances the timekeeping infrastructure
1386 * @ticks: number of ticks, that have elapsed since the last call.
1388 * Must be called with interrupts disabled.
1390 void xtime_update(unsigned long ticks)
1392 write_seqlock(&xtime_lock);
1394 write_sequnlock(&xtime_lock);